www.ijcrt.

org © 2023 IJCRT | Volume 11, Issue 5 May 2023 | ISSN: 2320-2882

LEAF DISEASE DETECTION USING DEEP

LEARNING TECHNIQUES
1
Dr. T. Praveen Blessington, 2Bhargav Krishnan, 3Yash Bharne, 4Kartiki Khoje, 5Prathamesh Padwal
1
Guide, 2,3,4,5Students
1,2,3,4,5
Department Of Information Technology
Zeal College of Engineering and Research, Narhe, Pune, India-411041

Abstract: Agriculture is an important business in order to meet the basic needs of food for the expanding global population. The
global economy and human nutrition both depend on the growth of grains and vegetables. Many farmers grow in remote areas of
the world and suffer large losses as a result of their reliance on manual monitoring of grains and vegetables and their lack of correct
knowledge and disease detection. The use of digital farming methods could be an innovative solution to easily and quickly detect
plant leaf diseases. In order to solve these issues, this research proposes a method for recognizing plant leaf illnesses and taking
preventive action in the agricultural industry using image processing and Convolutional Neural Networks (CNN). Leaf disease is a
major threat to agricultural production and food security worldwide. Early detection and identification of leaf diseases can
significantly reduce crop losses and improve crop yield. The use of deep learning in plant disease recognition can minimize the
drawbacks associated with the artificial selection of disease spot features, make the extraction of plant disease features more
objective, and accelerate the pace at which new technologies are developed. In this research paper, we propose a novel approach
for leaf disease detection based on deep learning techniques. This research paper contributes to the development of effective and
efficient methods for detecting and identifying leaf diseases, which can ultimately benefit farmers and help ensure global food
security.

Keywords – Disease Detection, Feature Extraction, Image Processing, Deep Learning, CNN
I. INTRODUCTION
In India, a developed country, around 70% of people are employed in agriculture. Farmers can select the ideal
insecticides for their plants from a wide range of qualifying crops. A considerable loss in productivity as a result of crop
damage would have an effect on the economy. The leaves of plants, which are their most delicate part, are where illness
symptoms first manifest. The crops must be examined for diseases from the very beginning of their life cycle until they
are prepared for harvest. The development of automatic and semi-automatic plant disease detection systems has utilized
a number of techniques in recent years, and automatic disease detection by merely monitoring the symptoms on the
plant leaves makes it both easier and more economical. These techniques have already shown to be more efficient,
affordable, and accurate than the traditional method of manual observation by farmers. Over the past ten years,
integrated pest management (IPM) strategies have largely supplanted conventional techniques for spraying insecticides.
Whatever the approach, precise disease detection at the point of onset is the first step in effective illness management.
Due of its success in identifying different outlines, deep learning algorithms are currently predominantly used for pattern
identification. DL automates feature extraction. When compared to other traditional machine learning techniques, the

IJCRT2305797 International Journal of Creative Research Thoughts (IJCRT) www.ijcrt.org g539

www.ijcrt.org © 2023 IJCRT | Volume 11, Issue 5 May 2023 | ISSN: 2320-2882
DL reduces error rate and computational time while achieving a high accuracy rate in the classification task. The main
objective of our work is to diagnose plant diseases and provide remedies using Residual Neural Network (ResNet). To
the advantage of both farmers and customers, the agriculture sector must incorporate technology and digitalization. The
use of technology and regular monitoring allows for the early detection and elimination of diseases. A greater
agricultural yield is sought. In recent years, deep learning has greatly outperformed traditional techniques in the area of
digital image processing. Utilizing deep learning for plant disease recognition can reduce the negative effects of
artificially selecting disease spot features, improve objectivity in the extraction of plant disease features, and quicken
the pace of technological development.

II. RISK ANALYSIS

Image size: The program's security depends on the size of the image. It takes less time to extract the key from smaller
photos since we have less options for choosing pixels. There are w*h pixels overall in an image with a height of h and
a width of w. The intrusive party or attacker must test all conceivable arrangements of the pixels they decide on in order
to extract the key from the image. By expanding the image, we may increase the application's security.

III. LITERATURE SURVEY

A CNN model is used in a prior study [1] to categorize the various plant diseases acquired from the Plant Village dataset. The
AlexNet architecture will classify the numerous plant diseases into 38 different distinct classes. Additionally, the suggested system
offers a decent way to anticipate plant diseases and can aid in their early detection. On the suggested system, alternative learning
rates could be explored in the future.
[2] It concentrated on applying a CNN model to forecast the pattern of plant diseases using images from a given dataset (a trained
dataset) in the field and historical data. The system will include the widest variety of plant leaves imaginable, enabling farmers to
learn about leaves that might not have previously been domesticated and providing a list of all potential plant leaves, which helps
them decide which crop to grow.
In [3] Convolutional neural networks and transfer learning are used in the working model to categorize various plant leaf diseases.
CNN is a kind of deep learning neural network and is effective at classifying images. Comparing the suggested technique to the
traditional method of manually observing each plant leaf, it is quicker and more accurate. The CNN model is employed to accurately
predict several plant diseases. Performance evaluation criteria for the model, such as accuracy, precision, recall, and F1 score, are
used during testing.
[4] The suggested Pomegranate Disease Classification System Using a neural network called Back Propagation that primarily
uses the method of Color and texture are employed as the features to segment the defective area. To avoid occlusion, the image is
typically taken with a plain background. The algorithm was compared to other machine learning models for accuracy.

IV. PROPOSED SYSTEM ARCHITECTURE

A system architecture is a conceptual representation of a system's structure, behavior, and other characteristics.
Designed subsystems and system components that work together to create the entire system might make up a system
architecture. The organization and structure of the software system are specified by the architecture. Additionally, it
discusses the links between the various software system components, levels of abstraction, and other features. An
architecture can be used to specify a project's objectives or to direct the creation of a brand-new system. This section
will look at the numerous steps that must be taken in order to create, employ, and get the most likely results from
different classifiers. Out of all the various results and accuracies provided by various models, we will use the model
that yields the best results and accuracy for identifying leaf disease.

IJCRT2305797 International Journal of Creative Research Thoughts (IJCRT) www.ijcrt.org g540

www.ijcrt.org © 2023 IJCRT | Volume 11, Issue 5 May 2023 | ISSN: 2320-2882

Fig 1: proposed system architecture

Data Collection: The first step is to collect leaf images that are healthy and infected with various diseases. This data can
be collected from various sources such as databases or by capturing images using a camera or smartphone.
Data Pre-processing: The collected data needs to be pre-processed to remove noise, adjust the brightness and contrast,
and resize the images to a uniform size. This step is important as it helps to improve the accuracy of the disease detection
algorithm

Fig 2: classification system architecture

ResNet:

The ResNet architecture is based on the idea of residual learning, which is the concept of adding shortcut connections
(or skip connections) between layers to help the network learn the residual mapping instead of the original mapping.
The residual mapping is the difference between the input and output of a layer, and it allows the network to learn a more
accurate representation of the input. The ResNet architecture has several different versions, but the most commonly
used is ResNet-50, which has 50 layers. The ResNet-50 architecture has a series of convolutional layers, followed by a
series of residual blocks. Each residual block consists of two or three convolutional layers, followed by a shortcut
connection that adds the input of the block to its output.

IJCRT2305797 International Journal of Creative Research Thoughts (IJCRT) www.ijcrt.org g541

www.ijcrt.org © 2023 IJCRT | Volume 11, Issue 5 May 2023 | ISSN: 2320-2882

In this diagram, the input is passed through a convolutional layer, followed by batch normalization and ReLU activation.
The output is then passed through a max pooling layer and a series of residual blocks. Each residual block consists of
two or three convolutional layers with batch normalization and ReLU activation, followed by a shortcut connection that
adds the input of the block to its output. The output of the final residual block is passed through average pooling, a fully
connected layer, and a softmax activation function to produce the final output.

V. PROJECT DESIGN

Fig 3: data flow diagram

Here, the input image is initially tested using the training dataset. After testing, features are extracted from the image.
The classifier then received information about these characteristics as well as whether the image showed a healthy or
diseased leaf as input. The classifier then establishes a link between the returned features and the likelihood that a
disease is present. The leaf illness is diagnosed in the subsequent stage by comparing the input image to the pre-existing
images in the collection. The best feasible treatment is advised together with the forecast of the sickness the plant will
contract.

VI. RESULTS
The entire dataset is divided into training and testing sets at random. The ResNet model is trained using the training
dataset. These sets are typically divided into portions of 20% to 80%, 40% to 60%, 60% to 40%, 80% to 20%, etc. The
training dataset can be expanded to include additional photos to produce the most accurate results. In this study, the
model is trained using 80% of the dataset, then tested using the remaining 20%. As shown in Table 1, this collection
includes 50257 images of plant leaves from several distinct categories, including images of prevalent plant diseases.
Since colored images provide more accuracy than grayscale images, all of the images utilized in this work are colored
images. The images were all taken at various perspectives and under various circumstances.

IJCRT2305797 International Journal of Creative Research Thoughts (IJCRT) www.ijcrt.org g542

www.ijcrt.org © 2023 IJCRT | Volume 11, Issue 5 May 2023 | ISSN: 2320-2882

TABLE I
performance of resnet model for healthy and unhealthy leaf

Fig 4: actual vs predicted labels of classes

IJCRT2305797 International Journal of Creative Research Thoughts (IJCRT) www.ijcrt.org g543

www.ijcrt.org © 2023 IJCRT | Volume 11, Issue 5 May 2023 | ISSN: 2320-2882
IMPLEMENTATION DETAILS

Fig 5: user interface

Fig 6: prediction of disease and its causes

Fig 7: preventive measures against the disease

VII. IMPORTANT LIBRARIES /PACKAGE

Keras: The open-source software program known as Keras offers a Python interface for artificial neural networks.
Keras offers the TensorFlow library interface.
TensorFlow: A machine learning and artificial intelligence software library called TensorFlow is open-source and cost-
free. Despite being applicable to a wide range of activities, deep neural network training and inference are given
particular emphasis.

IJCRT2305797 International Journal of Creative Research Thoughts (IJCRT) www.ijcrt.org g544

www.ijcrt.org © 2023 IJCRT | Volume 11, Issue 5 May 2023 | ISSN: 2320-2882
VIII. CONCLUSION
In conclusion, this study proposed a deep learning-based approach for automated detection and classification of plant
leaf diseases using ResNet architecture. Leaf diseases are detected and identified using Deep Learning and image
processing techniques. With the discovery, it will be simpler to spot plant diseases early on and stop crop loss and
disease transmission. The objective of this algorithm is to find anomalies on plants living in the wild or in greenhouses.
Usually, a plain background is used when taking the picture to prevent occlusion The proposed method achieved high
accuracy in identifying and differentiating between different types of diseases affecting plants. Through our experiments
and evaluation, we showed that the ResNet model outperformed other commonly used deep learning models, such as
VGG and Inception-v3. The proposed method has several advantages, including its ability to handle large datasets and
its capability to learn high-level features from images.Overall, this study provides a significant contribution to the
development of automated plant disease detection systems and paves the way for further research in this area.

REFERENCES
[1] Recognition of Plant Diseases using Convolutional Neural Network | G. Madhulatha, O. Ramadevi | Proceedings of
the Fourth International Conference on I-SMAC (IoT in Social, Mobile, Analytics and Cloud) (I-SMAC) | IEEE 2020

[2] Detection of Plant Leaf Diseases using CNN | S.Bharath, K.Vishal Kumar, R.Pavithran, T.Malathi | IRJET
2020

[3] Plant disease classification using deep learning | Akshai KP, J.Anitha | 3rd International Conference on
Signal Processing and Communication (ICPSC) | IEEE 2021

[4] Plant Disease Detection Using Machine Learning | Mr. P V Vinod, Mr. Ramachandra Hebbar, Niveditha M,
Pooja R, Prasad Bhat N, Shashank N | International Conference on Design Innovations for 3Cs Compute
Communicate Control | IEEE 2018

[5] “Machine Learning for Plant Leaf Disease Detection and Classification – A Review,” | L. Sherly Puspha
Annabel, | International Conference on Communication and Signal Processing | IEEE, 2019.

[6] “Detection of Potato Diseases Using Image Segmentation and Multiclass Support Vector Machine,” |
Monzurul Islam, Anh Dinh, Khan Wahid, | CCECE, IEEE, 2019.

[7] “A Review on Machine Learning Classification Techniques for Plant Disease Detection” | Mrs.Shruthi U,
Dr.Nagaveni V, Dr.Raghavendra B K | ICACCS, IEEE, 2022.

[8] “Detection and Classification of Groundnut Leaf Diseases using KNN classifier” | M. P. Vaishnnave, K.
Suganya Devi, P. Srinivasan, G. ArutPerumJothi | IEEE, 2019.

[9] “Plant Disease Detection Techniques: A Review,” | Gurleen Kaur Sandhu, Rajbir Kaur | ICACTM, IEEE,
2019

IJCRT2305797 International Journal of Creative Research Thoughts (IJCRT) www.ijcrt.org g545

www.ijcrt.org © 2023 IJCRT | Volume 11, Issue 5 May 2023 | ISSN: 2320-2882
BIBLIOGRAPHY
1
Dr. T. Praveen Blessington,
Professor, IT Department, Zeal College of Engineering and Research, Pune.
He is having 17 years of experience in teaching and research. His areas of interest are VLSI Design, IoT, Machine
Learning and Blockchain technology. He has published 25 research papers so far.

2
Bhargav Krishnan
Pursuing Bachelor of Engineering in Information Technology from Zeal College of Engineering and Research,
Pune-41.

3
Yash Bharne
Pursuing Bachelor of Engineering in Information Technology from Zeal College of Engineering and Research,
Pune-41.

4
Kartiki Khoje
Pursuing Bachelor of Engineering in Information Technology from Zeal College of Engineering and Research,
Pune-41.

5
Prathamesh Padwal
Pursuing Bachelor of Engineering in Information Technology from Zeal College of Engineering and Research,
Pune-41.

IJCRT2305797 International Journal of Creative Research Thoughts (IJCRT) www.ijcrt.org g546