Papers by Dr. Shailaja Uke
Brain Tumor Detection Using Convolutional Neural Network
Identification of brain tumor is a highly difficult task in early life stages. The presence of br... more Identification of brain tumor is a highly difficult task in early life stages. The presence of brain tumor among humans has increased in large amounts in recent years. Gliomas are one of the most common types of primary brain tumors that account for 30% of all human brain tumors and 80% of all malignant tumors. The World Health Organization (WHO) specified rating system is used as a basic method for medical diagnosis, prognosis and life. The main ideology is to propose and develop reliable, typical methods for detecting the brain tumor, extracting its characteristic and classifying the glioma using Magnetic Resonance Imaging (MRI). The model developed automatically assists in brain tumor detection and is implemented using image processing and artificial neural network. Detecting the tumors at starting point is very critical for a patient's healthy life. There are several literatures on identifying these kinds of brain tumors and enhancing the precision of detection. This method use Convolutional Neural Network algorithm to estimate the severity of the brain tumor which gives us accurate results.
Comparative Analysis of Multiple ML Models and Real-Time Translation
A huge population in the present world is suffering from hearing impairment. This hinders their a... more A huge population in the present world is suffering from hearing impairment. This hinders their ability to communicate and express their opinions. Implementing an accurate gesture capture and interpretation system through an image classification model can significantly reduce the impact of hearing impairment, enabling effective recognition of the user's gestures. The proposed project develops machine learning models based on computer vision and image processing to capture and interpret the hand gestures of differently-abled people. Various image classification algorithms are studied, and a comparative analysis using predefined evaluation parameters identifies the most effective solution for gesture recognition for individuals with hearing impairments. A real-time live video implementation model to recognize certain regularly used gestures is also executed. This video implementation will detect some of the frequently used signs, enabling easier communication for individuals with hearing impairments and muteness. The proposed approach is versatile, with applications in device control and user feedback.
Blockchain-Based Land Registries System
Zenodo (CERN European Organization for Nuclear Research), Aug 20, 2023
Systems and Methods: Image Encryption Using Steganography and XOR Operations
Cybersecurity is a significant part of safe data transmission, particularly through networks such... more Cybersecurity is a significant part of safe data transmission, particularly through networks such as the Internet. Confidential medical or military images and personal images of individuals and organizations are a big part of such data. These images need a very high level of security to prevent access by unauthorized users. Unauthorized access to such images can lead to significant financial and life losses. To prevent this, we incorporate various encryption-decryption algorithms and concepts like steganography into a system for safe, reliable, efficient, and secure image storage and transfer. We tune each algorithm implemented in the system individually to produce the best results upon integration with other algorithms. Further, we provide users with Graphic User Interface that helps them interact with this system conveniently and efficiently. Finally, we test the system using various medical images and present the results.
This paper introduces a sophisticated visual debugger that seamlessly integrates C++ code analysi... more This paper introduces a sophisticated visual debugger that seamlessly integrates C++ code analysis with C# WPF visualisation for an enhanced debugging experience. The tool operates by extracting symbol information from target C++ source files during debugging sessions, utilizing PDB files, specifically tailored for the Visual Studio environment. The core functionality of the tool involves the extraction and storage of symbol information within a meticulously designed stack frame and stack trace classes. Leveraging C++ capabilities, the project captures not only call stack information but also variables and data structures associated with each stack frame, providing a comprehensive view of the program's execution state.To facilitate interoperability and visualization, the project employs a marshalling mechanism to seamlessly transfer the extracted data from the C++ backend to a C# WPF frontend. The Windows Presentation Foundation application serves as an intuitive and interactive interface, presenting the call stack alongside associated variables and data structures in a visually appealing manner. This tool is positioned to significantly enhance the debugging process for developers working in C++ within the Visual Studio environment. By seamlessly integrating symbol extraction, data structuring, marshalling, and visualization, it provides a holistic and efficient solution for diagnosing complex code scenarios. The resulting visualization aids developers in gaining a deeper understanding of their code's execution flow, enabling quicker identification and resolution of issues during the debugging process.
Tumor is the abnormal development of cells in the body. The fairly significant overgrowth of brai... more Tumor is the abnormal development of cells in the body. The fairly significant overgrowth of brain cells is known as a brain tumor and classification manually takes time and is only possible at a few diagnostic facilities. Therefore, it is necessary to create a system that can identify the type of brain tumor based solely on the input MR images. To evaluate these malignancies, a common imaging approach is magnetic resonance imaging (MRI), however, manual segmentation cannot be done in a timely manner due to the volume of data generated by MRI, restricting the application of exact quantitative measurements in clinical practice. This study presents a novel hybrid methodology for the classification of brain cancers based on the analysis of magnetic resonance imaging (MRI) data. The methodology compares the performance of different classifiers, namely Convolutional Neural Networks (CNNs), K-nearest neighbor, Support Vector Machine and Logistic Regression in predicting the presence of brain tumors. The CNN model is specifically employed to get features from the MRI data and apply them to the classifiers. The results of the study show that the different CNN models, along with KNN, SVM, LR, and Long Short-Term Memory Networks were taken for tumor classification. The achieved accuracies were 82.35%, 78.43%, 61.26% and 74.28% respectively. Among these algorithms, KNN demonstrated the highest accuracy, indicating its effectiveness in accurately classifying brain tumors based on the MRI data. By integrating these classifiers and utilizing the CNN model for feature extraction, this hybrid methodology presents a promising method for classifying brain tumors from MRI data. The findings highlight the potential of combining deep learning techniques like CNNs with traditional machine learning algorithms to increase the reliability of diagnosing brain tumors and contribute to the field of medical image analysis.
In this digital era where social media is present everywhere, cyberbullying has emerged as a crit... more In this digital era where social media is present everywhere, cyberbullying has emerged as a critical issue. It adversely affects mental health and online interactions. A bilingual cyberbullying detection system has been proposed to address this issue. This system is capable of analyzing text in both English and Hindi (transliterated into English). The system employs advanced data extraction and preprocessing techniques. Leverages machine learning algorithms to detect and accurately classify cyberbullying. The dataset for the English language has been sourced from Twitter while the dataset for Hindi has been created by extracting comments from YouTube. The systems architecture integrates Python technologies, Streamlit, Pandas and Scikit-learn to result in a robust web application designed for real-time text analysis. Regular expressions are used to identify and filter offensive language. Despite obstacles like algorithm bias and data privacy, the system is committed to creating a safer digital environment. This research demonstrates the practical application of these technologies to create a multilingual tool aimed at mitigating cyberbullying on social media platforms.
A huge population in the present world is suffering from hearing impairment. This hinders their a... more A huge population in the present world is suffering from hearing impairment. This hinders their ability to communicate and express their opinions. Implementing an accurate gesture capture and interpretation system through an image classification model can significantly reduce the impact of hearing impairment, enabling effective recognition of the user's gestures. The proposed project develops machine learning models based on computer vision and image processing to capture and interpret the hand gestures of differently-abled people. Various image classification algorithms are studied, and a comparative analysis using predefined evaluation parameters identifies the most effective solution for gesture recognition for individuals with hearing impairments. A real-time live video implementation model to recognize certain regularly used gestures is also executed. This video implementation will detect some of the frequently used signs, enabling easier communication for individuals with hearing impairments and muteness. The proposed approach is versatile, with applications in device control and user feedback.
We are living the era of computers where nowadays everything can be done using computers i.e., Ar... more We are living the era of computers where nowadays everything can be done using computers i.e., Artificial Intelligence and Machine Learning, have made technology very easy that now humans can interact with machines in such a way that their task can be done only by interacting with Machines i.e., Computers. Yes, we are talking about Virtual Personal Assistants (VPAs). As of now there are many such VPAs like Alexa, Bixby, Echo, Siri, Google Assistant available on desktop, mobiles and in device also but as the technology becomes easier it becomes difficult to use for particular people such as senior citizens, blind people, also children below certain age. Also, this VPAs which are present nowadays do not provide as much facilities. To overcome this, we developed voice assistant in python on windows system which provide user to perform any task without using keyboard. Also, user can send email just by giving receiver's email and the message user wants to send. Also, now user can book a cab just by saying where he wants to go and also book bus and train tickets. Our main aim towards developing this is to make certain things more efficient.
Cybersecurity is a significant part of safe data transmission, particularly through networks such... more Cybersecurity is a significant part of safe data transmission, particularly through networks such as the Internet. Confidential medical or military images and personal images of individuals and organizations are a big part of such data. These images need a very high level of security to prevent access by unauthorized users. Unauthorized access to such images can lead to significant financial and life losses. To prevent this, we incorporate various encryption-decryption algorithms and concepts like steganography into a system for safe, reliable, efficient, and secure image storage and transfer. We tune each algorithm implemented in the system individually to produce the best results upon integration with other algorithms. Further, we provide users with Graphic User Interface that helps them interact with this system conveniently and efficiently. Finally, we test the system using various medical images and present the results.
In the current era of generative Artificial Intelligence (AI) boom, most image generation models ... more In the current era of generative Artificial Intelligence (AI) boom, most image generation models rely solely on prompt-based input to generate images. Often it is very challenging for a user to convey their ideas using textual prompts, thus resulting in images that differ significantly from user's intended vision. The proposed AI image generation system enables users to input a rough sketch, which is then transformed into a refined and accurate image closely aligned with the original sketch. This is achieved by integration of multiple state of art Generative Adversarial Networks (GANs) such as pix2pix (image translation), Deep Convolutional Generative Adversarial Network (DC GAN), Enhanced Super-Resolution Generative Adversarial Networks (ESRGANs) (image restoration). This AI system makes image generation tasks more intuitive and aligned with user expectations. This method could improve design processes in fields like fashion and architecture.
The advent of the digital age has brought about previously unheard levels of connectedness and in... more The advent of the digital age has brought about previously unheard levels of connectedness and information availability, as well as an increase in cyber threats. Unauthorized Uniform Resource Locators (URLs) are now a common threat and a preferred avenue for cyberattacks. Because of the inadequacy of traditional blacklist-based solutions to counteract shifting strategies, machine learning (ML) is being investigated for proactive and predictive cybersecurity measures. The goal of this research work is to improve cybersecurity defenses against URL-based vulnerabilities by utilizing ML and ensemble learning. Several machine learning methods are used, such as Random Forest, Decision Tree, Logistic Regression, XGBoost, SVM, and Naive Bayes. The intricacy of the digital threat landscape is addressed by combining the strengths of different models through the use of ensemble approaches like Weighted Voting and Soft Voting. This study highlights the significance of Weighted Voting and shows how well ensemble techniques may combine different machine learning model and emphasizes how important it is to keep improving cybersecurity defenses in line with changing threats. The results enable the possibilities for further advancements in the sector and offer workable answers to the emerging cybersecurity issues.
This paper reviews the evolution and trends of Hand Gesture Recognition systems. Through an exhau... more This paper reviews the evolution and trends of Hand Gesture Recognition systems. Through an exhaustive analysis of papers sourced from reputable academic databases, it delves into the current technologies, algorithms, efficiency metrics, and performance evaluations of these systems. Noteworthy advancements in recognition accuracy and processing speed underscore the growing efficacy of these solutions, albeit challenges persist in achieving real-time translation and robustness across sign language dialects. Looking forward, the review identifies opportunities for future research, including algorithm refinement, multi-modal approaches, and usercentric design considerations, thereby contributing to the ongoing development of inclusive communication technologies for the sign language community.
In a time of accelerating technology development, the fast-food business is constantly looking fo... more In a time of accelerating technology development, the fast-food business is constantly looking for new ways to improve client experiences. To improve and streamline the client order-taking procedure, this research study offers a revolutionary voice-activated fast food ordering helper. We show a strong and effective voice assistant system that is capable of accurately interpreting and responding to consumer queries by using the capabilities of DistilBERT which is a BERT NLP model. This pretrained model serves as the basis for the subsequent finetuning procedure, which improves performance by utilising domain-specific data from fast-food restaurant menus. User's responses and menus are stored in MongoDB which can be forwarded to the cooking section. The user's responses can be used for constantly training the voice assistant to better understand user intent. The Google-TTS and STT library converts Text to speech and speech to text respectively. After fine-tuning to DistilBERT the model gave an accuracy of 95%.
International journal of engineering research and technology, Jan 3, 2014
Mobile users like to use their own consumer electronic devices anywhere and at anytime to access ... more Mobile users like to use their own consumer electronic devices anywhere and at anytime to access multimedia data. Hence we expect that wireless ad hoc networks will be widely used in the near future since these networks form the topology with low cost on the fly. However, consumer electronic devices generally operate on limited battery power and therefore are vulnerable to security threats like data flooding attacks. The data flooding attack causes Denial of Service (DoS) attacks by flooding many data packets. However, there are few existing defense systems against data flooding attacks. The existing schemes may not guarantee the Quality of Service (QoS) of burst traffic since multimedia data are usually burst. Therefore we propose a novel defense mechanism against data flooding attacks with the aim of enhancing the throughput.
International journal of computer applications, May 17, 2013
Wireless sensor networks (WSNs) are growing extremely and becoming more and more attractive for a... more Wireless sensor networks (WSNs) are growing extremely and becoming more and more attractive for a variety of application areas such as surveillance of information, industrial secrets, air pollution monitoring, area monitoring, and forest fire detection, home automation, industry monitoring, and many more. As WSN is mostly used for gathering application specific information from the surrounding environment, it is highly essential to protect the sensitive data from unauthorized access. WSNs are vulnerable to various security attacks because of broadcast nature of radio transmission. The primary weakness shared by all wireless application and technologies is the vulnerability to security attacks/threats. The performance and behaviour of a WSN are vastly affected by such attacks. In order to be able to better address the vulnerabilities of WSNs in terms of security, it is important to understand the behaviour of the attacks. This paper aims at addressing behavioral modeling of critical security attack residing in the physical layer and data link layer of wireless sensor network. UML gives the finest diagrammatic representation of any system which is best for developers. Our efforts to synchronize WSN with UML are discussed in the paper. The security attacks are modeled by using state machine diagram of Unified Modelling Language (UML). This modeling of security attacks will help programmers to develop counter measures.
International Journal of Mobile Network Design and Innovation, 2015
Wireless sensor networks (WSNs) consist of sensor nodes and the sensor nodes are capable of colle... more Wireless sensor networks (WSNs) consist of sensor nodes and the sensor nodes are capable of collecting, sensing and gathering data from environment. These networks have huge application in disaster management, habitat monitoring, security and military, etc. Wireless sensor nodes are very small in size and very low battery power and have limited processing capability. We focus on data-aggregation in wireless sensor networks. Data aggregation is a very important technique in WSNs and helps in reducing the energy consumption by eliminating redundancy. The main goal of data aggregation is to collect and gather data in an energy efficient manner so that network lifetime is improved. In this paper we present a survey of data aggregation techniques and algorithms in wireless sensor networks. We compare and contrast different techniques as well as algorithms on the basis of performance measures such as lifetime, energy efficiency, latency, data accuracy and transmissions policies.
Optimal video processing and soft computing algorithms for human hand gesture recognition from real-time video
Multimedia Tools and Applications
An enhanced artificial neural network for hand gesture recognition using multi-modal features
Computer Methods in Biomechanics and Biomedical Engineering: Imaging & Visualization
International Journal of Science and Research (IJSR), 2017
Wireless Sensor Networks (WSNs) consist of sensor nodes and the sensor nodes are capable of colle... more Wireless Sensor Networks (WSNs) consist of sensor nodes and the sensor nodes are capable of collecting, sensing and gathering data from the environment. These networks have extensive application in disaster management, habitat monitoring, security, and military, etc. Wireless sensor nodes are very small in size and very low battery power and have limited processing capability. Data aggregation is a very important technique in WSNs and helps in reducing the energy consumption by eliminating redundancy. The main aim of data aggregation is to collect and gather data in an energy efficient manner and due to this the network lifetime is improved. One data aggregation method in a WSN is sending local representative data to the Sink node based on the spatial-correlation of sampled data. Based on this correlation degree, a data density correlation degree (DDCD) clustering method is presented in detail so that the representative data have a low distortion on their correlated data in a WSN. The proposed system uses a cooperate node or data centre node which cooperates among the sensor nodes in a particular area of WSN and which improves transmission policies as well as energy efficiency. To design a modified data density correlation degree clustering algorithm for energy balanced network by using clustered data aggregation methodology. The simulation results show that the resulting representative data achieved using the proposed modified data density correlation degree clustering method have better throughput, packet delivery ratio, dropping ratio, delay, normalized overhead and average energy consumption than those achieved using the data density correlation degree clustering method.
Uploads
Papers by Dr. Shailaja Uke