In what ways does the integration of IoT technology in industrial applications improve operational efficiency, and what are some potential security challenges involved?

The integration of IoT in industrial applications enhances operational efficiency by enabling real-time monitoring and control, thus allowing for timely decision-making and reduced downtime. For instance, IoT systems can continuously monitor substation parameters, providing automatic alerts and data to operators for efficient management . However, IoT systems face significant security challenges such as data confidentiality, integrity, privacy concerns, and potential attacks. To ensure secure deployment, various protocols like MQTT are employed, but issues like insecure network services and insufficient authentication remain major concerns .

What is the strategic importance of IoT in enhancing security and surveillance systems, and what architectural considerations must be addressed?

IoT enhances security and surveillance systems by facilitating comprehensive real-time monitoring, data collection, and automated response capabilities through interconnected devices. The strategic importance lies in its ability to provide continual oversight and immediate access control, improving the overall security posture of monitored environments. Architectural considerations include ensuring robust network layers that incorporate secure data transmission protocols like MQTT, resilient to cyber threats, and designed with privacy and access control to prevent unauthorized use. Proper data encryption and authentication mechanisms are essential to safeguard sensitive surveillance data .

What are the key considerations in designing wearable antennas for medical applications, and how do these considerations impact their performance?

When designing wearable antennas for medical applications, key considerations include material flexibility, surface conformability, radiation efficiency, and specific absorption rate (SAR) levels to ensure safety. Performance is impacted as these antennas must maintain high directivity and gain even when bent or crumpled. Textile materials like cotton or polyester can be used for their lightweight and flexible nature, enhancing comfort and ease of integration into garments. The dielectric properties and antenna design, such as E-shaped or patch structures, also influence frequency resonance and bandwidth, which are critical for maintaining reliable communication in medical settings .

Discuss the significance of the S-band and C-band frequencies in wireless communication and how textile antennas might be utilized effectively at these frequencies.

S-band frequencies (2-4 GHz) are crucial in applications such as airport surveillance and satellite communication where they serve as downlink channels. C-band frequencies (4-8 GHz) are primarily used in radar operations and Wi-Fi applications. Textile antennas, being flexible and lightweight, can be effectively integrated into clothing, making them suitable for Wireless Body Area Networks (WBAN) operating at these frequencies. For instance, an E-shaped microstrip textile antenna designed for S and C bands can provide high gain and efficient communication while being wearable and compact .

How does data augmentation help in improving CNN model robustness, and what are its specific benefits in medical image analysis?

Data augmentation improves CNN model robustness by expanding the diversity of training samples through techniques like rotation, zoom, and flipping, which help the model generalize well across different data variations and reduce overfitting. In medical image analysis, this is particularly beneficial because it allows CNNs to learn from a wider array of features, even with limited available data, increasing the model's ability to correctly identify pathologies across varied medical imaging scenarios, crucial for achieving high accuracy in diagnostics .

In what ways can CNN architectures be adjusted to improve the performance in complex image classification tasks such as medical diagnostics?

CNN architectures can be adjusted to improve performance in complex image classification tasks by optimizing the model's layers and refining hyperparameters for specific diagnostic features. Techniques such as increasing the depth of CNN layers and implementing sophisticated pooling strategies can enhance the network's ability to recognize intricate patterns. Utilizing modifications like depth-wise separable convolutions, akin to Xception or DenseNet architectures, can reduce computational load while maintaining high-level features extraction. Data augmentation strategies help diversify training inputs, potentially boosting robustness and generalization in diagnostics .

How does the application of deep learning, specifically CNNs, enhance the accuracy and performance in medical image analysis compared to traditional methods?

Convolutional Neural Networks (CNNs) improve the accuracy and performance in medical image analysis by automatically understanding complex features from data directly, in contrast to traditional methods which require manual feature extraction before classification. CNNs can learn both local and global features, adapting flexibly to various tasks like tumor detection, leading to superior accuracy as evidenced by a 97.87% accuracy rate in tumor image classification, surpassing traditional methods .

How can machine learning techniques like self-organizing maps (SOM) and support vector machines (SVM) enhance document classification, and what are their comparative advantages?

Self-organizing maps (SOM) and support vector machines (SVM) provide enhanced document classification by effectively organizing and categorizing texts into different classes based on content. SOM, as an unsupervised method, excels in yielding high accuracy by organically learning data distribution without pre-set labels. When combined with SVM, a supervised learning model, the classification accuracy improves further due to SVM's capability to define optimal hyperplanes for classification, making it superior in dealing with high-dimensional data and providing structured classification outcomes .

Evaluate how triplet loss functions in FaceNet contribute to improving face verification processes in comparison to inter-class metric learning.

Triplet loss functions in FaceNet contribute significantly to enhancing face verification by ensuring that the distance between a positive pair is minimized, while a negative pair is maximized within an embedded space. This technique contrasts with inter-class metric learning that focuses primarily on distinguishing between known classes rather than refining the intra-class and inter-class metric distances explicitly. As a result, triplet loss provides a finer granularity in differentiating subtle variations in features, enhancing the overall precision of face verification processes, which is critical in applications where high accuracy is paramount .

What are the advantages of using FaceNet over traditional face recognition techniques like Viola-Jones and Haar Cascade classifiers?

FaceNet offers significantly better results in face recognition compared to traditional techniques like Viola-Jones and Haar Cascade classifiers. This is because while methods like Viola-Jones and Haar Cascade classifiers focus primarily on object detection within OpenCV, FaceNet provides a more advanced approach, which can accurately identify faces in various conditions, such as different lighting scenarios and partially obscured or covered faces .

Open navigation menu

Upload

0% found this document useful (0 votes)

1K views1,001 pages

Techno-Societal 2020

This document provides the contents of the proceedings from the 3rd International Conference on Advanced Technologies for Societal Applications. The proceedings are split into two volumes and contain papers on various topics related to sensor image and data driven societal technologies, including office monitoring systems, document categorization using machine learning, antenna design, speaker identification, engine monitoring, disease detection, and more. The proceedings were edited by Prashant M. Pawar and five other scholars.

Uploaded by

Desalegn Dga

We take content rights seriously. If you suspect this is your content, claim it here.

Available Formats

Download as PDF, TXT or read online on Scribd

0% found this document useful (0 votes)

1K views1,001 pages

Techno-Societal 2020

Uploaded by

Desalegn Dga

We take content rights seriously. If you suspect this is your content, claim it here.

Available Formats

Download as PDF, TXT or read online on Scribd

Prashant M.

Pawar ·
R. Balasubramaniam ·
Babruvahan P. Ronge ·
Santosh B. Salunkhe · Anup S. Vibhute ·
Bhuwaneshwari Melinamath Editors

Techno-Societal
2020
Proceedings of the 3rd International
Conference on Advanced Technologies
for Societal Applications—Volume 1
Techno-Societal 2020
Prashant M. Pawar · R. Balasubramaniam ·
Babruvahan P. Ronge · Santosh B. Salunkhe ·
Anup S. Vibhute · Bhuwaneshwari Melinamath
Editors

Techno-Societal 2020
Proceedings of the 3rd International
Conference on Advanced Technologies for
Societal Applications—Volume 1
Editors
Prashant M. Pawar R. Balasubramaniam
SVERI’s College of Engineering Precision Engineering Center
Pandharpur, Maharashtra, India Bhabha Atomic Research Centre
Mumbai, Maharashtra, India
Babruvahan P. Ronge
SVERI’s College of Engineering Santosh B. Salunkhe
Pandharpur, Maharashtra, India SVERI’s College of Engineering
Pandharpur, Maharashtra, India
Anup S. Vibhute
SVERI’s College of Engineering Bhuwaneshwari Melinamath
Pandharpur, Maharashtra, India SVERI’s College of Engineering
Pandharpur, Maharashtra, India

ISBN 978-3-030-69920-8 ISBN 978-3-030-69921-5 (eBook)

[Link]

© The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature
Switzerland AG 2021
This work is subject to copyright. All rights are solely and exclusively licensed by the Publisher, whether
the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse
of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and
transmission or information storage and retrieval, electronic adaptation, computer software, or by similar
or dissimilar methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this book
are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or
the editors give a warranty, expressed or implied, with respect to the material contained herein or for any
errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional
claims in published maps and institutional affiliations.

This Springer imprint is published by the registered company Springer Nature Switzerland AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
Contents

Sensor Image and Data Driven Societal Technologies

Office Monitoring and Surveillance System . . . . . . . . . . . . . . . . . . . . . . . . . 3
Vishal Patil and Yogesh Jadhav
Categorizing Documents by Support Vector Machine Trained
Using Self-Organizing Maps Clustering Approach . . . . . . . . . . . . . . . . . . . 13
Vishal Patil, Yogesh Jadhav, and Ajay Sirsat
Bandwidth Improvement of Multilayer Microstrip Patch
Antenna by Using Capacitive Feed Technique for Broadband
Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
Anil K. Rathod, Md. M. Bhakar, M. S. Mathpati, S. R. Chougule,
and R. G. Sonkamble
Use of Median Timbre Features for Speaker Identification
of Whispering Sound . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
Vijay M. Sardar, Manisha L. Jadhav, and Saurabh H. Deshmukh
Intelligent System for Engine Temperature Monitoring
and Airbag Deployment in Cars Using . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
Akshay A. Jadhav, Swagat M. Karve, Sujit A. Inamdar,
and Nandkumar A. Admile
Analysis and Prediction of Temporomandibular Joint Disorder
Using Machine Learning Classification Algorithms . . . . . . . . . . . . . . . . . . 51
Roopa B. Kakkeri and D. S. Bormane
Machine Learning Approach in Cooperative Spectrum Sensing
for Cognitive Radio Network: Survey . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
Vaishali S. Kulkarni, Tanuja S. Dhope(Shendkar), Swagat Karve,
Pranav Chippalkatti, and Akshay Jadhav

v
vi Contents

Study on Detection of Leukemia in Human Blood Pattern Based

on Microscopic Photographs Using Image Processing Techniques . . . . . 73
Swagat M. Karve, Pravin Kishrsagar, Akshay A. Jadhav,
and M. Aravind Kumar
Brain Tumor Detection Using Deep Neural Network . . . . . . . . . . . . . . . . . 85
Rajshree B. More and Swati. A. Bhisikar
Design and Simulation of Different Structures of Micro Strip
Patch Antenna for Wireless Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
Anil. J. Kokare, Mahesh. S. Mathpati, and Bhagyashri. S. Patil
State Context and Hierarchical Trust Management in WSN
for Intrusion Detection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
Ranjeet B. Kagade and J. Santhosh
Portable Camera Based Assistive Text and Product Label
Reading from Hand Held Object by Using Android App
for Blind Person . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
Somnath Thigale and Ranjeet B. Kagade
Automatic System for Identifying Cholesterol . . . . . . . . . . . . . . . . . . . . . . . 125
Mohua Biswas, Pragtee Tathe, Geeta Unhale, and Papiya Biswas Datta
Design of an IoT Based System for Monitoring and Controlling
the Sub-Station Equipment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
Pranali Bodke and A. A. Kalage
Implementation of Iridology for Pre Diagnosis of Brain Tumor . . . . . . . 141
Pragtee Tathe, Mohua Biswas, Anup Vibhute, Geeta Unhale,
Mrunmayi Raut, and Papiya Biswas Datta
Wireless Communication Using Light Fidelity Network . . . . . . . . . . . . . . 153
Nimisha Deval, Prajakta Satarkar, Akshata Jadhav, and Rupali M. Shinde
Smart Trolley with Automatic Master Follower and Billing
System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
R. Arpita, K. S. Shashidhara, and Veerendra Dakulagi
Analysis and Design of E Shaped Dual Band Microstrip Textile
Antenna for Wireless Communication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
Husain Bhaldar, Sanjay Kumar Gowre, Mahesh S. Mathpati,
Ashish A. Jadhav, and Mainaz S. Ustad
Enhancement Technique for Early Stage Lung Cancer Detection
Using Foldscope Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
Vanita D. Jadhav and Lalit V. Patil
Contents vii

Foldscope to Detect the Growth of Microorganisms on Various

Materials and Vessels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
Vanita D. Jadhav, Richa Tamhane, Kiran Kedar, Shruti Kawade,
and Aboli Gaikwad
Implementation of A* Algorithm for Real-Time Evacuation
During Fire Situation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
Shilpa K. Rudrawar, Pallavi Ghorpade, and Dipti Y. Sakhare
Involuntary Traffic Control System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209
Shriniwas V. Darshane, Ranjeet B. Kagade, and Somnath [Link]
Early Detection of Diabetic Retinopathy Using Machine Learning . . . . . 215
Vishal V. Bandgar, Shardul Bewoor, Gopika A. Fattepurkar,
and Prasad B. Chaudhary
The Effective Use of Deep Learning Network with Software
Framework for Medical Healthcare . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225
Padmanjali A Hagargi
Detection of Brain Tumor Using Image Processing and Neural
Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241
Vanshika Dhillo, Dipti Sakhare, and Shilpa Rudrawar
Automatic Guided System for Image Resolution Enhancement . . . . . . . . 249
Neeta P. Kulkarni, J. S. Kulkarni, and S. M. Karve
Residual Network for Face Progression and Regression . . . . . . . . . . . . . . 257
Dipali Vasant Atkale, Meenakshi Mukund Pawar,
Shabdali Charudtta Deshpande, and Dhanashree Madhukar Yadav
Design and Simulation of 2-Element, Circular Shaped MIMO
Antenna for C-Band Frequencies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 269
Ashish Jadhav, Nagashettappa Biradar, Husain Bhaldar,
Mahesh Mathpati, Manoj Deshmukh, and Renuka Wadekar
Attribute Inspection of Product Using Image Processing . . . . . . . . . . . . . 281
Anup S. Vibhute, Reshma R. Deshmukh, P. S. Valte, B. D. Gaikwad,
and Shrikant Pawar
A Proposed Method for Audio Steganography Using Digital
Information Security . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 295
Pratik Kurzekar and Shrinivas Darshane
Spine Diseases Detection Using SVM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 309
Jyoti M. Waykule and V. R. Udupi
BCI Integrated Wheelchair Controlled via Eye Blinks and Brain
Waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 321
Sredha Prem, Jeswin Wilson, Shelby Mathew Varghese, and M. Pradeep
viii Contents

Classification of Soil Nutrients Using IoT . . . . . . . . . . . . . . . . . . . . . . . . . . . 333

Sandesh Koli, Dhaval Khobare, Amol Salunke, and Ranjeet B. Kagade
Non-invasive Methodological Techniques to Determine Health
of a Bone . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 343
Meghana R. Khare and Raviraj H. Havaldar
Face Detection and Recognition Using Raspberry Pi . . . . . . . . . . . . . . . . . 351
P. R. Dolas, Pratiksha Ghogare, Apurva Kshirsagar, Vidya Khadke,
and Sanjana Bokefode
Human Tracking Mechanism for Institutions Using RFID
and Facial Recognition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 355
Rameez Shaik and L. V. Patil
Monitoring Power Consumption and Automation Using IOT . . . . . . . . . 365
Riya Jain, Revati Awale, Neha Kothari, Swarali Shah, and Amit Kore
Light Fidelity (Li-Fi): Future 5G Wireless Connectivity
to Empower Rural India . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 377
Prajakta A. Satarkar and Girish V. Chowdhary
Optimized Dynamic Feature Matching for Face Recognition . . . . . . . . . . 387
Ganesh Gopalrao Patil and Rohitash Kumar Banyal
Novel Secure Routing Protocol for Detecting and Presenting
Sybil Attack . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 393
S. M. Sawant, S. M. Shinde, and J. S. Shinde
IoT Model for Heart Disease Detection Using Machine Learning
(ML) Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 399
Madhuri Kerappa Gawali and C. Rambabu
Security Threats and Their Mitigations in IoT Devices . . . . . . . . . . . . . . . 411
Saurabh Gupta and N. Lingareddy
Age and Gender Classification Based on Deep Learning . . . . . . . . . . . . . . 425
Tejas Agarwal, Mira Andhale, Anand Khule, and Rushikesh Borse
Handling of Auxiliaries in Kannada Morphology . . . . . . . . . . . . . . . . . . . . 439
Bhuvaneshwari C. Melinamath
Dysarthria Detection Using Convolutional Neural Network . . . . . . . . . . . 449
Pratibha Dumane, Bilal Hungund, and Satishkumar Chavan
AR for Maintenance Training During COVID-19 Pandemic . . . . . . . . . . 459
Jyoti Pawar and Trupti Bansode
Face Recognition Based Attendance System Using Cv2 . . . . . . . . . . . . . . . 469
Vedant Khairnar and C. M. Khairnar
Contents ix

IoT Enabled Secured Card Less Ration Distribution System . . . . . . . . . . 477

Shilpa K. Rudrawar, Kuldeepak Phad, and Prajwal Durugkar
Voice Assisted Bots for Automobile Applications . . . . . . . . . . . . . . . . . . . . . 489
Shilpa K. Rudrawar, Nikhil Choudhar, and Ankit Meshram
Content-Based Image Retrieval Using Color Histogram and Bit
Pattern Features . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 499
Nandkumar S. Admile, Akshay A. Jadhav, Swagat M. Karve,
and Anil A. Kasture
KNN and Linear SVM Based Object Classification Using Global
Feature of Image . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 509
Madhura M. Bhosale, Tanuja S. Dhope, and Akshay P. Velapure
MURA: Bone Fracture Segmentation Using a U-net Deep
Learning in X-ray Images . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 519
Komal Ghoti, Ujjwal Baid, and Sanjay Talbar
Effective Usage of Oversampling with SMOTE for Performance
Improvement in Classification Over Twitter Data . . . . . . . . . . . . . . . . . . . . 533
Deepak Patil, Poonam Katyare, Parag Bhalchandra, and Aniket Muley
Multi-Classification of Breast Histopathological Image
Using Xception: Deep Learning with Depthwise Separable
Convolutions Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 539
Suvarna D. Pujari, Meenakshi M. Pawar, and Madhuri Wadekar
Dense Haze Removal Using Convolution Neural Network . . . . . . . . . . . . 547
Mayuri Dongare and Jyoti Kendule

ICT Based Societal Technologies

Diabetic Retinopathy Detection with Optimal Feature Selection:
An Algorithmic Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 559
S. Shafiulla Basha and Syed Jahangir Badashah
Students Perception and Satisfaction Towards ICT Enabled
Virtual Learning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 569
Moshina Rahamat and B. Lavanya
An Appointment Scheduler: A Solution to Control Covid-19
Spread . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 579
Apeksha M. Gopale
Bandobast Allocation and Attendance System . . . . . . . . . . . . . . . . . . . . . . . 587
Prashant S. Bhandare, Somnath A. Zambare, Amey Bhatlavande,
and Shamsundar Bhimade
Fire Fighting Un-manned Air Vehicle for Remote Areas . . . . . . . . . . . . . . 593
N. Shashank Bhat, K. S. Shashidhara, and Veerendra Dakulagi
x Contents

Human Age Classification and Estimation Based on Positional

Ternary Pattern Features Using Ann . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 603
Shamli V. Jagzap, Lalita A. Palange, Seema A. Atole, and Geeta G. Unhale
Object Recognition Using Fuzzy Classifier . . . . . . . . . . . . . . . . . . . . . . . . . . 613
Seema A. Atole, Shamli V. Jagzap, Lalita A. Palange, and Akshay A. Jadhav
An Effective Approach for Accuracy of Requirement
Traceability in DevOps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 623
Vinayak M. Sale, Somnath Thigale, B. C. Melinamath, and Siraj Shaikh
Clustering of Fruits Image Based on Color and Shape Using
K-Means Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 639
Vidya Maskar, Kanchan Chouhan, Prashant Bhandare, and Minal Pawar
Modern Education Using Augmented Reality . . . . . . . . . . . . . . . . . . . . . . . 651
Vishal V. Bandgar, Ajinkya A. Bahirat, Gopika A. Fattepurkar,
and Swapnil N. Patil
OSS Features Scope and Challenges . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 661
M. K. Jadhav and V. V. Khandagale
Text Summarization and Dimensionality Reduction Using
Ranking and Learning Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 667
Dipti Bartakke, Santosh Kumar, Aparna Junnarkar, and Somnath Thigale
Properties of Extended Binary Hamming [8, 4, 4] Code Using
MATLAB . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 683
N. S. Darkunde, S. P. Basude, and M. S. Wavare
Identification of Fake News on Social Media: A New Challenge . . . . . . . 689
Dhanashree V. Patil, Supriya A. Shegdar, and Sanjivini S. Kadam
A Smart and Secure Helmet for Safe Riding . . . . . . . . . . . . . . . . . . . . . . . . 703
Ramesh Kagalkar and Basavaraj Hunshal
On Some Properties of Extended Binary Golay [24, 12, 8] Code
Using MATLAB . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 713
N. S. Darkunde, S. P. Basude, and M. S. Wavare
Encoding Using the Binary Schubert Code [43, 7] Using
MATLAB . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 725
M. S. Wavare, N. S. Darkunde, and S. P. Basude
Data Mining Techniques for Privacy Preservation in Social
Network Sites Using SVM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 733
Vishvas Kalunge and S. Deepika
Near Field Communication (NFC) Technology and Its
Application . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 745
R. D. Kulkarni
Contents xi

Performance Analysis of Convolution Neural Network

with Different Architecture for Lung Segmentation . . . . . . . . . . . . . . . . . . 753
Swati P. Pawar and S. N. Talbar
A Secure Data Sharing Platform Using Blockchain
and Fine-Grained Access . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 761
Shamsundar Bhimade, Prashant Bhandare, Amey Bhatlavande,
and Bharati Deokar
Efficient and Interactive Fuzzy Type Ahead Search in XML Data . . . . . 769
Laxman Dethe, Geeta Khare, Avdhut Bhise, and Somnath Zambare
IOT Based Interactive Motion Detection Security System Using
Raspberry Pi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 779
Geeta Khare, Subhash Pingale, Avdhut Bhise, and Sharad Kawale

Commercially Successful Rural and Agricultural Technologies

Cognitive Intelligence of Internet of Things in Precision
Agriculture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 789
Rahul Keru Patil and Suhas Shivlal Patil
Flooring: A Risk Factor for Fall-Related Injuries in Elderly
People Housing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 797
Unesha Fareq Rupanagudi
Analysis of Construction Readiness Parameters for Highway
Projects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 805
Harshvardhan R. Godbole and R. C. Charpe
Biogas Generation by Utilizing Agricultural Waste . . . . . . . . . . . . . . . . . . 819
Prathamesh Chaudhari and Shivangi Thakker
Effect of Change in the Resilient Modulus of Bituminous Mix
on the Design of Flexible Perpetual Pavement . . . . . . . . . . . . . . . . . . . . . . . 829
Saurabh Kulkarni and Mahadeo Ranadive
Analysis of Three-Stage Open Pan Heat Exchanger Working
on Dual Fuel for Jaggery Making . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 839
Abhijeet N. Kore and Sanjay S. Lakade
The Performance and Emission Analysis of Diesel Engine
with Sunflower Biodiesel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 847
Aniruddha Shivram Joshi and S. Ramesh

Deployable Environment or Healthcare Technologies

Experimental Analysis of Effect of Bio-lubricant Between
Tribological Systems of Piston Ring Under the Jatropha Oil . . . . . . . . . . 863
Mhetre Rahul Sanjay and L. B. Abhang
xii Contents

Garbage Monitoring and Collection System Using RFID

Technology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 873
Amol A. Kadam, AksahyAjadhav, Dhanraj P. Narsale,
Anil M. Kasture, S. M. Karve, and Manoj A. Deshmukh
A Survey on Mental Health Monitoring System Via Social
Media Data Using Deep Learning Framework . . . . . . . . . . . . . . . . . . . . . . . 879
Satyaki Banerjee and Nuzhat F. Shaikh
An Approach to Heart Disease Prediction with Machine
Learning Using Chatbot . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 889
Chinmay Nanaware, Arnav Deshmukh, Nikhil Chougala, and Jaydeep Patil
Automated Early Detection of Diabetic Retinopathy . . . . . . . . . . . . . . . . . 897
Supriya Shegdar, Ameya Bhatlavande, Dhanashree Patil,
and Sanjivani Kadam
Product Lifecycle of Automobiles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 905
V. K. Bupesh Raja, Ajay Shivsharan Reddy,
Suraj Ramesh Dhavanapalli, D. R. Sai Krishna Sanjay,
BH. Jashwanth Varma, and Puskaraj D Sonawwanay
Analysis of Heavy Metal Pollutants in the Sediments
from Coastal Sites of Al-Hodiedah Governorates, Yemen . . . . . . . . . . . . . 911
Majeed Hazzaa Nomaan, Dipak B. Panaskar, and Ranjitsinh S. Pawar
Development of Low Cost PCR Product Detection System
for Screening and Diagnosis of Infectious Diseases . . . . . . . . . . . . . . . . . . . 923
Patil Yogesh Navalsing, Suri Vinod Kumar, Suri Aseem Vinod,
Kar Harapriya, and Thakur Mansee Kapil
Determination of Viscous, Coulomb and Particle Damping
Response in SDOF by Forced Oscillation . . . . . . . . . . . . . . . . . . . . . . . . . . . 935
S. T. Bodare, S. D. Katekar, and Chetan Chaudhari
Assessment of Godavari River Water Quality of Nanded City,
Maharashtra, India . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 947
P. R. Shaikh, Girish Deore, A. D. Pathare, D. V. Pathare, and R. S. Pawar
Groundwater Quality Assessment in an Around Thermal Power
Plants in Central India . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 963
V. U. Deshmukh, D. B. Panaskar, P. R. Pujari, and R. S. Pawar

Micro, Nano Manufacturing, Fabrication and Related

Applications
XML Based Feature Extraction and Process Sequence Selection
in a CAPP System for Micro Parts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 973
G. Gogulraj and S. P. Leokumar
Contents xiii

To Study and Optimize the Effects of Process Parameters

on Taper Angle of Stainless Steel by Using Abrasive Water Jet
Machining . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 983
Meghna K. Gawade and Vijaykumar S. Jatti
Effect of Process Parameters on Response Measures of Cartridge
Brass Material in Photo Chemical Machining . . . . . . . . . . . . . . . . . . . . . . . 995
Bandu. A. Kamble, Abhay Utpat, Nitin Misal, and B. P. Ronge
Thermal Performance of Two Phase Closed Thermosyphon
with Acetone as Working Fluid . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1005
Shrikant V. Pawar and Abhimanyu K. Chandgude
Effect of Slip on Vortex Formation Near Two-Part Cylinder
with Same Sign Zeta Potential in a Plane Microchannel . . . . . . . . . . . . . . 1013
Souvik Pabi, Sumit Kumar Mehta, and Sukumar Pati
Optimization of WEDM Parameters During Machining of Ni-75
Using AHP-MOORA Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1023
S. A. Sonawane and S. S. Wangikar
Structural Analysis of Novel Mini Horizontal Axis Wind
Turbine (NMHAWT) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1031
Pramod Magade, S. P. Chavan, Sushilkumar Magade, and Vikram Gaikwad
Fabrication of Tree Type Micro-Mixer with Circular Baffles
Using CO2 Laser Machining . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1039
Sachin R. Gavali, Sandeep S. Wangikar, Avinash K. Parkhe,
and Prashant M. Pawar
Sensor Image and Data Driven Societal
Technologies
Office Monitoring and Surveillance
System

Vishal Patil and Yogesh Jadhav

Abstract Facial recognition is a biometric software category that mathematically

maps the facial features of a person and stores the data as a face-print. Using machine
learning algorithms, the software compares a live capture or digital image to the stored
face print to verify an individual’s identity and help automate authentication. Facial
recognition will increase protection, recognize unauthorized entry and keep a track
of visitors. ID passes are yesterday’s technology Our project’s main task is to identify
if the person is an employee or a visitor by using a face recognition system where in
security guards job is to watch over the process and stepping in only when the system
says that the person is not an employee or when they see something suspicious.

Keywords Face recognition · FaceNet · OpenCV · Office monitoring

1 Introduction

A facial recognition system is a technology capable of recognizing and validating

an individual from a digital picture or a video source. Facial recognition usually
functions in conjunction with captured facial features from a given picture using
faces in a database. It is also recognized as a biometric technology based on arti-
ficial intelligence, which recognizes the person by detecting patterns based on the
facial textures and shape of a person’s face. Even though a face recognition system
was initially implemented with the help of the computer, a wider implementation is
seen on mobile devices as well and also in different technologies such as robotics
wherein it is used as access control in the security system and can be different from
other biometrics such as fingerprinting or eye iris recognition systems. It has also

V. Patil (B)
Computer Engineering Department, K. J. Somaiya Institute of Engineering and Information
Technology, Mumbai, India
Y. Jadhav
Computer Science and Engineering Department, Amity School of Engineering and Technology,
Amity University, Mumbai, India

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 3

P. M. Pawar et al. (eds.), Techno-Societal 2020,
[Link]
4 V. Patil and Y. Jadhav

become popular recently as a marketing and commercial device. Many implementa-

tions include automated interaction between humans and computers, video analysis,
automatic indexing of images and video archives. In today’s application, facial recog-
nition can be approached in different ways. The primary component analysis (PCA),
Linear Discriminant Analysis (LDA), and deep learning are examples of such vari-
ables. PCA is a form of dimensional reduction sometimes used to reduce dimensions
on a fairly wide dataset. You do this by turning variables into smaller sections in large
datasets, while still using large datasets details. The PCA approach is very commonly
used as the principle is very straight forward and has a computer-efficient algorithm
as illustrated in following Fig. 1.
FaceNet is one of the implementations of facial recognition using deep learning.
This is a one-shot approach for learning with Euclidean space to measure each
face’s resemblance size. FaceNet is a fairly recent approach used with the Deep
Convolutional Network system in Google study in 2015. There were two separate
systems in the analysis, the network system for Zeiler and Fergus and the newest
network system for Inception. CNN in train practices Stochastic Gradient Descent
(SGD) implementing back propagation and AdaGrad standards. FaceNet provides
up to 99.63% accuracy from datasets that have labeled Faces in the Wild (LFW)and
95.12% accuracy on YouTube Faces Database. The benefit of using the FaceNet
approach is that this model requires only limited alignment to cut the face region
relatively tightly. FaceNet’s drawback is that the training cost is very costly because
it requires a CPU. Office Monitoring and Surveillance System is a face recognition-
based system for providing an efficient security solution for commercial buildings
and offices. The project does this by performing facial recognition of people entering
the premises and checking if the person is an employee of the company or a Visitor.
If the system detects the personas an employee, he will be granted access to the
company. If the system does not detect the person as an employee, they will be
provided with a welcome screen. The welcome screen will contain options whether
the person has reserved a meeting, and with which employee. The guest user will
choose the employee they want to meet and their video will be live-streamed to
the employee. If the employee recognizes the guest, they will be granted access;
otherwise, the guest will be directed to a nearby consultant. Along with this system
also does object detection and fire recognition. The system uses Google FaceNet,
OpenCV, SVC model to perform face recognition. The system contains a web-based
application that will be used by all the employees to get notified about the presence of
a visitor willing to visit them in the office premises. The same web-based application
will be used for storing the visitor details and monitoring them. The application is
developed using the Laravel framework. The proposed system will help to reduce the
physical procedures, time & efforts required for the same and will save the paperwork
as well; to provide security for offices; to ensure the safety of employees and also
maximize the efficiency of guards.
Office Monitoring and Surveillance System 5

Fig. 1 System flow

6 V. Patil and Y. Jadhav

2 Literature Survey

Wei Wu, Chuanchang Liu, Zhiyuan Su have presented the result of the analysis
and the methodology of applying FaceNet. Recent studies show that deep learning
approaches to this challenge can produce impressive results. To overcome these chal-
lenges, this program uses a profound learning approach. This comes with a module
to detect the faces and identification of the faces. The MTCNN facial detection
module is very simple, accurate and robust enough for variations in lighting and
contexts. This module is based on FaceNet, which teaches you to map the facial
images directly at the position of the Euclidean space, where the two-pointed space
inside the Euclidean space corresponds directly to the similarity of two facial images.
When this Euclidean space is generated, facial images are transferred as facial vectors
to the FaceNet embedding system. Feature vectors are then listed in a SVM to help us
easily recognize the face images. The experimental findings indicate that the device
is highly accurate, low computational complexity, and has a strong implementation
benefit for real time identification [1].
Many businesses have widely adopted video monitoring devices to increase
protection in their workforce. There are few visible cameras because of their suscep-
tibility to eye strain A kind of control method is used in this paper to solve the
problem. The network is made up of wireless and video monitoring devices. When
security threats are detected, an alert may be activated by the wireless control device
and the video stream from the camera to the computer is enabled. Therefore, the
system avoids dangers resulting from machine breakdown or user error and thereby
increases industrial efficiency [2, 3].
Facial recognition is an essential process to identify and separate the facial region
from the majority of the photos of every face detection device. There are numerous
techniques proposed, ranging from the simple detection of edges to the advancement
of techniques like pattern recognition. The review contrasts two facial recognition
approaches the characteristics of them and the local binary patterns focused on impact
rate recognition and speed detection. The algorithms have been tested on Microsoft
Visual C++ 2010 Express with OpenCV software. Test findings show the reliability
and trust of the functionality of the local binary template for implementing a facial
recognition device in real time [4].
Paper 4 introduces the presentation on Jetson TX2 with FaceNet and MTCNN
algorithms of an advanced and multi-camera face recognition surveillance device.
Presents the analysis of the modern, multi-camera facial recognition tracking system
FaceNet and MTCNN algorithms on Jetson TX2. The suggested portable system
tracks the target or suspect along with the ID/location of the camera and documents
its presence through various camera installations in the database. Triplet loss is one of
the most significant things we find in this article. The gap between the anchor and the
result, which implies a separate identity, is reduced using Triplet loss procedure, if the
result is accurate and indicates the same identity; the gap between the anchor and the
sample is then maximized. Triplet loss is one of the easiest ways to practice effective
128-dimensional embedding for any image. The representation of the anchor here is
Office Monitoring and Surveillance System 7

the frame from which the triplet loss was measured. Three pictures are required to
quantify the triplet loss: an anchor, positive and negative [4, 5].
Study of paper 5 describes a way of monitoring individuals and obtaining canon-
ical frontal pictures that fit the sensor network model in 2D world coordinates. In front
photos, monitoring and identity management have a specific desirability because they
are essentially invariant with daily appearance changes. This method has been applied
and tested on the FaceNet wired camera prototype network. Their key contribution
is to demonstrate how to sense the trajectories of moving objects to acquire canon-
ical vision of high quality while maintaining node capacity. The authors present an
overview of the method and display the task algorithm on data obtained from the
FaceNet camera network [6–11].
The purpose of project implemented in paper 6 is to create a complete hardware
and software framework for the surveillance of entry into a secure facility with
modern facial recognition technology (FRT) and lightweight, cheap components.
For visualization in the operation, FRT senses and recognizes individuals as they
enter and exit the area protected. The project uses FaceNet FRT to boost facial
recognition accuracy by means of a deep-learning approach. The FaceNet FRT is
implemented on a Raspberry Pi using the Neural Compute device Intel Movidius
with input from two portable cameras [7].
Substantial advancement in the field of identity testing and facial detection has
been made lately, taking advantage of advances in the convolution neural networks
(CNN). The findings on human level have been exceeded by FaceNet’s performance
particularly in terms of accuracy in “Labeled Faces in the wild (LFW)” datasets
and “Youtube Faces in the wild (YTF).” The triplet loss in the FaceNet has been
proven successful for facial verification. However, if a large dataset is used to train
the model, the amount of triples available is explosive. In this article, we propose a
simple class-by-class triple loss based on intra-class distance metric learning which
can substantially reduce the number of potential triplets to learn, but simplifying the
classic three-point loss function hasn’t deteriorated the approach. Experimental tests
of the most commonly used LFW and YTF benchmarks show that the model with
a class-wise simple triplet loss can achieve the latest results. And the simulation of
the dispersal of the learned features based on the MNIST data set has shown the
effectiveness of the proposed method of better class separation and discriminating
against the other state-of-the-art loss functionality [12–14].

3 Proposed System

Our system provides automated surveillance and monitoring for offices. This system
will work on employee’s faces for distinguishing a person from a visitor and an
employee of the organization using Face Recognition. The project subject matter
is a solution to orthodox methods of logging in employees in physical handwritten
logbooks, keeping security guards on duty 24 × 7, relying solely on sensors to fire
breakouts, etc. So, to make up for these fairly unreliable methods, we have come
8 V. Patil and Y. Jadhav

up with a plan that this project will increase the quality and efficiency of security in
office premises. The main objective of our project will be to recognize an employee
to let them in the premises and greet visitors respectfully keeping security as a
higher priority. The model is a deep convolutional neural network, which is learned
by means of a three-fold loss function, which induces vectors for one identity to
become more identical (smaller distance). A significant breakthrough in this research
was the emphasis on teaching a computer to establish embedding directly (instead
of removing them from an intermediate layer of a computer).
The system will be fed live image frames through a CCTV [15], which will be
used to detect faces coming up in the frame. For the detection of faces, we are using
OpenCV. Multiple faces can be detected at once. Once the faces are detected, the
faces will be cropped to size and by analyzing the eye orientation, the faces will be
aligned to a proper axis. Once Face Alignment is achieved, these images will be fed
to Google FaceNet. FaceNet will then extract the features from the image received.
Then, these face embedding systems are used as the basis for training classifier
systems on standard face recognition benchmark datasets, achieving state-of-the-
art performance. These preprocessed images will be used by the Support Vector
Classifier to train the model. This training will be done for each new employee that
joins the organization. Once the training phase is completed, the preprocessed data
received realtime, will be fed to the Support Vector Classifier Inference model which
will recognize the face as an employee or not. Changes (recent scars, beard/non-beard,
different lighting conditions) to faces are anticipated to expand its classification
performance as illustrated in Fig. 2.

4 Experimental Results

We require 40–50 facial data of each employee. Above images are used for training
the system. Accuracy achieved ranges from 95 to 100% as shown in Fig. 3.

5 Conclusion and Future Scope

We have seen various kinds of techniques in this paper that allow us to detect the
face and recognize it in various situations such as dark and light and even in various
clothing where the face is partially visible. All the techniques such as viola jones
used for object detection in OpenCV and haar cascade classifier which is also used
by the help of OpenCV. However, we observed that FaceNet provide significantly
better results than the previously mentioned techniques. Additional features in this
project can be further added after successful implementation, for example, visitors
who have been granted access in the office can be monitored for behavioral analysis;
sentiment analysis to check whether the visitor becomes hostile and creates nuisance,
etc. Furthermore, additional features in which employees can be monitored to keep a
Office Monitoring and Surveillance System 9

Fig. 2 Working of face recognition

Fig. 3 Face recognition

10 V. Patil and Y. Jadhav

check on the performance in work can be implemented. In restricted areas within the
company, only employees who are allowed access will be granted entry otherwise
no.

References

1. Wu W, Liu C, Su Z (2017) Novel real-time face recognition from video streams. In: Conference
on computer systems, electronics and control (ICCSEC), Dalian, pp 1149–1152. [Link]
org/10.1109/ICCSEC.2017.8446960
2. Xiugang G, Yang L, Ruijin Z, Liang Y (2011) Study on monitoring systems with video capture
cards and wireless sensors. In: IEEE conference on electronics, communications and control
(ICECC), 03 November 2011
3. Saputra DIS, Amin KM (2016) Face detection and tracking using live video acquisition in
camera closed circuit television and webcam. In: 1st international conference on information
technology, information systems and electrical engineering (ICITISEE), Yogyakarta, pp 154–
157. [Link]
4. Kadir K, Kamaruddin MK, Nasir H, Safie SI, Bakti ZAK (2014) A comparative study between
LBP and Haar-like features for face detection using OpenCV. In: 4th international conference
on engineering technology and technopreneuship (ICE2T), Kuala Lumpur, pp 335–339. https://
[Link]/10.1109/ICE2T.2014.7006273
5. Jose E, Greeshma M, Haridas MTP, Supriya MH (2019) Face recognition based surveillance
system using FaceNet and MTCNN on Jetson TX2. In: 5th international conference on advanced
computing & communication systems (ICACCS), Coimbatore, India, pp 608–613. [Link]
org/10.1109/ICACCS.2019.8728466
6. Heath K, Guibas L (2007) Facenet: tracking people and acquiring canonical face images in a
wireless camera sensor network. In: First ACM/IEEE international conference on distributed
smart cameras, Vienna, pp 117–124. [Link]
7. Boka A, Morris B (2019) Person recognition for access logging. In: IEEE 9th annual computing
and communication workshop and conference (CCWC), Las Vegas, NV, USA, pp 0933–0936.
[Link]
8. Chien Y-T, Huang Y-S, Jeng S-W, Tasi Y-H, Zhao H-X (2003) A real-time security surveillance
system for personal authentication. In: IEEE 37th annua international Carnahan conference on
security technology. Proceedings, Taipei, Taiwan, pp 190–195. [Link]
2003.1297558
9. Qezavati H, Majidi B, Manzuri MT (2019) Partially covered face detection in presence of
headscarf for surveillance applications. In: 4th international conference on pattern recognition
and image analysis (IPRIA), Tehran, Iran, pp 195–199. [Link]
8786004
10. Cao Y, Pranata S, Nishimura H (2011) Local binary pattern features for pedestrian detection
at night/dark environment. In: IEEE international conference on image processing, Brussels,
pp 2053–2056. [Link]
11. William I, Ignatius Moses Setiadi DR, Rachmawanto EH, Santoso HA, Sari CA (2019) Face
recognition using FaceNet (Survey, Performance Test, and Comparison). In: Fourth interna-
tional conference on informatics and computing (ICIC), Semarang, Indonesia, pp 1–6. https://
[Link]/10.1109/ICIC47613.2019.8985786
12. Ming Z, Chazalon J, Luqman MM, Visani M, Burie J (2017) Simple triplet loss based
on intra/inter-class metric learning for face verification. In: IEEE international conference
on computer vision workshops (ICCVW), Venice, pp 1656–1664. [Link]
ICCVW.2017.194
13. Geng X, Zhou Z-H, Smith-Miles K (2008) Individual stable space: an approach to face
recognition under uncontrolled conditions. IEEE Trans Neural Networks 19(8):1354–1368
Office Monitoring and Surveillance System 11

14. Ibrahim R, Zin ZM (2011) Study of automated face recognition system for office door access
control application. In: IEEE 3rd international conference on communication software and
networks, Xi’an, pp 132–136. [Link]
15. Celine J, Sheeja Agustin A (2019) Face recognition in CCTV systems. In: International confer-
ence on smart systems and inventive technology (ICSSIT), Tirunelveli, India, pp 111–116.
[Link]
Categorizing Documents by Support
Vector Machine Trained Using
Self-Organizing Maps Clustering
Approach

Vishal Patil, Yogesh Jadhav, and Ajay Sirsat

Abstract This paper mainly emphasis on the use of machine learning algorithms
such as self-organizing maps (SOM) and support vector machines (SVM) for clas-
sifying text documents. We have to classify documents effectively and accurately to
different classes based on their content. We tested classification of self-organizing
map on Reuters R-8 data set and compared the results to three other popular machine
learning algorithms: k-means clustering, k nearest neighbor searching, and Naive
Bayes classifier. Self-organizing map yielded the highest accuracies as an unsu-
pervised method. Furthermore, the accuracy of self-organizing maps was improved
when used together with support vector machines.

Keywords Document Categorization · Self-organizing maps (SOM) · Support

vector machines (SVM)

1 Introduction

It is very important to store information in a way that is easy for retrieval. This
becomes even more important in the case of electronic information. Every day we
search for documents on our computer is it at home and at work. With each passing
day the amount of information increases. The problem which we need to overcome is
manually storing all related information together and searching for that information.
This may take a lot of effort and time. The solution to this problem is to classify
and label documents. Once the documents are classified and ordered, we can go

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 13

P. M. Pawar et al. (eds.), Techno-Societal 2020,
[Link]
14 V. Patil et al.

through the documents and search for information according to our needs. Consider
a scenario where documents are already ordered and. In such a scenario, we can
browse through documents effectively. This is where machine learning comes into
focus.
Machine learning algorithms can classify documents automatically making it
easy to store them. These algorithms can be supervised or unsupervised. Supervised
machine learning algorithms involve human interference. Since it may not be the case
that class labels are already known for each document, this may give rise to human
error. Also, many a times the document has no title or has an ambiguous title, which
makes it difficult to understand the category of the document. Thus we have focused
on unsupervised machine learning algorithms for the task of document classification.
We have used Kohonen’s maps or self-organizing maps to classify documents. Self-
organizing maps (SOMs) [1] is an effective data Visualization method which can be
used to reduce the dimensions of data by using self-organizing neural networks [2].
It is a very popular method for clustering [3] and classification of text documents [4,
5]. We compared self-organizing map with one unsupervised [6] and two supervised
machine learning algorithms.
Self-organizing maps gave better performance as compared to k nearest neighbor
and k-means clustering with accuracies as high as 90–92%. However, the draw-
back of SOM is Curse of Dimensionality (COD) because of which its efficiency is
affected. Because of this, SOM is not able to differentiate between classes that give
the same Euclidean distance even though they have different features. Hence, we
have implemented a technique which uses SOM and SVM. SOM takes the unlabeled
text documents and forms clusters for these documents which form the training data
set for the Support Vect4or Machine (SVM) [7].
This method has a two-fold advantage; it reduces the effect of COD while also
eliminating human interference needed to label the training documents. Support
Vector Machine have been one of the most widely used machine learning algorithm
in most of the text classification application given as input for many a times for large
datasets manually classifying and labeling training documents in the preprocessing
stage leads to an increase in time and cost complexity. Thus, the unsupervised clus-
tering methods are used which eliminate the need of human involvement for grouping
together set of documents. The output of these methods is then Training phase of
SVM. This hybrid model gives excellent accuracy while also eliminating the effect
of COD. Moreover, we can see that the accuracy of text classification increases after
using this hybrid approach.

2 Literature Survey

Support vector machine is one of the most widely used supervised machine learning
algorithms. It is one of the most commonly used methods and has found use in a
variety of applications due to its excellent accuracy [8] (Fig. 1).
Categorizing Documents by Support Vector Machine … 15

Fig. 1 The optimal

separating hyper plane

However, for some of the classification domains, it might not always be possible to
label the training data set of Support vector machine. For certain domains, labeling the
training data might be difficult when human expertise is unavailable or when it may
be time-consuming and costly to involve humans. Hence, in cases where it may not
be possible to label training data, unsupervised clustering methods are introduced
as they can perform their task without training the dataset. For domains wherein
training data is unavailable, we can use unsupervised machine learning methods,
such as SOM and k-Means to group documents based on its content.
A SOM is a single layer neural network. The name artificial neural network is
because of the fact that they were originally inspired by the way how biological
neurons were thought to work. SOM is made up of multiple units lying along a grid
of n-dimensions. In the input layer for every unit has the number of weights and
input patterns are same. Training a Self Organizing Map when provided with an
input pattern requires calculating the distance between that pattern and every other
unit in the neural network [9]. A unit that has the least distance to Best Matching
Unit (BMU) is then selected and that pattern is then mapped onto that unit. There
will be a one to one correspondence between the mapping of patterns that lie nearby
in the input space and the units lying nearby in the output space. This will be visible
in the visual map generated by SOM [9, 10]. Thus, SOM is topology preserving as
neighborhoods of a node are kept intact through the mapping process (Fig. 2).

3 Proposed System

The data set we have selected is a variant of the Reuters- 21,578 collection. This
dataset from the year 1987 includes 21,578 English Reuters news documents. More
than one class labels to some the documents Therefore; we have discarded documents
16 V. Patil et al.

Fig. 2 SOM architecture

with multiple class labels and selected only those documents which have only one
class label.
We then selected the 10 largest classes to obtain the collection of 8008 documents,
collection of 8008 documents, consisting of 2254 test documents and 5754 training do
over the past two decades, many research works focusing on document classification
by machine learning algorithms have been carried out. A large number of these papers
focus on improving the efficiency of SVM. Many researchers have proposed a hybrid
method which combines unsupervised learning algorithms such as SOM with SVM
[7]. Wu’s research shows how we can improve the accuracy of SVM [11] analyzing
zebra fish gene expression [12] by analyzing data in the training data set by using
SOM. SOM filters the training data for Support vector machines thereby reducing
the train dataset by keeping only those neurons in every category of SOM that are
their BMUs while it eliminates the remaining we then performed conventional text
preprocessing on the dataset as shown in Fig. 3.
1. The dataset we have used is the Reuters R8 dataset having 5485 training and
2254 test samples respectively.
2. For preprocessing, firstly words we retransformed to their stems and then, stop
words were removed.
3. Next, we calculated the frequencies of resultant words and computed document
vectors for all documents.
4. Finally, tf-idf weights were calculated for all resulting word stems.
5. For using a self-organizing map in the document classification task we have to
label the map nodes with class labels of the training data set.
6. The labelled nodes then represent the document classes. The map can then
classify new documents by mapping them.
SOM is primarily used for labeling and organizing the data samples which are not
labeled. As mentioned earlier in this paper, the SOM [2] is affected by COD because
it measures similarity using Euclidean distance [7]. Reducing the dimensions of
original data in their principle components might be of some help. However, this
reduction is unsuitable for some datasets, because every feature consists of some
information. Using this may result in loss of information if any one of the dimensions
Categorizing Documents by Support Vector Machine … 17

Fig. 3 Flowchart showing

Preprocessing steps
performed on the data set
18 V. Patil et al.

is removed [7]. The obtained labels are coupled with the related original data samples
for creating a machine labeled training set for the SVM which was machine organized.
The SOM begins the step of clustering the documents by deciding which important
features are needed for efficient clustering of the data into groups [7]. A process of
statistical cluster analysis is usually used to better the feature extraction which is
iterative in nature. The dimensions of data are reduced to two dimensions which
simplify viewing and interpretation is obtained using the feature maps [7].
Self-organizing maps generate a map which represents the dataset [7]. When
different neurons are fired for each data point a greater number of neurons will give
larger number of classes. Thus, to prevent this, the neurons are first grouped in small
clusters using Euclidean distance. Many neurons together make a cluster [7]. Every
cluster is given a class number. Using the Euclidean distance all the data points are
grouped together around these clusters. A number of a particular cluster is assigned
to data points with the closest proximity to that cluster [7].
The combination of SOM and SVM which results in a method which uses unsuper-
vised learning and has better generalization ability [7]. This is assumed from the prop-
erties of the SVM. SVM is known to be excellent classification methods with respect
to performance [7]. The generalization attribute of Support Vector Machine because
we implement structural risk minimization which selects an optimal separating hyper
plane which guarantee classification with very high accuracy (Fig. 4).

4 Results and Conclusion

First we compared different algorithms used for clustering based on their accuracy.
The results are as shown in the Table 1 and the graph in Fig. 6. We can see from the
Table 1 that the accuracy of SOM is 92.45% which is very good when compared to
methods such as k means and k nearest neighbor whose accuracies are 90.23% and
89.9% respectively. The accuracy of Naive Bayes is 95.24%. As seen in Fig. 6 the
accuracy of the hybrid model of SOM and SVM is 96.44% which is more than the
accuracy obtained from Naive Bayes (Fig. 5).

5 Conclusion

We tested Self Organizing Maps for classification of text documents and compared
it with some of the commonly used text classification methods. Naive Bayes had
the best accuracy overall, but in unsupervised algorithms, SOM turned out to have
the best accuracy for text classification. Sometimes, the knowledge and expertise to
label and organize documents into separate classes might be unavailable and thus
unsupervised methods viz. SOM proves to be useful. Self-organizing maps gave over
90% classification accuracy in the Reuters R-8 Data Set making it is effective for the
document classification. Also, the accuracy of self-organizing maps was comparable
Categorizing Documents by Support Vector Machine … 19

Fig. 4 Block diagram implemented hybrid classification algorithm from [7]

Table 1 Comparison of
Algorithm Accuracy (%)
clustering methods
k means 90.23
kNN 89.9
Naive Bayes 95.24
SOM 92.45

to some of the supervised machine learning methods that we tested. The training
phase of SOM generates an intuitive visual map, which is useful in a variety of
applications.
The learning procedure of this method though costly, is advantageous. Theo-
retically, the learning phase of SOM is more time consuming as compared to the
learning of k nearest neighbor. However, practically it is the exact opposite. The map
has significantly less nodes compared to the training dataset. This result to 10 time’s
quicker classification when compared to that obtained from k nearest neighbor using
20 V. Patil et al.

Fig. 5 Clusters formed after using SOM

Fig. 6 Comparison of various machine learning algorithms

the same data set. We don’t have to build a new map whenever we get a new input
document in the dataset, we can map it and perform the learning phase afterwards
when there is no new input data available. The results presented above show that
when SOM is used along with SVM the accuracy obtained is near to that which we
get when the data used for training is classified by supervised machine algorithms.
Categorizing Documents by Support Vector Machine … 21

References

1. Saarikoski J, Laurikkala J, Jrvelin K, Juhola M (2011) Self-organising maps in document clas-

sification: a comparison with six machine learning methods. Adaptive and Natural Computing
Algorithms, pp 260–269
2. Haykin S (1999) In neural networks: a comprehensive foundation, Second Edition. Prentice
Hall, Upper Saddle River
3. Mary Amala Bai V, Manimegalai D (2010) An analysis of document clustering algo-
rithms. In: International conference on communication control and computing technologies,
Ramanathapuram, pp 402–406
4. Ko Y, Seo J (2000) Automatic text categorization by unsupervised learning. In: Proceedings
of the 18th international conference on computational linguistics, (COLING2000), pp 453459
5. Indu M, Kavitha KV (2009) Review on text summarization evaluation methods. In: International
conference on research advances in integrated navigation systems (RAINS), Bangalore, 2016,
pp 1–4
6. Rui W, Liu J, Jia Y (2016) Unsupervised feature selection for text classification via word
embedding. In: ICBDA
7. Shafiabady N, Lee LH, Rajkumar R, Kallimani VP, Akram NA, Isa D (2016) Using unsu-
pervised clustering approach to train the support vector machine for text classification.
Neurocomputing 211:4–10
8. Brucher H, Knowlmayer G, Mittermayer MA (2002) Document classification methods for
organizing explicit knowledge. In: Proceedings of the 3rd European conference on organiza-
tional knowledge, learning and capabilities, (ECOKLC02), Institute of Information Systems,
University of Bern, Engehaldenstrasse Bern, Switzerland, Athens, Greece, pp 124–126
9. Lobo VJAS (2009) Application of self-organizing maps to the maritime environment. In:
Popovich VV, Claramunt C, Schrenk M, Korolenko KV (eds) Information fusion and
geographic information systems. Lecture Notes in Geoinformation and Cartography. Springer,
Berlin, Heidelberg
10. Sigogne A, Constant M (2009) Real-time unsupervised classification of web documents. In:
IMCSIT, pp 281–286. IEEE
11. Li TS, Huang CL (2009) Defect spatial pattern recognition using a hybrid SOMSVM approach
in semiconductor manufacturing. Expert Syst Appl 36(1):374–385
12. Wu W, Liu X, Xu M, Peng J, Setiono R (2004) A hybrid SOM-SVM method for analyzing
zebra fish gene expression. In: Proceedings of the 17th international conference on pattern
recognition (ICPR04) vol 2, pp 323–326
Bandwidth Improvement of Multilayer
Microstrip Patch Antenna by Using
Capacitive Feed Technique
for Broadband Applications

Anil K. Rathod, Md. M. Bhakar, M. S. Mathpati, S. R. Chougule,

and R. G. Sonkamble

Abstract In this research paper, Bandwidth improvement is investigated in E-shaped

Micro strip patch antennas by using multilayer and capacitive feed techniques. Capac-
itive feeding technique has been used to cancel the inductive impedance of probe by
capacitive patch. The Rectangular capacitive patch has been used separately near the
radiating E-shaped design model. Performance analysis of E-shaped conventional
and suspended capacitive feed Micro Strip Patch Antenna is done to enhance its
parameters like Bandwidth, Gain, and Directivity. The proposed E-shaped multi-
layer capacitive feed Micro strip patch Antenna is shown improved bandwidth of
238 MHz and Gain of 6.071 dB as compared with conventional candidates. The
proposed antenna design models have a working center frequency in the range of
2.36 GHz to 2.59 GHz, which can be used for wireless applications. The Return
Loss and VSWR have been investigated in an acceptable range. E-field and Current
flow of the antenna are within the desired radiating patch. It has observed that the
capacitive feed technique E-shaped design model improves the Bandwidth, Gain,
and other antenna parameters over the conventional probe feed antenna.

Keywords E-shaped · Bandwidth · Gain · Directivity · Suspended techniques ·

Capacitive feed technique

A. K. Rathod (B) · Md. M. Bhakar

Department of Electronics & Communication Engineering, G.N.D. Engineering College, Bidar,
India
M. S. Mathpati
Department of Electronics & Telecommunication, SVERI’s COE, Pandharpur, India
S. R. Chougule
Department of Electronics & Telecommunication, KIT’s COE, Kolhapur, India
R. G. Sonkamble
Department of Computer Science and Engineering, SGU, Kolhapur, India

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 23

P. M. Pawar et al. (eds.), Techno-Societal 2020,
[Link]
24 A. K. Rathod et al.

1 Introduction

Nowadays, antennas are an essential part of any wireless communication system.

With technological advancements, antennas are playing a very vital role in commu-
nication [1, 2]. Micro strip antennas (MSAs) have been reported that they have
lightweight, small-volume several advantages and that they can be made conformal
to any type of host surface. Besides, PCB Technology can be used the physical
manufacturing of Micro strip patch Antenna so that a large amount of production
can be done at a small cost [1]. Defense and other commercial applications of Micro
Strip Patch Antenna are alternating choices to replace the conventional Antenna.
However, the MSA has been reported that they are suffering from Narrow Band-
width. Increasing the Bandwidth of the MSA has become a primary requirement of
research in the field. This is reflected in a large number of papers on the subject
published in journals and conference proceedings [1]. However, different types of
Broadband MSA design models have been investigated in the last few decades. The
coaxial or probe feed techniques can be used in many design methodologies. To
impedance matching, we have to do the soldering of the Patch and center conductor
of the coaxial connector. We can place at any desired location of feed point to match
input impedance; this is the advantage of the probe feeding method [1, 3]. A new
-shaped micro strip antenna is reported, which increased the bandwidth of the
antenna by converting rectangular to E-shape and -shaped micro strip antenna [4].
The stacking of two substrates and slotted patches were used to get the wide band-
width is investigated [5]. Defected Ground Structure (DGS) is a novel technique
used to modify the ground plane to improve the gain and bandwidth is reported
[6]. Multilayer stretchable conductors (SCs) are demonstrated for the first time as
an alternative cost-effective method for implementing conformal and reconfigurable
antennas over flexible substrates [7]. Some of Researchers have been designed patch
antenna on single, double, and multilayer dielectric substrate introduced by microma-
chining process [8]. A miniature patch antenna is proposed for satellite applications
in (C-band, X-band, and Ku-band). The designed multiband antenna is sensible for
satellite applications [9]. A fractal antenna is designed at 2.4 GHz frequency band
which is suitable for Wireless Local Area (WLAN) applications and FR-4 dielec-
tric substrate material is used [10]. To improve the bandwidth and gain of Micro
Strip patch antenna deferent methodologies investigated and anyone technique can
be adopted.

2 Proposed Antenna Design Model

Figure 1 shows the proposed E-shaped mulilayer MSA with capacitive feed. The
design Model consists of a copper ground plane, radiating Patch, air gap, and FR-4
as a dielectric substrate. The Capacitive probe feed technique has been used for the
impedance matching with the radiating patch.
Bandwidth Improvement of Multilayer Microstrip … 25

Fig. 1 E-shaped suspended MSA with capacitive feed patch

The air gap acts as a dielectric substrate having a dielectric constant is equal to
one. Here the proposed model has become Multilayer consisting of a copper ground
plane, air gap with a height of 6 mm, a dielectric substrate with a height of 1.6 mm
easily available in the market, and an E-shaped radiating patch [11]. The permittivity
can be selected near to 1 to obtain a wide bandwidth. If the dielectric substrate has
larger thickness then it gives rise to an increase in probe reactance for the excitation
of surface waves and coaxial feed, which reduces the efficiency of the antenna so that
dielectric height should be selected properly. This design model is used to investigate
the antenna parameters like Operating Frequency, Bandwidth, Return Loss, VSWR,
Directivity, Total Gain, Radiation pattern, and E-field.

3 Simulation Results of the Proposed Design Model

3.1 Bandwidth and Return Loss of E-shaped Suspended MSA

Figure 2 represents the Bandwidth and Return loss of E-shaped Suspended MSA
which gives the bandwidth of 238 MHz and Return Loss −12.98. The return loss of
Antenna is in acceptable range.
26 A. K. Rathod et al.

Fig. 2 Bandwidth and return loss of E-shaped suspended MSA

Fig. 3 VSWR of E-shaped suspended MSA

3.2 VSWR of E-shaped Suspended MSA

Figure 3 represents the VSWR of E-shaped Suspended MSA. It shows the VSWR
1.5778 must be less than 2 as per the standards. The given antenna can be radiate
and fabrication process can be done.

3.3 Directivity of E-shaped Suspended MSA

Figure 4 represents the Directivity of E-shaped Suspended MSA is 6.071 dB.

Bandwidth Improvement of Multilayer Microstrip … 27

Fig. 4 Directivity of E-shaped suspended MSA

3.4 Gain of E-shaped Suspended MSA

Figure 5 represents the Gain of E-shaped Suspended MSA is 6.0196 dB, which is
better than the conventional Antena.

3.5 Radiation Pattern of E-shaped Suspended MSA

See Fig. 6.

3.6 E-field of E-shaped Suspended MSA

Figure 7 represnts E-field of E-shaped Suspended MSA.

4 Result Analysis Table of E-shaped Suspended MSA

Table 1 consists of result analysis table for Hexagonal MSA & Suspended Hexagonal
MSA with performance Parameters like Operating Frequency in GHz, Bandwidth in
28 A. K. Rathod et al.

Fig. 5 Gain of E-shaped suspended MSA

Fig. 6 Radiation pattern of E-shaped suspended MSA

Bandwidth Improvement of Multilayer Microstrip … 29

Fig. 7 E-field of E-shaped suspended MSA

Table 1 Result analysis table

Parameters E-shaped suspended MSA
for E-shaped suspended MSA
Operating frequency (GHz) 2.361 − 2.599
Bandwidth (MHz) 238
Return loss(dB) −12.989
VSWR 1.577
Directivity (dB) 6.071
Total gain (dB) 6.019

MHz, Return Loss in dB, VSWR, Directivity in dB and Total Gain in dB as shown
in Table 1.
The Bandwidth can be defined as a reappearance extends over which VSWR is
always under two (which relates to an arrival loss of 9.5 dB or 11% reflected force).
The VSWR or impedance BW of the MSA is characterized as the recurrence extend
over which it is coordinated with that of the feed line inside determined cutoff points.
The Bandwidth of proposed E-shaped multilayer MSA is 238 MHz, which is
within the specified limits as shown in Fig. 2. Return Loss has been observed in an
acceptable range. VSWR can be defined in terms of the input reflection coefficient
(G) as:

1 + ||
V SW R = (1)
1 − ||

VSWR of E-shaped multilayer MSA is less than 2, which is shown in Fig. 3. As

shown in Fig. 4 for Directivity and Fig. 5 for Gain of E-shaped suspended MSA.
Figure 6 shows the radiation pattern for the proposed antenna. E-field of E-shaped
multilayer MSA is shown in Fig. 7. All the proposed antenna parameters have been
observed in the acceptable range.
30 A. K. Rathod et al.

5 Conclusion

In this paper, Bandwidth improvement is investigated in E-shaped Micro strip patch

antennas by using multilayer and capacitive feed techniques. The design model of
E-shaped multilayer MSA is operating at 2.45 GHz center frequency. The proposed
E-shaped multilayer Micro strip patch Antenna is shown improved bandwidth of
238 MHz and gains 6.019 dB, as compared with normal Antenna. The return loss
of the proposed designed model is −12.989 dB which is less than −10 dB. The
Voltage Standing Wave Ratio (VSWR) has been observed that 1.577, which is in the
acceptable range. FR-4 dielectric substrate material is used to model design having
relative permittivity (A) of 4.4. It has less cost compare to other materials. The
proposed E-shaped multilayer Micro strip patch Antenna can be considered a better
candidate for Broadband Applications.

Reference

1. Kumar G, Ray KP (2003) Broadband micro strip antennas. Artech House Boston, London
2. Al-Halawani MG, Al-Najjar M, Abdelazeez MK (2016) Microstrip antenna design for UWB
applications. IEEE. 978-1-5090-2886-3/16/$31.00
3. Faouri YS, Awad NM, Abdelazeez MK (2019) Hexagonal patch antenna with triple band
rejections. In: Proceedings on 2019 IEEE Jordan International Joint Conference on Electrical
Engineering Information Technology JEEIT 2019, pp 446–448
4. Deshmukh AA, Ray KP (2013) Analysis of broadband Psi ()-shaped micro strip antennas.
IEEE Antennas Propag Mag 55(2)
5. Tiwari K, Pujara D (2015) Multilayer slotted microstrip antenna for Wi-Fi application. IEEE.
978-1-4799-7815-1/15/$31.00
6. Kachare AB, Mathpati MS, Rathod AK (2015) Effect of size and location of the DGS on char-
acteristics of rectangular micro strip patch antenna. Int J Comput Appl (0975–8887), Confer-
ence on Emerging Applications of Electronics System, Signal Processing and Computing
Technologies (NCESC 2015)
7. Liyakath RA, Takshi A, Mumcu G (2013) Multilayer stretchable conductors on polymer
substrates for conformal and reconfigurable antennas. IEEE Antennas Wireless Propag Lett 12
8. Saha R, Maity S, Trigunayat N (2015) Enhancement of gain, bandwidth and directivity of a
patch antenna by increasing dielectric layers of the substrate through micromachining technique
for RFID application. In: 2015 international conference on advances in computer engineering
and applications (ICACEA), IMS Engineering College, Ghaziabad, India, IEEE. 978-1-4673-
6911-4/15/$31.00
9. Nageswara Rao L, Santhosh Kumar V (2020) A novel multiband micro strip patch antenna for
satellite applications. Int J Adv Sci Technol (IJAST) 29(3s):177–183. ISSN: 2005-4238
10. Ali KM, Nataraj C et al (2020) Bandwidth analysis of microstrip antenna with fractal design
for 2.4 GHZ wireless applications. Int J Adv Sci Technol (IJAST) 29(1):1344–1358. ISSN:
2005–4238
11. Hong S, Kim Y, Won Jung C (2017) Transparent microstrip patch antennas with multilayer
and metal-mesh films. IEEE Antennas Wireless Propag Lett 16(c):772–775
Use of Median Timbre Features
for Speaker Identification of Whispering
Sound

Vijay M. Sardar, Manisha L. Jadhav, and Saurabh H. Deshmukh

Abstract Identifying speaker from the whispered voice is difficult task contrasted
to neutral as voiced phonations are absent in the whisper. The accomplishment of
the speaker identification system for the most part relies on the selection of proper
audio features reasonable for the type of database and type of application. This paper
examines the various audio features available and emphasizes on the use of selected
timbrel features which are sorted by Hybrid Selection Algorithm. The limited number
of timbrel features namely MFCC, Roll-off, Brightness, Roughness, and irregularity
which are found outperforming when tested on CHAIN database. Likewise, the
possibility of using the MEDIAN based features is investigated by analysis. The
use of Median timbrel features reported an enhancement in speaker identification
accuracy by 2.4% compared to timbrel features only in whisper train-whisper test
scenario.

Keywords Whisper · Speaker identification · Timbrel feature · K-NN classifier ·

Median

1 Introduction

Speaker identification requires the main processes as (i) Pre-processing, (ii) Feature
extraction (iii) Training and (iv) Testing. Pre-processing may involve noise reduc-
tion, pre-emphasis, silence removal and equalization so forth. Audio features are
the speaker-specific information represented in a compact form. They will make it
possible to compare the speakers for identification purpose with minimum processing
speed and data. In the training phase, all extracted audio features are utilized for
modelling the speakers which will be used as the templates in the testing phase.

V. M. Sardar (B)
Jayawantrao Sawant College of Engineering, Pune, MS, India
M. L. Jadhav
MET’s Institute of Engineering, Nasik, MS, India
S. H. Deshmukh
Maharashtra Institute Of Technology, Aurangabad, MS, India

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 31

P. M. Pawar et al. (eds.), Techno-Societal 2020,
[Link]
32 V. M. Sardar et al.

While testing phase, features of query speaker is compared for the closest speakers
among the database. There are two types of variants in the speaker identification
system, in particular, closed-set and open set. In the open set scenario, the unknown
speaker not available in the database may also be used, but that task will be speaker
verification [1]. The speaker verification is a binary decision to whether a claimed
speaker is the same or not by comparing with the single audio template. While in
an identification task, every speaker is compared with every other speaker in the
database [2].
Identifying the person from the whispered voice is a challenging task due to
its hidden and multi-dimensional attributes. Unmistakable separation ability of the
rich phonation is misplaced from a whisper [3]. The ADs applicable for the neutral
database is not found effective for the whispered database. Hence, the adapted
approach as shown in Fig. 1 where a selection of the proper ADs is carried out
in the beginning only. This process of selecting the appropriate ADs reasonable for
the type of database and application will enhance the identification rate.
A majority of low-level audio features are discussed in the literature. It includes
primary features such as the zero-crossing rate, speech bandwidth, the spectral
centroid, and energy [4, 5]. A second high level and complex feature used for speech
and speaker recognition are like Mel-frequency cepstral coefficients (MFCC), roll-
off and brightness etc. Such features use the parametric analysis of the spectral
envelope [6]. It is expected that the features used for the classification need to be
more uncorrelated. Typical classification results using better non-correlated MFCC
features found correct up to 95% for music and speech classification which is a rela-
tively less complex task [7]. Besides, these results have reported for clean speech
scenario. For other databases including noise, music background, and telephone
speech etc. the performance degrades considerably [8, 9]. The block diagram of
speaker identification system is as shown in Fig. 1.
This paper is organized as follows. We portrayed the classification of audio
features, timbrel audio descriptors, MIR toolbox, and the impact of timbrel feature

Fig. 1 Block diagram of the speaker identification system

Use of Median Timbre Features for Speaker Identification … 33

selection on the identification accuracy in Sect. 2. The K-nearest neighbor classifier

with the most suitable variants for the whispered speaker database is described in
Sect. 3. The performance of the system with timbrel features and K-NN classifier
are presented in the result section (Sect. 4). The possibility of improvement based
on the MEDIAN of feature values is discussed in Sect. 5.

2 Audio Descriptors

2.1 Classification of Audio Descriptors

We can classify the audio features comprehensively in two classes as follows. The
global descriptor is the class of feature where computation is done on the complete
signal as a whole. E.g. attack-time of a sound is known from the complete duration
of an audio signal. Instantaneous descriptors is another class which computes on a
short period (say 40 ms) of audio signal called a frame. The spectral centroid of a
signal can vary with the time, hence it will be termed as an instantaneous descriptor.
As an instantaneous descriptor produces multiple values for the number of frames,
use of statistical operations (like the mean or median, the standard deviation, and or
inter-quartile range etc.) is essential to derive a single value representation. In the
CUIDADO project [10], a listing of 166 audio features is provided.
Depending upon the type of process used for extraction of the feature, we can
further differentiate:
• Features that are directly extracted from the time domain waveform of audio signal
like the zero-crossing rate.
• A transform like FFT, wavelet etc. is applied on a signal to extract the features.
E.g. MFCC.
• Feature extracted based on a signal model like the source or filter model.
• Features which converges to the human auditory response (response on bark or
ERB scale).

2.2 Timbrel Audio Descriptors

Timbre is the perceptual and multi-dimensional feature of sound. As the exact defi-
nition of timbre is very difficult, it can be analyzed by following the attributes
[11].
1. Harmonic analysis: The number, relative strengths, structure of harmonics.
2. Partial analysis: Phase, inharmonic partials, the content of partials.
3. Time-related parameter like rise-time.
4. Steady-state and attack slices.
34 V. M. Sardar et al.

2.3 Timbrel Audio Descriptors MIR Toolbox for MATLAB

Musical Information Retrieval (MIR) toolbox is for the most part intended to enable
the study of the relation between musical attributes and music-tempted sensation.
MIR toolbox uses a modular outline. It is well known that the common algorithms are
used in audio processing like segmentation, filtering, framing etc. with an addition of
one or more distinguished algorithms at some stage of processing. These algorithms
are available in a modular form and the individual blocks can be integrated to capture
some features [12].
The philosophy to integrate the appropriate modules is proposed in Fig. 2. For
example, to measure irregularity and brightness, we need the implement the algo-
rithms like reading audio samples, segmentation, filtering, and framing as the
common processes between them. In the final stage, due to inherent differences, irreg-
ularity needs peaking algorithm and brightness needs spectrum analysis (Fig. 2).
Even, the integration of different stages depends upon parameter variations. E.g.
mirregularity (…, ‘Jensen’), where the adjoining partials are taken into considera-
tion and mirregularity (…, ‘Krimphoff’) which considers the mean of the preceding,
same and next amplitude [13]. The flow diagram of algorithm modules and their
integration to extract the selected timbrel features are shown in Fig. 2.
Roll-off frequency: Roll-off is assessed from the foremost energy (85% or 95% as
a standard) contained below the predefined frequency.
Roughness: It estimates the average disagreement between all peaks of the signal.
It is also an indicator of the presence of harmonics generally higher than the 6th
harmonic.
Brightness: It is the measure of the percentage of energy spread above some cut-off
frequency.

Fig. 2 The logical flow of timbrel feature implementation in MIR

Use of Median Timbre Features for Speaker Identification … 35

Irregularity: It may be calculated as the sum of the square of the difference in

amplitude between adjoining partials or the sum of the amplitude minus the mean of
the past, the same component and subsequent amplitude.
Miraudio: This command loads the appropriate format of an audio file. E.g.
miraudio (‘[Link]’).
Mirsegment: This process splits a continuous audio signal into homogeneous
segments.
Mirfilterbank: A set of filters are required which are useful to select neighboring
narrow sub-bands that cover the entire frequency range. The effect like aliasing in the
reconstruction process is avoided e.g. mirfilterbank (…, ‘Gammatone’) processes a
Gammatone filterbank decomposition. The frame decomposition can be performed
using the mirframe command. The frames can be specified as follows: mirframe
(x,…, ‘Length’, w, wu).
mirspectrum: Discrete Fourier Transform decomposes the energy of a signal (be
it an audio waveform, or an envelope, etc.) along with the frequencies.
Mathematically, for an audio signal x;

N −1
−2ikn
Xk = xne N k = 0, . . . , N − 1 (1)
n=0

This decomposition is performed using a Fast Fourier Transform by the mirspec-

trum function.
Mirpeaks: Many features like irregularity require the Peaks analysis. Peaks are
calculated from any data x produced in MIR toolbox using the command mirpeaks
(x).

2.4 Timbrel Features for Better Speaker Identification

For better speaker identification, the accompanying conditions should be satisfied:

(i) The least variation in the inter-speaker feature (ii) Maximum discrimination in an
inter-speaker feature. We have selected a limited well-performing feature (MFCC,
Roll-off, Roughness, Brightness, and Irregularity) from MIR toolbox by using Hybrid
Selection Algorithm. It uses the iterative process of testing for the performance of
audio features. The accuracy of the system is tested with the independent features
and then by successively appending the new feature in conjunction with the previous
combination. To conclude, it is proven that a combined vector having timbrel features
namely MFCC, Roll-off, Brightness, Roughness and, Irregularity is found to be the
best [14]. An association among the intra-speaker samples and dissociation among
the inter-speaker samples are confirmed by correlation analysis. It predicts that the
selected timbrel features are well-performing for speaker identification in a whispered
voice. It is validated by the identification experiments in subsequent sections.
36 V. M. Sardar et al.

3 System Description

The system block diagram is presented in Fig. 3 which is described in succeeding

sub-sections.

3.1 Speaker Database

This undertaking makes use of the CHAIN database developed at the School of
Computer Science and Informatics University College Dublin [15]. It consists of
a total of 36 speakers with 33 samples each, with a good mix of male and female
voice samples (20 males and 16 females). The speech samples of 2–3 s duration are
recorded at a sampling frequency of 44.1 kHz. The phrases/sentences are selected
from CSLU and TIMIT database which ensures the phonetic balance in the corpus.

3.2 Hybrid Selection Algorithm

Hybrid selection is the iterative process which starts with the targeted timbre class
of ADs and further investigates the AD which maximizes the identification result
[16]. This algorithm also used for liver tissue pathologicssal image classification in
[17]. After every iteration, the feature which maximizes the classifier accuracy is
appended in combination with the previous feature/s. The process continues until no
further increase in accuracy is observed. The algorithm proved that MFCC, Roll-off,
Roughness, Brightness, and Irregularity are found to be best suitable.

Feature
Hybrid
Database

Extraction Classifier
Selection

Speaker
ID

Fig. 3 System block diagram

Use of Median Timbre Features for Speaker Identification … 37

3.3 K-Nearest Neighbor (K-NN)

K-NN is a simple and non-parametric algorithm which separates the data points
into several classes and a new sample point is used to anticipate the class. The non-
parametric technique of classification does not make any assumptions on the distri-
bution of data. This methodology is helpful as the greater part of the real world data
which does not obey the generally assumed pattern (e.g. linear regression models).
While managing with this classifier, the following parameters are used: the number
of nearest neighbors (k), a distance function (d), decision rule and n labelled samples
of audio files Xn . A class label to the unknown sample is identified based on the
minimum distance from the training classes. In another word, K-NN calculates a
posteriori class probabilities P (wi|x) as

ki
P( wi |x) = .P(wi ) (2)
k

where ki is the number of vectors which belongs to class within the subset of k vectors
[18]. For our system, all the variants of KNN classifier are verified to maximize the
identification accuracy. The variations tested for the number of nearest neighbor
are 1-NN, 2-NN, and 3-NN. Two distance functions Euclidean and City-block are
investigated. The rules namely nearest and consensus are also tested. After a variety
of experiments, it is concluded that a combination of 3-NN neighbors, City-block
distance and Nearest rule give the maximum identification accuracy in [19].

4 Median Features for Improvement in System

Performance

The intra-speaker variability is observed from the observations of the extracted

feature values. The extent of variations is analyzed for the selected features in [15]
using Standard Deviation (σ). However, incorporating standard deviations in the
feature modifications is difficult. As the value of standard deviation is to be either
added or subtracted from the feature value (i.e. feature value ±σ). It may need a
complex decision algorithm to decide the addition or subtraction operation for every
feature value of every speaker sample. Besides, modifying the feature values with σ
may exceed the normalization range (i.e. all distinct feature values are normalized in
0 to 1). The values of selected features (Roll-off, Brightness, Roughness and, irregu-
larity) given in Table 1. Till now, we have used the absolute values of all these features
in a vector form for the conducted experiments above. Here, we have analyzed all
these feature values for the MEDIAN value (Table 2). The median of a finite list of
numbers can be found by arranging all the numbers (elements) in ascending order.
If there are an odd number of elements, the middle one is picked. If there is an even
number of observations, then there is no single middle value; the median is then
38 V. M. Sardar et al.

Table 1 Feature values with their median

Features speaker sample Roll-off Roughness Brightness Irregularity
1_1 0.121857 0.322853 0.004915 0.083185
1_2 0.458324 0.486716 0.069048 0.127183
1_3 0.523269 0.664555 0.072613 0.459385
1_4 0.46465 0.369261 0.023994 0.296143
1_5 0.72602 0.76634 0.048737 0.402269
1_6 0.339668 0.372256 0.027058 0.465064
1_7 0.424056 0.428354 0.023742 0.351125
1_8 0.452389 0.518894 0.042518 0.465195
1_9 0.206797 0.367267 0.041907 0.394578
1_10 0.476325 0.357569 0.108497 0.199835
MEDIAN 0.455356 0.400305 0.042212 0.372852

Table 2 Intra-speaker distance absolute values and the median values

City block distance in inter-speaker sample
Speaker sample 1_1 1_2 1_3 1_4 1_5
1_1 0 0.944 (0.544) 1.321 (0.467) 0.860 (0.220) 1.762 (0.871)
1_2 – – 0.779 0.434 0.859
1_3 – – – 0.670 0.603
1_4 – – – – 0.902
1_5 – – – – 0
Values in bracket shows the distance from MEDIAN
– (empty spaces) are due to repeated distance values between combinations

usually defined to be the MEAN of the two middle values. So MEDIAN may be
considered as the fully sheared mid-range.
MEDIAN can be calculated by the following formula which will be generalized
for odd and even number of values in the set:

a# 2x + a# 2x /0.5
MEDIAN = (3)
2
Comparing the feature values of query (test) speaker with the MEDIAN values
of the particular assigned class (at the time of training), distance from MEDIAN
is recommended. As the intra-speaker features are distributed in some range of
its variability, the median as a reference of class value will certainly improve the
identification accuracy. The following analysis will confirm our conclusion.
In the above illustration, five samples (1_1, 1_2, 1_3, 1_4, and 1_5) of the same
speaker (speaker_1) are listed for the intra-speaker City-block distance. Then in
another approach, the distances of the speakers from the MEDIAN of a class are
Use of Median Timbre Features for Speaker Identification … 39

Table 3 Comparative identification accuracy by using MFCC, timbre only and timbre (Median)
Speech mode % Accuracy
Training–Testing MFCC Timbre Timbre (Median)
Neutral–Neutral 91.5 95.0 95.6
Whisper–Whisper 78.8 86.5 88.1
Neutral–Whisper 66.0 73.0 76.4
CHAIN database of 35 speakers, MFCC only/timbre feature/timbre (median), and K-NN
Classifier

calculated (shown in bracket). It is seen that the distances of the inter-speaker samples
are less when calculated concerning MEDIAN as compared the absolute values. It
means that the possibility of identification of all samples of speaker_1 increases as the
intra-speaker distances are reduced when calculated from the reference of MEDIAN
values.

5 Results

This section presents the results with the following parameters. Total 35 speakers
with whispered samples from a CHAIN database are used for the experiments. The
samples of each speaker are selected with a choice of 70% samples for training
and 30% for testing. KNN classifier settings are as—Rule: nearest, Neighbor: 3,
and distance-Metric: City-Block distance. The selected Timbre features by Hybrid
Selection Algorithm are Roll-off, Brightness, Roughness and, irregularity.
Mel Frequency Cepstral Coefficient (MFCC) is widely used features for speaker
identification task. Hence, the performance using MFCC features is compared with
timber features and the Median values of it (Table 3).
The above results in Fig. 4 are compared with a baseline speaker identification
system for whispered speech using the CHAIN database [20]. Further, an enhance-
ment of 2.75% in the identification accuracy system in whisper train-whisper test
scenario is reported in [21] are reproduced in Table 4.
The proposed timbre features using Median value shows that the identification
accuracy is increased by 2.4% in whisper train-whisper test scenario which is the
focus of this undertaking.

6 Conclusion

A large number and variety of audio descriptors are available which are selected
according to the application. As the attributes of whisper are drastically change
compared to the neutral voice, the perceptually motivated, multidimensional timbre
40 V. M. Sardar et al.

100

0
Neutral-Neutral Whisper-Whisper Neutral-Whisper
% Accuracy MFCC % Accuracy Timbre % Accuracy Timbre(Median)

Fig. 4 Comparative results using MFCC only/timbre feature/timbre (median)

Table 4 Baseline results of

Speech mode % Accuracy
speaker identification
accuracy by timbre features Training Testing
Neutral Neutral 95.0
Whisper Whisper 86.5
Neutral Whisper 73.0

features are expected to be most suitable. However, it is customarily recommended

to use limited and only well-performing features for high speed and performance. By
Hybrid Selection Algorithm, a set of limited features namely MFCC, Roll-off, Bright-
ness, Roughness, and irregularity is used for the speaker identification with a whis-
pered voice. When compared with the traditional MFCC features, the identification
accuracy using selected timbrel features is found to be enhanced by 7.7%. Taking into
consideration, the intra-speaker variability, the performance can be further increased
if the MEDIAN value of training class is used to identify the new speaker query. It is
due to decreased intra-speaker distance according to the analysis in this undertaking.
Hence, the use of median values of timbrel features further enhances the results by
2.4%.

References

1. Hourri S, Kharroubi J (2020) A deep learning approach for speaker recognition. Int J Speech
Technol Springer 23:123–131
Use of Median Timbre Features for Speaker Identification … 41

2. Jahangir R et al (2020) Text-independent speaker identification through feature fusion and deep
neural network. IEEE Access 8:32187–32202
3. Singh A, Joshi AM (2020) Speaker identification through natural and whisper speech signal.
In: Janyani V, Singh G, Tiwari M, d’Alessandro A (eds) Optical and wireless technologies.
Lecture Notes in Electrical Engineering, vol 546. Springer, Singapore
4. Fan X, Godin KW, Hansen JHL (2011) Acoustic analysis of whispered speech for phoneme
and speaker dependency. In: Proceedings of the annual conference of the international speech
communication association, INTERSPEECH, pp 181–184
5. Bhattacharjee M, Prasanna SRM, Guha P (2018) Time-frequency audio features for speech-
music classification. Project: Broadcast Video Analytics
6. Davis SB, Mermelstein P (1980) Comparison of parametric representations for monosyllabic
word recognition in continuously spoken sentences. IEEE Trans Acoust Speech Signal Process
ASSP s28:357–366
7. Hermansky H, Malaya N (1998) Spectral basis functions from discriminant analysis. In:
International conference on spoken language processing
8. Toonen Dekkers RTJ, Aarts RM (1995) On a very low-cost speech-music discriminator.
Technical Report 124/95, Nat. Lab. Technical Note
9. Dobrowohl FA, Milne AJ, Dean RT (2019) Timbre preferences in the context of mixing music.
Appl Sci (2076–3417) 9(8):1695–1695
10. Peeters G (2004) A large set of audio features for sound description (similarity and
classification) in the CUIDADO project
11. Park TH (2004) Towards automatic musical instrument timbre recognition, PhD thesis, the
department of music, Princeton University
12. Albert-Ludwigs-Universität Freiburg (2007) A Matlab Toolbox for Music Information. In:
Proceedings of the 31st annual conference of the Gesellschaft für Klassifikation e.V., March
7–9, pp 261–268
13. Lartillot O (2011) MIR toolbox 1.3.3 (Matlab Central Version)—User’s Manual. Finnish centre
of excellence in interdisciplinary music research, University of Jyväskylä, Finland
14. Sardar VM, Shirbahadurkar SD (2019) Timbre features for speaker identification of whispering
speech: selection of optimal audio descriptors. Int J Comput Appl (Taylor Francis, U.K.). ISSN
1206–212X
15. Cummins F, Grimaldi M, Leonard T, Simko J (2006) The chains speech corpus: characterizing
individual speakers. School of Computer Science and Informatics University College, Dublin
16. Deshmukh S, Bhirud SG (2012) A hybrid selection method of audio descriptors for singer
identification in North Indian Classical Music. In: Fifth international conference on emerging
trends in engineering and technology, pp 224–227
17. Liu H, Jiang H, Zheng R (2016) Computational and mathematical methods in MedicinePB—
Hindawi Publishing Corporation
18. Shah JK, Smolenski BY, Yantorno RE, Iyer AN (2015) Sequential k-nearest neighbor pattern
recognition for usable speech classification. Signal Processing Conference, IEEE Xplorer
19. Sardar VM, Shirbahadurkar SD (2018) Speaker identification of whispering sound: effective-
ness of timbral audio descriptors. In: International conference on power, communications and
sustainable energy system, Chennai
20. Wang J-C, Chin Y-H, Hsieh W-C, Lin C-H, Chen Y-R, Siahaan E (2015) Speaker identification
with whispered speech for the access control system. IEEE Trans Autom Sci Eng 12:1191–1199
21. Sardar V, Shirbahadurkar S (2018) Speaker identification of whispering speech: an investigation
on selected timbrel features and KNN distance measures. Int J Speech Technol 21. [Link]
org/10.1007/s10772-018-9527-4
Intelligent System for Engine
Temperature Monitoring and Airbag
Deployment in Cars Using

Akshay A. Jadhav, Swagat M. Karve, Sujit A. Inamdar,

and Nandkumar A. Admile

Abstract The CAN bus has emerged as vital means of communication within the
automotive sector. Intrinsically varied applications are being implemented using the
Controller Area Network Protocol. In this paper, two major automotive applications
namely engine temperature monitoring and control and airbag deployment mecha-
nism are implemented with the assistance of ARM7 based microcontroller nodes.
Thus the essential CAN protocol has been implemented for real time embedded
automotive applications.

Keywords ARM7 · Basic CAN Protocol · CAN Bus · Nodes · Real-Time data

1 Introduction

Controller Area Network Bus is an everlasting version of the communication bus

systems utilized in automotive industry. CAN BUS may be a serial digital commu-
nication protocol invented by German BOSCH Corporation within the early 80s to
understand the info exchange between numerous controllers and measuring instru-
ments in modem automobile. It’s a multi-master bus; the communication medium
are often a double stranded wire, coax or optical fiber. Communication speed is
up to 1 MBPS. Bus communication interface integrates the CAN protocol physical
level and therefore the link layer function, and it can complete the framing of the
communication of knowledge processing, including the position filled, the block data
code, the circulation parity check, the priority distinguished and other works. CAN
communication protocol feature is to encode the info block. Length of the info is up
to eight bytes, which may meet the electrical bus control commands, working status
and test data requirement. Meanwhile, the 8 bytes won’t take the bus for an extended
time, so it ensures real-time communication [1]. In this paper, two major automotive
applications namely engine temperature monitoring and airbag deployment mech-
anism are implemented. The project consists of two nodes namely sensing node
and therefore the controlling node. The sensing node and therefore the controlling

A. A. Jadhav (B) · S. M. Karve · S. A. Inamdar · N. A. Admile

SVERIs College of Engineering Pandharpur, Solapur, Maharashtra, India

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 43

P. M. Pawar et al. (eds.), Techno-Societal 2020,
[Link]
44 A. A. Jadhav et al.

node consists of the ARM7 based microcontroller (with inbuilt CAN Controller) and
may Transceiver each. The important time data from sensors of the sensing node
is processed by the microcontroller LPC2129. The CAN Controller takes care of
the CAN broadcast message data. This data is then sent onto the CAN Bus. This
data is then received by the controlling node from the CAN bus. The controlling
node then further takes appropriate action. For controlling the temperature, a lover
is often used; while the airbag deployment mechanism is often achieved by turning
on a relay connected to the controlling node. This ensures the engine temperature
being monitored and controlled further and therefore the driver’s safety; that has
drawn special attention thanks to increasing number of accidents day by day caused
by high risky speeds. Thus the essential CAN protocol has been implemented for
real time embedded automotive applications.

2 System Design

The Fig. 1 shows system block diagram.

The sensing node and therefore the controlling node consist of LPC2129 micro-
controller each. The sensing node senses the sensor data. The sensed data is then sent
onto the bus. This data is then received by the controlling node. The controlling node
then takes the controlling actions accordingly. Here the fan is often wont to control
the temperature of the engine while the relay depicts the deployment of airbag. The
amount of nodes is often added to the bus to extend the amount of applications served.

Fig. 1 Block diagram

Intelligent System for Engine Temperature … 45

2.1 System Hardware Components

A CAN bus automotive electronic system consists of every controller through the
CAN bus interconnected together to exchange information (such as speed, engine
speed, engine ambient temperature etc.). This mutual connection helps the controllers
to use the info at an equivalent time. On the opposite hand it helps to extend the
throughput of the system, to develop updated features within the system and reduce
the system cost. Electronic automotive instrument cluster node can fully reflect the
car factor, product design and technical standards. Intelligent sensor nodes accurately
receive various sorts of signals and at an equivalent time can eliminate interference
signals [2].
1. CAN Bus Electrical Characteristics: CAN transmission medium formed by
the two, One is called high-level transmission line CANH and the other one is
called low-level transmission line CANL.
2. Hierarchical structure of CAN protocol based BUS: According to CAN BUS
network structure, Classified into the following five levels, as shown in Fig. 2.
3. CAN Message Transmission and Frame Type: CAN message use frame as
transmission unit. In the CAN 2.0B specification is provided in standard format
data frame, Given in two different 2.0B frame format, the difference is the length
of different identifiers: Frame with n-bit identifier called the standard frame, the
frame contains 29 identifiers is called extended frame. Message transmission
has the following four different types of frames: data frame, remote frame, error
frame, overload frame [3].
4. Can Bus Node Design: Figure 3 shows the CAN Bus node bus which consists
of two nodes namely sensing node and the controlling node. The sensing node
consists of following sensors: temperature sensor and shock sensor. The real
time data from the sensors is sent to the ARM7 based LPC2129 microcon-
troller and then to the CAN Bus via CAN Transceiver MCP2551. This data is
then received by the controlling node and appropriate action is taken. Thus the
ambient temperature of engine can be monitored. Also to control it further, a

Fig. 2 Hierarchical structure of CAN bus

46 A. A. Jadhav et al.

Fig. 3 Circuit design of CAN node

cooling fan can be started. The other application is of Airbag deployment mech-
anism. As soon as the data is received from the shock sensor, the controlling
node takes appropriate action. The action can be making a relay on as way to
deploy the airbag further.
The sensor used for sensing temperature is LM35 while the sensor used for shock
sensing is 801S. The shock sensor is micro vibration detection sensor and is useful
for the demo purpose. The microcontroller used for the node is LPC2129 (ARM7
based and Philips made microcontroller). It has several advantages including inbuilt
CAN Controller, supports high speed CAN operation, low power consumption etc.
The CAN Transceiver plays a major role here. It is used for communication between
the node and the CAN bus and then the other nodes. The data is sent onto the bus
via the high speed CAN Transceiver. The CAN Transceiver MCP2551 is compatible
with CAN 2.0B. It can provide differential transmit capability to the bus and differ-
ential receive capability to the CAN controller. The chip was originally designed for
automotive high-speed communications is fully compatible with ISO/DIS standard,
and the moment of interference in the anti-car environment, protecting the ability
of the bus. It is compatible with ISO-11898 standard at speeds up to 1 Mb/s, can
resist the strong interference the moment of interference such as the automotive
environment. General internal bus of circuit protection and limiting circuit, and has
a low current standby mode and slope control to reduce radio frequency interfer-
ence [4]. The sensing node may control the temperature by turning a cooling fan
on. Also the airbag deployment is depicted by turning on a relay. Thus two appli-
cations namely engine temperature monitoring and airbag deployment mechanism,
have been implemented.
Intelligent System for Engine Temperature … 47

3 System Software Components

The system software is written in embedded C and is compiled using ARM Keil Micro
vision v4 software. Flash Magic is used to upload the program to the Microcontroller
chip. The software flow consists of CAN transmission and Reception of messages.
The flow is depicted below with the aid of flowcharts as shown in Figs. 4, 5.

3.1 Transmission flow

See Fig. 4.

Fig. 4 Transmission flow

48 A. A. Jadhav et al.

Fig. 5 Reception flow

3.2 Reception Flow

The CAN Controller is initialized first. The data from one node is sent onto the bus
and then received by the other node. The flowcharts of transmission and reception
of data are shown in figs. 4 and 5 [5]. The received messages are then monitored on
the terminal of a personal computer (Fig. 6).

4 Results

As shown in Fig. 6 the real time data can monitored through UART on a personal
computer. The messages with corresponding ids can be seen on the terminal. A
specific message can be dedicated to certain sensor data. The real time data thus
can be monitored and corresponding controlled actions can be taken further. The
engine temperature can be controlled by turning on a cooling fan; while the airbag
deployment mechanism can be achieved by turning on a relay.
Intelligent System for Engine Temperature … 49

Fig. 6 Data monitored at the terminal of PC

5 Conclusion and Future Scope

The paper proposes implementation of two salient automotive applications namely

engine temperature monitoring and airbag deployment mechanism. The engine
temperature thus is often controlled further which is one among the factor that helps to
extend its efficiency. Also the airbag deployment mechanism ensures driver’s safety.
The scope of the paper is often extended further by increasing the amount of nodes
like respective applications. Thus more and more automotive applications are often
deployed in cars to form them as efficient as possible.
50 A. A. Jadhav et al.

References

1. Piao C-H, Chen L, Cao J (2010) A design for controller area network bus real-time monitoring
system. In: 2011 International conference on computer science and network technology, pp
1621–1624
2. Bo L, Tao J (2012) The design of monitoring system based on CAN bus. In: International
conference on measurement, information and control (MIC), pp 137–140
3. Ran L, Junfeng W, Haiying W, Gechen L (2010) Design method of CAN BUS network
communication structure for electric vehicle. In: IFOST Proceedings. 978-1-4244-1849-7/08/
4. Li R, Liu C, Luo F (2008) A design for automotive CAN bus monitoring system. In: IEEE
vehicle power and propulsion conference (VPPC), Sept 3–5, Harbin, China
5. Quanqi W, Jian W, Yanyan W Design of vehicle bus data acquisition and fault diagnosis system.
978-1-61284-459-6/11/
Analysis and Prediction
of Temporomandibular Joint Disorder
Using Machine Learning Classification
Algorithms

Roopa B. Kakkeri and D. S. Bormane

Abstract Temporomandibular joint disorder (TMD) includes specifically a series

of musculoskeletal disorders that may affect the masticating system. Roughly 30–
40 percent of adults today have oral problems, and the most common cause of oral
problems is TMJ. This disorder is very prevalent in the general population, but it
affects more women and young people. The focus of this research review was on the
methods for detecting TMJ disorder using machine learning algorithms. Propelled
with the rise in use of machine learning techniques in the research dimensions of
medical diagnosis, in this paper there is an attempt to explore different classification
for predicting the TMJ disorder. The proposed techniques are evaluated on real time
TMJ datasets. Dataset related to TMJ screening in subjects had 84 instances and 11
attributes. After applying different machine learning techniques, results suggest that
Naïve Bayes and Adaboost models work better with higher accuracy of 93% and
92%.

Keywords Temporomandibular joint disorder · Surface electromyography ·

Machine learning · Classification algorithm · SVM · Naïve Bayes · Adaboost

1 Introduction

Today, the issue of TMJ disorder has been growing systematically across all ages
of the human population. Early diagnosis of this musculoskeletal condition will
significantly aid in maintaining the mental and physical health of the subject [1–
3]. With the increase in the use of machine learning-driven models in the forecasts
of various human diseases, their early detection now seems to provide promising
results based on different health and physiological constraints. We were inspired
by this reason to raise awareness in the diagnosis and study of TMJ condition to

R. B. Kakkeri (B)
AISSMS Institute of Information Technology, Sinhgad Academy of Engineering, Kondwa, Pune,
India
D. S. Bormane
AISSMS College of Engineering, SPPU, Kennedy Road, Pune, India

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 51

P. M. Pawar et al. (eds.), Techno-Societal 2020,
[Link]
52 R. B. Kakkeri and D. S. Bormane

enhance improved technique of treatment. The round end of the lower jaw, known
as condyles, slips around the socket of the temporal bone joint when we open our
mouth [3]. TMJ disease diagnosis becomes a problem when there are many other
facial and muscle conditions whose few signs are very similar to those with signs
of TMJ disease, which is why it is considered a complex disorder. That makes this
job a challenging one. Early detection and treatment [4] are most imperative steps
to be taken to decrease the symptoms of TMJ disorder problem and to improve the
quality of life of suffering people. However, there is no procedure of medical test
for detection of TMJ disorder. In dentistry, One of the most common disorder in the
oral cavity that mostly affects women is temporomandibular joint disorder. Popular
Temporomandibular Joint Disorder (TMD) symptoms include,
• Problems with opening your mouth wide as you try.
• Jaws in the open-mouth or closed-mouth role that get “stuck” or “stall”.
• When you open or shut your mouth or chew, cracking, ringing, or grating sounds
in the jaw joint.
• Inflammation or swelling on the side of your face [1–3].
• Dental pain, headaches, neck aches, dizziness, earaches, hearing issues, upper
shoulder pain, and ear ringing (tinnitus) can also be present in patients [4].
The method of diagnosing this is conducted manually with the use of clinical tools
that with improper use becomes unsafe for the patient. The existing clinical methods
are also affluent for or common citizens. Arising technologies aided with applications
through the use of machine learning algorithms. Signal processing is practically
applied in medical diagnosis Digital acquisition techniques with processing of signals
becomes an important option especially in medical diagnosis [3, 5]. Moreover, the
clinical diagnostic procedure for TMJ disorder is costly. In the long run, lack of
awareness or neglecting of early signs and symptoms of TMJ disorder may lead to
even costlier clinical treatment. The general objective of the study is
To detect Temporomandibular Joint Disorder (TMJD) using surface electromyo-
graphy of masticatory muscles.
To introduce current prospects of diagnosis and treatment of Temporomandibular
Joint Disorder.
To mention advantages of using surface electromyography of masticatory
muscles for detection of Temporomandibular Joint Disorder using machine leaning
classification algorithms.
The current paper aims to predict TMJ disorder. This paper is arranged in the
following sections: Sect. 1 gives introduction, Sect. 2 discuss materials and method-
ology. In Sect. 3, experimental results are described. Finally, the conclusions and
future scope of improvement are presented in Sect. 4.
Analysis and Prediction of Temporomandibular … 53

Fig. 1 Block diagram of the proposed work

2 Literature and Methods

This research study is categorized as a form of qualitative research as it determines

the suitability of masticatory muscle surface electromyography for the diagnosis of
TMJ disorder through machine learning techniques [6, 7].

2.1 Data Acquisition and Subjects

The first part in the research process conducted is the data assembly [3, 8]. Literature
survey is carried out by reading of articles and related journals from the internet
and clinical study was conducted to get details about TMJ disorder, its signs and
symptoms with radiographic evaluation through Orthopantomogram, Conversation
with Neurologist and dental surgeons were directed to alleviate the early findings
in selecting the subjects. A consent form is taken from all the subjects to willfully
participate in the study. Following Fig. 1 shows block diagram of proposed work.

2.2 Materials and Methods

The present study was conducted with total 84 subjects. Forty one subjects were
included in the control group and the rest were in experimental group. While
selecting the subjects, those with the history of dysfunction syndrome and orthodontic
management, lip incompetence, soreness in any muscle of mastication and previous
rebuilding in tooth were excluded.
A complete dental history is made for each subject and intraoral photographs
(OPG) are taken to identify the TMJ parameters [6]. Frontal belly of masseter and
frontward belly of temporalis of both sides (which are for convenience mentioned
here as masseter and temporalis, respectively) were studied.
For all the subjects, bipolar surface electrodes were used. The EMG recording
of these muscles was done in both groups at most calm position of jaw i.e., at
54 R. B. Kakkeri and D. S. Bormane

rest position and during maximum clenching at maximum intercuspal position [9–
11]. In both control group and experimental group, during the activity of maximum
clenching and during rest, the recording was done. Before recording, volunteers were
seated comfortably on a normal chair. Position of the head was kept erect and no
undertaking was allowed as jaw muscles respond to change in crown position. The
subject was settled and not able to see the computer screen. Facial areas were prepared
using a Medi-swab. EMG signals were composed using bipolar throwaway surface
electrodes of 6 mm diameter (Blue Sensor A-50-VS, Medico test, lstykke, Denmark)
placed on the volunteer’s temporalis and masseter muscle areas on both sides of
the face, thus giving four data collection channels (temporalis-right, masseter-right,
masseter-left, temporalis-left). Electrode cream (Grass EC2, USA) was functional
to give a flat surface on the electrode face and then the protective support detached.
Two surface electrodes were applied to the muscle, next to each other, the centers
16 mm apart laterally along each muscle, which was located by palpation. Reference
electrode was involved to an ear lobe.
The electrodes were associated via a Polygraph amplifier (Model 7E, Grass Instru-
ment Co., Massachusetts, USA) to an ADC interface comprising 12-bit, 333 kHz
and sampled using Spike2 data collection software (Cambridge Electronic Design
Ltd., Cambridge, UK), sampling at a rate of 1 kHz. The Polygraph 7P3 preamplifier
modules used filters (26 dB) of 3 Hz and 20 kHz, thence to 7DA driver amplifier
modules with output frequency response to 40 kHz (26 dB). The 7DA module had a
notch 50 Hz filter to suppress mains electrical interference. Recording was continued
in each case for up to 1 min.

2.3 Data Pre-processing

Pre-processing of data is a method in which the raw data is changed into a substantial
and practical format. Real-world evidence is typically partial and unreliable since
it includes a lot of mistakes and negligible values [5, 9]. A pre-processed upright
details still vintages to a decent performance. Various types of data pre-processing
[8] are used to adjust inaccurate and conflicting data, such as managing missing
information, minimizing data, etc.

3 Methodology

The next move was to develop the algorithm for the diagnosis of TMJ disorder
with obvious results on how to diagnose TMJ disorder by feature selection [12]. A
series of Computer Vision outlines is referred to and the probability of using them
to create the algorithm was checked. In the development of the algorithm [6, 7],
python libraries were attempted. This system libraries and documents were analyzed
Analysis and Prediction of Temporomandibular … 55

Fig. 2 Final training and testing

on which of these may provide the simplest but most efficient method for obtaining
the outcomes.

3.1 Training and Testing Model

The whole dataset has been split into two parts, i.e. one part is the training dataset
other part is the dataset testing with a ratio of 70:30 respectively. Training data has
once again been broken into two parts for cross-validation purposes. One component
is the preparation dataset, and another component is the 70:30 ratio evaluation dataset.
Fig. 2 shows training & testing.

3.2 Support Vector Machine (SVM)

SVM is an algorithm for supervised machine learning which can be used for problems
of classification or regression [13, 14]. Fig. 3 shows SVM To transform the results,
it utilizes a technique called the kernel trick, and then finds an optimal boundary
between the possible outputs based on these transformations. Simply put, it performs
some highly complex transformations of information, then works out how to isolate
the information depending on the labels or outputs you have identified. We used a
linear model of the kernel [12].
56 R. B. Kakkeri and D. S. Bormane

Fig. 3 Support vector machine

3.3 Logistic Regression

It is used to test discrete values depending on a given collection of independent

variable(s) (Binary values such as 0/1, yes/no, true/false). Logistic regression is
similar to linear regression, but the curve is built using the normal logarithm of the
target variable’s “odds” rather than the likelihood [7]. Fig. 4 shows logistic regression.
K Nearest Neighbor (KNN) KNN is just a broad view of the nearest neighbor’s
algorithm. Include k-nearest neighbors in their place of considering the closest
neighbor from a dataset containing n data points, k communicates how many neigh-
boring neighbors will have an influence on the classification process [13, 14]. The
distance from Euclidean, Manhattan, Minkowski and Hamming may be such distance
functions. The first three functions are used for continuous functions, and the fourth
one (Hamming) is for categorical variables. Based on the square root, the mainstream
labelled data point to which the unlabeled point must be classified is determined by
the number of samples, depending on the K.

Fig. 4 Logistic regression

Analysis and Prediction of Temporomandibular … 57

Fig. 5 Adaboost classifier

3.4 Adaboost

A machine-learning algorithm expressed by Yoav Freund and Robert Schapire is the

Ada Boost (short for “Adaptive boosting”) widget it is shown in Fig. 5. To boost their
presentation, it can be used with other learning algorithms [7]. This is achieved by
the tuning of poor learners. Boosting refers to an all-purpose and demonstrably true
means of creating a very precise classifier by mixing rough and mildly inaccurate
thumb laws.

3.5 Naïve Bayes Algorithm

It is a classification technique with a presumption of freedom between predictors

based on Bayes’ theorem. A Naïve Bayes classifier believes, in basic terms, that
the inclusion of a certain feature in a class is equivalent to the existence of any
other feature. The Bayes theorem provides a way for P(c) posterior likelihood to be
determined from P(c), P(x) and P(x) [7].

4 Result and Discussion

The result is measured in terms of specificity, sensitivity, and accuracy by using the
confusion matrix and classification report. The result depends on how accurate the
model is trained. The most relevant and important features selected were Masseter
and Temporal muscles for left and right during rest and Maximal voluntary clenching
(MVC).

4.1 Performance Evaluation Metrics

Gauging performance is key to check how fine a classification model work to attain
an objective. Performance valuation metrics are used to evaluate the efficacy and
58 R. B. Kakkeri and D. S. Bormane

performance of the classification model on the test dataset. It is important to choose

the exact metrics to evaluate the model performance such as confusion matrix, accu-
racy, specificity, sensitivity, etc. Following formulas are used to find the performance
metrics [13]:
Accuracy—Accuracy is the most innate performance quantity and it is simply a
ratio of acceptably predicted observation to the total observations.
Precision—Precision is the ratio of correctly predicted positive observations to the
total predicted positive observations. High precision relates to the low false positive
rate.
Recall (Sensitivity)—Recall is the ratio of correctly predicted positive observa-
tions to the all observations in actual class—yes.
F1 score—F1 Score is the biased average of Precision and Recall. Therefore,
this score takes both false positives and false negatives into account. Naturally it is
not as easy to understand as accuracy, but F1 is usually more useful than accuracy,
especially if you have an uneven class distribution.
In the biomedical field, classification has acquired a lot of importance in machine
learning. Classification techniques help to learn a model from a set of training data
and to classify one of the classes with test data. The analysis of the few existing clas-
sification algorithms [12, 15] is related to this study and their comparative efficiency
parameters will help to study the new algorithms.
The main kernel used in the analysis of EMG signal features to achieve better clas-
sification accuracy is the selected features. This article’s key purpose is to determine
the consistency of the classification [13, 14]. This article discusses a few classification
strategies for Supervised Machine Learning (ML) [12], compares and determines the
most successful classification algorithm based on the data collection, the number of
instances and variables (features). Five separate machine learning algorithms were
considered: k-Nearest Neighbor (KNN), Decision tree, Naïve Bayes, Logistic regres-
sion and Support Vector Machine. The TMJ disorder data set was used for the clas-
sification to apply the equations with 84 instances with ten characteristics as an
independent variable and one as a dependent variable for the study.
Cross-validation is a method for testing machine learning models by training
many machine learning models on the available input data sub-sets and testing them
on the sub-set of complementary data. Using cross-validation to avoid over fitting,
i.e. failing to generalize a pattern.

4.2 K-Fold Cross-Validation

Cross-validation shown in Fig. 6, is a re-sampling practice used to measure machine

learning models on a partial data set. To achieve cross-validation, we may use the
method of k-fold cross-validation. You have fragmented the input data into k data
subsets (also known as folds) during k-fold cross-validation.
Analysis and Prediction of Temporomandibular … 59

Fig. 6 k-fold cross-validation model

On all but one (k − 1) of the subclasses, train ML model, and then test the model
on the sub-set that was not used for training [14]. This procedure is repeated k times,
each time with a different sub-set intended for evaluation (and omitted from training).

4.3 Results

The final result is shown in following Fig. 7.

Fig. 7 Overall results of models with tenfold cross validation

60 R. B. Kakkeri and D. S. Bormane

5 Conclusion

Temporomandibular disorder is a condition which can put the individual in severe

pain, agony as well as unable to chew food properly. Presently, a proper diagnosis
and a structured treatment plan is not available as the cost of advanced equipment,
technique sensitivity of handling the equipment as well as understanding of multidis-
ciplinary approach is inadequate. It was clear after the literature survey that several
machine learning algorithms can be applied to the extracted and chosen features. Out
of the five supervised training of classification algorithms, the results indicate that
SVM was not introduced to be the most reliable and effective algorithm. Relative
analysis of the classifiers shows that Decision tree, Logistic regression and Random
Forest classification algorithms stood to be the reliable with cross validity of ten
folds.

References

1. Uddin F, Baten RBA, Rita SN, Sadat SA, Chowdhury NM (2017) Management of temporo-
mandibular joint dysfunction syndrome: an overview. J Bangladesh Coll Physicians Surg
35(3):133–141
2. Klasser GD, Okeson JP (2006) Electromyography in the diagnosis and treatment of temporo-
mandibular disorders. Oral Health 137:763–771
3. Mapelli BC, Zanandréa Machado LD, Giglio CS, De Felício CM (2016) Reorganization of
muscle activity in patients with chronic temporomandibular disorders. Arch Oral Biol 72:164–
171
4. Bianchi J et al (2020) Osteoarthritis of the temporomandibular joint can be diagnosed earlier
using biomarkers and machine learning. Sci Rep 10(1):1–14
5. Spiewak C (2018) A comprehensive study on EMG feature extraction and classifiers. Open
Access J Biomed Eng Biosci 1(1):1–10
6. Khezri M, Jahed M (2007) Real-time intelligent pattern recognition algorithm for surface EMG
signals. Biomed Eng Online 6:1–12
7. Sweeney EM et al (2014) A comparison of supervised machine learning algorithms and feature
vectors for MS lesion segmentation using multimodal structural MRI. PLoS ONE 9:4
8. Chowdhury RH, Reaz MBI, Bin Mohd Ali MA, Bakar AAA, Chellappan K, Chang TG
(2013) Surface electromyography signal processing and classification techniques. Sensors
(Switzerland) 13(9):12431–12466
9. Latif R, Sanei S, Shave C, Carter E (2008) Classification of temporomandibular disorder
from electromyography signals via directed transfer function. In: 30th annual international
conference of the ieee engineering in medicine and biology society, vol 2, no 3, pp 2904–2907
10. Pinho JC, Caldas FM, Mora MJ, Santana-PenÍn U (2008) Electromyographic activity in patients
with temporomandibular disorders. J Oral Rehabil 27(11):985–990
11. Suvinen TI, Kemppainen P (2007) Review of clinical EMG studies related to muscle and
occlusal factors in healthy and TMD subjects
12. Khammas BM, Monemi A, Bassi JS, Ismail I, Nor SM, Marsono MN (2015) Feature selection
and machine learning classification for malware detection. J Teknol 77(1):243–250
13. Khan MMR, Arif RB, Siddique AB, Oishe MR (2019) Study and observation of the variation
of accuracies of KNN, SVM, LMNN, ENN algorithms on eleven different datasets from UCI
machine learning repository. In: 4th international conference on electrical engineering infor-
mation and communication technology iCEEiCT 2018, pp 124–129
Analysis and Prediction of Temporomandibular … 61

14. Kucuk H, Tepe C, Eminoglu I (2013) Classification of EMG signals by k-nearest neighbor
algorithm and support vector machine methods, pp 1–4
15. Al-Faiz MZ, Ali AA, Miry AH (2010) A k-nearest neighbor based algorithm for human arm
movements recognition using EMG signals, EPC-IQ01 2010. In: 1st conference energy, power
control, vol 6, no 2, pp 159–167
Machine Learning Approach
in Cooperative Spectrum Sensing
for Cognitive Radio Network: Survey

Vaishali S. Kulkarni, Tanuja S. Dhope(Shendkar), Swagat Karve,

Pranav Chippalkatti, and Akshay Jadhav

Abstract In cognitive radio network some of the important functionalities is spec-

trum sensing. It plays a very vital role for unlicensed system to operate efficiently and
to provide the required improvement in spectrum efficiency. If the spectrum, which is
sensed is in idle state allow the unauthorized users (secondary users) to use the spec-
trum. Machine learning algorithms are used for spectrum sensing in cognitive radio
networks. They are weighted K-nearest neighbor, Support Vector Machine (SVM)
which comes under supervised learning and Gaussian Mixture Model (GMM), K-
means clustering which comes under unsupervised learning-based classification tech-
niques. In this paper rigorous survey is done by using machine learning algorithms
to review various methodologies used in spectrum sensing like K-nearest -neighbor,
GMM, K-means clustering and SVM.

Keywords Machine Learning · Support Vector Machine (SVM) · Primary user ·

K-means clustering · Secondary user · Spectrum sensing · Cognitive radio

1 Introduction

Due to rapid development in wireless network, a technology called Cognitive Radio

(CR) is developed to overcome the spectrum scarcity. Using this technology both
primary users (licensed) and secondary (unlicensed) users can utilize the spectrum.
In cognitive radio networks spectrum utilization increases with the spectrum allo-
cated dynamically. The unlicensed users can transmit in the vacant spectrum already
assigned to licensed users with minimum level of interference. To find the vacant
spectrum it senses and selects the spectrum by which it can meet the QoS. The licensed
users get back the spectrum from the unlicensed users whenever the licensed users
return. In cognitive radio networks spectrum sensing technique is one of the major

V. S. Kulkarni (B) · T. S. Dhope(Shendkar) · P. Chippalkatti

Department of Electronics & Telecommunication, JSPM’s Rajarshi Shahu College of
Engineering, Pune, India
S. Karve · A. Jadhav
SVERI’s College of Engineering, Pandharpur, India

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 63

P. M. Pawar et al. (eds.), Techno-Societal 2020,
[Link]
64 V. S. Kulkarni et al.

Fig. 1 Classification of
spectrum sensing

techniques [1] in which the harmful interference is prevented with licensed users and
the available spectrum is identified to improve the spectrum’s utilization.
In cognitive radio network one of the important techniques is Spectrum sensing
and is also the primary task for establishing a cognitive radio system [1].
Spectrum Sensing is of three types. Interference-based sensing, non-cooperative
sensing and Cooperative sensing.
In cognitive radio networks, machine learning methods of Cooperative spec-
trum sensing is the major leading technique. The capability of decision making can
be enhanced by combining machine learning with cognitive radio networks from
the previous experiences [2]. Compared to traditional methods unstable cognitive
spectrum resources can be used effectively.
Figure 1 shows the classification of Spectrum sensing. Non-cooperative sensing,
interference-based sensing and Cooperative sensing Spectrum sensing are the
classification of spectrum sensing.
The performance of detection of spectrum holes is enhanced by cooperative
sensing in which secondary user senses the spectrum with the collaboration with
primary user. Also known as receiver detection technique.
In non-cooperative spectrum sensing, individual CR scans for primary signal and
based on its detection CR decides the presence or absence of primary user.
Cognitive radio measures the interference environment and their transmissions
are adjusted such that the interference to PU is not above the regularity limit in
interference-based sensing [3].
Cyclo-stationary feature detection, energy detection and match filter detection are
the classification of non-Cooperative sensing.
One of the intelligent cooperative sensing methods is machine learning which has
the special feature of self-adaptation to environment, high sensing.
Table 1 shows the comparative advantages and disadvantages of spectrum sensing
methods [4]. The author has created this table to familiarize spectrum sensing and for
better understanding the state-of-art techniques to the researchers. The performance
of any spectrum sensing algorithm is determined by the probability of detection,
probability of false alarm, the energy detection for cognitive radio network follows
the following formula [5, 13].
Machine Learning Approach in Cooperative Spectrum … 65

Table 1 Spectrum sensing methods

Sr. No Techniques of sensing Pros Cons
1 Energy detection 1. There is no requirement 1. Noise uncertainty is
of pre knowledge of sensitive
primary signal 2. False alarm rate is high
characteristics
2. Implementation is easy
2 Cyclostationary feature 1. Signal and noise can be 1. When the sample size is
detection distinguished large, energy
2. Probability of false alarm consumption is high
decreases at low SNR
3 Matched filter-based 1. Detection is better at low 1. Knowledge of primary
detection SNR region user signal is required in
2. Sensing is optimal advance, hence
impractical
4 Detection based on It is not essential to have the Computation is very
Covariance knowledge of primary complex
signal characteristics in
advance
5 Spectrum sensing based on 1. Complex model can be 1. Techniques are complex
Machine learning used in easy manner 2. Detection rate is affected
2. Delay of detection is by feature selection
minimized

1. Let time of sensing be τs, at the sensing node, when the PU is absent, the signal
that is received is given by

s(u) = σ n(u) (1)

where σn(u) is the noise power at duration d.

2. When the PU is present, signal is given as

s(u) = g(u). r(u) + σ n(u) (2)

where g(u) is gain of the channel, r(u) is primary signal which is received.
3. Assuming some of the parameters, detection of energy is given by

τ s
1
ED = |s(u)|2du (3)
P0
0

where P0 is the spectral density of power τs is the time of sensing.

4. Probability of detection and probability of false alarm using energy detection can
be given by considering N cooperative sensing nodes. Probability of detection
is given by
66 V. S. Kulkarni et al.

αj − (2τsCB + βjτsCB)
pj (det)(αj , βj , τs , CB ) = Q
4τsCB + 4βjτsCB
∞ (4)
1 −t2
where Q(u) = √ exp
2π u 2

where pj (det) is the probability of detection and is given by P (β ≥ αj | H1) at j

belonging to 1 to N, αj = threshold of detection, H1-presence of PU.
Probability of false alarm is given by

αj − 2τsCB
pj (false)(αj , τs ,CB ) = Q √ (5)
4τsCB

where pj (false) = P (β ≥ αj | H0) at j belonging to 1 to N, h2-absence of PU.

The organization of the paper is as follows. Literature survey of various method-
ologies of cooperative spectrum sensing are described in Sect. 2. Conclusion in Sect. 3
is preceded by a table of comparative methodologies.

2 Literature Survey

Zan Li et al. [6] proposed CSS model of machine learning in which user grouping
concept is shown. CSS framework based on optimization model which had four
modules. Group scheduling module, SVM classification module, user grouping
module and SVM training module are the four modules of grouping concept. Without
the reduction in accuracy of sensing and the functions like grouping of optimized
users, redundant and abnormal are achieved by using three algorithms of grouping.
In the algorithm of first grouping there are groups of two which are normal and
abnormal. Then in second grouping algorithm redundant and non-redundant users
are distinguished and finally optimized model is established. As per the requirement
of cooperative sensing optimization of specific number of groups are done where
the users are divided, selecting each time. Finally, the user group are non-redundant
and secured that involve in the recognition of pattern. Optimization problem can be
solved by Binary particle swarm optimization. By using this grouping algorithms,
the CR networks can be avoided the harm of abnormal and the redundant users.
Assigning silent state for long term or temporarily for many users reduces the unnec-
essary energy consumption. The efficiency of sensing can be improved by increasing
the grouping. This is done by increasing the speed of classification in model of SVM
in machine learning. Centralized CSS model network consisting of fusion center is
used in the simulation with 10 or 20 CRs. Here the simulation result shows that there
is 17.98% rate of improvement for the speed of average classification. Along with the
entire network security they also showed the operational efficiency of SVM model
increased using group algorithm 1. Few non-highly redundant users and highly redun-
dant users are used in grouping algorithm 2 with network of higher signal quality.
Machine Learning Approach in Cooperative Spectrum … 67

SVM achievement 100% accuracy. In algorithm 3, the users of CR are divided into
cooperation groups of two or more which are optimized. Cooperative sensing can be
performed independently by each group. The classification speed increases with the
number of groups being more.
Yingqi Lu et al. [7] proposed machine learning based classification which uses the
probability vector of dimension being low. This method has the time for classifica-
tion shorter and duration for training smaller. So, energy vector with N-dimensions
is not used here. Keeping the same spectrum performance and by converting a high
dimensional feature vector, a feature vector of constant two-dimensional is achieved.
With M secondary users and primary user selecting single SVM and K-means algo-
rithms of machine learning are used. Small scale CRN uses primary user as single
and M = 2SUs and large-scale CRN using single primary user (PU) and M = 9SUs
are considered.
In Small scale CRN, the techniques of machine learning are trained and then
new feature vectors are tested, which gives the result of (PFA) probability of false
alarm and (PD) the probability of detection. They showed that The accuracy of
detection compared to algorithms of AND rule and OR rule which are hard fusion,
classification based on machine learning is better. In terms of energy vector, there is
better performance by SVM based classification using probability vector. With the
highest transmitting power of PU, performance is better.
In large scale CRN, through plot of Receiver operating characteristics curves
accuracy is shown better in both SVM-ploy and SVM-linear when the probability
vector is combined. Here energy vector is replaced by probability vector and proved
the accuracy to be better.
Through simulation they proved that
1. With the equal training samples, the duration of training for K-means clustering
is longest.
2. SVM-ploy takes more training time than SVM-linear as more time is spent on
feature vector mapping by functions of polynomial kernel.
3. SVM-linear is more efficient than SVM-ploy.
4. With the probability vector, the performance is best of SVM linear compared
to various decision-making scheme such as K-means, SVM-ploy. These deci-
sions making schemes have duration of training and classification delay low,
probability of detection high.
Xiaolin Ma et al. [8] proposed Extreme Learning machine (ELM) for numerous
primary users to implement the algorithm of sensing cooperative spectrum with the
network of cognitive radio. ELM which is an algorithm of machine learning that
has three step free tuning process is used to make very accurate the channel sensing
and identification which involves. In this paper initially the identification of multiple
primary users is established by spectrum sensing model. A classification scheme
of channel pattern can be obtained by ELM with the help of energy detection and
channel model. Channel model as well as sensing model, energy model, fusion center
is also implemented.
68 V. S. Kulkarni et al.

In channel model, the primary user which uses the channel detects the signal
consists of an authorized signal and noise which is called Rayleigh channel. Using
multiple primary users, the signals can be detected by secondary user in the network
if the number of primary users is more than one. In energy model, usage of channel
by primary users can be sensed by energy. The fusion center aims at determination
of usage of channel. Based on the transmission of energy vectors from secondary
user channel usage is determined.
Simulation is carried out with multiple primary users and the results are compared
with SVM, it was shown that training time of ELM is shorter. The detection
probability being higher the probability of overall detection will also be better.
SVM and ELM algorithms were proposed to train the samples with 1000 and 2000
as the sample size. It is proved that ELM needs only 92 ms to reach probability of
detection compared to SVM which requires 8 times more time than ELM. Accuracy
of SVM is also less than ELM.
Olusegun Peter Awe et al. [9] worked in cognitive radio network in order to
solve the problem of sensing the spectrum on the condition of multiple primary
users and proposed algorithm based on based multi-class support vector machine
(SVM) which is ECOC (error correcting output codes). Spectrum holes is detected
by Joint spatial–temporal detection which is implemented using this algorithm. To
study the attributes of each state, through one-versus-all (OVA) and one-versus-
one (OVO) Multi-class support vector algorithms are used [10–12]. Error correcting
output codes (ECOC) scheme is used to solve the optimization problem. The multi
class problem can be dealt with ECOC which is a framework that creates multiple
binary classification task by breaking it [9] and [10]. Initially the condition is consid-
ered as detection of multiple class signal in which more than one sub-class are
formed from each class. Then multi-class SVM algorithms checks for performance
of the energy-based features and ECOC (error correcting output codes). Detection
of performance is mainly based on accuracy of classification and curves of receiver
operating characteristics. They proved that unused spatial spectrum can be detected
by using ECOC.
Karaputugala Madushan Thilina et al. [1] proposed technique of machine learning
for cognitive radio in which both supervised and un-supervised methods are used.
weighted K-nearest-neighbor and support vector machine learning is used in super-
vised and gaussian mixture model, K-means clustering is used in unsupervised where
cooperative spectrum sensing is implemented. Initially by using the energy levels of
vector the availability is checked for channel which is considered as feature vector.
The decision is taken by feeding this to the classifier. A classifier categorizes into
two categories the availability of channel with available class and the unavailable
class for each feature vector. The training phase has to be covered by classifier before
the classification being online. Partitioning of training feature vectors into K clus-
ters is done by implementing the algorithm of K-means in which test energy vector
class and primary user state represented by each cluster is determined by classifier.
Mapping of each cluster determines the available and unavailable class of channel.
Training feature vectors are obtained from Gaussian mixture distributions in GMM
which corresponds to a cluster. The SVM and the KNN is also proposed due to the
Machine Learning Approach in Cooperative Spectrum … 69

capability of higher prediction. In the SVM, the margin between separating hyper-
planes and feature vectors are maximized to obtain support vectors. Performance is
determined by evaluating all the classification techniques such as the ROC curve, the
classification delay and training time.
The result shows that
1. Capability of classification type K-nearest neighbor for 1000 energy vectors is
shown by the time taken for uploading the training energy vector to the classifier
as 50 μsas compared to GMM having high training duration of 12,796 s for
1000 samples and SVM having highest training duration of 1. 65,817 s for 1000
samples
2. KNN classifier has high classification delay even though training time is lowest
compared to Fisher linear discriminant which shows lowest classification delay.
Also, no change is observed in classification delay in GMM, K-means clustering
and Fisher linear discriminant classifiers with the training energy vector with
different batch.
3. K-means clustering is better approach compared to another classifier because of
capability PU detection being higher and classification delay and lower training.
4. Compared to other algorithms performance of detection is found highest in
SVM classifier by using Kernel functions like polynomial kernel and linear
kernel which is used to map feature space with higher dimensional space.
They also concluded that the classifiers can be trained by obtaining the energy
vectors one to one by improving the CSS approaches. Comparative Table Method-
ologies.
Table 2 shows the comparative methodologies used by various authors and their
findings which led to a better result.

3 Conclusion

In wireless communication systems, spectrum plays avital role as a valuable resource.

The challenging concept in radio network is Cognitive radio. The cognitive radio
which provides the channel wireless, can be dynamically configured and also be
programmed in its vicinity to avoid interference and congestion.
In this survey different methodologies using ML are presented with their results.
various spectrum sensing methods with their pros and cons are presented in this
paper. Various methods have been studied for Dynamic spectrum sensing like K-
nearest-neighbor, GMM, K-means clustering and SVM. From this literature survey
it is observed by comparing to various methodologies, support vector machine gives
better result. Further work will be carried out with support vector machine algorithm.
70 V. S. Kulkarni et al.

Table 2 Comparative table methodologies

Sr. No Related work Topic Methodology Results
1 Zan Li et al. Improved SVM based CSS User grouping
cooperative spectrum model is used for algorithm for
sensing model based group scheduling multi-band
on ML for CRN and User grouping cooperative sensing
modules
2 Xiaolin Ma et al. CSS using Extreme To obtain channel compared to SVM
learning machine for pattern classification detection probability
cognitive radio ELM (Extreme is higher in ELM
networks with learning machine) is and also training
multiple primary used for energy time is shorter in
users detection model and ELM
Wireless channel
model
3 Yingqi Lu et al. ML techniques with System model and The performance of
probability vector for probability vector for K-means clustering
CSS in CRN SVM and K-means or SVM algorithms
clustering algorithms can be improved by
probability vector
4 Olusegun Peter Cooperative Multi class support Successfully
Awe Spectrum Sensing in vector machine detected unused
Cognitive Radio spatial spectrum
Networks using opportunities
Multi-clss Support
Vector Machine
Algorithms”
5 Karaputugala Machine Learning Unsupervised Unsupervised
Madushan thilina Techniques for classifiers-GMM, classifiers-K-means
et al. Cooperative K-means clustering and Supervised
Spectrum Sensing in Supervised classifiers-SVM are
Cognitive Radio classifiers-K-nearest accurate approaches
Networks neighbor and SVM for CSS

References

1. Thilina KM, Choi KW, Saquib N, Hossain E (2013) Machine learning techniques for
cooperative spectrum sensing in cognitive radio networks. IEEE J Sel Areas Commun
31(11):2209-2221
2. Kim KJ, Kwak KS, Choi BD (2013) Performance analysis of opportunistic spectrum access
protocol for multi-channel cognitive radio networks. J Commun Netw 15(1):77–86
3. Arjoune Y, Kaabouch N (2019) Review-a comprehensive survey on spectrum sensing in cogni-
tive radio networks. In: Recent advances, new challenges, and future research directions,
sensors
4. Bkassiny M, Li Y, Jayaweera SK (2013) A survey on machine-learning techniques in cognitive
radios. IEEE Commun Surv Tutor 15(3):1136–1159 (Third Quarter)
5. Bae S, So J, Kim H (2017) Article on optimal cooperative sensing with energy detection in
cognitive radio
6. Li1 Z, Wu W, Liu X, Qi P (2018) Improved cooperative spectrum sensing model based on
machine learning for cognitive radio networks. J Inst Eng Technol 2485–2492. ISSN 1751–8628
Machine Learning Approach in Cooperative Spectrum … 71

7. Lu Y, Zhu P, Wang D, Fattouche M (2016) Machine learning techniques with probability vector
for cooperative spectrum sensing in cognitive radio networks. In: IEEE wireless conference
and networking conference track 1: PHY and fundamentals, pp 1–6 (WCNC)
8. Ma X, Ning S, Liu X, Kuang H, Hong Y (2018) Cooperative spectrum sensing using extreme
learning machine for cognitive radio networks with multiple primary users. In: 2018 IEEE 3rd
advanced information technology, electronics and automation control conference, pp 536–540
9. Awe OP, Lambotharan S (2015) Cooperative spectrum sensing in cognitive radio networks
using multi-class support vector machine algorithms. In: IEEE 2015, pp 1-7
10. Arthy A, Periyasamy P (2015) Review on spectrum sensing techniques in cognitive radio
network. In: Proceedings of UGC ponsored national conference on advanced networking and
applications, pp 80–83
11. Hsu C, Lin C (2002) A comparison of methods for multiclass support vector machines. Neural
Netw IEEE Trans 13(2):415–425
12. Allwein E, Schapire R, Singer Y (2001) Reducing multiclass to binary: a unifying approach
for margin classifiers. J Mach Learn 1:113–141
13. Yucek T, Arslan H (2009) IEEE communications surveys & tutorials
Study on Detection of Leukemia
in Human Blood Pattern Based
on Microscopic Photographs Using
Image Processing Techniques

Swagat M. Karve, Pravin Kishrsagar, Akshay A. Jadhav,

and M. Aravind Kumar

Abstract For the time being, blood disorders are defined by examination of micro-
scopic images of blood cells. It may lead to identity the blood disorders class of
blood-associated diseases. This paper discusses an initial look at the creation of
Microscopic Blood Pattern Images to detect leukemia forms. Pixel reading can be
very critical since it is possible to detect and classify pixel diseases at an earlier stage.
From there, more steps, such as disease control, surveillance and prevention, can be
completed. Photos are used because they can be fairly priced and do not require
costly check-out and laboratory equipment. White blood cell deficiency, leukemia,
will be identified by the gadget. The computer can use microscopic image capabilities
to look at texture, geometry, colour and statistical evaluation changes. Changes in
these characteristics may be used as feedback to a classifier. A literature review has
been undertaken and it is suggested to categories types of leukemia by reinforcement
learning to know. A brief debate about the problems involved has also been prepared
by researchers.

Keywords White blood cell · Microscopic images · Leukemia · Reinforcement

learning · ALL

1 Introduction

Over the past decade, scientific imaging has proved to be one of the most vital
visualisation and analysis techniques in biology and medicine. A remarkable creation
of new, powerful devices for the identification, storage, transmission, reading and
display of scientific images has occurred at this period. This has contributed to a

S. M. Karve (B) · A. A. Jadhav

SVERIs College of Engineering, Pandharpur, Maharashtra, India
e-mail: smkarve@[Link]
P. Kishrsagar
AVN Institute of Engineering & Technology, Hyderabad, India
M. Aravind Kumar
GVVRIT, Bhimavaram, India

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 73

P. M. Pawar et al. (eds.), Techno-Societal 2020,
[Link]
74 S. M. Karve et al.

substantial increase in the utility of virtual photo processing strategies [1] for clinical
problem solving. In advisement of System used for medical field, the toughest part of
medical imaging lies. A close interdisciplinary partnership between physicians and
engineers involves the design, implementation and evaluation of complex clinical
frameworks. Reading via pix is most importantly aimed at accumulating information,
detecting diseases, analyzing diseases, manipulating and remedying, monitoring and
evaluating [2]. Meanwhile, blood disorders are classified by visual examination of
microscopic blood cell snap shots. It could contribute to a group of safe blood-
associated diseases from the detection of blood disorders. Most cancers are one of
the most feared via The Human Disease. Leukemia is a form of blood cancer, and it
can result in death if it is discovered past due. Leukemia takes place when the bone
marrow contains a number of strange white blood cells. If there are lots of exceptional
white blood cells, the blood system’s equilibrium can be interrupted. When the blood
sample is obtained and analyzed by haematologists, the life of ordinary blood may
be identified. Hematologists must visually examine the microscopic pix, and the
procedure is time consuming and tiring [3–5].
The method needs human professional and sensitive to errors due to emotion
Disruption and Course’s human physical capacity have its own limit. In addition, it is
difficult to obtain a steady impact from visible inspection [3]. For additional research,
visual inspection may also have the simplest qualitative effects [3]. Research shows
that all data is used for maximum of recent techniques Approximately blood for e.G.
Red blood cell variety, degree of haemoglobin, stage of hematocrit, mean corpuscle
extent and much greater as the parameter for classifying diseases like thalassaemia,
most cancers, and many others. The early and fast detection of the type of leukemia
greatly facilitates the presentation of the correct treatment for the precise form of
leukemia [6]. The diagnostic strategies currently used depend on the reading of
Immunephenotyping, Fluorescence In Situ Hybridization (FISH), DanCytochem-
istry Cytogenetic Assessment [6, 7]. State-of-the-art and high-priced laboratories are
required on the way to run the diagnostic techniques and a high ratio of misidentifica-
tion was mentioned; as suggested in [8]: “The first-rate laboratories rely on as many
as a dozen distinct, hard-work-intensive technologies, all of which require relatively
educated professionals.” Having said that in relation to the subtype, maybe 50% of
sufferers are misdiagnosed [6]. With this unit, higher images can be processed, study
time reduced, the impact of subjective elements removed and the accuracy of the
identity system at the same time increased [9]. The leukaemia category and inspec-
tion will be focused on the texture, form, length, coloration and white blood cell
statistical analysis. This research is hoped to help boom productivity globally and
can benefit and be a huge contribution in the field of medical and trend popularity at
the same time. The primary aim is to decorate algorithms that can extract information
from human blood Where human blood is the primary source for detecting diseases
at an earlier stage and can save it quickly [10]. This system should be powerful for
the diversity that exists among individuals, protocols for pattern selection, time and
many others [10]. In this article, in classifying leukaemia types, we will advise on
the use of encouragement to learn (RL). As in [11], a variety of medical picture
problems can be solved by Reinforcement Mastering. Among the objects concerned,
Study on Detection of Leukemia in Human Blood … 75

clinical photographs have a very similar degree of grey and texture. Additionally,
segmentation blunders can appear and boom. Any other issue could be the lack of
a sufficient number of samples of education if a Supervised Mastering Method is
hired. Background. Blood is the main source of records that provide an indicator of
fitness changes and specific disease improvements. Adjustments to the amount or
presence of materials that have been produced can direct a person’s health situation.

1.1 Leukemia

Most blood cells are formed from cells called stem cells within the bone marrow. A
soft material contained in the center of any bone can be the bone marrow. Stem cells
are going to expand and become some pretty blood cells. Each blood group has a
role of their own. Blood components are made up of:
(a) Red blood cells (erythrocytes)—It humps oxygen to the tissues and carbon
dioxide back to the lungs.
(b) White blood cells (leukocytes)—To Contaminate-Safeguard the organism.
Several forms of white blood cells are available.
(c) Platelets—helps control bleeding via blood coagulation.
(d) Plasma—The dissolved ion-containing fluid in the blood needed for cell
activity, consisting of sodium, potassium, chloride, hydrogen, magnesium and
iron. The cells will die when the blood cells are old or damaged and born cells
will replace them [12].
Figure 1 demonstrates how damaged cell got replaced with new cells and grown
into multiple blood components. They grow into either somatic myeloid cells or
somatic lymphoid cells. Myeloid stem cells have now matured and become myeloid
blast cells. This explosion would produce a red blood corpuscle, a platelet, and a
number of other white blood corpuscles. Lymphoid stem cells may also develop and
can form a lymphoid blast, and many kinds of white blood cells may ultimately form
this blast. The lymphoid blast differs from the white blood cells in the myeloid blast.
Since the disease is dangerous and may cause death, the research would specialize in
leukaemia. Bone marrow produces irregular white blood corpuscles for someone who
has leukaemia. Abnormal white blood cells will not die until they have to, contrary
to normal cells. Therefore, there are several irregular white blood corpuscles that
conflict with normal white blood cells in order to perform their functions. It constructs
a blood system imbalance within the human body. Leukemia is also clustered to
promote the rapid progression and severity of this disease. Leukemia is either acute
or chronic.
(a) Leukemia-Leukemic cells may perform tasks like normal white blood cells at
an earlier level. They will eventually develop into serious leukaemia.
(b) Leukemia—Leukaemia cells are unable to perform functions such as normal
white blood cells. Leukemia cell count can develop rapidly and become serious
76 S. M. Karve et al.

Fig. 1 Production of blood

cell [12]

within a short period of time. Typically, leukaemia is also classified into four
types [8].
(c) Acute leukemia (ALL)—In children aged 2–10 years, acute leukaemia (ALL)
typically occurs. This is the most common form of leukaemia. In adults, they
still appear.
(d) Figure 2 shows acute chronic myelocytic leukaemia (AML) is common in
children under 1 year of age with this form of leukaemia. It’s very unusual for
adolescents. Yet it’s mostly in Adults who are 40 years of age.

Fig. 2 Acute lymphocytis

Leukemia (All) [8]
Study on Detection of Leukemia in Human Blood … 77

1.2 Blood Cell Research

Some study has been wiped out to automate the process of detection of blood
corpuscles and the patient can be correctly diagnosed next. A lot of them are [13]
to establish a method for the identification and classification of plasmodium by
microscopic images of blood cells. They use the morphology approach, so the key
criteria for improving this technique are the best methods for segmenting images
of blood corpuscles [14]. Identify patients with thalassemia using genetic program-
ming and neural networks. Instead of microscopic pictures, matured blood corpuscle,
platelet and reticulocyte data is used for eg cell percentage, hemoglobin level, hema-
tocrit level, mean volume corpuscle, distribution of hemoglobin distance, etc. to
spot patients with thalassemia, thalassemia characteristics, and usual. The multi-
layer perceptron (MLP) with 2 hidden layers is the result obtained. Training data
outcomes are 99.27% maximum accuracy, 98.16% mean accuracy, and 0.64% vari-
ance. Although the results of data testing are 88.72% overall accuracy, 84.44% mean
accuracy and 2.41% variation. While [4] a method called Leuko is being developed
and textural information is used to increase differences between leukocytes. They
used textural parameters, which are energy, inertia, homogeneity and correlation,
based on grey level occurrence matrices (GLCM). In designing the Leuko method,
Due to tedious process of features selection data reductioncan be achieved in the
view of learning the classification algorithms quickly and accurately. Classifier can
help standardize from available data, it is easier to interpret results as well as reduce
the time [15]. A device was then developed to identify leukemia cells by using
photographs of the bone marrow. Using the Support Vector Machine (SVM) clas-
sifier, the framework was designed to exploit features related to texture, geometry
and statistical analysis in blood cell images. In order to get the best recognition, they
stress on producing and selecting characteristics. They only pick 30 best features
and this generates a 13.07% training data error and 18.71% testing data errors. In
medical imaging, there are many applications that use reinforcement learning [11]
and the In their work, Kubota et al. [16] have used reinforcement learning (RL). They
use RL in order to Beat the medical photos with some issues. Among the objects
concerned, medical images have a very similar degree of grey and texture. Errors
in segmentation can occur and increase. If a supervised learning technique is used,
another concern may also be the lack of a sufficient number of coaching samples. A
minimum training dataset is required using the RL method [11]. Using Q-learning to
segment images from computed tomography (CT). They use images of cranial CT.
They noticed that they were ready to segment a picture into certain distinct regions
at the same time. The images are split into multiple sub images. An intervention
for sub images to differ and change the Qmatrix is selected by the Reinforcement
Learning (RL) agent. There is an evaluator who compares the outcomes and gives a
gift to the RL agent. The precision of segmentation that they achieve is above 95%
[16]. Apply the RL method to segmental prostate ultrasound images. They still use
Q-learning and the findings indicate a high potential for reinforcement learning to
be implemented in medical segmentation. By using an RL agent, their technique is
78 S. M. Karve et al.

to control the local threshold and hence the post-processing parameter [17]. Extract
the area of the kidney as a preprocessor for the diagnosis of renal disease. They use
X-ray CT abdominal images. Inside the rough kidney area, Q-learning is used and
the edge of the kidney contour is detected. However, for an actual contour, there are
a few error margins and it is corrected by the snake process. The chance of success
is a sort of that’s a poor 53%, another application for RL is used by [18].

2 Research Methodology

Research methodology and steps which will be utilized in this research includes
(Fig. 3):
1. Image Acquisition: With efficient magnification, blood images from slides will
be collected from the nearby hospital.
2. Preprocessing: The photos will be acquired during image processing and unnec-
essary staining during you’ll be distracted by noise. As our ROI would be white
blood cells, the history is going to be removed. Image enhancement will be
performed during this preprocess because the contrast enhancement technique
is capable of enhancing the quality of the medical image [19].
3. Segmentation: White blood corpuscle (WBC) segmentation and determination
of the nucleus ROI for WBC only. This is because the cytoplasm is sparse in
leukaemia cell images [7]. So the emphasis will only be on the nucleus of the
WBC. Determination of the WBC forms from the nucleus should be completed.
Only lymphocytes and myelocytes should be considered, and whether or not
they are blast cells or not should be determined. Others should be omitted, such
as neutrophils, basophils and eosinophils.

Fig. 3 Typical steps in

process of automating blood
recognition
Study on Detection of Leukemia in Human Blood … 79

4. Feature Extraction: The most important problem in generation of features of

blood cells that characterize them during a way enabling the popularity of
various blast types with the very best accuracy [15]. For the nucleus of lympho-
cytes and myelocytes, the features to be included are: Geometric characteris-
tics, which include area, radius, perimeter, symmetry, concavity, compactness,
strength, excentricity, elongation, form factor, will be acquired. Texture charac-
teristics, which include homogeneity, energy, correlation, entropy [7], contrast,
second angular momentum, will be obtained.
5. Color Features: The RGB colour spaces will be converted into colour spaces
for HSV or L*a*b. Their average colour values will be collected. Statistical
characteristics: it is important to obtain the mean, variance, skewness, kurtosis of
the histograms of the image matrix and thus the gradient matrix for RGB or HSV
or L*a*b colour space (whatever the case may be). ALL is small based on [19],
blast cells are uniform, cytoplasm is scarce, round and usually includes single
nucleoli within the nucleus. Although the blasts are larger and irregular shape in
AML, and usually with the involvement of Auer rode, several nucleoli. Qiang
and Zhongh [20] said that although red blood tends to be rather darker than the
background, the WBC appears The (RBC) corpuscle appears at an intermediate
intensity. Cseke [21] shows that white cells are the darker components that tend
to be pale in the RBC images. Platelets are much smaller than cells that are
white and red.

3 Classifications

Classification is the process of assigning to a known class an undefined test vector.

A reinforcement learning algorithm will be proposed during this process. The RL
strategy will identify the leukaemia groups into Both, AML, CLL, and CML as in
Fig. 4 [22].
The essential RL model is used. The world receives the agent’s behavior and
changes it to a replacement state. The agent receives the current state of the world
and the environment receives a gift or punishment. The reward/punishment will be
received by the agent from the environment accepted the environmental action. The
agent gets information and learns how to react to the very best reward. Actions that

Fig. 4 Basic model of RL

80 S. M. Karve et al.

aim to expand the general amount of incentive values should also be selected. The
agent would use a strategy that we have called as a policy of action to decide on
an environmental action. Otherwise the trend would be reduced by the mechanism
generating this behavior [22]. Q-learning can be a soft strategy that suggests that its
Q-values, regardless of discovery, approximate the optimum Q-values that are Q(s,a).
There will be Q-values contained in the Q-matrix. Q(s,a) is that r, obtained by taking
action, a from the economy, s, the estimated amount of potential payoffs. The one
with the best Q value is going to be the optimum action. Q(s,a) will be revised to
support the experience as follows:

Q(s,a) = (1 − α)Q(s, a) + α r + γmaxQ(s , a )

where an is the rate of learning and 0 < γ < 1 is the discount factor. As a conclusion,
as in Fig. 5, the research methodology can be viewed.

Fig. 5 Proposed research

methodology [11]
Study on Detection of Leukemia in Human Blood … 81

4 Discussion

Some blood cell concerns need to be addressed in order to solve them. One of the
issues is the matter of the blood corpuscle itself. Capell [6] say that their method
fails to classify some of the blood cells in the process of classification. Thanks to
environmental strain, a number of cells are also deformed into an arbitrary shape [3].
The Notes on their algorithms that do not distinguish cells that overlap. Overlapping
cells may also be joined by disease-induced cells [23].
He argued that identification between two neighboring cells in their development
line is the most difficult problem since the cells are very close and thus the boundary
point between two neighbors is not well defined. But overlapping problems are also
solved using the watershed approaches used in their studies [6]. Another challenge
is information collection. Blood sample images should be adequate to ensure that
generalization properties are always expressed and that unseen data can be identified
correctly [6]. Lacking samples means that the info [5] can only be represented by
a few key components [24]. Believes that separate data sets are the perfect route
should be used for each point. However, cross validation or bootstrap sampling can
also be used as it is difficult to collect a sizeable amount of samples. It is hoped that
the approach of RL in the classification phase would reduce the issue of inadequate
data. Once we are to build the method, all the problems posed by scientists have to
be taken into account. By applying effective strategies, we should always strive to
resolve them.

5 Conclusion

This study includes using microscopic blood sample images to classify the forms of
leukaemia. By using features in microscopic images, the device will be constructed by
analysing changes in texture, geometry, colours and statistical analysis as a classifier
input. The system should be effective, accurate, less time interval, smaller error, and
high precision, cheaper cost and robust for individual varieties, sample collection
protocols, time and so on. Knowledge derived from microscopic blood sample images
will support individuals by rapidly predicting, resolving and treating blood disorders
for a particular patient.

References

1. Valencio CR, Tronco MN, ACB Domingos, Domingos CRB (2004) Knowledge extraction
using visualization of hemoglobin parameters to identify thalassemia. In: Proceedings of the
17th IEE symposium on computer based medical systems, pp 1–6
2. Adollah R, Mashor MY, Nasir NFM, Rosline H, Mahsin H, Adilah H (2008) Blood cell image
segmentation: a review. In: Proceedings of biomedical, vol 21, pp 141–144
82 S. M. Karve et al.

3. Ritter N, Cooper J Segmentation and border identification of cells in images of peripheral blood
smear slides. In: 30th Australasian computer science conference, conference in research and
practice in information technology, vol 62, pp 161–169
4. Sabino DMU, Costa LDF, Rizzatti EG, Zago MA (2004) A texture approach to leukocyte
recognition. Real Time Imaging 10:205–206
5. Colunga MC, Siordia OS, Maybank SJ (2009) Leukocyte recognition using EM algorithm.
MICAI, LNAI 5845. Springer Verlag, Berlin Heidelberg, pp 545–555
6. Capell K (2012) Meeting leukemia’s diagnostic challenge. Business Week, The Mc
Graw Hill Companies, London. [Link]
49007_mz001.htm
7. Halim NHA, Mashor MY, Hassan R (2011) Automatic blasts counting for acute leukemia based
on blood samples. Int J Res Rev Comp Sci 2(4):971–976
8. Bergen T, Steckhan D, Wittenberg T, Zerfab T (2008) Segmentation of leukocytes and erythro-
cytes in blood smear images. In: 30th annual international IEEE EMBS conference, Vancouver,
Canada, pp 3075–3078
9. Shitong W, Min W (2006) A new detection algorithm (NDA) based on fuzzy cellular neural
networks for white blood cell detection. IEEE Trans Inf Technol Biomed 10(1):5–10
10. Shin H, Markey MK (2006) A machine learning perspective on the development of clinical
decision support system utilizing mass spectra of blood samples. J Biomed Inform 39:227–248
11. Chitsaz M, Woo CS (2011) Software agent with reinforcement learning approach for medical
image segmentation. J Comput Sci Technol 26(2):247–255
12. Bergen T, Steckhan D, Wittenberg T, Zerfab T (2008) Segmentation of leukocytes and erythro-
cytes in Blood Smear Images. In: 30th annual international IEEE EMBS conference, Vancouver,
Canada, 20–24 August, pp 3075–3078
13. May P, Ehrlich HC, Steinke T (2006) ZIB structure prediction pipeline: composing a complex
biological workflow through web services. In: Nagel WE, Walter WV, Lehner W (eds) Euro-Par
2006. LNCS, vol 4128. Springer, Heidelberg, pp 1148–1158
14. Foster I, Kesselman C (1999) The grid: blueprint for a new computing infrastructure. Morgan
Kaufmann, San Francisco
15. Piurri V, Scotti F (2004) Morphological classification of blood leukocytes by microscope
images. In: IEEE international conference on computational intelligence for measurement
systems and applications, Boston, MA, USA, pp 103–108
16. Kubota Y, Mitsukura Y, Fukumi M, Akamatsu N, Yasutomo M (2005) Automatic extraction
system of a kidney region based on the Q-learning. Springer-Verlag, Berlin Heidelberg, pp
1261–1267
17. Markiewicz T, Osowski S, Marianska B, Moszczynski L (2005) Automatic recognition of
the blood cells of myelogenous leukemia using SVM. In: Proceedings of international joint
conference on neural networks, Montreal, Canada, pp 2496–2501
18. Sadeghian F, Seman Z, Ramli AR, Kahar BHA, Saripan M (2009) A framework for white blood
cell segmentation in microscopic blood images using digital image processing. Biol Proced
Online 11(1):196–206
19. Sahba F, Tizhoosh HR, Salama MMA (2006) A reinforcement learning framework for medical
image segmentation. In: International Joint Conference on Neural Networks, Vancouver,
Canada, 16–21 July, pp 511–517
20. Qiang W, Zhongli Z (2011) Reinforcement learning, algorithms and its application. In: Inter-
national conference on mechatronic science, electric engineering and computer, Jilin, China,
pp 1143–1146
21. Cseke I (1992) A fast segmentation scheme for white blood cell images. IEEE, pp 530–533
22. Birndorf NI, Pentecost JO, Coakley JR, Spackman KA (1996) An expert system to diagnose
anemia and report results directly on hematology forms. Comput Biomed Res 29:16–26
23. Using visualization of hemoglobin parameters to identify thalassemia (1981) Proceedings of
the 17th IEE symposium on computer based medical systems, pp 1–6. M.S.: 2004 identification
of common molecular subsequences. J Mol Biol 147:195–197
Study on Detection of Leukemia in Human Blood … 83

24. Srinivisan KS, Lakshmi D, Ranganathan H, Gunasekaran N (2006) Non invasive estimation
of hemoglobin in blood using color analysis. In: 1st international conference on industrial and
information system, ICIIS 2006, Sri Lanka, 8–11 August, pp 547–549
Brain Tumor Detection Using Deep
Neural Network

Rajshree B. More and Swati. A. Bhisikar

Abstract Brain tumor identification is an essential task for assessing the tumors
and its classification based on the size of tumor. There are various types of imaging
strategies such as X-rays, MRI, CT-scan used to recognize brain tumors. Computed
Tomography (CT) scan images are used for in this work for Brain tumor Image Iden-
tification. CT-scan images are used, because as it gives size, shape and blood vessels
detailing and is non-invasive technique. CT-scan is commonly utilized because of the
superior quality of image. Deep learning (DL) is the most recent technology which
gives higher efficiency results in recognition, classification. In this paper, the model
is developed by using Convolution neural network to detect the tumor of brain image
from a dataset from Kaggle. The dataset contains near about 1000 images. Tumor is
identified by image processing algorithm using CNN, time complexity is 90 m sec,
and the accuracy of the present system is 97.87%.

Keywords Brain tumor · CT scan images · Deep neural network · Convolution

neural network · Spyder (Python 3.7)

1 Introduction

Brain Tumor can be characterized as an artificial and abnormal increase in brain cells.
The human skull is inflexible, and its size is restricted in this manner any undesired
development in the tumor may influence on human brain activity. Besides, it gets
increased into other body organs and impact human body functionality [1, 2]. The
World Health Organization (WHO), says that the brain malignant growth influences
under 2% of the human population, which causes serious complications and intrica-
cies [3]. A brain tumor can be gathered from multiple points of view, primary and
secondary tumor. The previous findings show 70% of all brain tumor are primary,

R. B. More (B) · Swati. A. Bhisikar

Department of Electronics and Telecommunication Engineering, JSPM’s Rajarshi Shahu College
of Engineering, Tathawade, Pune, India
Swati. A. Bhisikar
e-mail: sabhisikar_entc@[Link]

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 85

P. M. Pawar et al. (eds.), Techno-Societal 2020,
[Link]
86 R. B. More and Swati. A. Bhisikar

while secondary (optional) tumor are the residuals 30%. This categorization of the
primary tumor first originates in the brain at the early stage. On the contrary, tumors
that first emerge in some other pieces of the body and afterwards moved to the cere-
brum are called optional tumors, and most of them are dangerous [4]. A brain tumor
is an aggregation of irregular cells in the brain. Any development inside the skull,
such a restricted space can cause issues. Brain tumors can be harmful (malignant) or
non-harmful (benign). At the point when the harmful tumor is developed, it increases
the pressure inside the skull. This may cause brain damage, and life expectancy is
very low [2]. Various imaging strategies can be utilized to distinguish and arrange a
brain tumor. In imaging modalities, MRI is widely recognized non-invasive methods
[5–7]. CT scan shows a tumor shape, size and location, blood vessels that feed the
tumor. X-ray images show different types of body parts in different shades of black
and white; the X-ray images are widely used to checking fractures (broken bones)
[8].
At an early stage, tumor detection is done by using image processing provides a
fundamental understanding of the presence of tumor and its type which is advanta-
geous for the treatment strategy and diagnosis to save patients life [9]. Moreover, the
detection of a stage of tumor is a difficult task for radiologists and physicians. This
needs specialists to detect tumor by comparing tumor tissues and by areas, to make
Images clear for human vision, image processing filters are used. Thus, it is a require-
ment for Computer-Aided Diagnosis (CAD) framework to early diagnose brain tumor
in reasonably less time to reduce radiologist burden [2]. Machine Learning (ML) is the
science that allows software applications to predict more accurate outcome depend
on training without being specifically programmed [10]. ML algorithms play a vital
role in the medical imaging field as a part of artificial intelligence [11]. Supervised
and unsupervised learning are two types of ML. In supervised learning techniques,
the outcome is based on input variables it analyses the training data and produces
output from the labelled data, in unsupervised learning, the data is not labelled, and
the algorithm is applied on input data without any prior training of data. Artificial
Neural Network (ANN) is based on a collection of connected artificial neurons. It
is used to simulate input database internal weighting system [12], Support Vector
Machine (SVM) is used for characterization problem it uses the kernel to transform
input data based on these transformations it gives optimum boundary between the
possible outputs, and K-Nearest Neighbors (KNN) is the non-parametric approach
used for the classification. Classification is based on nearest neighbour rule [13]. In
comparison, the outcome of the unsupervised technique is dependent only on the
input variables [14] and Self-Organization Map (SOM) [15]. For segmentation of
tumor the image is to be converted into grayscale used this information for detection
of tumor. From the handcrafted features, meaningful information can be extracted
by the expert in the medical field and which requires a considerable time and may
cause an error while handling large database [16].
Feature learning and accuracy is achieved by increasing training samples instead
of machine learning and neural systems which are the main advantages of Machine
learning which promote extremely powerful and exact model [17, 18]. The approach
Brain Tumor Detection Using Deep Neural Network 87

to train and developed the classification of brain tumor using CNN architecture was
developed connected convolution layers and achieved an accuracy of 91.43% [19].
In this paper, the identification of brain tumor is obtained using CNN classifier
the paper is focused on pre-processing, segmentation and feature extraction. For
simulation Python programming language is used. It is used for general purpose,
and it is a high-level programming language, and it’s more efficient than any other
languages. Accuracy of the proposed method is of 97.87%.

2 Methodology

Figure 1 shows the block diagram for Brain tumor identification using Convolution
Neural Network (CNN). Convolution Neural Network (CNN) is used in Medical
image processing field [2]. A four-layer CNN is introduced for tumor detection, as
shown in Fig. 1. Following is the block diagram with an explanation.

2.1 Pre-processing

There is a lot of non-essential noise in the image, so we need to improve the quality of
the images in pre-processing ways. The main function of the processing is to improve
the single-volume ratio, improve image appearance, remove noise and undesirable
backgrounds, smoothing the interior parts and keeping the edges.

Fig. 1 Block diagram of brain tumor identification

88 R. B. More and Swati. A. Bhisikar

Fig. 2 CNN architecture

2.2 Extraction of Features

Extraction of features is the method of gathering higher-level image details such as

structure, texture, color and contrast. Analysis of textures is an important parameter
for human visual perception and machine learning. Texture finding and analysis at
different stages of tumor detection can improve the diagnosis. It is used to enhance
diagnostic system accuracy effectively by selecting statistical characteristics such as
mean, contrast, energy, entropy, standard deviation, kurtosis, etc.

2.3 CNN Architecture

The convolution neural network contains inputs and output layers, as well as many
hidden layers. CNN’s hidden layers usually include a series of convolution layers
that convince duplication or other dot product. Four types of layers available on CNN
as represented in Fig. 2, layers are
(1) Input layer, (2) Convolution Layer, (3) Pooling Layer and (4) Fully Connected
layer.
Input Layer: The input of the network is a 2 Dimensional digital image. The
network has an input layer which takes the image as the input.
Convolution layer: These are the most important blocks used in convolution
neural networks. Convolution is a filter applied to an input that results in an activation
function. Neural Networks consists of a series of hidden layers that transform the
input image. Every layer is a collection of a set of neurons, where each layer is fully
connected to all neurons in the next layer.
Pooling layer: Pooling is the process of extracting the features from images output
of a convolution layer. Maximum pooling operation calculates the maximum value
in every patch of the feature map.
Fully connected layer: It forms the last few layers in a neural network. The input
to the fully connected layer is the output from the final Pooling layer. This process
determines the most accurate weights. The output Eq. 1 below,
Brain Tumor Detection Using Deep Neural Network 89

y1 = x1ω1 + x2ω2 + · · · xnωn (1)

The primary CNN shown in Fig. 2. CNN advantage is it performs well, and it
gives better accuracy. It is covering local and global features. It also learns different
features from images. In algorithm-based image classification, we need to select the
features (local, global) and classifiers. Other learning algorithms or models can also
be used for image identifications and classification.
Convolution Neural Network is used in object recognition and biological image
segmentation which enhances the performance of deep learning methods. In tradi-
tional methods, features are to be extracted first and then these features are fed
into the training model for the desired result of classification. CNN’s automatically
understands complex characteristic features directly from the data given.
Thus, in this paper, the main focus is on CNN architecture design instead of image
processing for extracting features. CNN takes patch image from the entire image as
input and uses convolution filters which are flexible and trainable; also, it makes use
of local subsampling to extract features.
Figure 3 shows the four-layer CNN model gives us a remarkable result for the
detection of the tumor. We observe the performance in two ways based on the sepa-
ration of the dataset. First, we tried to partition data in 70:30 formats that means 70%
data goes to the training and remaining data goes for testing by using this splitting
ratio, the accuracy is 97.5% [20]. And secondly, 80% of the images are used for
training, and some of the images are given for testing where we get 97.87% accuracy
and 98.47% training accuracy. In line with these lines, the proposed model provides
a positive effect on a scale of 80:20. Table 1 shows the above details in the form of
a table.

Fig. 3 A convolution
network architecture

Table 1 Performance of the proposed CNN model

Sr no Total images Training image Testing image Splitting ratio Accuracy (%)
1 1000 700 300 70:30 92.98
2 1000 800 200 80:20 97.87
90 R. B. More and Swati. A. Bhisikar

Table 2 Performance
Methodology Accuracy (%)
Comparison
Seeta et al. [20] 97.5
Hussain [17] 96.87
Proposed CNN model 98.87

Related to testing the performance, as shown in Table 2 of our model database,

we used the database in the Brain Tumor diagnostic field, and that was taken from
Kaggle. It consists of two classes: phase 0 and Phase 1. Class-0 represents non-tumor
images, and class 1 represents Tumor images. 800 and 200 images containing tumor
and non-tumor respectively classified as class 1 and class-0. All images are CT scan
images. We obtained a positive result by dividing the 80 to 20 datasets by going to
a CNN image test.

3 Flowchart

Figure 4 shows the complete process of Image identification of Brain tumor; CT

digital images from the database are loaded and labelled. Further pre-processing,
validation is done on test sets. The database images are split into training, testing
and validation. The digital brain tumor images are given as input to the optimiza-
tion algorithm. Finally, network training is done, and performance computations are
presented.

Fig. 4 Process of image

identification of brain tumor
Brain Tumor Detection Using Deep Neural Network 91

4 Results

In this work, CNN is used, which provides image data classification. The reason why
CNN is best is CNN doesn’t require feature extraction because CNN layers extract
features from an image by itself. CT scan image data set, which is made available
from Kaggle open-source dataset portal.
The system captures the brain images and verifies the brain tumor is present or
not in that particular brain image. Then it will find a brain tumor is present or not.
Figure 5 shows the GUI system of the complete process, in that process first we
Train the model using CNN algorithm after completing the training process select
the image of brain tumor and then press the prediction button. Finally, the system
Display the result in that result display a message whether the tumor is detected or
not.
Figure 6 shows the system GUI. There are 3 button search buttons have a different
functionality and Image Process windows for image preprocessing technique. In that
GUI first train the model after completing training procedure, select brain tumor
image and after that click the prediction button then we get results on above image
Brain tumor is not detected.
Figure 7 shows the result of the proposed system. It shows the detected tumor,
and Fig. 8 shows the non-tumor image detection.
Train Here CNN Module (Fig. 9) will be training with an accuracy of 98.87%.

Fig. 5 GUI of system & image process window

Fig. 6 Tumor detection

92 R. B. More and Swati. A. Bhisikar

Y- Y- Y-

Fig. 7 Tumor image detection

N- N- N-

Fig. 8 Non-tumor image detection

Fig. 9 CNN module

5 Conclusion

The main aim of this research project is to design an effective system for identifying
brain tumors with high accuracy, performance and low complexity. This paper shows
tumor identification. The Kaggle database contains tumor and non-tumor CT scans
Brain Tumor Detection Using Deep Neural Network 93

images. In this work, brain tumor detection is accomplished through a convolution 4-

layer neural network. Editing is done in the python language. The accuracy compared
to other methods, such as the Support Vector Machine (SVM) is 100%. Processing
time is high. For the proposed CNN based on identification and recognition, the
accuracy of the detection is 95%, and the processing time is 90 ms.

6 Future Scope

(a) By using Web Application patient can upload their brain image and get the
result, whether the tumor is present or not.
(b) By developing Mobile Application it’s easy to use them for the evaluation for
all other Kind of diseases.
(c) In Feature, this technique can be developed to classify the tumors based on
Feature Extraction.

References

1. DeAngelis LM (2001) Brain tumors. N Engl J Med 344(2):114–123

2. Sultan HH, Salem NM, Al-Atabany W (2019) Multi-classification of brain tumor images using
deep neural network. Special section on Deep Learning for computer Aided diagnosis. IEEE
Access 7. [Link]
3. Stewart BW, Wild CP (2014) World cancer report 2014. IARC, Lyon, France
4. Brain, Other CNS and Intra cranial Tumors Statistics. Accessed: May 2019. [Online]. Available:
[Link]
5. Hunnur SS, Raut A, Kulkarni S (2017) Implementation of image processing for detec-
tion of brain tumors. In: 2017 international conference on computing methodologies and
communication (ICCMC), Erode, pp 717–722. [Link]
6. Gurbină M, Lascu M, Lascu D (2019) Tumor detection and classification of MRI brain image
using different wavelet transforms and support vector machines. In: 42nd international confer-
ence on telecommunications and signal processing (TSP), Budapest, Hungary, pp 505–508.
[Link]
7. Parveen, Singh A (2015) Detection of brain tumor in MRI images, using combination of fuzzy
c-means and SVM. In: 2015 2nd international conference on signal processing a integrated
networks (SPIN), Noida, pp 98–102. [Link]
8. Behin A, Hoang-Xuan K, Carpentier AF, Delattre J-Y (2003) Primary brain tumors in adults.
Lancet 361(9354):323–331
9. Litjens G, Kooi T, Bejnordi BE, Setio AAA, Ciompi F, Ghafoorian M, van der Laak JAWM,
van Ginneken B, Sánchez CI (2017) A survey on deep learning in medical image analysis. Med
Image Anal 42:60–88
10. Gooden berger ML, Jenkins RB (2012) Genetic so fadult glioma. Cancer Genet 205(12):613–
621
11. Drevelegas A (2002) Imaging of brain tumors with histological correlations. Springer, Berlin,
Germany
94 R. B. More and Swati. A. Bhisikar

12. Louis DN, Perry A, Reifenberger G, von Deimling A, Figarella-Branger D, Cavenee WK,
Ohgaki H, Wiestler OD, Kleihues P, Ellison DW (2016) The 2016 world health organiza-
tion classification of tumors of the central nervous system: a summary. Acta Neuropathol
131(6):803–820
13. Bishop C (2006) Pattern recognition and machine learning. Springer-Verlag, Berlin, Germany
14. Machhale K, Nandpuru HB, Kapur V, Kosta L (2015) MRI brain cancer classification using
hybrid classifier (SVM-KNN). In: Proceedings of international conference on industrial
instrumentation and control (ICIC), pp 60–65
15. Shasidhar M, Raja VS, Kumar BV (2011) MRI brain image segmentation using modified fuzzy
C-means clustering algorithm. In: Proceedings of international conference on communication
systems and network technologies (CSNT), pp 473–478
16. Goswamiand S, Bhaiya LKP (2013) Brain tumour detection using unsupervised learning
based neural network. In Proceedings of international conference on communication
systems and network technologies (CSNT), pp 573–577
17. Hossain T, Shishir FS, Ashraf M, Al Nasim MDA, Shah FM (2019) Brain tumor detection using
convolutional neural network. In: 2019 1st international conference on advances in science,
engineering and robotics technology (ICASERT)
18. Siar M, Teshnehlab M (2019) Brain tumor detection using deep neural network and machine
learning algorithm. In: 2019 9th international conference on computer and knowledge engi-
neering (ICCKE), Mashhad, Iran, 2019, pp 363–368. [Link]
8964846
19. Paul J (2017) Deep learning for brain tumor classification. In: 13 March 2017 SPIE 10137,
Medical imaging 2017: biomedical applications in molecular, structural, and functional
imaging, 1013710. doi: [Link]
20. Seetha J, Selva kumar Raja S (2018) Brain tumor classification using convolutional neural
networks. Biomed Pharmacol J 11:1457–1461. [Link]
Design and Simulation of Different
Structures of Micro Strip Patch Antenna
for Wireless Applications

Anil. J. Kokare, Mahesh. S. Mathpati, and Bhagyashri. S. Patil

Abstract The need for multiband, bigger addition and low profile radio wires to help
numerous remote applications prompted the plan of Microstrip reception apparatuses.
Microstrip radio wires because of their little profile configuration take less zone. This
paper presents a straightforward rectangular Microstrip Patch Antenna, E-Shaped,
U-Shaped, +-Shaped radio wires work at 2.2 to 3.8 GHz. The Proposed reception
apparatus will be in lightweight, keen and conservative unit contrast and comprises
of metallic fix and ground between which is a dielectric medium called the substrate.
This various structures of MSA are utilized for military, remote and common appli-
cations. The CADFEKO programming is utilized to register the increase, power,
radiation example and S11 of receiving wire.

Keywords Microstrip · Fractal · Multiband · WLAN · LTE · CADFEKO

1 Introduction

Taking into account the advancement of the ongoing remote correspondence frame-
works and its application, more extensive data transfer capacity, multiband and low
profile radio wires are in incredible interest for both business and military applica-
tions. The quick increment of remote interchanges prompts an enormous interest in
the planning of a multiband radio wire. Expectedly, every reception apparatus works
at single or double recurrence groups, where distinctive receiving wire is utilized for
various applications. The plan of proposed receiving wires is utilized for the fast,

Anil. J. Kokare (B) · Bhagyashri. S. Patil

Electronics & Telecommunication Department, SMSMPITR, Akluj, Maharashtra, India
Mahesh. S. Mathpati · Bhagyashri. S. Patil
GNDEC, Bidar, India
Mahesh. S. Mathpati
Electronics & Telecommunication Department, SVERI’s College of Engineering, Pandharpur,
Maharashtra, India
Bhagyashri. S. Patil
Visvesvaraya Technological University, Jnana Sangama, Belagavi, Karnataka, India

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 95

P. M. Pawar et al. (eds.), Techno-Societal 2020,
[Link]
96 Anil. J. Kokare et al.

versatile correspondence and furthermore advancement of microwave frameworks,

for example, WLANs, WiMAX alongside the conveyance of rapid information.
Various receiving wire plans, for example, E-formed radio wires have been introduced
for such multi-standard portable terminal application. A tale GPS fix reception appa-
ratus with fractal EBG structure utilizing an attractive natural substrate is introduced,
which can meet the prerequisites of scaling down and elite of GPS [1]. Hexagonal
fractal receiving wire configuration begins with the first emphasis of square fix and
is portioned by eliminating the centre square from it. For the second emphasis, the
square is cut into nine harmonious sub squares by 3-by-3 evaluation, and the focal
sub-square is eliminated. A similar system is then applied recursively to the excess
eight sun squares, and for the third emphasis again, we take 33% of second sub
squares [2]. An enhanced and minimized printed double band fractal reception appa-
ratus reasonable for WLAN applications. The proposed reception apparatus takes a
shot at 2.4 and 5.2 GHz [3]. A composite scaled-down fractal radio wire as a mix of
Minkowski and Koch bends is introduced. The structure of the proposed reception
apparatus is after the effect of the alterations made with the essential fractal square
and three-sided bends. The radio wire can be utilized for most handheld gadgets
and subsequently finds wide applications in the field of remote and portable appli-
cations [4]. A test took care of E-formed fractal fix reception apparatus (EFPA)
for heptads band LTE/WWAN (GSM850/900/1800/1900/UMTS/LTE2300/2500)
activity is proposed, and different cycles of this fractal receiving wire are looked
at, and an improved plan is introduced.

2 About Fractal Geometries

The term fractal was begotten by the French mathematician B.B. Mandelbrot during
the 1970s after his spearheading research on a few normally happening sporadic
and divided calculations not contained inside the domains of customary Euclidian
math [5]. The term has its underlying foundations in the Latin word fractus, which
is identified with the action word finger (which means: to break). These calculations
were commonly disposed of as amorphous; however, Mandelbrot found that specific
uncommon highlights can be related to them. He found a typical component in huge
numbers of these apparently sporadic calculations and figured speculations depen-
dent on his discoveries [6]. Two instances of normally happening fractal calculations
are snow-chips and limit of geographic mainlands. The fractal receiving wires are
not the same as customary radio wires since it is fit for working at various frequencies
all at once. The vital favourable circumstances of fractal radio wires are diminished
reception apparatus size, uphold multiband and wideband activity with improved
receiving wire execution. These can be accomplished utilizing fractal calculation
like Hilbert, Sierpinski, Koch and Minkowski are the different kinds of fractal calcu-
lations [7]. All these fractal calculations are utilized to plan a little size multiband
and wideband reception apparatuses.
Design and Simulation of Different Structures … 97

3 Methodology

The patch antenna has been designed for the following dimensions to achieve the
requirements of wireless communication applications.
Resonant frequency (fr ) = 2.4–4.8 GHz, Dielecrtric constant (ε) = 4.4.

3.1 Rectangular and E-Shaped Microstrip Patch Antenna

The radio wire is planned at 2.4 GHz recurrence and built as a fix on the substrate.
Model number one in Fig. 1 shows a rectangular fix receiving wire configuration to
work near 3.8 GHz will be demonstrated. The model is first developed as a fix on
an infinitely huge substrate since it rushes to make and to recreate. Figure 2 shows
an E-formed fix radio wire plan fractal math is in gendered in an iterative style,
prompting self-comparative structure.

Fig. 1 Rectangular patch

antenna

Fig. 2 E-shaped patch

antenna
98 Anil. J. Kokare et al.

3.2 U-Shaped and Plus Shaped MSA

Figure 3 shows Plus-formed structure. This is a reference receiving wire or base

shape reception apparatus. Further, this base shape radio wire is adjusted by embed-
ding’s level spaces on the two sides with the separate focus of fix can be utilized in
remote correspondence applications. Figure 4 shows U-molded structure can work
at 2.5 GHz for some applications in ongoing remote correspondence.

Fig. 3 Plus-shaped patch

antenna

Fig. 4 U-shaped patch

antenna
Design and Simulation of Different Structures … 99

Table 1 Comparison of different structures of microstrip patch antennas

Parameters Rectangular Plus shape U shape E shape
Resonant frequency (GHz) 2.7 3.8 2.5 2.4
Reflection coefficient (dB) −40 −36 −38 −11
VSWR 1.5 1.83 2 1.85
Impedance (ohm) 50 48 49.5 51.2
Return loss (dB) −22 −23 −22 −18

4 Simulation Results

Table 1 shows the comparison of different shapes of the microstrip antenna.

For the examination of proposed Different shape of microstrip patch fractal recep-
tion apparatus, the receiving wire boundaries like reflection coefficient, VSWR,
addition and transmission capacity are recreated utilizing reproduction program-
ming CADFEKO. CADFEKO is a full-wave electromagnetic field test system that
depends on the Method of Moments (MoM). It is a business programming device that
can be utilized for receiving wire plan, radio wire situation examination, RF struc-
ture execution forecast, EMC just as dissipating issues and bio-electromagnetics.
The initial four cycles of the middle feed E shape fractal reception apparatus are
mimicked by utilizing CADFEKO programming, and results are demonstrated as
follows.
The Reflection coefficient for the of the E shape fractal receiving wire is plotted
in Fig. 5. Reflection coefficient esteems—11 dB, at particular thunderous frequen-
cies 2.4 GHz. The E formed fractal radio wire creates a low reflection misfortune
contrasted with the standard qualities needed for portable application at GSM band.
Impedance and VSWR plots are shown in Figs. 6 and 7 with the appropriate value
required for wireless application. The antenna radiation pattern is plotted in a 3D
view, as shown in Fig. 8.

5 Conclusion

It would have been a major bit of leeway to know the CADFEKO 14.0 reproduction
programming already as a lot of estimations might have been applied. Miniature strip
radio wires and their hypothesis get substantially more perplexing as you need to
make more proficiency. The various sorts of reception apparatuses are explored and
effectively reproduced in this paper. The reproduced reflection coefficient, impedance
and radiation design demonstrated well execution. Miniature strip reception appa-
ratuses have become a quickly developing territory of examination. Their poten-
tial applications are boundless, on account of their lightweight, conservative size,
and simplicity of assembling. E-formed micro strip reception apparatus is entirely
100 Anil. J. Kokare et al.

Fig. 5 Reflection coefficient

of the patch antenna

Fig. 6 Impedance of patch

antenna
Design and Simulation of Different Structures … 101

Fig. 7 VSWR of E-shaped patch antenna

Fig. 8 E-shaped patch

antenna radiation

planned with an expanded transmission capacity as compared to rectangular minia-

ture strip receiving wire antenna. Comparison of different shapes of the microstrip
antenna is simulated and presented.

References

1. Ranjan A, Singh M, Sharma MK, Singh N (2015) Analysis and simulation of fractal antenna
for mobile Wi-max. IEEE Trans Future Gener Commun Netw 7(2):57–64
2. Jannani A (2011) Design of E-shape fractal simple multiband patch antenna for S-band LTE
and various mobile standards. IEEE Trans Antennas Propag 8:126119–126126. [Link]
10.1109/access.2020.3006831
102 Anil. J. Kokare et al.

3. Permana CG, Munir A (2011) Printed multiband antenna for mobile and wireless communica-
tions. IEEE Trans Antennas Propag 236–240
4. Sreelaxmi K & Mithun Megha Sarkar TP (2014) Multiband miniaturized fractal antenna for
mobile communications. Int J Res Eng Technol (IMPACT: IJRET) ISSN(E): 2321–8843;
ISSN(P): 2347–4599 2(4)
5. Khidre et al (2012) Presented U slot micro-strip antenna for higher mode applications. IEEE
Access Digit Object Identifier 8:112840–112845. [Link]
6. Dhillon SS, Marwaha A, Nagpal A (2013) Multiband E-shaped fractal microstrip patch
antenna with DGS for wireless applications. In: Proceedings of international conference on
computational intelligence and communication networks, Mathura, pp 22–26
7. Lotfi P, Azarmanesh M, Soltani S, Bayatmaku N (2016) Design of simple multiband patch
antenna for mobile communication applications using new E-shape fractal. IEEE Antennas
Wirel Propag Lett 10
State Context and Hierarchical Trust
Management in WSN for Intrusion
Detection

Ranjeet B. Kagade and J. Santhosh

Abstract Wireless sensor network is defined as homogeneous or heterogeneous

system containing large number of sensors namely called as nodes used to monitor
different environments in cooperatives. WSN is composed of sensor nodes (SN),
base stations (BS) and cluster head (CH). The popularity of wireless sensor network
has been increased day by day exponentially due to its wide range of application. The
applications of wireless sensor networks are air traffic control, healthcare systems,
home services, military services, industrial & building automations, network commu-
nications, VAN etc. The advantage of WSN is that it is very easy to install in critical
regions where normal network cannot be set. Thus the wide range of applications
attracts attacker. To secure from different types of attacks mainly intruder, intrusion
detection system based on dynamic state context and hierarchical trust in WSNs
(IDSHT) is proposed. The trust evaluation is carried out in hierarchical way. The
trust of sensor nodes is evaluated by cluster head (CH) whereas trust of cluster head
is evaluated by neighbor cluster head or base stations. Hence the content trust, honest
trust and interactive trust is put forward by combining direct evaluation and feed-
back based evaluation in the fixed hop range. In this way the complexity of trust
management is carried in hierarchical manner and trust evaluation overhead is mini-
mized. This proposed work addresses the security issues of wireless sensor network.
A more prominent intrusion detection system based on context level and trust level
is introduced. This mechanism achieves more than 90% accuracy in detection of
routing attack and sinkhole attack. The architecture suggested in this paper is used to
develop two level of trust model. Accuracy of 90% and 95% is expected in intrusion
detection and context text detection respectively.

R. B. Kagade (B)
Department of Computer Science and Engineering, SVERI’s College of Engineering, Pandharpur,
Maharashtra, India
PAH Solapur University, Solapur, Maharashtra, India
R. B. Kagade · J. Santhosh
Department of Computer Science and Engineering, Veltech Rangarajan Dr Sagunthala R & D
Institute of Science and Technology (Deemed to Be University), Avadi, Chennai, Tamilnadu, India

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 103
P. M. Pawar et al. (eds.), Techno-Societal 2020,
[Link]
104 R. B. Kagade and J. Santhosh

Keywords Hierarchical trust management · Base station · Cluster head · Intrusion

detection mechanism · P2P network · WSN

1 Introduction

Wireless sensor network (WSN) is collection of sensor nodes that are equipped
with environmental sensors for heat, moisture, humidity, pressure, air/water quality,
weight, sound etc. WSN does not have fixed topology. WSN has wide range of
applications in field of agriculture, health, home, industrial, military and natural
calamities for monitoring and data collection purpose. The advantage of WSN is
that it is very easy to install in critical regions where normal network cannot be set.
Sensor nodes contain five components trans-receiver, processor, battery, hardware
and memory. The wide range of applications can be elaborated as follows:
Wild Habitat Monitoring
Sensors can be used to monitor the conditions of wild animals or plants in wild
habitats. Sensors can be used to collect information in the water and above water.
Disaster Monitoring
Information collection in natural and non-natural disaster areas is very important task.
A normal network cannot be set in such areas. WSN becomes prominent solutions
in such situations.
Warfield Monitoring
Sensors can be deployed in a war field to monitor the presence of forces and vehicles,
and track their movements, enabling close surveillance of opposing forces.
Sensitive Place Protection
Sensor nodes can be deployed around sensitive objects, for example, atomic
plants, strategic bridges, oil and gas pipelines, communication centers, and military
headquarters, for protection purpose.
Vehicle Monitoring
Sensors can be mounted on unmanned robotic vehicles, tanks, fighter planes,
submarines, missiles, or torpedoes to guide them around obstacles to their targets
and lead them to coordinate with one another to accomplish more effective attacks
or defenses.
Remote Sensing
Sensors can be deployed for remote sensing of nuclear, biological, and chemical
weapons, detection of potential terrorist attacks, and reconnaissance [1]. Due to its
vital range of applications it has several ranges of attackers at different levels. In
WSN sensor nodes have limited battery power, limited power to communicate and
State Context and Hierarchical Trust Management … 105

compute. WSN is susceptible to so many attacks, because of broadcasting nature of

network.
An intrusion is defined as a sequence of related actions performed by a malicious
adversary that results in the compromise of a target system. An Intrusion Detection
System specially looks for something wrong operation and events that might be the
outcome of a cause of attack, worm or system expert for pleasure.
Intrusion detection system consists of four methods which detects the system is
described below [2].

1.1 Network Based Intrusion Detection System

NIDS will monitor the traffic, malicious changes that are happening in the network
which causes changes that leads to system degradation.

1.2 Host Based Intrusion Detection System

HIDS may also be able to make out malicious business trade that starts with malware
and is attempting to put out on top for other systems.

1.3 Signature Based Detection System

Signature-based detection systems guide all the small data traversing the network
and make a comparison against a knowledge-base of signatures or given properties
of experienced violent behavior, like antivirus software.

1.4 Anomaly Based Detection System

Anomaly-based detection systems guide network business trade and make a compar-
ison against a started baseline, to come to a decision about what is taken into account
normal for the network with respect to bandwidth, protocols, harbors and other
devices [3, 4].
Trust is the degree of belief that a node can have on another node in the network
based on trust metric or trust rating o the node. Different types of attacks due to
intrusion are illustrated as follows [5].
GTMS: group based trust management scheme GTMS works in two types of
topologies intergroup topology and intra-group topology. In intergroup topology
106 R. B. Kagade and J. Santhosh

centralized trust management is adopted and in intergroup topology distributed trust

management is adopted.
ATRM: agent based trust and reputation management scheme An agent based
trust and reputation management scheme [6] (ATRM) is based on a clustered WSN
with mobile agent system. It requires every node to hold the mobile agent which is
administrating the trust and reputation of the hosting node.
PLUS Parameterized and localized trust management scheme for sensor network
security [7]. It uses distributed approach to adapt to different operational environ-
ments and different applications. The trust is calculated based on either direct or
indirect observations.
RFSN reputation based framework for sensor networks they have proposed a
framework where each node maintains reputation metrics which includes the past
behavior of other sensor nodes of the network and the metrics used for predicting the
future behavior. The values of the trust are evaluated on the basis of that reputation
and for representing the values of reputation, Bayesian formulation is used.
TRGR Trust management scheme for resilient geographic routing Trust manage-
ment scheme for resilient geographic routing [8] (TRGR) is a simple trust manage-
ment scheme which uses resilient geographic routing. Geographic routing consists
of two parts: geographic forwarding and complementary routing [9]. The trust algo-
rithm works in a fully distributed manner, in which each node monitors the behavior
of one hop neighbors. The basic idea of this trust management scheme is to favor
well behaving honest nodes by giving them the credit for each successful packet
forwarding, while penalizing suspicious nodes that doesn’t route packet according
to route.
BRMSN Behavior reputation method for sensor networks Behavior reputation
method for sensor networks [10] (BRMSN) measures the spatial information between
the nodes as reputation measurement. The model is a reflection of the node’s compre-
hensive ability about the actual physical properties and conduct essentially. Nodes in
the network not only include the identity of the trust but also the trust in the behavior
of the node. The model focuses on the local testing.
In this paper Intrusion detection system is introduced which depends on trans-
mission time and relieving time. Two level trust mechanism is proposed to reduce
overhead of base station and cluster head. The number of successful and unsuccessful
transmission of data at WSN decides the trust value of sensor nodes.
This paper is organized as follows: In Sect. 2 the existing systems are discussed
while Sect. 3 introduces the proposed work. Section 4 decides expected results and
Sect. 5 concludes the proposed.

2 Related Work

Hossein Jadidoleslamy proposed hierarchical intrusion detection mechanism [11].

The mechanism is designed and applicable. It works in one or two levels; it is consis-
tent to application domain and required security level. The research is focused on
State Context and Hierarchical Trust Management … 107

clustering of WSN and deploying cluster based intrusion detection system. The mech-
anism works in static and heterogeneous network, hierarchical and clustering struc-
ture. Hierarchical cluster based and leach based routing protocol is used for efficient
communication. The mechanism is proved to be as robustness and fault tolerant
design. The research provides reliable service but is an expensive mechanism. The
intrusion detection architecture supports real-time detection property almost 80.6%.
The mechanism results 80.6% and 55.7% accuracy for content based detection and
context based detection capabilities respectively.
Daojing He proposed a distributed trust evaluation model for medical sensor
networks [5]. The traditional cryptographic methods are not sufficient for trust
evaluation in medical sensor networks. The research work uses transmission rate
and leaving time into trust evaluation to detect malicious nodes. They proposed an
application-independent and distributed trust evaluation model for MSNs. The trust
management is carried out through the use of simple cryptographic techniques [5].
A novel distributed trust evaluation model for MSNs, where each node manages
trust records of other nodes about performing some activities. Centralized malicious
node detection and secure unicast routing is presented. It proved to be as improved
packet delivery with effective malicious node identification mechanism. IDS mech-
anism depends on transmission time, relieving time and packet dropping ratio. 88%
accuracy is shown by researchers with limited number of malicious nodes in WSN.
Fenye Bao proposed hierarchical trust management for WSNs and applied it
to routing and intrusion detection to detect selfish or malicious nodes [12]. The
work focused on multidimensional trust attributes and the trust value was calculated
through social trust and QoS trust, including intimacy, honesty, energy, and unselfish-
ness; meanwhile, subjective trust and objective trust were taken into consideration
to validate the proposed protocol [1]. However, the node with the maximum number
of interactions with neighbors was considered as the most trustworthy in the process
of the calculation of the intimacy trust inspired by social networks. The difference is
the consideration of the reasonable range of the maximum number of interactions,
as interaction that exceeds the range indicates malicious behavior. The mechanism
is proved to have IDS detection capabilities with 90% accuracy with false positive
probability is zero and method is scalable. Trust based geographic routing is used in
WSN.
Xiaoyong Li put forward a lightweight and dependable trust system for clustered
WSNs, which aims at decreasing resource consumption and enhancing the relia-
bility of CHs’ trust evaluation [13]. A self-adaptive weighting mechanism is used
to calculate trust value of CH which is better than subjective weight method. A
series of theoretical proofs were given in the research to verify the effectiveness of
the mechanism. In the process of trust evaluation, only successful and unsuccessful
interactions were taken into consideration, with no other trust evaluation factors taken
into account. The mechanism takes interactive trust, honesty trust and content trust
into account, addressing problems of consuming energy maliciously and tampering
multidimensional observing data with lower resource overhead, which is described in
108 R. B. Kagade and J. Santhosh

the performance evaluation [1]. The research is focused on minimizing memory over-
head and transmission overhead. It works better to protect from garnished attack and
bad mouthing attack. Results show that intrusion detection capabilities are 87.5%.
Shaikh proposed GTMS; a group-based trust management scheme for clustered
WSNs. GTMS evaluates the trust of a group of nodes in contrast to traditional
trust schemes that always focus on the trust values of individual nodes [14]. In this
approach WSN requires small memory to store trust value at each node. The mech-
anism achieves significant reduction of the cost associated with the trust evaluation
of distant nodes. But it depends on a broadcast-based strategy to collect feedback
from the CMs of a cluster, which requires a significant amount of resources and
power. This mechanism worked for wired and wireless mechanism. It focused on
reducing cost of trust evaluation. Trust value is calculated depending on time based
past interaction. Timing window is used to measure the number of successful and
unsuccessful interactions. Trust evaluation cost is minimized by 14.6–15.7%.
Ismail Butun has presented intrusion detection system for mobile Ad Hoc
networks [15]. Agent based distributed and collaborative IDSs are emphasized in
research. Two types of classifiers are used for detection of intrusion; Decision Tree
and support vector machine. Dynamic Source Routing, Ad hoc On-demand Distance
Vector and Destination Sequenced Distance Vector protocol is used for routing data
in WSN.
Limitations of Classical Approaches
Signature-based detection approaches are relatively easy to implement, require no
learning curve. This eliminates the risk of over-training or voluntary deformation
of the profile that can be observed in behavior-based approaches. However, these
approaches require an active maintenance and very frequent updates of the signa-
ture database to integrate any new attack discovered. Indeed, the update cannot be
performed automatically as in the case of behavior-based detection. This fact implies
a higher rate of false negatives.
The problem arises especially with very recent attacks for which signatures have
not been included in the database yet. Also, the absence of a standard pattern descrip-
tion language limits the usefulness of signatures described in a given language since
interoperability between different detectors is probably not possible. If signature
values are too simplified it can lead to detection of false intrude which corresponds
to legitimate actions and therefore to trigger false positives.
Anomaly-based detection approaches have several interesting features. First, as
the hypotheses are made only on the normal behavior of the system and not on
possible attacks, detection is exhaustive. Indeed, “the system allows a prior to detect
all that differs” from established normal behavior. Thus, it becomes possible to
envisage detection of unknown attacks and no specific knowledge about the attacker
is required. All necessary information is collected within the system. On the other
hand, once the learning phase terminates, the IDS does not require particular update.
The definition of normal behavior evolve only slightly if any.
Nevertheless, a high rate of false positives is the main weakness of these
approaches because it is sometimes difficult to define the “normal behavior”. Sudden
State Context and Hierarchical Trust Management … 109

changes in the environment can have an impact on behavior. This sudden change in
behavior will be considered as an anomaly and an alert will be generated. Also,
since the first phase is dedicated solely to the development of the definition of
“normal behavior”, this one is particularly vulnerable to attack. Indeed, the pres-
ence of signals related to an attack in the learning trace will result in skewing the
definition of behavior. Thereafter, any similar attack will be treated as a normal
behavior. The information used during this first phase in the optimal condition must
be totally free from damage. In practice, it is frequently impossible to have such
perfect environment.

3 Proposed Methodologies

Problem Statement: To Develop State Context and Hierarchical Trust Mecha-

nism in Wireless Sensor Networks for Intrusion Detection
Following assumptions are considered [1]:
(i) The WSN is cluster-based, and SNs in a cluster could communicate with
the CH directly, whereas CHs communicate with BSs directly or indirectly
through other CHs.
(ii) Each SN has a unique ID and belongs to a unique cluster, and CHs have more
energy than SNs.
(iii) The data transmission model in a WSN is hybrid, including continuous and
event-driven.
(iv) The states of SNs include hibernation, monitoring and active, and the transition
between monitoring and active is taken into consideration during the trust
evaluation of SNs.
(v) Sensor nodes are deployed densely and redundantly for reliability.

Cluster Creations
In wireless sensor network large numbers of application specific sensor nodes are
connected with each other on ad hoc purpose. If each node starts communication,
computation and routing then energy consumption, collision and congestion can
occur. Hence it may become reason for performance degradation. Node clustering
can solve these issues. In clustered network wireless sensor network is divided into
small units. Cluster head is elected. Sensor nodes in each cluster will communicate
with respective cluster head CH and CH aggregates data and will transfer data to
a central base station. The cluster head is considered to have maximum energy. To
maintain the maximum energy at cluster head re-election after certain interval of
time is best possible solution.
LEACH [16] is the first clustering scheme. It is used for periodical data gathering
in WSNs. It assumes the sensor nodes will remain static and communicate with
each other by single-hop only, and they can transmit data to the sink node directly.
110 R. B. Kagade and J. Santhosh

Its operation is divided into rounds and each round is composed of two phases
[17]. In cluster formation phase, LEACH elects some cluster heads according to the
probability shown below.

k
N −k(r mod N /k)
, Ci (t) = 1
Pi(t) =
0, Ci (t) = 0

where k is the desired number of cluster heads, Ci (t) is the indicator function deter-
mine whether or not node i has been a cluster head in most recent (r mod N/k) rounds.
Pi (t) is the probability for node i to become a cluster head. The rest of sensor nodes
join the proper cluster according to the signal strength from the cluster heads. In
the data transmission phase, the cluster heads aggregate the data from their cluster
members. Since cluster head is chosen by probability in each round, the load is
balanced to certain extent.
Trust Calculations
Interactive trust refers to the trust value computed by the number of interactions
between nodes, and an interaction means a node sending/receiving a packet or a
request to/from another node. The trust value is mapped to the integer number in
the range of [0, 10], where 0 demonstrates the most distrustful, while 10 implies the
most trusted, and 5 is the medium trust [1] shown in Fig. 1.
Sensor nodes trust is evaluated by the CH in a cluster, i.e., CH-to-SN trust,
which considers multidimensional trust, including interactive trust, honesty trust
and content trust, during the procedure of trust calculation.
Interactive trust SITij (t) is calculated by the number of interactions between
node j and its CH i in t. In the proposed method, interaction refers to all commu-
nication behavior including sending and receiving of request and data packets. The
greater the number of interactions of two nodes, the higher is the trust value [18].
However, in WSNs, if the number of interactions exceeds a threshold, the trust value
will decrease because there may exist malicious interactions such as attacks that send
a large amount of packets or requests to exhaust the energy of the node. Therefore,
unlike trust evaluation in social networks, the interactive trust evaluation method in

Fig. 1 Wireless sensor network

State Context and Hierarchical Trust Management … 111

WSNs is put forward. Inspired by Normal Distribution in Statistics, the probability

density function, which is normalized to [0, 1] to calculate the interactive trust, is
adopted when the number of interactions exceeds a threshold.
Interactions between CH and SNs are abstracted as an undirected weighted
graph, the weight of which represents the number of interactions between them.
The interactive trust value of SN j evaluated by CH, SITij (t) can be defined as

Wi j
S I T i j(t) = 10 × , j ∈ G, W i j ≤ λμ;
Wi j

|W i j − μ|
10 × ex p ex p − j ∈ G, W i ≥ λμ;
θ

where x denotes the largest integer that is equal to or less than x, μ is the mean value
of the number of interactions between CH and SNs in the same state, λμ is taken as
the threshold of the interaction range, in which λ is a parameter used to define the
upper limit of normal interactions, and θ is a significant factor, which values 1, 10
and 100 when wij is a single digit, tens digit or hundreds digit, correspondingly, and
so on.
Example:
The WSN is assumed to contain eight nodes and one cluster head. Cluster head is
selected depending on computation power, energy, honesty and distance between
CH and base station (Fig. 2).

Fig. 2 Trust calculation in

WSN
112 R. B. Kagade and J. Santhosh

Calculation of Trust value

(i) λ is set to be 2.
(ii) The value of θ is 1,10,100 depending on value of wij . If the value of wij is two
digits θ is 10. If wij is three digit number then θ is 100.
(iii) μ is mean of weights.
μ = ((3 + 5+4 + 13 + 5+6 + 4+16)/8) = 7
(iv) Maximum weight max (wij ) = 13 because weight 16 is greater than λμ(14).
The maximum weight is discarded and 13 is considered as maximum weight.
(v) μ = ((3 + 5+4 + 13 + 5+6 + 4+16)/8) = 7
Trust value is calculated as follows SITij (t) =

(a) For node A = 10 ∗ (3/13) =10 ∗ 0.2307

=2.307
=2

(b) For node D = 10 ∗ (13/13) =10 ∗ 1

=10 ∗ 1
=10

(c) For node H. The weight is 16; it is greater than threshold λμ which is 14. As
the number of is greater than 14 it is considered as distrustful. θ is 10 as the
wij is two digit number. The trust value is calculated as follows (Table 1)

SITij (t) =10 × exp(−|16 − 7|/10)

=10 × exp(−|0.9|)
=10 × 0.4065696
=4.06
=4

Table 1 Evaluation results of

Nodes Evaluation results of trust values
examples in Fig. 1
A 2
B 3
C 3
D 10
E 3
F 4
G 3
H 4
State Context and Hierarchical Trust Management … 113

Honesty Trust of SNs

Honesty trust SHTij (t) is calculated by the number of successful and unsuccessful
interactions between CH i and a non hibernating SN j in t. The CH i overhear the
SN j if j does not deliver a packet in t or transmits the packet to another node that is
not in its routing table, or if the packets from j do not reach the CH i, the interaction
between them is considered an unsuccessful interaction.
The number of successful and unsuccessful interactions between active nodes and
CH i in t is denoted as s and f, and the trust value is evaluated using formula given
below.

10 × (s + 1)
S I T i j(t) = , when f = 0
(s + f + 2)

10 × (s + 1)
−1 , when f = 0
(s + f + 2) × f 2

When there are no interactions between active members, i.e., s = f = 0, the trust
value is 5. If there are unsuccessful interactions, the honesty trust value will decrease
sharply because of the punishment executed by [13]. For non active members, they
inherit the trust value of their lasno hibernating state.
Content Trust of SNs
Content trust is the trust evaluation based on observing data, which is data-oriented
trust calculated by CH. Content trust is introduced because the WSN is a data-
centric network and the observing data are the factor of most concern for applications.
Tampering attacks often occur in WSNs to interfere with the network and applications
and can be identified by content trust.

C T i j(t) = 10 × ex p(−Di j)

1
dm 2

Di j = (Xik − X jk)2̇
k=1

The overall trust of SN j evaluated by CH i is calculated by formula given below,

which aggregates the interactive trust, honesty trust and content trust.

S O T i j = αS I T i j + β S H T i j + (1 − α − β)SC T i j

Parameters α, β ∈ [0, 1] are weights for each sub trust value. The higher the
weight, the more important that sub trusts is to overall trust and vice versa.
114 R. B. Kagade and J. Santhosh

Intrusion Detection at SN level

Malicious SN detection is executed by the respective CH. The CH c evaluates and
maintains the trust value of SN j in the same cluster and selects a trust threshold TSthi
according to the trust value of SNs in cluster i, which is calculated as (Table 2) :

T S thi = avg j∈C Land S O Tcj ≥5{ S O Tcj } ; ∃ j, s.t S O Tcj ≥ 55; other

Intrusion Detection at CH level

T C th = avg j∈C H SandC O Tbj ≥5{C O Tbj } ; ∃ j, s.t C O Tbj ≥ 55; other

4 Expected Results and Outcomes

The proposed methodology will be tested on network simulator (ns2.35). Large

numbers of sensor nodes are considered as components of WSN. The WSN is
assumed as clustered WSN.
There are two major types of attacks in wireless sensor network; active attack and
passive attack [19].
Major intrusion attacks can be listed as follows:
(i) Routing attacks
(ii) Selective forwarding
(iii) Sinkhole attack
(iv) Sybill attack
(v) Wormholes attacks.
The proposed mechanism will work efficiently on routing attacks and sinkhole
attack; as the method is depending on transmission time and relieving time.
Detection Rate (DR) of an Intruder attack

Detection rate
It is defined as ratio of total number of attacks detected and total number of
attacks appeared [19].

P K T detect
W S N acc =
P K T detect + F P K T detect

W S N acc = Accuracy of proposed method

P K T detect = Accurately detected packets as intruder.
F P K T detect = False detected packets as intruder.
State Context and Hierarchical Trust Management … 115

False Positive Rate (Frate )

FP
F Pdetect =
FP + T N

FPdetect is the rate at which false packet detection.

FP is number of false packet detected.
TN is true negative packet detected, the no. of legitimate records.

5 Conclusion

This paper addresses the security issues of wireless sensor network. A more promi-
nent intrusion detection system based on context level and trust level is introduced.
This mechanism achieves more than 90% accuracy in detection of routing attack and
sinkhole attack. The architecture suggested in this paper is used to develop two level
of trust model. Where first trust level is sensor node to sensor node communication
and second trust level is between sensor node and cluster head. The overhead of
cluster head and base station is minimized in this way. Architecture will be tested
on simulator ns2. Accuracy of 90% and 95% is expected in intrusion detection and
context text detection respectively (Table 2).

Table 2 Database at WSN

Items Implications
ID The ID of a SN
wij The number of interactions between CH i and SN j
μ The mean value of interactions between a CH and SNs that are at the same state in a
cluster
θ The significant factor based on the value of wij
s The number of successful interactions between i and j
f The number of unsuccessful interactions between i & j
Dij The Euclidean distance of data between average and j
SITij The interactive trust of SN j evaluated by CH i
SHTij The honesty trust of SN j evaluated by CH i
SCTij The content trust of SN j evaluated by CH i
SOTij The overall trust of SN j evaluated by CH i
TSthi The threshold of malicious SN detection
116 R. B. Kagade and J. Santhosh

References

1. Zhang Z, Zhu H, Luo S, Xin Y, Liu X (2017) Intrusion detection based on state context and
hierarchical trust in wireless sensor networks. IEEE Trans Content Min
2. Benisha RB, Raja Ratna Dr S (2018) Prevention of cyber attacks in control systems: a review.
In: Proceedings of the 2nd international conference on trends in electronics and informatics
(ICOEI 2018)
3. Ntalampiras S (2014) Detection of integrity attacks in cyber-physical critical infrastructures
using ensemble modelling. In: IEEE transactions on industrial informatics, vol 11
4. Sadi Y, Ergen SC, Park P (2014) Minimum energy data transmission for wireless networked
control systems. IEEE Trans Wirel Commun 13(4):2163–2175
5. He D, Chen C (2012) A distributed trust evaluation model and its application scenarios for
medical sensor networks. In: IEEE transactions on information technology in biomedicine, vol
16, no 6
6. Boukerche A, Li X, El-Khatib K (2007) Trust-based security for wireless ad hoc and sensor
networks. Comput Commun 30:2413–2427
7. Maddar H, Kammoun W, Youssef H (2017) Trust intrusion detection system based on location
for wireless sensor network. Springer International Publishing AG
8. Liu K, Abu-Ghazaleh N, Kang K-D (2007) Location verification and trust management for
resilient geographic routing. Parallel Distrib Comput 67(2):215–228
9. He T, Stankovic JA, Lu C, Abdelzaher T (2003) SPEED: a stateless protocol for real-time
communication in sensor networks. In: Proceedings of the 23rd international conference
ondistributed computing systems (ICDCS’03), Washington, DC, USA, pp 46–55
10. Zhou M-Z, Zhang Y, Wang J, Zhao S-Y (2009) A reputation model based on behavior trust
in wireless sensor networks. In: Eighth IEEE international conference on scalable computing
and communications
11. Jadidoleslamy H (2011) A hierarchical intrusion detection architecturer for wireless sensor
networks. Int J Netw Secur Its Appl (IJNSA) 3(5)
12. Bao F, Chen I, Chang M, Cho J (2012) Hierarchical trust management for wireless sensor
networks and its applications to trust-based routing and intrusion detection. IEEE Trans Netw
Serv Manage 9(2):169–183
13. Li X, Zhou F, Du J (2013) LDTS: a lightweight and dependable trust system for clustered
wireless sensor networks. IEEE Trans Inf Forensics Secur 8(6):924–935
14. Shaikh RA, Jameel H, d’Auriol BJ, Lee H, Lee S, Song Y (2009) Group-based trust management
scheme for clustered wireless sensor networks. IEEE Trans Parallel Distrib Syst 20(11):1698–
1712
15. Butun I, Morgera SD, Sankar R (2014) A survey of intrusion detection systems in wireless
sensor networks. IEEE Commun Surv Tutor 16(1):234–241
16. Heinzelman WR, Chandrakasan A, Balakrishnan H (2002) An application-specific protocol
architecture for wireless microsensor netw orks. IEEE Trans Wirel Commun 1(4):660–670
17. Ye M, Li C, Chen G, Wu J (2005) An energy efficient clustering scheme in wireless
sensor networks. In: PCCC 2005, 24th IEEE international performance, computing, and
communications conference
18. Hao F, Min G, Lin M, Luo C, Yang L (2014) MobiFuzzyTrust: an efficient fuzzy trust inference
mechanism in mobile social networks. IEEE Trans Parallel Distrib Syst 25(11):2944–2955
19. Kalnoor G, Agarkhed J (2018) Intrusion threats and security solutions in wireless sensor
networks. Int Robot Autom J
Portable Camera Based Assistive Text
and Product Label Reading from Hand
Held Object by Using Android App
for Blind Person

Somnath Thigale and Ranjeet B. Kagade

Abstract We propose a camera-based mechanical man app. This app helps the blind
persons to browse the text on explicit objects. In this system the camera captures the
actual text on the object. Multiple techniques square measure applied to its text.
Such as Optical Character Recognition that supply the operation of scanning and
recognition of text and a few have integrated voice output. From a grayscale image,
thresholding are often accustomed produce binary pictures i.e. image with solely
black or white colors, Filtering are often accustomed cut back the noise of image, Next
image segmentation technique is employed to perform the method of partitioning a
digital image into multiple segments. The goal of segmentation is to modify and/or
amendment the illustration of a picture into one thing that’s a lot of significant and
easier to analyses. Image scaling is the method of resizing a digital image. Next
technique employed in this project is template matching. Temples matching is a way
in the digital image process for locating tiny components of a picture that match
a template image. Also template extraction are often employed in producing as a
vicinity of internal control, some way to navigate a mobile golem or as some way to
notice edges in images then finally voice output are going to be generated then blind
man will simply listen to the text on it explicit object.

Keywords Assistive devices · Blindness · Distribution of edge pixels · Hand-held

objects · Optical character recognition (OCR) · Stroke orientation · Text reading ·
And text region localization

1 Introduction

The 314 million visually impaired folks worldwide, forty five million square measure
blind handicap that was free by “World Health Organization” in ten facts concerning
sightlessness. Even during a developed country just like the US, the 2008 National
Health Interview Survey (NHIS) reportable that AN calculable 25.2 million adult

S. Thigale (B) · R. B. Kagade

SVERI’s College of Engineering, Pandharpur, Maharashra, India
e-mail: sbthigale@[Link]

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 117
P. M. Pawar et al. (eds.), Techno-Societal 2020,
[Link]
118 S. Thigale and R. B. Kagade

Americans (over 8%) square measure blind [1, 2]. The human baby generation ages
square measure quickly increasing. all over written text square measure obtainable
as an example receipts, bank statements, building menus, room handouts, report,
product packages, directions on medication bottles, room handouts etc. While screen
readers, optical aids and video magnifiers will facilitate blind users and people
with low vision to access documents. The few devices offer sensible access to
common hand-held objects like product packages and objects written with text as an
example prescription medication bottles. The visually handicapped person to scan
written labels and merchandise packages can enhance freelance living and foster
economic and social self-reliance. Today, there square measure already has some
systems for transportable use however they can’t handle product labeling. The trans-
portable Universal Product Code readers designed to assist blind people identify
totally different products in an in depth product info will change the user for visu-
ally handicapped person to access data regarding these products through speech and
Braille. during this system an enormous limitation is that it’s terribly exhausting for
blind users to seek out the position of the Universal Product Code and to properly
purpose the Universal Product Code reader at the Universal Product Code however
Some reading-assistive systems like pen scanners could be used in these and similar
things. This technique integrates OCR computer code used for scanning and recog-
nition of text and a few have integrated voice output. The OCR is optical character
recognition. These systems square measure typically designed and perform best with
document pictures with appropriate backgrounds, commonplace fonts, a tiny low
vary of font sizes, and well-organized characters. This technique is used instead of
business product boxes with multiple decorative patterns. The OCR computer code
cannot directly handle scene pictures with complicated backgrounds and therefore
the document to be scan should be nearly flat, placed on a transparent, dark surface
and contain principally text [3, 4].
Even though variety of reading assistants square measure designed specifically for
the blind folks, to our knowledge, there’s no existing scanning assistant will read text
from the varieties of difficult patterns and Even though variety of reading assistants
square measure designed specifically for the blind folks, to our knowledge, there’s
no existing scanning assistant will read text from the varieties of difficult patterns
and backgrounds that found on many everyday business product. Like text data will
obtainable in varied scales, fonts, colors, and orientations. Images. Initial is Milk
box (Top) and Men lavatory signage (Bottom).
Above Figs. 1 and 2 shows (a) camera captured images. (b) Localized text regions
that show in fig mark blue. (c) Text region scropped from image. (d) Text codes
recognized by OCR (optical character recognition). Text at the top-right corner of
bottom image is shown during a increased callout. Mobile accurately reads black print
text on a white background however there some issues to recognizing colored text
or text on a colored background. It cannot scan text with complicated backgrounds
as a result of they can’t simply detected the text from background. Text printed
on cylinders with crooked or incomplete pictures as an example as soup cans or
medication bottles. These systems require a blind user to manually localize areas
of interest and text regions on the objects. As shown in Fig. 1, such text data will
Portable Camera Based Assistive Text … 119

Fig. 1 Examples of printed text from hand-held objects with multiple colors backgrounds

appear in multiple scales, fonts, colors. to help blind persons to scan text from hand-
held objects, we have planned of a camera-based helpful text reading frame work
to track the item of interest at intervals the camera read and extract print text data
from the item. the prevailing rule will effectively handle complicated background and
multiple patterns and extract text data from each hand-held objects and assemblage as
shown in Fig. 2. The hand-held object extract camera image and we tend to develop
a motion-based technique to obtain region of interest (ROI) of the item. During
this ROI perform the text recognition technique. The localization of text regions
in scene pictures they’re divide in 2 classes: rule-based and learning–based. Rule
primarily based rule to used component level image process to extract the text data
120 S. Thigale and R. B. Kagade

Fig. 2 Two examples of text localization and recognition from camera captured Images. First is
Milk box (Top) and Men bathroom signage (Bottom)

from predefined text layout like character size, ratio, edge density, character structure,
color uniformity of text and learning primarily based rule square measure used model
text structure and extract representative text options to make text Classifiers [5, 6].

2 Existing System

In these days society, there square measure already a couple of systems that square
measure moveable use for blind persons. For instance, moveable code readers. The
blind those who want to access data regarding these product, the moveable code reader
helps the blind individuals to spot totally different product in an intensive product
information. However properly position the code reader at the code is incredibly
arduous task for blind users.
Con jointly there some system that uses camera-based helpful text reading frame-
work. Foremost object of interest inside the camera read is track so written text data
from the item is extracted by this framework. During this framework, the item of
interest is positioned at the middle of camera’s read. This object ought to be posi-
tioned specified, it ought to be seem within the camera read. These wide angles
of this camera accommodate the users with solely particular/approximate aim. The
system extract hand-held object from the camera image. To get the region of interest
(ROI) of the item this framework uses the motion primarily based technique. Then
text recognition is performed just for that region of interest (ROI) [7, 8].
Portable Camera Based Assistive Text … 121

Fig. 3 Region of interest

Sometimes multiple scales, fonts and colors square measure won’t to write the
text characters. Conjointly immense quantity noise is contained within the captured
pictures that contain text instead of industrial product boxes with multiple ornamental
patterns the pictures with straightforward background, commonplace fonts, tiny vary
of font sizes and well organized character square measure used for these systems,
systems provide higher performance however scene pictures with advanced back-
grounds aren’t directly handled some optical character recognition(OCR) software
system shown in Fig. 3.

3 Proposed System

In existing system camera capture the text on image and so process is finished in
portable computer and voice output is created however it’s not possible for visually
handicapped person to hold the portable computer every time. To overcome the
matter of existing system the new system is introduced that’s transportable camera
primarily based golem app to helps the visually handicapped person. We projected
golem app this app facilitate the visually handicapped person to browse the text on
explicit object. In this system camera capture the actual text on image and multiple
techniques square measure applied on it text and finally voice outputs are generated
and so visually handicapped person will simply listen the text on it explicit object.
122 S. Thigale and R. B. Kagade

4 Framework and Algorithm

Gray scale
In Fig. 4 photography and computing, a grayscale or gray scale digital image is a
picture during which the worth of every component could be a single sample, that is,
it carries solely intensity data. Pictures of this type, conjointly referred to as black-
and-white, area unit composed solely of reminder grey, varied from black at the
weakest intensity to white at the strongest.
How does one convert a color image to grayscale? If every color component is
represented by a triple (R, G, B) of intensities for red, green, and blue, however
does one map that to one range giving a grayscale value? There area unit following 3
algorithms. The lightness methodology averages the foremost distinguished and least

Fig. 4 Framework and algorithm

Portable Camera Based Assistive Text … 123

distinguished colors: (max(R, G, B) + min(R, G, B))/2. The average methodology

merely averages the values: (R + G + B)/three.
The luminousness methodology could be a additional refined version of the
common methodology. It conjointly averages the values, however it forms a weighted
average to account for human perception. We’re additional sensitive to inexperienced
than different colors, therefore inexperienced is weighted most heavily.
The formula for luminousness is 0.21 R + 0.71 G + 0.07 B.
The example sunflower images in Fig. 5.
Threshold
Thresholding is that the simplest methodology of image segmentation.

Original image

Lightness

Average

Luminosity

Fig. 5 Flower
124 S. Thigale and R. B. Kagade

• From a grayscale image, thresholding is accustomed produce binary pictures i.e.

image with solely black or white colours.
• It’s typically used for feature extraction wherever needed options of image area
unit regenerate to white and everything else to black. (Or vice versa) refer above
Fig. 5.

5 Conclusion

In this paper we’ve represented a system that scans written text on hand-held objects
for aiding visually handicapped people. So as to resolve the common aiming draw-
back for blind users. We tend to plan mobile apps. The mobile camera captured
specific images and applied the text recognition on the image to extract and localize
the written text. We planned an OCR software package. This OCR software package
performs varied techniques for manufacturing the audio output for the visually hand-
icapped person. Then visually handicapped people simply scan the text from the
object.

References

1. Rajkumar N, Anand MG, Barathiraja, Portable camera-based product label reading for blind
people. Department of electrical and electronics engineering, The Kavery Engineering College,
Mecheri, Tamil Nadu, India
2. Yi C, Tian Y, Arditi A Portable camera-based assistive text and product label reading from
hand-held objects for blind persons
3. Text detection in natural scene images by Stroke Gabor words, Yi C Dept of Computer Science,
The graduate center, City Univ of New York, New York, U.S.A., e-mail: CYi@[Link].
Tian Y Dept of Electrical Engineering, The City College and Graduate Center, City Univ of
New York, New York, U.S.A., e-mail: ytian@[Link]
4. Chen X, Yang J, Zhang J, Waibel A Automatic detection and recognition of signs from natural
scenes
5. Kumar S, Gupta R, Khanna N, Chaudhury S, Joshi SD Text extraction and document image
segmentation using matched wavelets and MRF
6. Yang X, Tian Y, Yi C, Arditi A Context-based indoor object detection as an aid to blind
persons accessing unfamiliar environments. Dept of Electrical Engineering, The graduate Center
Arlene R Gordon Research Institute, the city college of New York. The City University of New
York Lighthouse International NY 10031, USA NY 10016, USA New York, NY 10222, USA
{xyang02,ytian}@[Link] cyi@[Link] aarditi@[Link]
7. Experiments with a New Boosting Algorithm., Yoav Freund and Robert E. Schapire,
AT&T Laboratories 600 Mountain Avenue, Murray Hill, NJ 07974-0636, yoav,
schapire@[Link]
8. Assistive Text Reading from Complex Background for Blind Persons, Yi C Dept. of
Computer Science, The Graduate Center, City Univ of New York New York, U.S.A. e-mail:
CYi@[Link], Tian Y Dept of Electrical Engineering. The City College and Graduate
Center, City Univ of New York, New York, U.S.A
Automatic System for Identifying
Cholesterol

Mohua Biswas, Pragtee Tathe, Geeta Unhale, and Papiya Biswas Datta

Abstract At the outer edge a solid white ring circling iris exist which is known
as sodium ring or deposit normally located around iris of an eye which symbolizes
the existence of high level cholesterol in the human body. There is a half-circle of
gray, white and yellow deposits in the outer edge of cornea known as Arcus senilis
or arcus senilis cornea occurred because of fat and cholesterol deposits in people
under 45 years of age. The high cholesterol is also an indication of the presence of
hyperlipidemia which indicates the increased amount of fats in the blood. There is a
risk of developing heart disease which leads to stroke and death. There is a chance to
occur a common type of heart disease known as coronary artery disease (CAD) when
the arteries that supply blood to heart muscle become hardened and narrowed. This
is also due to the buildup of cholesterol. Iridology approach is the different type that
also helps to identify diseases with the help of pattern of an iris. Whenever there is a
excess deposition of the cholesterol occur in the body it will create a whitish sodium
ring around the iris, By analyzing the sodium ring we can identify the existence of
cholesterol in the human body.

Keywords Cholesterol · Coronary · Arcus senilis · Iridology

1 Introduction

In the human body there is a presence of a slippery or waxy substance known as

Cholesterol. It is required to develop cells in human body. But excessive of it may
affect in human body. Normally it occurs in two ways. First the human liver gener-
ates the required amount of cholesterol in human body and the remaining amount of
cholesterol is from food which we digest. Excessive amount of it leads to heart attack.
It is a combination of steroid and alcohol having the chemical formula C27H46O. An

M. Biswas (B) · P. Tathe · G. Unhale

SVERI’s College of Engineering, Pandharpur, Solapur, India
e-mail: msbiswas@[Link]
P. B. Datta
Devi Mahalaxmi Polytechnic College, Titwala, Kalyan, Thane, India

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 125
P. M. Pawar et al. (eds.), Techno-Societal 2020,
[Link]
126 M. Biswas et al.

approach based on laser helps to calculate the cholesterol in human blood through
skin. In order to classify the cholesterol in the skin the infrared (IR) absorption spec-
troscopic is proposed [1]. Generally, there is a presence of cholesterol of approx-
imately 11% in skin because of weight. It gets increased if there is a symptom of
coronary artery disease. In that case for analyzing coronary artery disease palm test
is not so effective as a screening tool. The blood test or lipid profile test is the modern
method to analyze the level of cholesterol. Before lipid profile test there is a need of
12 h fast. This test is tested by observing variations in the pattern of iris which helps
to identify the increase level of cholesterol known as Arcus senilis. Arcus senilis is
a half-circle of gray, white and yellow deposits in the outer edge of cornea. Arcus
senilis is caused due to the lipid deposit in the inner layer of the cornea peripheral.
Because of increase amount of cholesterol there is chance of occurring arcus juve-
nilis problem which indicates discoloration of an eye [2]. Thus for analyzing and
detecting the level of cholesterol in the body of humans’ iris pattern is the alternate
and effective approach known as iridology which was found by Bernard Jensen’s.
Many iridologists admitted that the occurrence of heart disease is due to cornea
having yellowish-white deposit. The high level of cholesterol in the blood is known
as Hypercholesterolemia. It is necessary to check the level of cholesterol regularly
which helps to get rid of the formation of gallstones with bile pigments that normally
occurs because of high level of cholesterol. Protein in human body helps cholesterol
to flow in human body and it is known as lipoprotein. Cholesterol is distinguished in
two ways depend upon the amount of protein present: (i) Low Density Lipoproteins
(LDL) and (ii) High Density Lipoproteins (HDL).

2 Human Iris

The iris of the eye is a thin having 12 mm in diameter and rounded in shape consists of
connective tissue and muscle that encircles the pupil. The Fig. 1 shows the schematic
diagram of a human eye. There is a lens behind the iris of the eye. At the back of the
eye on the inside there are linings covered by cells for light-sensing known as retina
[3]. Retina converts light into electrical impulses. These impulses are transferred to
the brain by optic nerve. The coloring portion of the human eye manages the light to

Fig. 1 Schematic
representation of human eye
Automatic System for Identifying Cholesterol 127

enter. The iris of the eye shut the pupil if the light is bright and because of that less
amount of light passes. If the light is low then iris opens the pupil that allows light to
pass. The size of the pupil is managed by the sphincter and the dilator muscles that
vary from 10% to 80% [4].

3 Iridology

The term iridology which is also referred as iridodiagnosis or iridiagnosis is an impor-

tant approach or method that helps to test iris of a human eye in order to find out
patient’s health. Many experts test their estimation to iris diagram which is distributed
into various regions that correlate to different human body parts. Iridologists consid-
ered human eyes as “windows” that reflect the present condition of human health.
The color of iris is greenish-yellow that encircles the pupil of black color. Sclera is
the white portion of the eye. Usually for examining iris iridologists use torch or hand
held lens for tissue change. The iris of an eye is the basic representation of human
body that speaks about specific transformation in human organs. The iris at the right
hand side as shown in Fig. 2 demonstrates the situation of the organs at the right side
of human body and the left hand side of the iris demonstrates the situation of the
organs at the left side of human body [5].
In order to examine the cholesterol in human body, Iridology is the substitute
for treatment where individual’s health and the condition of organs in the body are
reflected. As per the iridology chart, the detection of cholesterol in human body is
simply by observing the sodium whitish ring present in the patient’s eye. Also, due
to the presence of cholesterol in human body there is a variation in the iris pattern
and known as Arcus Lipoides (Arcus Senilis or Arcus Juvenilis) [6]. Some of them
is shown in Fig. 3.

Fig. 2 Relationship cholesterol presence with arcus senilis

128 M. Biswas et al.

Fig. 3 Example of iris image

4 Methodology

The following the approaches that are supported by the automated system in this
paper are shown in Fig. 4:

A. Image Recognition or Acquisition:

Generally, image recognition or acquisition is the initial phase of vision system.
It is a process by which the image is gathered with the help of photographic
equipment and that images are saved in the database. But the authentic sample
of an eye image is gathered from sufferer at the clinic of cardiology and ophthal-
mology. Database consists of both the cases normal as well as exceptional cases
gathered from clinic or from internet. The existence of cholesterol in the body
is detected by a sodium ring which is normally occurred in corneal part of the
iris [7].

Fig. 4 Methodology
Automatic System for Identifying Cholesterol 129

Fig. 5 Captured eye image

B. Pre-processing:
The second phase is image preprocessing. The Fig. 5 shows captured image for
second phase. As we normally gathered images from various sources there may
be existence of distortion. So the purpose of image pre-processing is to improve
the distorted images to get better quality image with the help of enhancement
techniques. In this phase histogram equalization method is used to enhance the
contrast of an image.
C. Image Localization or Segmentation:
The image localization or image segmentation is the most critical phase. It is
a method of detecting interior and exterior border regions of the iris. By using
subtraction approach circular shaped iris is detected and extracted from it.
D. Image Normalization:
As the phase of segmentation ends the image normalization phase starts where
it reconstruct the image of an iris. In this phase it unwrap the circular region
of the iris to rectangular block with the help of Daugman’s rubber sheet model
method. Evaluation can be done either from top to bottom or from bottom to
top.
E. Region of Interest:
In this phase of extraction the portion of the iris image is extracted which we
want to refine by simply cropping the normalized iris image. The region of
sodium ring is only extracted.
F. Feature Extraction:
After the region of interest is carried out the next phase is to find out the various
features such as Mean, Entropy, Standard Deviation, Smoothness, Kurtosis,
Variance and Homogeneity related to this region. The results are displayed in
two ways; one is the normal conditions and other is the exceptional cases of
cholesterol.
G. Classification: (Keep text style as normal)
Figure 6 show the classifier used for next stage of image processing based on the
result of feature extraction we classify the level as normal level and cholesterol
level which is beyond normal level [8].

5 Result

In Fig. 7, the above result shows the implementation of our methodology tested on
patient suffering from cholesterol. In this two images of an iris are captured; one is
colour image and other is black and white. In the case of localized iris image the
130 M. Biswas et al.

Fig. 6 SVM classification

Fig. 7 Stages for cholesterol detection for Abnormal Iris

white portion i.e. sodium ring is present in brown fiber like area. The sodium ring is
segmented and by analyzing the ROI we get the conclusion that there is a presence
of cholesterol in human body beyond the safe level.
In Fig. 8, there is absence of sodium ring in brown portion which means no
cholesterol in the body.
Automatic System for Identifying Cholesterol 131

Fig. 8 Stages for cholesterol detection for Normal Iris

6 Conclusion

In this paper automatic identification of cholesterol in human body through iris is

presented. This system requires less time to identify the existence of cholesterol in
human body. In biomedical field it is an important application.

References

1. Li L-H, Dutkiewicz EP, Huang Y-C, Zhou H-B, Hsu C-C (2019) Analytical methods for
cholesterol quantification. Sci Direct 27(2)
2. Arora A, Paggowal N, Chawla D, Sharma E (2018) A non-invasive framework for detecting
hypercholesterolemia using computer vision. IJEDR 6(4). ISSN: 2321–9939
3. Songire SG, Joshi M (2016) Automated detection of cholesterol presence using iris recognition
algorithm. Int J Comput Appl
4. Shanker NR, Ezhil A, Archana S (2012) Non-invasive method of detection of cholesterol using
image processing. Int J Med Eng Inform
5. Silambarasan T, Sethupriyan P (2015) Identification and cholesterol estimation of skin using
hand pattern image. Int J Adv Res Electr Electron Instrum Eng 4(8)
6. Mahesh Kumar SV, Gunasundari R (2013) Comparative Analysis and Earlier Detection of
Cholesterol Based on the Measurement of Various Lipid Level Deposits in the Eye. CIIT 5(7)
7. Adi KG, Rao PV (2016) A non-invasive method of detection of cholesterol by level set segmen-
tation and ANN as classifier using MLP-BP and RBF. Int J Adv Electr Comput Eng (IJAECE)
3(3)
8. Adi KG, Rao PV, Adi VK (2015) Analysis and detection of cholesterol by wavelets based and
ann classification. In: 2nd International conference on nanomaterials and technologies (CNT
2014)
Design of an IoT Based System
for Monitoring and Controlling
the Sub-Station Equipment

Pranali Bodke and A. A. Kalage

Abstract In the era of modern digitalization world, it is a simple to monitor and

control the substation equipment remotely using expensive PLC and SCADA system,
but it is desirable to design a system which is cost-effective, smart and reliable. So that
IoT is an effective solution as the real-time capability of IoT is considered as a key
feature for monitoring and control applications of power systems. The IoT is a system
of interrelated computing devices, mechanical and digital machines, objects, animals
or people that are provided with unique identifiers and the ability to transfer data over a
network without requiring human to human or human to computer interaction. These
devices capable of interacting with one another directly or indirectly and also data
collection are performed locally or remotely via centralized servers or cloud-based
applications. This paper aims to design a low-cost energy monitoring and controlling
system using IoT devices. This paper shows a result of the effective use of IoT devices
in a power system to obtain better efficiency with less time. The use of IoT devices
improves the power system performance remotely without any human intervention.
In this project, the prototype of the system using Raspberry pi has been designed. The
use of Raspberry pi reduces manpower and maintenance cost. It performed mainly
two functions such as oil quality and oil level sensing and transformer differential
protection.

Keywords Internet of things · Power system monitoring and controlling · Oil

quality and quantity sensing · IoT based differential protection

1 Introduction

Remote monitoring and controlling the sub-station equipment is an essential issue

for the power/energy management department which is normally done manually or
using an expensive PLC and SCADA system. With the emergence of the internet
and computational era, a smart monitoring and reliable controlling system over the
entire sub-station equipment is highly desirable that can be achieved by introducing

P. Bodke (B) · A. A. Kalage

Sinhgad Institute of Technology, Lonavala, India

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 133
P. M. Pawar et al. (eds.), Techno-Societal 2020,
[Link]
134 P. Bodke and A. A. Kalage

the Internet of Things (IoT) technology [1, 2]. IoT is the network of physical devices
embedded with electronics, software, sensors, actuators and network connectivity
which have the ability to identify, collect and exchange the data. Each thing is
uniquely identifiable through its embedded computing system and able to inter-
operate within the existing internet infrastructure [3]. An IoT based network strategy
for monitoring and controlling the sub-station equipment provide efficient time and
resource management [4, 5]. This project mainly focused on to indicate the quality
and present amount of oil in the transformer & oil circuit breaker (OCB) at a remote
location without being physically present so that proper corrective actions must be
taken by the operator. This proposed system represents an alternative way to provide
differential protection where the distance between two sides of the equipment does
not affect system performance.
This system’s objectives are to proposed IoT-based smart monitoring & control-
ling system using Raspberry-Pi module, which will enable the large data storage on
the cloud at a very low cost so that live tracking can be done.

2 Proposed System Block Diagram

Figure 1 shows the block diagram of the proposed system. The system aims to provide
an operator at a remote location with sufficient information to determine the quality
and present amount of oil in the transformer & oil circuit breaker (OCB).
Even though monitoring & controlling, the system represents an alternative way to
provide differential protection where the distance between two sides of the equipment

Fig. 1 Proposed system block diagram

Design of an IoT Based System for Monitoring … 135

has no effect on system performance. An alarm message with what corrective action to
be taken will be provided to the operator at any dangerous condition. Any authorized
person can get mail alert & perform the same tasks by visiting the power system’s
official webpage. It uses the different sensors to sense the transformer oil quality and
quantity as well as the current sensor has been used, and the data is sent to the cloud
server through Wi-Fi network. The advantage of this system is that the data can be
saved and the real-time data can be obtained at any time. Once the data is received
it analyses and checks for the respective action and signal are sent to the actuator.
The major part of the system is as follows:
A. Sensor
The sensors are used here to sense the transformer/OCB oil quality and oil level
also the current sensor to sense the current of a current transformer.
• Transformer or OCB Oil Level Sensor: The HC-SR04 Ultrasonic distance
sensor is used to sense the oil level of the transformer or OCB.
• Transformer oil quality sensor: The TSSP40 infrared sensor is used. An IR
sensor consists of an IR LED and an IR Photodiode. When the IR transmitter
emits radiation, it reaches the object, and some of the radiation reflects back
to the IR receiver.
• Current sensor: The ACS712 is a fully integrated, hall effect-based linear
current sensor with 2.1kVRMS voltage isolation and an integrated low-
resistance current conductor.
B. Actuators
Actuator performs the tasks that are required by the operator at a remote loca-
tion. Servo motor is connected to Raspberry-pi module which performs the
operation as per the instructions received.
C. IoT based differential protection
The differential protection is one of the most commonly used protection
schemes for electrical equipment in power system substation. But due to the
cost of pilot wire, this protection scheme can’t be useful for every system.
One of the major disadvantages of pilot wire differential protection is that as
the length of pilot wire increases the effect of capacitance on system perfor-
mance increases. IoT based differential protection is a smart solution where
two modules send the secondary current data to the webserver. When the data
coming from two different terminals of the protected zone differs from each
other beyond permissible limit, the webserver sends a command to an internet
module which initiates the CB operation to protect the system. So thus, the
problem associated with a pilot wire can be eliminated. To implement practi-
cally IoT based differential protection, two current sensors readings are taken
and the difference of this current is continuously monitoring.
136 P. Bodke and A. A. Kalage

Fig. 2 Proposed system circuit diagram

3 IoT Technology

Internet of things IOT consists of two words Internet and Things. The term things
in IoT refers to various IoT devices having unique identities and have capabilities to
perform remote sensing, actuating and live to monitor of certain sort of data [6, 7].
IoT devices are also enabled to have a live exchange of data with other connected
devices and application either directly or indirectly, or collected data from other
devices and process the data and send the data to various servers.

4 Proposed System Circuit Diagram

In this system, we are using a Raspberry-Pi 3 Model B+ controller with IoT tech-
nology using Wi-Fi 8266 module. From Fig. 2, the ultrasonic sensor and current
sensor, both data’s are stored in the cloud, and these data’s are controlled using
Raspberry-Pi. Finally, the sensors provide the data continuously to the consumer [8].

5 Proposed Methodology

The proposed system is to monitoring and controlling of substation equipment

remotely using IoT. After the successful connection of all the sensors to the rasp-
berry pi module system check for instructions continuously [9]. Once the fault occurs,
instructions are received by mail and corrective action must be provided by actuators.
This module sends the data’s to raspberry-pi through IoT and to operate using Email
or web page. This IoT system monitoring the substation parameters like current,
voltage, and oil level inside the transformer or OCB and provides continuous data to
Design of an IoT Based System for Monitoring … 137

the operator. After getting the correct data from the sensors, they can easily able to
operate substation parameters such as CB and relay [10]. For data exchange MQTT
protocol is used. MQTT is fast becoming one of the main protocols for IoT (Internet
of Things) deployments. The Linux operating system is used.
The proposed system uses different sensors to sense the transformer oil quality
and quantity as well as the current sensor has been used, and the data is sent to the
cloud server through Wi-Fi network. The advantage of this system is that the data
can be saved and the real-time data can be obtained at any time. Once the data is
received, it analyses and checks for the respective action and signal are sent to the
actuator.

6 Results

After the successful connection of communication devices to the server, oil tank
vacancy level and CTs current are sent to the webserver in an automatic manner. At
the violation of their normal values, an alarming message with a corrective step to be
taken to solve the specific problem was sent to the webserver. We can continuously
monitor the substation parameters on ThingSpeak UI.
Figures 3 and 4 show the command Email in order to control the servo motor (CB
state) and to change the tap position of the transformer from the webpage. Based on
the experiment results, it is terminated that the proposed new IoT based sub-station
controlling and monitoring system worked successfully.

Fig. 3 Email alert for current differential protection

138 P. Bodke and A. A. Kalage

Fig. 4 Email alert for low oil level indication

7 Conclusions

In this paper, the IoT based smart system has been developed for remote monitoring
and controlling the entire sub-station equipment, which is very reliable, user-friendly
and low cost as compared to the conventional system. The proposed approach is
an absolutely automatic system that includes self-checking of oil level from the
transformer/oil circuit breaker, continuous sensing of two CTs secondary current,
sending data to the webserver, storing and displaying data in the web page, and
sending a comment to an internet module for performing the specific task such as
initiating the CB operation, tap changing the transformer, and so on.
Also, it created a new pathway to take necessary measures priory in case of any
emergencies that may occur in substations. In future, the system can be enhanced
with additional features for automation of the tasks at the substation.

References

1. Hossain MS, Rahman M, Sarker MT, Haque ME, Jahid A A smart IoT based system for
monitoring and controlling the sub-station equipment
2. Balamurugan S, Saravanakamalam D (2017) Energy monitoring and management using
internet of things. In: Proceedings of the international conference on power and embedded
drive control (ICPEDC), Chennai, pp 208–212
3. Morello R, Capua CD, Fulco G, Mukhopadhyay SC (2017) A smart power meter to monitor
energy flow in smart grids: the role of advanced sensing and IoT in the electric grid of the
future. IEEE Sens J 17(23):7828–7837
4. Hlaing W, Thepphaeng S, Nontaboot V, Tangsunantham N, Sangsuwan T, Pira C (2017)
Implementation of Wi-Fi-based single-phase smart meter for the internet of things (IoT). In:
Design of an IoT Based System for Monitoring … 139

Proceedings of the international electrical engineering congress (IEECON), Pattaya, pp 1–4

5. Arun Chandra P, Mohith Vamsi G, Srimanos Y, Mary GI (2018) Automated energy meter using
Wi-Fi enable raspberry Pi. In: IEEE international conference on recent trends in electronics,
information & communication technology
6. Ramakrishnan R, Gaur L (2016) Smart electricity distribution in residential areas: internet of
things (IoT) based advanced metering infrastructure and cloud analytics. In: Proceedings of
the international conference on internet of things and applications (IOTA), Pune, pp 46–51
7. Pau M, Patti E, Barbierato L, Estebsari A, Pons E, Ponci F, Monti A (2018) A cloud based
smart metering infrastructure for distribution grid services & automation. Sustain Energy Grids
Netw 15
8. John A, Varghese R, Krishnan SS, Thomas S, Swayambu TA, Thasneem P (2017) Automation
of 11 kV substation using raspberry pi. In: Proceedings of the international conference on
circuit, power and computing technologies (ICCPCT), Kollam, pp 1–5
9. Li L, Ota K, Dong M (2017) When weather matters: IoT-based electrical load forecasting for
smart grid. IEEE Commun Mag 55(10):46–51
10. Chooruang K, Meekul K (2018) Design of an IOT energy monitoring system. In: International
conference on ICT & knowledge engineering, November 2018
Implementation of Iridology for Pre
Diagnosis of Brain Tumor

Pragtee Tathe, Mohua Biswas, Anup Vibhute, Geeta Unhale,

Mrunmayi Raut, and Papiya Biswas Datta

Abstract Human body is the magical creation of god. It carries many intercon-
nected systems. Changes in one system replicate changes in another system. Due to
such interconnection of systems, we can analyze one system by observing changes
in another system. Iridology supports the same theory. Iridology tells the relation
between iris and other systems present in the body. Evaluation can be done in the
form of the iris that speaks about the physical condition of different body parts. In
the proposed method by observing different iris images without doing any compli-
cated and time-consuming test, we can perform diagnosis for the brain tumour. This
method can be used as a pre diagnosis tool.

Keywords Iris · Iridology · Pre diagnosis · Brain tumour

1 Introduction

One of the best creations of god is the human body. The human body contains many
organs, and these organs are linked with each other. One of the important organs in
the human body is the eye which contains many parts like sclera which is whitish
outer background, iris which is middle central ring may be brown, black or blue and
the black centre pupil. The branch of science which focuses on the study of the iris is
called Iridology. Iridology, also known as iridodiagnosis, is a technique in which the
iris images are analyzed with different aspects. The technique of iridology is based

P. Tathe (B) · M. Biswas · A. Vibhute · G. Unhale

SVERI’s College of Engineering, Pandharpur, India
e-mail: pbtathe@[Link]
Solapur University, Solapur, India
G. Unhale
e-mail: ggunhale@[Link]
M. Raut
FTC COER Sangola, Solapur University, Solapur, India
P. B. Datta
Devi Mahalaxmi Polytechnic College Titwala, Kalyan, Thane, India

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 141
P. M. Pawar et al. (eds.), Techno-Societal 2020,
[Link]
142 P. Tathe et al.

Fig. 1 Iris chart for right and left iris

on the theory that different body parts are related to different regions in the iris. The
human iris is categorized into 90 different regions, and each region replicates the
status of different human body parts. Whenever there is a change in the body organs,
it is reflected in the iris. The form of the iris varies according to the condition of body
parts. The right eye iris replicated the status of right side body organs and left eye
iris replicates the status of left body organs [1, 2]. With several nerves connected,
the iris is linked with body parts as shown in the iris chart in Fig. 1. It indicates the
distribution of different organs over the iris circle.
The well solidly organized fibers of iris are the indication of a healthy body. The
unhealthy body has disturbed solidly organized fibers and makes them slackly as
shown in Fig. 2.
A brain tumour is a major disease that has affected over many million people across
the globe; the rate of people getting affected will exponentially increase in the coming
years. There are different methods for detecting the brain tumour like MRI scan, CT
scan etc.; these techniques are time-consuming; it also needs specially trained persons
to handle it. The major thing related to these techniques is that the patients at the initial
stage are not going to look forward towards these techniques. The proposed project
aims to construct a graphical user interface that allows any physician to predict brain
health with user-friendly and fast diagnosis techniques. These techniques integrate
various image processing steps together in order to complete the process of diagnosis
as shown in Fig. 3.
Implementation of Iridology for Pre Diagnosis of Brain Tumor 143

Fig. 2 Schematic representation of normal and abnormal human eye

2 Methodology

A. Capturing of eye images

In this approach, the images of the eye are collected by using a high-resolution
camera and with these images a database is formed as shown in Fig. 4. The database
is segregated into two parts: one part contains images of iris of a healthy person, and

Fig. 3 Methodology for iris

diagnosis 1. Capturing of Eye Image

2. Pre-processing of Eye Image

3. Sectionalisation

4. Standardisation of iris Image

5. Taking ROI

6. Feature Extraction

7. Classification
144 P. Tathe et al.

Fig. 4 Captured eye image

another part contains images of iris of brain tumour suffered person. The database
can also be formed by using images collected from an online database such as CASIA
database, MMU database, and UBIRIS database [1]. Normally, in-hospital ophthal-
mology department is dealing with the iris images. For the proposed work database
is created by taking the total 60 images out of those 30 images are of a normal
person, and 30 images are of persons with a brain tumour. The below figure shows
the captured image of an eye.
B. Pre-processing of eye images
The pre-processing is the next step in the methodology of the proposed system. In this
stage, whatever the captured image is there, it has to pass through various filtering
algorithms. After passing through these filtering algorithms, the image quality gets
improved because this stage scales back the presence of noise within the iris image.
This stage is very important to make end results more relevant than first. For the iris
images, adaptive filtering is used [1, 3–5].
C. Sectionalisation
Sectionalisation is the process to seek out interior and exterior precincts of the iris.
The circular Hough transforms often won’t to comprehend the radius and centre
coordinates of the pupil and iris sections. By subtracting pupil from the sclera, we will
be able to fix the iris. Once the iris region is segmented from an eye, the subsequent
step is to modify the iris region into fixed dimensions. After subtraction, we will get
the iris pattern into a circular shape ring [1, 6, 7].
D. Standardization of the iris image
Daugman’s rubber sheet model (as represented in Fig. 5) is employed for standard-
ization of Iris ring after standardization of iris ring; the ring is transformed into a
rectangular shape were to represent the points on the ring we can use x and y coordi-
nates. The relation between preceding and the following image is specified with the
relation between the coordinate systems [1, 3, 8].

Fig. 5 Daugman’s rubber sheet model

Implementation of Iridology for Pre Diagnosis of Brain Tumor 145

Fig. 6 Taking out ROI from a standardized iris image

E. Taking out ROI

With reference to the point mentioned in the introduction part that different sections
of the iris are related with specific body parts so for performing analysis of brain
tumour we need to crop the area from the normalized iris. As we have divided the
iris area into sectors, we only need to crop the sector, which represents the brain
area. In the circular iris, if we consider a specific point as a starting point then with
respect to that, we have to select coordinates for a brain tumour. Depending upon
which point we have selected as a starting point to convert into a rectangular shape,
the coordinates values are going to change. If we compare the iris with the image of
a clock, then the zone from 11 to 1 represents the brain area. So accordingly, we can
take the region of interest [2, 9] as shown in Fig. 6.
F. Feature extraction
Once we get the significant region subsequently, we can find out different
features for that region. The extracted features are as follows [1, 3].
1. Mean: The average level of intensity of the image.

G−1
Mean: μ = i p(i)
i=0

2. Variance: Variance describes the variation of intensity around the mean

G−1
V ariance: σ 2 = (i − μ)2 p(i)
i=0

3. Skewness: It is the measure of the asymmetry of the probability distribution

of the random variable.
146 P. Tathe et al.

G−1
Skewness: μ3 = σ −3 (i − μ)3 p(i)
i=0

4. Entropy: It is the amount of information present.

G−1
Entr opy: H = − p(i)ln[ p(i)]
i=0

5. Kurtosis: It is the measure of the flatness of the histogram of the image.

G−1
K ur tosis: μ4 = σ −4 [(i − μ)4 p(i)]−3
i=0

6. Energy: It is the measure of the brightness of the histogram of the image

G−1
Energy: E = [ p(i)]2
i=0

G. Classification
Once we have calculated different features for normal as well as abnormal iris ROI
images, the next step is to train the classifier. Nowadays there are different classi-
fiers which we can use, but out of all support vector machines is more preferable
because it has the kernel trick to transform our data and with the help of which an
optimal boundary is detected for preferable outcomes. So nonlinear SVM classified
the images into two classes; one indicates healthy human iris, and other indicates iris
affected by a brain tumour [1, 10].

3 Result

Figure 7 shows the various operations which are mentioned in the methodology on
normal eye images. If we consider the sequence of operations, then in the first step
eye image is taken with the help of the high-resolution camera. Then that image is
converted into grayscale or in black and white format. Then using the segmentation
method iris area is detected. Once the iris ring is detected, then it is converted into
a rectangular shape, and then the specified area for a brain tumour is separated
by the cropping method. Once we get the ROI then the different features listed in
methodology are calculated, Table 1 shows the values of different features.
In Fig. 8 once our features get calculated depending upon the values, classifier
gives the result. Figure 9 shows the result of the classifier for normal Iris images.
Figure 10 shows the various operations which are mentioned in the methodology
Implementation of Iridology for Pre Diagnosis of Brain Tumor 147

Fig. 7 Various operations on input eye image to find out ROI of brain tumour

Table 1 Values of different

Features Values
features calculated for normal
iris Mean 94.9621
Standard deviation 64.7603
Entropy −0.3473
Smoothness −0.0645
Skewness 0.0319
Contrast 1.0000
Kurtosis 0.0508
Homogeneity 0.0145
Variance 0.5039

on brain tumour affected people eye images. If we consider the sequence of opera-
tions, then in the first step eye image is taken with the help of a high-resolution camera.
Then that image is converted into grayscale or in black and white format. Then using
the segmentation method iris area is detected. Once the iris ring is detected, then it is
converted into a rectangular shape, and then the specified area for a brain tumour is
148 P. Tathe et al.

Fig. 8 Colour feature and texture feature distribution for normal iris images

Fig. 9 Decision of classifier

separated by the cropping method. Once we get the ROI then the different features
listed in methodology are calculated, Table 2 shows the values of different features.
Figure 11 shows the plot of values of different features calculated for abnormal
iris ROI image.
Once our features get calculated depending upon the values of features, classifier
gives the result. Figure 12 shows the result of the classifier for brain tumors affected
person’s Iris image.
Once we train the classifier with 30 iris images and we check the system for the
remaining 30 images then out of that for 25 images, we will get accurate results. So,
we can say that for the above system we get about 83% of accuracy.
Implementation of Iridology for Pre Diagnosis of Brain Tumor 149

Fig. 10 Various operations on input eye image to find out ROI of brain tumour

Table 2 Values of different

Features Values
features calculated for brain
tumor affected iris Mean 121.5513
Standard deviation 68.3303
Entropy −0.4531
Smoothness −0.0718
Skewness −0.0317
Contrast 1.0000
Kurtosis 0.1207
Homogeneity 0.0102
Variance 0.5177

4 Conclusion

About brain tumour as pre diagnosing stage in an easy and accurate way. As this
technique is non-invasive and less time consuming, it is user friendly to the patients.
If we train the classifier with different variety of iris images, we can also move
towards further accuracy. The proposed system is developed for the pre diagnosing
150 P. Tathe et al.

Fig. 11 Colour feature and texture feature distribution for brain tumour affected iris image

Fig. 12 Decision of
classifier

of brain tumour using Iris images by using MATLAB simulation software. This
method allows any physician to predict.

References

1. Chaskar UM, Sutaone MS (2012, Apr 3) On a methodology for detecting diabetic presence
from iris image. IEEE Xplore. ISSN 12654029
2. [Link] Accesses May 2010
3. Tathe PB, Patil MM (2015) Analysis of health condition based on iris image. IJIERT
ICITDCEME-2015. ISSN: 2394-3696
4. Zheng J (2018) An iris recognition algorithm for identity authentication. IEEE Xplore. https://
[Link]/10.1109/ICITBS.2018.00162
5. Kaur N, Juneja M (2014) A review on iris recognition. IEEE Xplore, INSPEC Accession
Number: 14254396
6. Lozej J, Štepec D (2019) Influence of segmentation on deep iris recognition performance. IEEE
Xplore. [Link]
Implementation of Iridology for Pre Diagnosis of Brain Tumor 151

7. Jayalakshmi S, Sundaresan M (2013) A survey on iris segmentation methods. In: International

conference on pattern recognition, informatics and mobile engineering ear: 2013. Conference
Paper, Publisher IEEE
8. Mohammed A, Al-Gailani MF (2019) Developing iris recognition system based on enhanced
normalization. IEEE Xplore INSPEC Accession Number: 19029194
9. Shanker NR, Ezhil A, Archana S (2012) Non-invasive method of detection of cholesterol using
image processing. Int J Med Eng Inf IJMEI2012.048384
10. Ali H, Salami MJE (2008) Iris recognition system by using support vector machines. IEEE
Xplore, INSPEC Accession Number: 15385486
Wireless Communication Using Light
Fidelity Network

Nimisha Deval, Prajakta Satarkar, Akshata Jadhav, and Rupali M. Shinde

Abstract In current period, Light fidelity network is becoming very famous and
catching the interest of some of customers with that one advanced technology-based
totally features. It brings a brand new decision inside the Wi-Fi communication
exchange and those are showing so much curiosity to recognize approximately it.
Light fidelity is a wireless communication-primarily based expertise which transmits
records over the community through a light source like led’s as opposed to radio
frequency indicators (RF) with very excessive information charge. The aim of paper
Wireless and optical networks are widely used nowadays because the network perfor-
mance is a very important issue to supply services to a good variety of users whereas
reassuring users’ quality of service necessities. It aims to analysis the wireless and
optical networks performance. The learning toward research the act as an Light
fidelity Network system in wireless communication consumes remained created and
achieve a knowledge speed of 10 k rate over 40 cm distances.

Keywords Li-Fi · Network performance · Optical transmission system ·

Simulink · OFDM

1 Introduction

Inside the period of overloaded (statistics conversation) global, light fidelity stands
an innovative technique of wireless communication exchange like stands usages
managed lighting to transfer information wireless. The communication of statistics
remains one among maximum essential each day accomplishments within imme-
diate developing international. The present wireless communication networks that
is connect North America Country toward net square measure very gradually while
multiple gadgets. Also using amount of tools which get entry to the remaining, the

N. Deval (B) · P. Satarkar · A. Jadhav · R. M. Shinde

SVERI’s College of Engineering, Pandharpur, India
e-mail: nddeval@[Link]

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 153
P. M. Pawar et al. (eds.), Techno-Societal 2020,
[Link]
154 N. Deval et al.

availability of secure bandwidth varieties it a good deal greater hard to experience

high records switch fees and to attach a comfy network. Li-fi consumes a miles wider
range for communication compare toward traditional procedure of ratio communica-
tions network that depend on wireless influences. In the simple system over due this
period is that the statistics may be transmitted done controlled light through using
variables light strengths quicker than the social appreciations container observe. This
era makes use of part of the spectrum that is withal not will utilize- the color spec-
trum, in preference to gigahertz wireless communication waves used for proceedings
control.
The knowledge of li-fi became brought aimed at the main time via a German
physicist Harald Hass within the TED Laboratory international speak scheduled
seen wireless communication exchange (VLC) in July 2011, via mentioning toward
the situation as “information through illumination”. They said a standard lamp with
an light to communicate a audio visual of a flourishing bloom that convert then
expected on an show. Now easy relations, li-fi can remain through to light-based
[Link] an alternative of wireless communication waves it makes use of
light to communicate information. Now room of Wi-Fi modems, light fidelity might
us age transmit prepared through controlled able lamp that in addition to transmit
and acquire records [1].
Through including innovative and utilized bandwidth of evident light to the at
present to be consumed wireless communicate impressions for Statistics switch, li-fi
can performance am a in purpose in releasing the light masses which the cutting-
edge wireless appliance is finish. Hence the situation can suggestion extra occurrence
ensemble of the direction of 400 THz as associated to that to be consumed in Radio
Frequency communicate which is around 300 GHz. Similarly, because the li-fi vari-
eties use and observable range, the situation resolve assist improves apprehensions
that the optical can be convey as a waves approaching through wireless fidelity should
to unfavorably have an effect on our health [2].
By means of conversation via visible light communication, li-fi era consumes the
opportunity to exchange in what way charge to the remaining, movement movies,
acquire emails and plenty more. Protection could no longer be a effort as data cannot
be retrieve intimate the absence of light. As per a conclusion consequence, the situ-
ation can be used in excessive safety services areas wherein RF communicate is at
risk of bug somebody’s room.
To aim of paper to analysis the performance optical and wireless network built
proceeding optical transmission method analyze the parameter of performance
network in Light Fidelity [3].
Wireless Communication Using Light Fidelity Network 155

2 Working

2.1 Visible Light Communication

To purpose of paper toward study the presentation optical and wireless network
established on optical transmission method analyze the parameter of performance
network in Light Fidelity as shown in Fig. 1 [3].

2.2 Methodology

Li-Fi primarily built communication method is totally different from light communi-
cation system as a result of Visible Light Communication is simply a point-to-point
communication system. If Li-Fi technology into pictures then by using sender to
receiver through digital transmission [4].
The working principle is simple are using optical communicate to sender and
receiver through messages shown in Fig. 2.

Fig. 1 Light spectrum Wi-Fi or Li-fi

Fig. 2 Optical communicate

to sender and receiver
through messages
156 N. Deval et al.

Fig. 3 Components of light

fidelity

2.3 How Li-Fi Work

There running of li-fi as following. There are two simple components are used:
led light source and light sensor (picture-detector). The light source is picture and
detector on every other end. When led bulb is on it starts off evolved sparkling and
the light sensor senses the light from light source and gets signals now the shape of
binary signals i. e. 1 or 0 [5]. When a few statistics is transmitted over the community
from the net to the consumer tool, it transmits over the network and flashing of led
bulb is a sign of the message after which picture detector feel mild and get hold of
the message and ahead to its vacation spot end by way of displayed in Fig. 3.
• LED, Photo Detector (Light Sensor), Optical Wireless Communication (OWC),
End-user Devices (Laptops, computer), Radio Frequency Signal’s Amplifier, Line
of sight.
More the brighter LED bulb, More frequently data will transmit light signals over
the network and highly reliable.

2.4 Wi-Fi Versus Li-Fi

Comparison of Wi-Fi and Li-Fi [6] is mentioned in Table 1.

Wireless Communication Using Light Fidelity Network 157

Table 1 Parameter
Parameter Wireless fidelity Light-fidelity
Speed 54–250 Mbps 1–3.5 Gbps
Ranges 20–100 m 10 m
IEEE standard 802.11b 802.15.7
Network topologies Point to multipoint Point to point

2.5 Speediness

For Wireless Fidelity, we have rate margins for information transmission Although
Light Fidelity offers exceptionally high-speed of the network, and that we resolve
transfer large records in exactly limited minutes of your period. Quickness for Li-Fi is
10,000 creases quite wireless fidelity abundant more than 1 Gbps are often achieved.

2.6 Security

Radio waves will penetrate the complete manner via walls. These ends in varied
safety worries as they’ll be intercepted while not issue. At an equivalent time as facts
switch aimed at light fidelity is extremely enclosed then innocent. As we to tend
to use seen light-weight not at all sign separates via partitions. Such seen delicate
communication may be used firmly in airplanes without worrying airways indicators.

3 Related Work

The author described this report possesses statistics approximately the visible mild
communicate the use of li-fi technology. This modern-day generation li-fi describes
transmission of “records via illumination” taking the fiber out of fiber optical by way
of sending records through a controlled light bulb that varies in intensity quicker than
the human eye can observe. Discuss of this paper on a Visible Light Communication
(Li-Fi) performed work done by Wate [3].
The author described li-fi is a label for wireless-communication structures using
mild as a service rather of traditional radio frequencies, as in wireless. Li-fi has the
benefit of being able to be used in touchy regions along with in aircraft without
inflicting interference. Discuss of this paper on a The New Era of Transmission and
Communication Technology: Li-Fi (Light Fidelity) LED and TED Based Approach
performed work done by Prakash [7].
The author defined the short boom in devices utilization and application devel-
opments has brought on many researchers to appearance past wireless. The oppor-
tunity of the usage of visible mild for brief variety verbal exchange is explored with
158 N. Deval et al.

the aid of many researchers. Discuss of this paper on a Performance Analysis of

Visible Light Communication System for Free Space Optical Communication Link
performed work done by Shamsudheen [5].
The term li-fi refers to visible mild communicate (vlc) technology that makes use
of as medium to deliver excessive-velocity communicate in a way just like wireless.
Discuss this paper on a Li-Fi–The Future Technology in Wireless Communication
performed work done by Chavan [8].
The author defined this era is quality for high density wireless facts communicate
in restricted area and to overcome radio intervention trouble. In this paper on a Li-
Fi—A Revolution in the field Of Wireless-Communication performed work done by
Parth Vora [1].

4 Li-Fi Network

In light-Fidelity might be a wireless communication expertise, that is uses light to

communicate information then situation among procedures. Now this real-world
passé Li-Fi is a light communiqué method that is skilled and transferring data at
great quickness finished the visible light communication (VLC). On dealings of its
completion users, the expertise are related to Wireless Fidelity the key procedural
transformation existence that Wireless Fidelity uses to radio rate to communicate
information.
The area is struck by number of light attachment, which delivers light for radiance.
Every graceful is focused through a Li-Fi modem before a Li-Fi chip and, then,
similarly works as per a visual improper location otherwise access point (AP). The
optical wireless network positions are related to the core network by great rapidity
compliment networks. The mild fixtures additionally need to include electromagnetic
sensor toward acquire indicators since such stations. The revealing light illuminations
are controlled at in elevation amounts as shown in Fig. 4.

Fig. 4 Light Fidelity environment network

Wireless Communication Using Light Fidelity Network 159

The resultant great radio rate of recurrence flickers which remain abundant devel-
oped than the recharge amount of a laptop display are not evident to the residents of
the area.
Influence as fine as records dismiss remain condition to each light fixture using a
quantity of dissimilar methods.

5 Features of Li-Fi

5.1 Volume

In Wireless communication evidence is communicated over wireless influences that

square measure limited and valued. It’s a restricted information amount, concerning
Li-Fi. By the quickly increasing real world growth of capability like now in era, and
then arranged we have a trend to square measure organization out of range [8].

5.2 Effectiveness

Everywhere stand an out sized variety of cellular wireless improper positions that
put away large amount of drive. Record of the drive is working aimed at conserving
depressed the end position rather than communication. So, effectiveness of such
Radio corrupt positions is extremely little.

5.3 Safety

Numbers appropriation or riding is insignificant compared to Wireless Fidelity

meanwhile the vary of data communication is limited to exact universe and optical.

6 Applications

6.1 RF Range Assistance

Extra ability difficulties of cell systems could also remain fresh burdened to light
fidelity networks wherever offered. That’s especially powerful at the allow wherever
restricted access tend to sit down [9].
160 N. Deval et al.

6.2 Smooth Illumination

Some personal or community light beside road spots may remain used to deliver
li-fi hotspots and then the equal transportations then method foundation could also
remain accustomed show and manipulate illumination and information.

6.3 Harmful Surroundings

Light Fidelity delivers a secure different toward magnetism intervention from

incidence infrastructures in surroundings like sources then organic compound
flowers.

6.4 Clinic and Healthcare

Li-Fi produces no magnetic force intervention consequently doesn’t interfere with

medicinal instruments, neither remains the situation affected through by tomography
scanners [6].

6.5 Subsurface Infrastructures

Since of sturdy sign interest in seawater, roof use is impractical. Audio waves must
extraordinarily small bandwidth then interrupt sea life. Li-fi presents an answer aimed
at quick-variety infrastructures.

6.6 Transportable Connectivity

Workstations, functional phones, doses then different cell devices will interconnect
directly the employment of li-fi. Fast selection hyperlinks provide terribly excessive
records quotes and additionally offers security.
Wireless Communication Using Light Fidelity Network 161

7 Result and Discussion

A sound permit 3 * 3 arrival connected with1 KB existed transferred. In the research

remained accepted out in a 3 m × 3 m × 3 m possibility below usual day bright
situation. The communicated appearance is shown in Fig. 5 [10].
Sender
The Sender contains is a connected with a laptop in distinction to and which statistics
resolve remain communicated. Facts may stand a man and woman otherwise a writing
record or picture reserved at the processor. On the present art effort, rs232 port srem
ain re used to pass the documents since the pc. Rs232 anchorages may remain set
with MATLAB surroundings then may remain second hand to communicate the
information successively. The controlled use exists a silver managed, producing
comprehension now seen range. The varieties method are used is an on–off keying
(ook).
Receiver
Photograph sensors remain recycled aimed at discovered the gray mild, like incor-
porates the communicated statistics. Picture diodes are used as detectors. The sensed
indicator remains before improved the use of transfer impedance amplifier. Inter-
preting the received statistics is executed on the receiver pc the use of MATLAB
2017 programme [11].
Model standon Matlab Simulink R2017b
The whole Model is an carried out in matlab r2017b simulink software program
software. There’s displayed model of optical transmission machine shown in Fig. 6.

Fig. 5 Transmitted image

162 N. Deval et al.

Fig. 6 Model stand on optical transmission system

8 Conclusion

Light fidelity has standard generation in the area of Wi-Fi information transmission.
It’s miles superior techniques of Wi-Fi Communications that use light as facts carrier.
This paper is part of advent of model stand simulation stage of optical spread method
in matlab Simulink model. To investigate maintain and enhance their community
overall performance light fidelity network.

References

1. Vora P, Bhanushali P, Goswam P (2015) Li-Fi—a revolution in the field of wireless-

communication. Int J J Adv Res Field Eng 5
2. Wu X, Basnayaka D, Safari M, Haas H (2016) Two-stage access point selection for hybrid
VLC and RF networks. In: IEEE 27th PIMRC
3. Wate K, Mattani N, Gole A (2013) Visible light communication (Li-Fi). J: IJERT 2(10)
4. Ahriz I, Douin J-M, Lemoine F, Wei A (2018) Performance evaluations in optical and wireless
networks for project 130
5. Shamsudheen P, Kumar S (2016) Performance analysis of visible light communication system
for free space optical communication link. ScienceDirect 24
6. Chatterjee S, Agarwal S, Nath A (2015) Scope and challenges in light fidelity (Li-Fi) technology
in wireless data communication. J: IJIRAE 2(6) (June)
7. Prakash R, Agarwal P (2014) The new era of transmission and communication technology:
Li-Fi (Light Fidelity) LED and TED based approach. J: IJARCET 3(2)
8. Chavan H, Josh A (2016) i: Li-Fi–the future technology in wireless communication. J: ICI2IM
9. Saeed N, Celik A, Al-Naffouri TY, Alouini MS (2019, Nov) Under water optical wireless
communications, networking, and localization: a survey. In: Science Direct, vol 94 Ad hoc
Network
10. Wang Y, Basnayaka DA, Wuand A, Haas H (2017) Optimization of load balancing in hybrid
LiFi/RF networks. IEEE
11. Šalík P, Róka R, Gorazd T (2018) Simulation platform of optical transmission system in matlab
simulink. ScienceDirect
12. Hadi HA (2016) Wireless communication tends to smart technology Li-Fi and its comparison
with Wi-Fi. J: AJER 5(5)
Smart Trolley with Automatic Master
Follower and Billing System

R. Arpita, K. S. Shashidhara, and Veerendra Dakulagi

Abstract The system “Smart Trolley with Automatic Master Follower and Billing
System”, has been developed for shopping malls. Once the client enters the mall, the
trolley follows the client with the help of Bluetooth and Ultrasonic sensors placed
in the cart. All the trolleys are designed with a RFID reader, digital display screen,
Bluetooth and ultrasonic sensors. All the products within the shopping complex
are connected with RFID tags. Once the customer puts any item into the cart, its
distinctive ID range is detected. At the same time, the chosen product name with the
cost, expiry date, discount amount if any available are displayed on the liquid digital
display screen, thereby the value gets added to the whole bill. If the client wants to
take away the item from the cart, the client must press the button which is available
on the trolley and then the client has to scan the product once again and therefore
the cost of that specific product gets subtracted from total quantity bill. Once the
purchase is completed, the customer has to select the upload button and the whole
quantity bill is send to the billing counter. With the assistance of payment app the
payment is completed. Once the bill amount is paid, the data is distributed to the exit
gate and the trolley gets disconnected to the customer and the client can exit the gate.

Keywords Radio frequency identification reader (RFID) · RFID reader · RFID

tags · Atmega 328P microcontroller · Liquid crystal display (LCD) · Mobile
application

1 Introduction

In Indian cities one can see a very big queue at shopping malls on holidays, festi-
vals and weekends. This queue becomes a lot longer throughout large offers and
discounts. As the client enters the mall one must carry the trolley until the end of

R. Arpita · K. S. Shashidhara
Department of ECE, Nitte Meenakshi Institute of Technology, Bangalore, India
e-mail: [Link]@[Link]
V. Dakulagi (B)
Department of E&CE, Guru Nanak Dev Engineering College, Bidar, Karnataka, India

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 163
P. M. Pawar et al. (eds.), Techno-Societal 2020,
[Link]
164 R. Arpita et al.

the shopping that becomes difficult for older individuals. When purchase of all the
things is completed the client must reach the billing counter for billing purpose. If
the barcode scanner techniques are used to scan the items, then this may be a time
intense method. So this leads to long queues at the billing counter. Therefore this
system provides solutions to all these problems. The main objective of this system
is to help the older individuals and to reduce the time spent by the customer in the
shopping mall.

2 Related Work

The automatic billing system has been implemented with RFID technology [1, 2].
The add-on features such as included product detection, product recommendation,
budget setting and automatic billing [1]. Further in [3], the android application has
been developed, where barcode scanner was used to scan the products. This paper
projected a way to decrease the time of the client. In [4], the aim of this paper was
to cut back the shopping time. The barcode scanner and the touchscreen display
were used and the payment was done using payment app like Paytm, UPI. In the
paper [5], the aim of the paper was to modify and accelerate the method by the
utilization of RFID tags on every product and Android application connected to the
microcontroller employing a Bluetooth module [6, 7].

3 Proposed System

The complete system contains of AT mega 328P small controller, RFID reader, RFID
tags, 16*2 LCD, ultrasonic sensing element, Bluetooth, WIFI module is as shown
in Fig. 1. This projected system works as client in the shopping mall, she/he first
takes a trolley car. Each trolley car has ultrasonic device and Bluetooth, as he/she
comes closer to the trolley car, the trolley car gets connected to explicit client and
also the trolley car follows the client. Each cart is connected with a RFID reader,
a microcontroller and LCD screen. Once the client starts dropping product into
the trolley car, tags are scan by the reader and also the reader sends the data to the
microcontroller. The microcontroller compares the data with the information already
keep in it. If the information matches then the price of that product are displayed on
the LCD screen for user. If the user would like to get rid of any product from the cart
then they’ll subtract that product from trolley car and value of that specific product
are subtracted from the overall quantity in a flash and the product information with
total quantity gets transmitted to the central Billing system through wireless network
module. With the assistance of the payment app, the overall bill quantity is often paid.
Once the payment is completed, the data is shipped to the exiting gate. Therefore the
gate opens and also the trolley car gets disconnected.
Smart Trolley with Automatic Master Follower and Billing System 165

Fig. 1 Block diagram of the Trolley system

Arduino, UNO, Bluetooth HC-05, RFID reader module, RFID tags, Ultrasonic
sensor HC-SR04, Power supply, WIFI module. Arduino IDE, Google Firebase, MIT
Inventor.

4 Design and Implementation

The functionality of the system has been explained with the help of flow chart shown
in Fig. 2. Initialize the system by scanning the RFID tags to the RFID reader. Once
the tag is read, it reads the memory of the tag from the data which is already assigned
to the particular tags. The cost, name, the expiry date and the discount of the products
gets displayed on the LCD. All the selected products are scanned in the similar order
and the respective cost of all the items are added to the total bill amount. If any items
are to be removed from the cart, the particular item should be scanned again and
hence the price of the product gets reduced from the total bill amount. At the last,
the final amount is displayed and the total bill is generated.

5 Experimental Results

The Drone unit and the Dropping Mechanism unit have been developed and the
system is successfully coordinated. Initially the drone without payload as shown in
Fig. 3 has been tested successfully.
166 R. Arpita et al.

Fig. 2 Flowchart of the system

Fig. 3 Scanned details

displayed on the LCD
Smart Trolley with Automatic Master Follower and Billing System 167

The client enters the shopping complex and he/she picks the smart trolley. Each
smart trolley has ultrasonic sensors and Bluetooth placed in it so the smart trolley
follows that individual client. Every product within the shopping mall is hooked up
with the RFID tags and every RFID tags have distinctive number that is employed
to unambiguously establish the merchandise. Once the merchandise placed upon the
RFID reader the tag gets known by the reader once it comes inside the antenna vary
and so the product-ID beside the end date and value is shipped via server. Google
Firebase is employed to form and maintain the information at the server finish.
At the server finish, the information is compared with the RFID ID range that
is placed on the smart trolley and therefore the details of those things like Tag ID,
price, and expiry date and discount percentage are compared. If it matches then
the information gets displayed on the liquid crystal display screen. The information
communicated is saved for a flash within the small controlled memory. The client
is able to choose the things with the assistance of the information displayed on the
screen.
If the client would really like to get rid of any item then he/she must scan the
merchandise once more that removes the actual item. The actual item and its price
are aloof from the whole bill quantity. The ultimate quantity displays on the liquid
crystal display.
After finishing the shopping, the client must press the “upload” button. Once the
button is ironed, the whole bill quantity of all things gets generated. The client must
select the payment technique and so created invoice is transferred to payment web
site and so the payment is finished.
Once the payment is finished, the knowledge is transferred to the exit gate. Thus
the client has completed the shopping while not standing in long queues. When
the system is initialized the first message displayed is “SHOPPING CHART” as its
initialization message. When the RFID tags of the products are read by the reader,
the details of the product gets displayed and the amount of each product gets added
to the total bill as shown in Fig. 3.
The Firebase Database is used for assigning and storing the value of the products.
Google Firebase is used to create and maintain the database at the server end as
depicted from Fig. 4.
With the help of ultrasonic sensors, the calculation of the distance between the
trolley and the Customer is done and the calculated value gets displayed on the LCD
screen.
The Ultrasonic sensor situated on the trolley handles establishes the connection
between client and trolley with the help of Bluetooth module. Further with the help
of ultrasonic sensors, the distance between the trolley and the Customer is calculated
and value gets displayed on the LCD screen as shown in Fig. 5. The trolley follows
the customers when the Bluetooth gets connected to that particular person and the
trolley follows that client till the end of the shopping with the aid of ultrasonic sensor
as shown in Fig. 5. In Fig. 6, the calculated distance value is displayed on LCD.
The mobile application shown in Fig. 7 has been developed can be used by both
the shopkeeper and the Customer. Using this app, the seller can update the details of
any products whereas the customer can use this app to make the payments.
168 R. Arpita et al.

Fig. 4 Database created for the products

Fig. 5 Trolley following the customer

Fig. 6 The calculated

distance value displayed on
LCD
Smart Trolley with Automatic Master Follower and Billing System 169

Fig. 7 The Login page of

the mobile app

The trolley follows the customers when the Bluetooth gets connected to that
particular person and the trolley follows that client till the end of the shopping. As
shown in Fig. 5. The mobile application shown in Fig. 7 has been developed can
be used by both the shopkeeper and the Customer. Using this app, the seller can
update the details of any products whereas the customer can use this app to make the
payments.
Once the purchase of items is completed, then the bill can be paid using the mobile
app developed. The Bill summary is displayed in the app as shown in Fig. 8.
The items being purchased is logged along with details such as Product name,
Quantity and Total cost as shown in Fig. 8.

6 Conclusion

The objective is effectively attained in the prototype model developed. The developed
product is of low cost, amiable to use and does not require any specific practice.
The ability to take a decision can be done in the cart itself which can be used in
the shopping complexes for effortless and clever way of purchasing items to save
vitality, time and money of the customers. The project is evaluated with different
trial cases with distinct items assessed for all the practical trials. The system ought to
170 R. Arpita et al.

Fig. 8 The bill payment

page of the mobile app

be generated that scale back the time of the customers in looking out the placement
of the product. The client simply sorts the name of the item; the trolley ought to
mechanically guide them to the location of the item. The system ought to be hooked
up with buzzer; if the load of trolley exceeds certain weight unit then it will facilitate
the older customers. Additional economical and long vary RFID readers are often
used. Thefts are often avoided by use of robotic ARM for choosing the items. Since,
it’s a prototype it has limited load carrying capacity around 3 kg.

References

1. Prasiddhi K (2017) Innovative shopping cart for smart cities. In: 2017 2nd IEEE interna-
tional conference on recent trends in electronics information and communication technology
(RTEICT), 19–20 May, India
2. Ravindranath KR, Sanjay AI, Prashant CM (2017) RFID based supermarket shopping system.
In: 2017 international conference on big data, IoT and data science (BID) Vishwakarma Institute
of Technology, Pune, 20–22 Dec
3. Wankhede SS, Nikose A, Radke DP, Khadse DB, Tiwari S (2018) Electronic shopping trolley
for shopping mall using android application. In: Proceedings of the international conference on
communication and electronics systems (ICCES 2018)
4. Viswanadha V, Pavan Kumar P, Chiranjeevi Reddy S (2018) Smart Shopping Cart. 2018 IEEE
5. Kumar A, Srivastava S, Gupta U (2019) Internet of things (IoT) based smart shopping centre
using RFID. 2019 IEEE
6. Raghunandan, Mallikarjun, Satheesh Rao (2017) Billing automation for supermarkets. In: Inter-
national conference on current trends in electronics information and communication technology
(ICCTCEEC-2017), India
7. Sutagundar A, Ettinamani M, Attar A (2018) Iot based smart shopping mall. 2018 IEEE
Analysis and Design of E Shaped Dual
Band Microstrip Textile Antenna
for Wireless Communication

Husain Bhaldar, Sanjay Kumar Gowre, Mahesh S. Mathpati,

Ashish A. Jadhav, and Mainaz S. Ustad

Abstract The microstrip antenna play important role in wireless communication,

because microstrip antennas are light in weight and compact in size. Now days,
the wearable microstrip antenna is emerging technology widely used to perform
the communication between the one human body to another human body and this
technology is called Wireless Body Area Network (WBAN). The wearable microstrip
antennas are used in various wireless applications such as medical, military and
satellite etc. The geometry of the designed microstrip textile antenna consist of E
shaped rectangular patch the top and the full ground plane at bottom. The dimension
of designed antenna is of 35 mm × 31 mm by its width and length. These two planes
are separated by the cotton fabric as dielectric material of (Er ) of 1.6 and the thickness
of 1 mm. In this paper the E shaped microstrip textile antenna is designed which is
resonating at dual frequencies of 3.7 and 7.4 GHz with high directivity. The proposed
antenna is used for wireless communication in S and C band. The proposed antenna
has been simulated in CST software and the various antenna parameters such as
return loss, VSWR and directivity have been analyzed.

Keywords C band · Cotton fabric · E shaped textile · S band and WBAN

H. Bhaldar (B) · A. A. Jadhav

Department of Electronics and Communication Engineering, Bheemanna Khandre Institute of
Technology, Bhalki, affiliated to Visvesvaraya Technological University Jnana Sangama,
Belagavi, Karnataka, India
e-mail: hkbhaldar@[Link]
H. Bhaldar · M. S. Mathpati · A. A. Jadhav
Department of Electronics and Telecommunication Engineering, College of Engineering, Shri
Vithal Education and Research Institutes, Pandharpur, India
S. K. Gowre
Department of Electronics and Communication Engineering, Bheemanna Khandre Institute of
Technology, Bhalki, India
M. S. Mathpati
GNDCE Bidar Visvesvaraya Technological University Jnana Sangama, Belagavi, Karnataka, India
M. S. Ustad
SVIPE, ICMS Kasegaon, Pandharpur, India

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 171
P. M. Pawar et al. (eds.), Techno-Societal 2020,
[Link]
172 H. Bhaldar et al.

1 Introduction

The microstrip antenna is widely used for wireless application due to its lightweight,
compact size. Due to use of textile material as in the development wearable antennas,
wireless body area network (WBAN) has been rapidly increasing in the wearable
devices which are used for wireless communication in the field of medical, sport wear,
civil and satellite applications. The wearable antennas easily fabricated into clothes.
In this paper E shaped microstrip textile antenna is proposed for dual frequencies for
wireless application at 3.7 and 7.4 GHz. The wearable antenna can be designed by
using various textile materials such as jeans, cotton, polyester, leather and PDMS.
The relative permittivity (Er ) of these textile materials varies from 1.25 to 1.7 except
PDMS material. It is easier now days to study the performance characteristics such
as bending and crumpling conditions of antenna, due to use of textile material in
designing of wearable antenna.
The two frequency bands of 3.7 and 7.4 GHz; these frequencies come under S
band of frequency range 2–4 GHz and the C band of frequency range 4–8 GHz in the
electromagnetic spectrum. The application of S band is used for airport surveillance
system and to control the air traffic. The S band also used for satellite communication
as downlink frequency of range 3.4–4.2 GHz from space to earth station. The C
band is mostly used over Ku band for satellite communication. This band is also
used for multiple applications such as weather forecasting radars and Wi-Fi devices
as IEEE802.11a. The medical accelerators are operated at the S band frequencies.
They are designed at high frequencies to reduce the size and weight. The medical
accelerators are used in radiation therapy field and it has been implemented in the
Intra Operated Radiation Therapy (IORT).
The multiband E shaped microstrip antenna has been designed with ground and
defective ground structure. The proposed antenna is operating at 3 frequency bands
of 3.7, 6.7, 7.9 and 8.7 GHz and antenna provided the 120, 495, 225 and 315 MHz
bandwidths at these resonating frequencies [1].
The proposed antenna has been designed with the rectangular shape which is
resonating at 2.45 GHz frequency and provides the high gain of 8.05dBi. The
designed microstrip textile antenna with jeans as dielectric material been used for
the Wi-Fi modem and achieved the return loss of −15.76 dB [2]. TERA founda-
tion has designed the antenna which is resonating at S and C bands. The designed
antenna used for medical accelerators in hydrotherapy which radiates the protons
or light ions. The designed medical accelerators have been used at 3 and 5.7 GHz
frequencies [3].The designed antenna has simulated at 2.45 GHz with various dielec-
tric substrate materials such as Cotton, Polyester, Cordura and Lycra. The values of
return losses at 2.45 GHz frequency for respective textile material have studied −
32, −35, −29 and −31 dB [4, 5].
The multiband microstrip antenna has designed with moon structure patch on top
and proposed antenna is resonating at the frequencies from 5.3 to 10.15 GHz and
provided the bandwidth of 62.78%, return loss of −25 dB at 5.44 GHz and −24 dB
at 8.05 GHz. The author used the moon strip line structure of patch [6]. The Idellyse
Analysis and Design of E Shaped Dual Band Microstrip Textile … 173

and Rama Reddy have proposed the circular polarized textile antenna and U slot
conical antenna which is designed at ISM band of frequency range 2.4–2.45 GHz,
the values of S11 parameter −35 and −20 dB are observed at resonating frequency
[7, 8].
The proposed multiband wearable antenna has been radiated at frequencies of
3.42, 9.73 and 11.17 GHz with directivities of 3.35, 4.23 and 5.19 dBi [9]. The
author has observed the miniature feeding network for aperture coupled antennas
designed at frequency of 2.4–2.483 GHz with gain of 5.6 dBi. The author have
studied various shapes of coupling aperture such as ring, H-shaped, rectangular, E
shaped, cross etc. at various frequencies in ISM band [10].
The proposed microstrip antenna have designed with circle shaped radiator with
tapered microstrip feed line and has rectangular ground plane. The proposed antenna
used at ultra-wide band (UWB) application for frequency range of 2.8–16 GHz [11].
From the literature survey, it is stated that the gain and directivity of proposed
antenna is enhanced in this paper. The proposed antenna can be used for high
directional antenna for S and C band application.

2 Antenna Configuration and Design

The microstrip textile antenna is most preferred antenna over the printed antenna due
to ease in fabrication and light in weight. The microstrip textile antenna is similar
in structure with simple microstrip patch antenna. The microstrip textile antenna
consist of top patch made with copper foil and bottom ground plane made with
same copper foil of thickness 35 micron. These two planes are separated by textile
material as dielectric substrate as shown in Fig. 1. This methodology of antenna
design is demanding for flexible textile (fabric) antennas, in which designed antenna
can easily attached to clothes and antenna can be wore on the body. This technology
introduced the one body to other body communication called Wireless Body Area
Network (WBAN).
In this proposed study, The rectangular E shaped microstrip textile antenna is
designed for 3.7 GHz frequency also the proposed antenna is resonating at 7.4 GHz
additional frequency band due to structure designed E shaped. The inset line feeding

Fig. 1 Microstrip textile

antenna
174 H. Bhaldar et al.

method used proposed antenna and excited with voltage source connected to the port
of an antenna. The microstrip antenna is designed at the frequency band of 3.7 GHz
for S band and 7.4 GHz for C band wireless communication. The top patch of antenna
consists of E shaped rectangular structure of dimension 35 mm × 31 mm of width
and length. The E shaped slot is designed by dividing the patch width in equal 5
slots which is shown in Fig. 2. The bottom ground plane is designed with full ground
and made of copper foil. The cotton shirt is used as substrate material with dielectric
constant of 1.6 with substrate height of 1 mm, so that designed antenna is wearable,
light in weight and compact in size.
The width and length microstrip patch antenna is calculated for resonant frequency
of 3.7 GHz and dielectric constant of 1.6 by using standard equations of rectangular
microstrip patch antenna design [2]. The Table 1 shows that the list of antenna
parameter used design E shaped rectangular microstrip textile antenna.

a b

c d

Fig. 2 Proposed antenna a E shaped patch of simulated antenna b Ground plane of proposed
antenna. c E shaped patch of fabricated antenna d Ground plane of fabricated antenna
Analysis and Design of E Shaped Dual Band Microstrip Textile … 175

Table 1 List of antenna parameter

Antenna parameter Value
Dielectric constant Er cotton shirt 1.6
Length of patch (L) 31 mm
Width of patch (W) 35 mm
Size of E shaped slot 7.2 mm
Length of ground plane (Lg) 70 mm
Width of ground plane (Wg) 70 mm
Substrate height (hs) 1 mm
Substrate thickness (ht) 0.035 mm

The proposed microstrip antenna resonating at two frequency bands 3.8 and
7.49 GHz. These frequency bands are used for S band and C band communica-
tion. The downlink frequency of satellite communication used in this frequency S
band 3.4–4.2 GHz and the C band mostly used for radar communication 4–8 GHz.
The medical accelerators also used in the S band in the radiation filed theory.
The various parameters of antenna has been analyzed at these two frequency
bands such as return loss, VSWR, directivity and bandwidth as shown in Table 2.
The designed antenna is excited with voltage source connected to microstrip feed
line of thickness around 2 mm of impedance 50 . The proposed antenna have been
designed and simulated in CST Microwave Studio the various antenna parameters
have been observed.

Table 2 Simulated results of E shaped microstrip textile antenna

S. No. Parameters Resonating frequency
F1 = 3.86 GHz F2 = 6.47 GHz
1 S11 dB −33.87 −19.65
2 VSWR 1.14 1.28
3 Directivity in dBi 8.74 7.32
4 Radiation efficiency (%) 115 88.44
5 Total efficiency (%) 33 34
6 Bandwidth 90 MHz 120 MHz
7 %fractional BW 2.33 1.60
176 H. Bhaldar et al.

3 Result and Discussion

(I) Return Loss: The return loss should be kept as low as below −10 dB, to achieve
the best radiation. From Fig. 3a, it has been observed that designed antenna
have two frequency bands at 3.86 and 7.49 GHz with return losses of −33.877
and −19.652 dB. The S11 parameter values of proposed antenna has been
improved and compared with paper implemented. The tested results of fabri-
cated antenna provide the two bands at 2.45 and 3.5 GHz, the wide bandwidth
of 1.9 GHz is obtained at 3.5 GHz frequency band which is shown in Fig. 3b.

The fractional bandwidths are also calculated at these frequency bands using Eq. 1.

%Fractional BW = FH −FL /FC (1)

where FH = High cut off frequency, FL = Low cut off frequency and Fc = Centre
frequency.
The % fractional bandwidths have been calculated 2.33 and 1.6 at these
frequencies which is shown in Table 2 of simulated results.

Fig. 3 Return loss, a Simulated and b Fabricated/tested

Analysis and Design of E Shaped Dual Band Microstrip Textile … 177

Fig. 4 3D radiation pattern the antenna at 2 frequency bands: a 3.7 GHz and b 7.4 GHz

(II) 3D Radiation pattern: From the Fig. 4, it has been observed that 3D radi-
ation pattern of theantenna at resonating frequencies provides the directivity
of 8.74 dBi at 3.7 GHz frequency and 7.32 dBi at 7.4 GHz frequency. These
values of directivity have been improved 7.02–8.74 and 6.37–7.4 compared
with paper implemented and these values fulfil the need of wireless commu-
nication. The proposed antenna can be used for high directional antenna for S
and C band application.
(III) 2D Radiation pattern of proposed antenna: From Fig. 5, it has been
concluded that polar plot of designed antenna has the main lobe magnitude of
8.74 dBi and angular bandwidth of 68.8° at 3.7 GHz frequency and with the
main lobe magnitude of 6.02 dBi and angular bandwidth of 44.1° at 7.4 GHz
frequency.
It has also observed from Fig. 5, that the designed antenna also provided the back
radiation of −21.6 and −15.3 dB at these frequencies.
(IV) VSWR of antenna: From Fig. 6, it has been observed that the value of
VSWR for designed microstrip textile antenna lies in between 1 to 2, which
states the proper impedance matching between excited source and the feed-
line. The values of VSWR at resonating frequencies are 1.14 and 1.28 at 3.7

Fig. 5 2D Radiation pattern of the proposed antenna at 2 frequency bands a 3.7 GHz and b 7.4 GHz
178 H. Bhaldar et al.

Fig. 6 VSWR of proposed antenna

and 7.4 GHz. These vaules of VSWR have been compared with the paper
implemented and it is has been observed values are better.

(V) Efficiency: The efficiency of anenna isdefiend as the ratio of radiated powerof
anantenna to the applied power of antenna. The radiation efficiency antenna
differs from the total efficiency of antenna due to loss takes place because of
the impedance mismatching.
The total efficiency is given by Eq. 2.

Et = Zl ∗ Er (2)

where Et = Total Efficiency, Er = Radiation Efficiency and Zl = Impedance loss.

It has been observed that the proposed antenna radiated at 2 frequancy bands. The
radiation efficiency and total efficiency of antenna at 3.7 GHz are 115 and 33%. The
radiation efficiency and total efficiency of antenna at 7.4 GHz are 88.44 and 33%.
From the values of total efficiency it has been concluded that the ohmic losses takes
place due to impedance mismatch.
The Table 2 shows that the comparison of simulated results of proposed antenna
for two frequency bands.

4 Conclusion

In this proposed study, the E shaped microstrip textile antenna has been designed for
3.7 GHz frequency. The proposed prototype is simulated and analyzed for the various
antenna parameters VSWR, Return Loss and Bandwidth. The gain of antenna has
been improved to good scale at resonating frequencies, if we compared with base
paper implemented. The designed antenna is used for S band and C band commu-
nication. From the simulated results, it has been observed that, antenna is radiating
very properly at 3.7 and 7.4 GHz frequency bands. The total efficiency is reduced
due to impedance loss in comparison of radiated efficiency.
Analysis and Design of E Shaped Dual Band Microstrip Textile … 179

Acknowledgements The author would like to thank the review committee and my guide Dr. Sanjay
Kumar Gowre for his valuable guidance and support. The author also thankful to SVERI’s College
of Engineering Pandharpur for providing support to use antenna laboratory and Research center,
Department of Electronics and Communication, BKIT Bhalki for support and motivation.

References

1. Nagpal A, Dillon SS, Marwaha A (2013) Multiband E-shaped fractal microstrip patch antenna
with dgs for wireless applications. Int Conf Comput Intell Commun Netw 978-0-7695-5069-
5/13 $26.00 © 2013 IEEE. [Link]
2. Bhaldar H, Goware SK, Mathpati MS, Jadhav AA (2020) Design of high gain wearable rect-
angular microstrip textile antenna for wireless application. Int J Innov Technol Exploring Eng
(IJITEE) 9(5). [Link] ISSN: 2278-3075 (March 2020)
3. Mendes C, Peixeiro C (2018) On-body transmission performance of a novel dual-mode
wearable microstrip antenna. IEEE Trans Antennas Propag 66(9) (September 2018)
4. Chen SJ, Ranasinghe DC (2018) A robust snap-on button solution for reconfigurable wearable
textile antennas. IEEE Trans Antennas Propag 66(9) (September 2018)
5. Potey PM, Tuckley K (2018) Design of wearable textile antenna with various substrate and
investigation on fabric selection. In: 3rd international conference on microwave and photonics
(ICMAP 2018), 9–11 Feb 2018
6. Saxena A, Singh VK (2018) A moon-strip line antenna for multi-band applications at 5.44 GHz
resonant frequency. In: 4th international conference on advances in electrical, electronics,
information, communication and bio-informatics (AEEICB-18)
7. Martinez I, Werner DH (2018) Circular-polarized textile based antenna for wearable body area
networks. IEEE Trans 978-1-5386-7102-3/18/$31.00© 2018 IEEE
8. Reddy RS, Kumar A (2018) Dual band circular polarized wearable antenna for military appli-
cations. In: 2018 international CET conference on control & communication, July 2018.
978-1-5386-4966-4/18/$31.00 ©2018 IEEE
9. Purohit S, Raval F (2014) Wearable textile patch antenna using jeans as substrate at 2.45 GHz.
Int J Eng Res Technol 3(5). ISSN 2278-0181 (May 2014)
10. Zhang J, Yan S, Vandenbosch GAE (2017) Miniature network for aperture-coupled wearable
antennas. IEEE Trans Antennas Propag 65(5) (May 2017)
11. Sun Y, Cheung SW, Yuk TI (2014) Design of a textile ultra-wideband antenna with stable perfor-
mance for body-centric wireless communications. Published in IET Microwaves, Antennas &
Propagation. Accepted on 3rd July 2014. [Link]
Enhancement Technique for Early Stage
Lung Cancer Detection Using Foldscope
Methodology

Vanita D. Jadhav and Lalit V. Patil

Abstract Due to ever increasing population and growth in pollution there has been
considerable increase in the cancer-causing agents due to this there is a need to
detect cancer at its early stages so that it becomes easier to cure it. The precautions
to be taken to cure or prevent cancer or other tumor is that cause death very less or
completely negligible this is why there is a need to create a system that can suggest
the early stages of cancer and the precautions that will be taken in order to prevent
this cancer from growing more. Cancer as it is cannot be cured hundred percent
when it is in the moderate stages and above as there are very few treatments for
undertaking or curing cancer completely. We have undertaken the job of detecting
lung cancer as it is the major cause for death in Cancer diseases. Lung cancer when
it is in the initial stages is very accurate and is very difficult to detect and it spreads
rapidly and the chances to stop this growth are very less or completely negligible so
to avoid this, we have decided to create a system that can detect lung cancer at its
early stages. Lung cancer comes in stages and to cure this we need to understand
each and every stage precisely the lung cancer has around Four stages. After looking
at the stages of cancer we can clearly see that the first stage is very important in the
recovery process and for cancer treatment. So, if we detect the cancer at its early
stages it becomes easier for us to cure it and the patient suffers very less due to
this cancer. In some cases, the microscopic examination of this cancer shows wrong
results as the cancer is very less and cannot be easily detected even at microscopic
level. If we record such occurrences it becomes easier for us to detect for the next
time and the chances of error reduces. Here comes the role of Technology and our
system. As we are using machine learning for the development of a system which
is Computer aided design for early lung cancer detection it becomes easier to train
the machine for intangible occurrences of results which cannot be neglected as life
of each and every patient is precious. Our system is completely based on the data
analysis of multiple cases of cancer where lung cancer differs from first stage to

V. D. Jadhav (B)
CSE Department, SVERIs COE, Pandharpur, India
e-mail: vdjadhav@[Link]
L. V. Patil
CSE Department, SKN COE, Sinhgad Road Vadgaon Bu, Pune, India

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 181
P. M. Pawar et al. (eds.), Techno-Societal 2020,
[Link]
182 V. D. Jadhav and L. V. Patil

last stage. By considering the occurrences of results we can analyze and design the
pattern for this disease and can easily predict whether the underlying patient being
monitored is suffering from cancer. We are using machine learning based on Python
programming language as it becomes easier to monitor data as well as the detection
process of cancer. Python makes it easier to use machine learning and we can train
the system for the future. Python has multiple directories which we can use for our
project and which makes our project very simple and accurate as we are going to
take the data set which includes arrays of data.

Keywords Image processing · Image enhancement · Foldscope

1 Introduction

Machine learning is a branch of science that deals with training of the machine for
future occurrences. For some cases the cancer detection process gives wrong results
as the cancer occurrence is very rare and for this, we use machine learning which
helps us to detect the cancer for such occurrences in future proceedings. Even for the
microscopy detection of Cancer lungs we need some predetermined data set which
helps us in detection of that cancer so that we can clearly say that it is a normal cell
or paranormal cell. We have taken images of multiple cancer as well as normal cells
and taken the data into consideration of a data set. Due to this multiple use of data
we can clearly say that we have covered multiple cases as the cancer which we are
going to detect is intangible to the system. We have covered around 1000 different
cases where cancer is in early stages.
The data is it is completely unusable when it is not properly managed and precisely
developed. We have taken search algorithms that are very precise in itself and create
very accurate data that can be used easily and precisely to take tests under serious
conditions. This data from the data set is used for matching the occurrences of given
sample images for the system which will detect the occurrences of Cancer if present
or absent. Before creating the detected the images collected of the cells need to be
processed at, they need some sort of image segmentation and image processing as
it contains noise as well as lots of error which are due to the microscopic detection
using human help. Machine learning makes it easier to calculate the situation and
monitor the rare case executions and enter it into the data set as the rare cases should
be properly managed in the future occurrences. In some cases, the cancer in the cells
will not be detected due to its slightly lesser intense present in the cancer cells. These
rare occurrences will be recorded and sent to the data set so that it can store these
occurrences for further convenience and avoid for the inconvenience due to absence
of the data for calculation [1]. Our system will record every occurrence of the result
and send it to the data set so that the system for every next calculation becomes more
pieces and continuously gets more Precision for this testing process. This system will
work fine and will definitely be a boon to the medical world as it is more precise and
can detect cancer at its early stage and make it easier to cure the patient and treat the
Enhancement Technique for Early Stage Lung Cancer Detection … 183

patient easily. This is the reason we have selected this task and problem statement
for making the world a better place.
Foldscope is an origami based paper microscope which can assemble from
punched sheets of card-stock and lens [2]. The foldscope is firstly printed on a
card sheet and according to the marking on the card sheet the foldscope is assembled
the same way as in origami craft works. The lens is also assembled in the fold-
scope. It is used for the magnification required for getting a microscopic view of
the subject or sample being tested by the foldscope user. There are also magnetic
couplers used for connecting a sophisticated camera or other imaging device which
is used for capturing images of videos of the sample. The magnification of the fold-
scope ranges from 140X to 2000X and 2 micron resolution. We made samples of
specimens of different test subjects for the purpose of microscopic testing of the
subjects under foldscope. The created specimen samples were loaded on slide and
then were mounted on foldscope for getting a microscopic view of the samples.
The magnification is 140x+ and resolution is 2 microns [3]. We can actually see and
examine microscopic cellular activities of the test specimens. By this we can propose
the actual underlying abnormality of the cells of the specimens.

1.1 Computerized System for Lung Cancer Detection

Lung cancer remains the leading cause of Cancer related problems in the world.
Early diagnosis can improve the effectiveness of treatment and increase the patient’s
chance of survival. Positron emission tomography (PET), computed tomography
(CT) [4], low dose computed tomography (LDCT), and contrast-enhanced computed
tomography (CE-CT) are the most common noninvasive imaging modularities for
detecting and diagnosing lung nodules [5]. Pet CT scans can be done at various
centers available in the country but due to machinery requirement and cost of the
Machines, very few centers are available [6, 7].
Due to the cost of the machinery and the required technology used for the detection
the cost of this testing is very high and cannot be done by a normal middle class family.
With high cost level how the current systems available make this system completely
invisible to every common man living in the country. Is there a need to make this
system somewhat cheaper so that every citizen in the country being rich or poor can
use this system for their help. Also the reach of the centers is limited to cities and
metropolitan areas. Due to this the people from rural areas and the people living in
the villages and sub urban areas are not benefited by these centers. The average cost
of the pet CT scan at any Centre across India ranges from 15,000 to 20,000 rs for
tests and this does not assure the complete detection of lung cancer [8]. This is just a
preliminary examination done in order to check whether the person’s lung is affected
with cancer or not. Further is the test negative in searching for lung cancer or in case
of unavoidable circumstances which lead to inconsistent results the patient is sent
for other scans which cost lakhs and lakhs of rupees. Even if the patient manages to
collect this money the result would never be different if cases of cancer are very rare.
184 V. D. Jadhav and L. V. Patil

For such cases the money as well as the time that a patient should be treated is wasted
which time could prove to be crucial in the recovery of the patient. To avoid such
circumstances, we have decided to build projects in such a way that every person
living in this country would be benefited by this system. Power system is not much
cost efficient as it requires just one cellular Level Examination of the patient and the
rest of the work will be done by the machine. The system which we are going to
develop will tell you the result of the comparison of the cellular Level examination
of the patient with the predefined data set we have developed in order to detect if
the patient is suffering from lung cancer or not. The cost of the system will not be
more and it will be beneficiary for every patient had the patient will require no direct
visiting to the center of examination [9].
Because the patient can visit the nearby facilities in order to take the samples of
the cells microscopic level and directly send this to the centers so that we can check
the samples and tell whether the patient samples of specimens are suffering from
cancer or affected with any other cases.
This could even be a remote management system for cancer detection and can
prove to be a simple and efficient way for the people living in rural areas or remote
areas where small medical facilities are also not available for cancer detection. The
fight against cancer will become easier and more widely available due to this system.
The reach of cancer awareness and recovery will increase day by day, if this system
comes into existence this is the need for the development of the system.

1.2 Objectives

(a) To implement methodology that works with cellular level detection and
determination of cancer.
(b) To predict the possibility of lung cancer using the data-set facilitated by
machine learning [10].
(c) To compare between cancerous and non cancerous nodules [11].
(d) To create awareness among the masses for early screening of the disease.
(e) To reduce the need for higher-level machinery implementation in order to
develop low cost systems with high output.

2 Methodology

With reference to Fig. 1, we are going to provide the solution to resolve the above
stated problem by designing and developing a CAD system for early lung cancer
detection by inculcating python integrated with machine learning as a key technique
for evaluation of the collected data [12]. Our system has different modules for clear
segmentation of the detection of cells specimen which are stated below:
Enhancement Technique for Early Stage Lung Cancer Detection … 185

Fig. 1 Architecture of
proposed system

A. Search Module

This module is specifically designed and propagated to locate the testing specimen
which is going to be submitted to the system for further evaluation procedure. This
module is a key benefit which facilitates the user to easily locate the test image and
submit it to the system.
B. Image Processing Module

The image of the cell specimen that is passed by the user through the Search Module
is later entered into this module for further processing. The image that is passed by
the search module is not ready for evaluation and needs precise processing. In this
module the image is assessed for noise as well as other impairments and are removed
from the image set. The image is later contrast adjusted for sharpening and better
clearance. Later on this processed image is passed to the next module.
C. Thresholding Module

The image that is received from the previous module is parsed by Otsu’s thresholding
algorithm. Image segmentation takes place in this algorithm and now the image is
ready to be examined and compared with the data-set.
D. Classification Module

Here, the processing ready image is examined and tallied with the data-set. The
image is completely compared with all the existing data-set images and the result
is sent back to the data-set in order to increase the accuracy for further evaluation.
In case, if the result is not found satisfactory then this situation is recorded so that
186 V. D. Jadhav and L. V. Patil

it does not repeat for the next time. In this way the module never makes the same
mistake twice.
Details of the Proposed Implementation
• Search Module

This module is specifically designed and propagated to locate the testing specimen
which is going to be submitted to the system for further evaluation procedure.
• Image Processing Module

In this module the image is assessed for noise as well as other impairments and is
removed from the image set. The image is later contrast adjusted for sharpening and
better clearance.
• Thresholding Module

The image is completely compared with all the existing data-set images and the result
is sent back to the data-set in order to increase the accuracy for further evaluation.
Working
The user needs to browse through the files and select the target image that is to be
entered in the system for the evaluation process. The image that is taken initially
has certain agents of error that need to be processed in order to get a more feasible
image for the testing process. The image also needs to be in a required format as the
prescribed data-set is built of.
Later, when the image is brought in the system then it needs to be further processed
in order to remove the noise and the errors. The noise may be of different sorts like
salt noise, smoke, etc. This noise is removed by using certain techniques that enhance
the image by removing such noise and the errors. The next parameter that is to be
enhanced is the sharpening and the contrast that is to be maintained and enhanced
accordingly in order to make the image more reliable and perfect for evaluation.
After the image processing the, image is next taken into the algorithmic evaluation
so as to increase its compatibility with the system. Otsu’s Thresholding technique is
used for making the image more and more suitable to work with the system. This
algorithm does significant scaling and other required jobs that make the image more
precise and more efficient in terms of details.
This processed image is then taken to evaluate and determine the cancer causing
agents presence in the sampling image. If the given image has some features that are
similar to that of any of the details in the data-set then the system records that data
Enhancement Technique for Early Stage Lung Cancer Detection … 187

and enters it into the data-set for next successive evaluation. All the results that are
evaluated by the system are successively sent to the data-set so as to improve the
quality of the data-set and increase the precision of the system. The more results we
evaluate the more the system works effectively.
The entire development process has been subdivided into two: the front end devel-
opment and the backend development. The front end comprises the visually visible
parts such as the Browse Button, Image Path, Predict Button and the Result. The
back end contains the data-set and its interaction with the trained data.
Front End Development
The front end is coded using PAGE Python GUI Generator. PAGE is a cross-platform
drag-and-drop GUI generator, bearing a resemblance to Visual Basic. It allows one
to easily create GUI windows containing a selection of Tk and ttk widgets. Required
are Tcl/Tk 8.6 and Python 2.7+. I am actually using Tcl/Tk 8.6 and Python 2.7. PAGE
springs from Virtual Tcl, a Tcl/Tk program, forked to generate Python modules that
realize the desired GUI. Tcl is required for running PAGE but is not required for
executing the generated Python code.
PAGE is not an end-all, be-all tool, but rather one that attempts to ease the burden
on the Python programmer. It is aimed at the user who will put up with a less than
completely general GUI capability in order to get an easily generated GUI. A helper
and learning tool, it does not build an entire application but rather is aimed at building
a single GUI class and the boilerplate code in Python necessary for getting the GUI
on the screen.
Back-end Development
The back-end development is totally done with Python. Some of the image pre-
processing techniques like Noise Removal, Gray Scaling and Otsu’s Thresholding
are used. One of the fundamental challenges in image processing and computer
vision is image denoising. What denoising does is to estimate the original image by
suppressing noise from the image. Image noise may be caused by different sources
(from sensors or from environment) which are often not possible to avoid in practical
situations. Therefore, image denoising plays an important role in a wide range of
applications such as image restoration, visual tracking, image registration, and image
segmentation. While many algorithms have been proposed for the purpose of image
denoising, the problem of image noise suppression remains an open challenge, espe-
cially in situations where the images are acquired under poor conditions where the
noise level is very high.
In global thresholding, we used an arbitrary value for threshold value. So, how
can we know if a value we selected is good or not? Answer is trial and error method.
But consider a bimodal image (In simple words, bimodal image is an image whose
histogram has two peaks). For that image, we can approximately take a value in
the middle of those peaks as a threshold value. That is what Otsu binarization does.
So in simple words, it automatically calculates a threshold value from an image
histogram for a bimodal image (For images which are not bimodal, binarization
won’t be accurate.). For this, our [Link] () function is used, but passes an
188 V. D. Jadhav and L. V. Patil

extra flag, [Link] OTSU. For threshold value, simply pass zero. Then the
algorithm finds the optimal threshold value and returns you as the second output,
retVal. If Otsu thresholding is not used, retVal is the same as the threshold value you
used.

3 Results and Discussion

The above Fig. 2 is the unprocessed image that is selected by the user. The image
contains sources of error and noise and needs to be processed.
The above Fig. 3 is the processed image after noise removal. Salt-noise and other
types of noise are removed from the image. The image is also sharpened so as to
make it compatible with the data-set.
The above Fig. 4 is gray-scaled to hide the uncertain parameters and con-
concentrate on the required part of the image. This image comes under 2-D image
spectroscopy of array data.
The Fig. 5 shown above is the processed image by the algorithm we have used
known as Otsu’s Thresholding Technique. The algorithm enhances the image details
and makes the image more accurate and precise [13]. This algorithm does image
enhancement.

4 Conclusion

Our enhancement technique effectively implements methodology that works with

cellular level detection and determination of cancer. It predicts the possibility of
lung cancer as cancerous or non cancerous using the data-set facilitated by machine

Fig. 2 Original image taken using foldscope

Enhancement Technique for Early Stage Lung Cancer Detection … 189

Fig. 3 Image after noise removal

Fig. 4 Processed image

learning. Helps to create awareness among the masses for early screening of the
disease. Reduces need for higher-level machinery implementation in order to develop
low cost systems with high output.
190 V. D. Jadhav and L. V. Patil

Fig. 5 Image after applying Otsu’s Thresholding algo

References

1. Gonzalez Rafael C, Woods Richar E, Eddins Steven L Digital image processing using MATLAB
2. Cybulski JS, Clements J, Prakash M (2014) Foldscope: origami-based paper microscope. PLoS
ONE 9(6):e98781. [Link]
3. Ferlay J, Soerjomataram I, Ervik M, Dikshit R, Eser S, Mathers C, Rebelo M, Parkin DM,
Forman D, Forman D (2015) [online]. Available: [Link]
4. Grove O, Berglund AE, Schabath MB, HJWL Aerts, Dekker A, Wang H, Gillies RJ (2015)
Data from: quantitative computed tomographic descriptors associate tumor shape complexity
and intratumor heterogeneity with prognosis in lung adenocarcinoma. The Cancer Imaging
Archive [online]. Available: [Link]
5. Grove O, Berglund AE, Schabath MB, Aerts HJWL, Dekker A, Wang H, Gillies RJ, Muoz-
Barrutia A (2015) Quantitative computed tomographic descriptors associate tumor shape
complexity and intratumor heterogeneity with prognosis in lung adenocarcinoma. PLOS ONE.
Public Library of Science (PLoS)
6. Lee JH et al (2017) Anatomically and functionally distinct lung mesenchymal populations
marked by Lgr5 and Lgr6. Cell 170(11491163):e1112
7. Storvall H, Ramskold D, Sandberg R (2013) Efficient and comprehensive representation of
uniqueness for next-generation sequencing by minimum unique length analyses. PLoS ONE 8
8. Clark K, Vendt B, Smith K, Freymann J, Kirby J, Koppel P, Moore S, Phillips S, Maffitt D,
Pringle M, Tarbox L, Prior F (2013) The cancer imaging archive (TCIA): maintaining and
operating a public information repository. J Digital Imaging 26(6):1045–1057
9. Armulik A, Genove G, Betsholtz C (2011) Pericytes: developmental, physiological, and
pathological perspectives, problems, and promises. Dev Cell 21:193215
10. Simon H (1994) Neural networks. Mac Millen College Pub co, New York
11. Marques S et al (2016) Oligodendrocyte heterogeneity in the mouse juvenile and adult central
nervous system. Science 352:13261329
12. Zepp JA et al (2017) Distinct mesenchymal lineages and niches promote epthelial self-renewal
and myofibrogenesis in the lung. Cell 170(11341148):e1110
13. Vanlandewijck M et al (2011) A molecular atlas of cell types and zonation in the brain
vasculature. Nature 554, 475480
Foldscope to Detect the Growth
of Microorganisms on Various Materials
and Vessels
Vanita D. Jadhav, Richa Tamhane, Kiran Kedar, Shruti Kawade,
and Aboli Gaikwad

Abstract Dr. Manu Prakash Professor, Department of Biotechnology, Stanford

University invented foldscope known to be paper microscope. A low cost, portable
and foldable microscope. It is ultra-affordable and durable and quality is almost
similar to conventional microscopes. It has various applications in various fields.
However, its importance is still obscured by many. Foldscope can be used as an
efficient tool to study pollen viability and stomata, to visualize cells and detect seed
viability was demonstrated. Foldscopes are used in projects by students to detect,
visualize and learn real life examples. In this paper we will use Foldscope to detect the
growth of microorganisms on various materials and on vessels. It can be a preventive
measure and various diseases can hence be prevented.

Keywords Foldscope · Research tool · Biology · Image-processing ·

Microorganisms · Detection

1 Introduction

The easiness of Foldscope attracts the attention of many of us. The educationalists
to researchers and the common people to scientists. The thing that pulls them is this
modern origami based microscope with its wide applications. Foldscope, however,
has a big hand in solving the microbial problems. During the twenty-first century,
Manu and his team put forth a modern hand microscope—Foldscope and it helps
lots of young minds to solve the mysteries and puzzles of the microbial world [1].
This modern world microscope is sure to take up other dimensions of the microbial
world with its applications [2]. In this paper, we shall discuss the use of foldscope
to detect the growth of microorganisms on different construction materials and on
vessels [3].

V. D. Jadhav (B) · R. Tamhane · K. Kedar · S. Kawade · A. Gaikwad

CSE, SVERI’s College of Engineering, Pandharpur, India
e-mail: vdjadhav@[Link]
R. Tamhane
e-mail: richadtamhane@[Link]

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 191
P. M. Pawar et al. (eds.), Techno-Societal 2020,
[Link]
192 V. D. Jadhav et al.

Foldscope to detect the growth of microorganisms on construction material

Corrosion is a natural process that converts a refined metal into a more chemically
stable form such as oxide, hydroxide, or sulfide. It is the gradual destruction of mate-
rials by chemical and/or electrochemical reaction with their environment. Corrosion
is one of the leading economic losses in the constructive world [4]. At construction
sites, though there are many factors in retardation of metals, microbes are found to be
a major factor to corrode the metals. A scientific report says that almost 20% of the
reason for corrosion is due to microbes [5]. The microorganisms which have been
associated with corrosion involve many genera and species. They may be divided
into three groups: bacteria, fungi, and algae. The most important bacteria that play a
significant role in the corrosion process are those involved in the sulfur cycle [6].
The corrosion starts with the formation of biofilm on the surface of the metal. A
biofilm comprises any syntrophic consortium of microorganisms in which cells stick
to each other and often also to a surface [7]. A microenvironment is created above
the surface and it can be different from bulk with properties such as pH or dissolved
oxygen. Biofilm is a complex structure. It is a tridimensional structure consisting of
nutrient gradients [8].
When a multispecies of biofilm is caused, it creates the most destructive corrosion.
These biofilms are caused due to the interaction between different species and a
metabolic product formed leads to corrosion. Also, there are several methods to
detect microbes on metals [9].

1.1 Process to Detect Microbes

This study consists of the samples from Cement, Wood, Iron plates, Gravel Asbestos
sheets, brick, glass and tiles. The considered sample was diluted one by one using
sterile distilled water. The diluted sample is now treated with agar medium by
pour plate technique and is incubated for 24–48 h. Petri plates were considered
for enumeration and finally, foldscope was used to detect the bacteria on the sample.

1.2 Detection of Microbes

1432 microbes were found enumerated among 174 samples. From the considered
sample material, 6 bacterial genus and 3 fungal genus were found isolated. Foldscope
images were later compared with the microscopic images and it was found that it
was the best alternative, cheap and durable. Detection of microbes on construction
materials had always been a challenging task in the science out there. But foldscope
has now proved to be the best possible solution for such problems.
Foldscope to Detect the Growth of Microorganisms on Various … 193

Table 1 Food and the

Grains and grain products Number of mycotoxin types
bacteria that affects it
Milk and milk products Aflatoxin
Peanuts, nuts and pulses Aflatoxin
Fruits and vegetables Patulin

2 Foldscope to Detect the Growth of Microorganisms

on Cooking Vessels

Microbes can be found almost everywhere in nature including air, water and land.
They are ubiquitous in nature. In the other cause, microbes are also important in our
day to day life and culture in ways to treat sewage, produce fuels, enzymes and even to
ferment the food. They play a role as an essential biological compound. On the other
hand, microbes are those pathogens responsible for spreading diseases and polluting
the hygiene [2]. Many disease causing pathogenic microorganisms are present in
the environment and they are easily exposed to the food products by handling and
growing on the surface of vessels used for food preparation [10].
Foldscope can be used to detect different types of microorganisms and bacteria
but can’t help to detect the viruses. The percentage of bacteria is highest with respect
to other reasons to cause illness and is up to 66%. Table 1 shows some of the bacterias
affecting food.
Moulds like Aspergillus produce aflatoxin, ochratoxin, citrinin and patulin are
vulnerable to humans. However, these are even difficult to get rid of. To prevent the
growth of such microbes, cleansing and sanitizing of utensils, vessels and equipment
must be done from time to time to avoid consequences [4].
Detection of microbes
As the studies have clearly proved that foldscope has a major role in the application to
detect microbes in cooking vessels. Beyond utensils, growth of microbes is found in
kitchen sponges and dish-clothes too. If it is not cleaned timely, it may lead to cross-
contamination and thus contaminate the food too. This study consists of different
types of cooking vessels such as copper, brass, stainless steel, tin, plastic.

3 Procedure

3.1 Preparation of Starch

Starch of 1000 ml was prepared and sterilized. The sterilized starch solution was
poured onto the vessels 100 ml each and kept for incubation at room temperature 27
°C. This starch has been applied over all the cooking vessels.
194 V. D. Jadhav et al.

3.2 Preparation of Nutrient Agar

It was made of sterile distilled water with the composition of nutrient agar 14, 2.5 g in
500 ml water. A plate was kept over it and it was incubated for about 24 h. Hence, the
presence and absence of colonies above it was monitored. This showed the growth
of microbes on the plate.

3.3 Gram Staining of Bacteria

The slides were prepared by simple staining of selected colonies. The cell wall
composition made it clear about the bacteria whether it was Gram positive or gram
negative. Thus, gram staining was done.

3.4 Isolation and Characterization of Microbes

Staining was performed at various time intervals to differentiate between gram posi-
tive and gram negative. The slide was placed in the paper based microscope interfaced
with a mobile phone. After staining the cells were observed under foldscope, the
recorded images were further analyzed using a microscope. Morphological obser-
vations were done upon simple staining methods and microbes were seen using
foldscope. Foldscope images clearly have shown the presence of bacteria and fungi
from the collected samples (Refer Figs. 1 and 2).

4 Conclusions

We came to know that foldscope unlike other microscopes is durable, affordable and
easy to carry anywhere. It is a best choice to detect microbes and other micro particles
for educational or research purposes. It gives best possible results and proves to be
an ideal tool to detect and monitor them. Experiments clearly showed that foldscope
gives upto 140X magnifications. Low cost and high outputs, made this tool a best
choice among the society. It has various applications and its specific advantages in
each new application.
Foldscope to Detect the Growth of Microorganisms on Various … 195

Fig. 1 Microbial growth on nutrient agar plates

Fig. 2 Microbes observed using foldscope

196 V. D. Jadhav et al.

References

1. Jadhav VD (2018) Foldscope: a low cost magnification device and its applications in various
fields
2. Rathod D. Foldscope: a educational cum research tool used in identification of microorganisms
3. Cybulski JS. Foldscope: origami based microscope
4. Calder R. Preliminary studies in the use of foldscope paper microscope for diagnostics analysis
of crystals in urine
5. Muthukumaran E. An experimental investigation on the quality of air and growth of microor-
ganisms in various construction materials at various conditions and in various cooking
vessels
6. Ephraim RKD. Diagnosis of Schistosoma haematobium infection with a mobile phone-mounted
foldscope and a reversed-lens CellScope in Ghana
7. Sharmila S. Anatomical Characterization of medicinal plants using foldscope: a paper based
origami microscope
8. [Link]
9. Raghavamma STV. Applications of paper microscope for the early detection of parasitic
pathogens in blood smear
10. Banerjee S. Foldscope, the frugal innovation and its application in food microscopy—a review
Implementation of A* Algorithm
for Real-Time Evacuation During Fire
Situation

Shilpa K. Rudrawar, Pallavi Ghorpade, and Dipti Y. Sakhare

Abstract Building architectures are growing towards increased complexity, with

countless people moving through them. Not all amongst the crowd could possibly
be familiar with the building to escape a fire danger zone. Even if the infrastructure
complies to safety standards, decision making for fire evacuation, while ensuring
safety, is utmost critical. Tailoring to these constraints, it is essential to protect lives
by efficient and complete evacuation. For fire emergency, the proposed evacuation
routing system is inputted by a group of wireless sensor nodes present across the
considered floor plan; a MATLAB based central server to find/calculate better safe
evacuation routes for the imperiled people, at a remote location in the building; a Wi-
Fi based network that communicates this calculated route from the sensor network
to server and server to the occupant, on evacuee’s cell phone. The information from
the sensors is transmitted by a Wi-Fi network and is aggregated by the Thingspeak
server. The real-time evacuation route is calculated by the server, towards the nearest
and safest exit door from the occupant’s instantaneous location, by deploying A*
algorithm for route optimization, along with data from sensor network that informs
about origin and fire spread regarding hazard’s location. The server transmits the route
information to the occupants through Wi-Fi connectivity. The endangered evacuees
aretherebyenabled to view and follow this information of dynamic and real time active
maps using a Smartphone. The proposed framework is prototyped and analyzed for
their future inclusion into existing fire evacuation systems.

Keywords Fire emergency · Real-time evacuation · Routing · A* algorithm ·

Wi-Fi · MATLAB

S. K. Rudrawar (B) · P. Ghorpade · D. Y. Sakhare

Department of Electronics and Telecommunication Engineering, MIT Academy of Engineering,
Alandi (D), Pune, India
e-mail: skrudrawar@[Link]

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 197
P. M. Pawar et al. (eds.), Techno-Societal 2020,
[Link]
198 S. K. Rudrawar et al.

1 Introduction

Living in skyscrapers has caused significant issues and difficult and unpleasant chal-
lenges in occupant evacuation during fires or any other emergencies. Predefined
evacuation maps are static in nature, because they showcase some main evacuation
routes which could be prone to blockage as a fire hazard evolves [1]. Evacuation
management regards the exit choice to be critical; some prominent evacuation strate-
gies generally consider egress by shortest distance, or balanced use of exits/stairs for
proper crowd dispersion, or prioritization of Fire floor or Fire layer evacuation [2].
To alleviate casualties, it is important to have emergency evacuation plans in place
before fire outbreaks occur. However, there considerably too many factors which
involve uncertainties and complexities in evacuations. Previous studies unfold that
amongst others, prominently, building structures, fire scenarios and evacuees share
a common physical and social interaction during evacuation activities for fire inci-
dents. Some factors why these components indulge in the evacuation are listed in
Table 1.
Smoke inhalation kills many, compared to the fire. The prominent dangers are
smoke and high temperatures, too that accompany fires, potentially of causing large
scale casualties. Therefore, critical signaling to the evacuees needs ensuring that
evacuation routes are safe and free from smoke to the maximum possible extent [3].
Hence the main aim of this paper is to inculcate such a mechanism which ultimately
is a real-time fire-safe pathfinder that helps improving the efficiency of evacuation.
Many projects in the past limit the real-time delivery of fire incident information
to fire services providers and first responders only.

Table 1 Components that interact during fire evacuation incidents

Interactional components for evacuation Reasons why the basic components obstruct the
systems safe evacuation during fire situation
Building structures 1. For high rise buildings, inoperable lifts or
escalators
2. Behavior of construction materials and building
structures
3. Inadequate/limited way outs or exits for escape
Fire scenarios 1. Insufficient evacuation practice through
evacuation drills
2. Impact of fire scenarios different in nature even
for same evacuation strategy can be different
3. Different fire materials can possibly produce
various smoke layers and thermal radiation
Evacuees 1. Unguided, self-directed evacuation might jam
the exits
2. Movement only towards known emergency
exits
Implementation of A* Algorithm for Real-Time Evacuation … 199

But the efficiency of evacuation can be improved if evacuees are enabled to take
the correct decision on time through digital signage/mobile devices. It is aimed to
be a guidance system for fire evacuation that will employ smart phones to provide
appropriate as well as timely information. This approach will be more reliable than
just some static data on walls shown by guide boards or say any announcements.

2 Literature Survey and Related Work

2.1 Market Survey

There is advanced automatic fire protection equipment available now such as

dampers, control doors, fans, and fire suppression pumps meant for fire extinction
in buildings. The market also has multi-sensor detectors responsive to ranges of
fires and with high stability to skip occurrences of unwanted alarming [4]. There
are also IOT enabled detectors offered by suppliers like Nest Labs, Kidde Systems,
Roost, First Alert, Leeo, etc. [5]. But many of these fire detection, protection and
extinguisher systems fall short of evacuation intelligence.

2.2 Literature Survey

‘Egress models’ mean computer simulation models. They basically anticipate

required time for building inhabitants to evacuate. Egress models could be similar to
zone models, and thereby determine time to the outset of illogical conditions arising
at a building [6]. Egress models are reliably utilized in design analyses (performance-
based) for code compliance towards alternative design. Egress models are also
put to use for figuring out where during egress, probably, the zones of blockages
would develop. PATHFINDER, EESCAPE, EGRESS, ELVAC, building EXODUS,
EVACNET4, EVACS, Simulex, EXIT89, EXITT, are some examples developed for in
the past. Many studies have attempted to analyze fire evacuation modeling by intense
study of Building Information Modeling (BIM), Agent-based modeling, Behavior
based modeling of crowd and Network-optimization based algorithms [7].
Amongst different others, A* is efficient routing algorithm [8]. The A* path-
calculation algorithm is just extended Dijkstra algorithm, so it similarly uses distance
as a shortest path calculation metric. It simplexes the area of search by its tile-
segmentation, calculating scores for then possible paths from a starting point to a
destination, circumventing each obstacle during this [9]. The shortest path is then
constructed, backtracking from the final point via intermediate tiles to starting point.
A* algorithm has been majorly used in computer games navigation, robot navigation,
and in that of Google’s self-driving car too, because of its efficiency in time [10].
Comparison of Prominent Path Search algorithms is shown in Table 2.
200 S. K. Rudrawar et al.

Table 2 Comparison of prominent path search algorithms

Criteria Genetic algorithm Artificial neural A* algorithm
network
Time complexity Greatest time More time needed than Efficient in chaotic
A* algorithm, but is paths too, ensuring no
best if input size is collisions
more
Optimality Overall path travelled Travels longer distance Ensures shortest path
is less than ANN always
Space complexity Memory requirements Better than Genetic Average memory
are very high Algorithm requirements
Reasons for failure Poorest performance if Might Fail for highly Might take too long
the path is like a Maize chaotic environment time to calculate moves
applied for too large input is
and exceeds the limits
of memory

A revised system is needed that can reorganize the existing safety infrastructure
of large buildings to become effective in saving people, reliably, by fast response.
The proposed system is not meant to replace existing fire-safety systems, but
rather it can incorporate evacuation intelligence into them. And has better future
scope due to its efficiency.

3 Methodology

3.1 Sensor Nodes

Fire sensing nodes have Node MCU to command and are prototyped as a module
consisting of 3 main sensors: a fire-detecting flame sensor, a smoke and gas MQ-2
sensor and aDHT-11 temperature sensor.
Congestion monitoring nodes also have Node MCU to command are additionally
consisting of 4th main sensors: IR (infrared) sensor for congestion detection, along
with others: a fire-detecting flame sensor, a smoke and gas MQ-2 sensor, and a
DHT-11 temperature sensor. Hence respective sensors are used at each node for data
acquisition by the Node MCU [11].

3.2 Central Hub and Path Planning

An algorithm was to be developed that would majorly take 2 primary inputs: Building
Map and the Fire situation. The system needed to be triggered by fire situation from
Implementation of A* Algorithm for Real-Time Evacuation … 201

the network meshed; and should reliably identify location of fire based sensor data.
Then using strategies in path planning, we aimed to develop evacuation maps starting
at various points as needed from inside the building. Hence sensor inputs have been
routed to MATLAB via Thingspeak server and A* is used for sensor data optimization
in evacuation path planning [12].
A* algorithm calculates the cost function f(n), as in Eq. 1 where h(n) is the distance
between the start position and node n and g(n) is the distance between n and target.

f(n) = h(n) + g(n) (1)

3.3 User Interface

So Thunkable App platform is used to create an Android Application—a fast, yet

informative app for finding most appropriate evacuation-route.

3.4 Prototype Implementation and Testing

We applied this approach to a close indoor floor map; which is of a single-floor

building and we observed effect of this infrastructure that provides evacuation routing
information to occupants. A sensor node in network covers a considered area of floor
plan. Detecting if that area is blocked/transitable. The sensor acquired data is carried
by a WiFi, collected at the central server. Here calculation of evacuation route, in real
time, to the closest exit to the occupant’s location begins. When done computing,
server sends this assessed info to the occupant through the communication network.
The user may follow this suggested path when he queries help for himself by the
Mobile App. Testing was verified by whether a flame brought close to the IR Flame
Sensor or not, sends a digital 1 or 0 to the Node MCU. Power-save/sleep mode
is exited by Node MCU if flame comes near IR sensor. Then interpreted data is
packaged and sent to the ESP8266 Wi-Fi transceiver by the Node MCU.

3.5 Algorithm Implementation in MATLAB

The experimental stages of algorithm Implementation in MATLAB are shown in

the following figures. Targets, starting point and obstacles were assumed in different
cases provide an acceptable output for the shortest path. The basic A star algorithm
implemented was capable of finding the shortest route, irrespective of the starting
point and the target.
202 S. K. Rudrawar et al.

Fig. 1 Prototypes sensor nodes

To have the input of Sensor Node onto the on-line server Thingspeak Platform, a
bread-board implementation of the sensors is connected to a Wi-Fi. This is depicted
by Fig. 1.

3.6 Input of Sensor Node onto the On-Line Server

Thingspeak Platform

Once Wi-Fi connection is sensed by the nodes, the monitored data is transmitted to
the Thingspeak Server. Hence, these sensor nodes continuously send the data online
to Thingspeak. This is shown in Fig. 2.
This data is further imported in MATLAB and incorporated in to the A* algorithm.
For instance, data captured by Node1, Node2 and Node3 is imported in MATLAB
Script by the following lines of code from the respective thingspeak channel.

node1 = thingSpeakRead(723592, ‘Fields’, 2)

node2 = thingSpeakRead(723592, ‘Fields’, 5)
node3 = thingSpeakRead(723592, ‘Fields’, 8)
Implementation of A* Algorithm for Real-Time Evacuation … 203

Fig. 2 Data sent to a channel on Thinkspeak platform

3.7 Android Application Development

To have the output of calculated evacuation route as a recommendation on Android

application Platform, the basic buttons & app connectivity is accomplished as shown
in Fig. 3. The various instances of evacuation map or images, provided timely at
hand for the end user are displayed in Fig. 4.

4 Results and Discussion

Presence of fire (flame) brought nearer to the Flame Sensor, sends a digital 1 or
0 to Node MCU. It in turn analyses this data by appropriate thresholding, for the

Fig. 3 Android
application—front page
204 S. K. Rudrawar et al.

Fig. 4 Android application

changed reading, packages this data and sends it to Thingspeak channel. Node MCU
by its ESP8266 Wi-Fi transceiver sends this data wirelessly out to the Wi-Fi network
aligned field-wise to the Thingspeak Channels with its information. Hence fire is
imitated by candle flame as sensed by fire sensor. For the prolonged presence of fire,
it is simulated to aggravate, and this spreading of fire is assumed as the dispersion
Wi-Fi specifications over which the nodes were tested is described below:
• Security: WPA2 PSK
• Frequency: 2.4 Ghz band
• Identification: Pixel Graphics
• SSID: uitech333.

Test Case 1:
Path calculated in case of absence of fire danger: In this instance, there was no fire
present and hence no fire detected by the sensor nodes. So the evacuee was provided
with the shortest path to escape from the room which he queried over the app from.
MATLAB calculation of optimized route is shown in Fig. 5a and b.
Test Case 2:
Decision of Correct Choice of Available Exits: In this instance, fire was present
at one door of a room. Even if that path was shortest, but it was overridden by a
comparatively safer path. So the evacuee was provided with the shortest path to
escape from the room which he queried over the app from. MATLAB calculation of
optimized route is shown in Fig. 6a and b.
In this instance, fire was now detected at another door of the same room. The
second door had no fire. Hence the shortest and safest path was suggested by opti-
mization. So the evacuee was provided with the shortest path to escape from the
Implementation of A* Algorithm for Real-Time Evacuation … 205

Fig. 5 a Path calculated in case of absence of fire danger. b Real time Output on Mobile App

Fig. 6 a Decision of correct choice of available exits. b Real time output on mobile app

room which he queried over the app from. MATLAB calculation of optimized route
is shown in Fig. 7a and b

4.1 Execution Time

Execution time of modified A* Algorithm implemented in MATLAB for one query

from user or evacuee is pictured in Fig. 8.
206 S. K. Rudrawar et al.

Fig. 7 a Decision of correct choice of available exits. b Real time output on mobile app

Fig. 8 Profile of M-script mapped from MATLAB

5 Conclusion

Proposed work gives a framework for calculating better evacuation routes for occu-
pants of an indoor scenario threatened by a fire caused, and a case of single occu-
pant considered. The A* algorithm via MATLAB uses in-formation collected from
the sensor network. Information from the sensors indicates door area blockage. A
Implementation of A* Algorithm for Real-Time Evacuation … 207

network provides indication on the direction to exit to the occupant for a quick evac-
uation. We tested a version of routing using the information about transit able spaces
and fire location and another that adds a distance as safety margin between the occu-
pant path and fire to keep the occupant far from the f