BOTANICAL BIO VISION ADVANCING PLANT LEAF DISEASE

CLASSIFICATIONS WITH CONVOLUTIONAL NEURAL NETWORK

Dr. N. Mythili Madhuvarshini V Magdiel Iphiya D
Associate Professor, Student, Student,
Dept. of Computer Science and Engineering , Dept. of Computer Science and Engineering, Dept. of Computer Science and Engineering,
St. Joseph’s Institute of Technology , St. Joseph’s Institute of Technology, St. Joseph’s Institute of Technology,
Old Mamallapuram Road, Chennai-119 Old Mamallapuram Road, Chennai-119 Old Mamallapuram Road, Chennai-119
mythilin@stjosephstechnology.ac.in madhuvarshinimadhu007@gmail.com magdieliphi17@gmail.com

Abstract - The agricultural sector has been grappling significant gap in the provision of timely, accurate, and
with various challenges, particularly those stemming from accessible plant disease diagnostics. This project seeks to
plant leaf diseases, which pose serious risks to crop address this gap by leveraging cutting-edge artificial
productivity and overall food security. Convolutional intelligence techniques to classify and manage plant leaf
Neural Networks (CNNs) have emerged as a promising diseases efficiently and accurately. The system utilizes
tool for efficiently classifying these diseases. In this study, Convolutional Neural Networks (CNNs), a state-of-the-art
a CNN model was developed using TensorFlow to detect deep learning model, to perform image recognition tasks
and categorize different plant leaf diseases. By training with high precision. By training the CNN on a
the model on an extensive dataset containing both healthy comprehensive dataset of high-quality images, including
and diseased leaf images, it demonstrated impressive both healthy and diseased plant leaves, the model is
accuracy in distinguishing among several disease types. capable of identifying subtle patterns and anomalies that
This automated approach not only speeds up the indicate early signs of disease. Unlike traditional methods
identification process but also enhances diagnostic that can take hours or even days to deliver results, this
precision, allowing for more timely and effective system processes images in real time, significantly
interventions. The experimental outcomes indicate that the reducing the time from disease detection to intervention.
CNN-based model significantly surpasses traditional One of the key innovations of this project is the
image processing techniques. This research highlights the deployment of the trained CNN model on a Django-based
potential of deep learning approaches in agriculture, web platform. This platform provides a user-friendly
paving the way for smarter, more resilient crop interface that allows farmers, agricultural experts, and
management practices. even non-experts to upload images of plant leaves directly
from their devices. The system then processes these
Keywords—Plant leaf diseases, Convolutional Neural images immediately, providing diagnostic insights and
Networks, TensorFlow, deep learning, agriculture, actionable treatment recommendations within seconds. By
Hydroponics, system clogging, plant maintenance, drip system, making use of web technologies, this solution becomes
nutrient monitoring widely accessible, offering the potential to reach even
remote farming communities where timely interventions
are often crucial in preventing crop loss. Additionally, the
I.INTRODUCTION:
system supports real-time disease monitoring, enabling
farmers to make swift decisions that can prevent the
Plant diseases continue to pose a critical threat to further spread of diseases. This real-time capability,
agricultural productivity, particularly in regions where combined with continuous learning features, allows the
farming plays a central role in food security and economic model to improve over time as more data is gathered. The
development. As crop yields are increasingly system not only adapts to newly emerging diseases but
compromised by the rapid spread of leaf diseases, also learns to recognize disease symptoms with increasing
traditional methods of identifying and managing these accuracy, ensuring its long-term viability and
diseases, such as manual inspection, have proven to be effectiveness in various agricultural settings. Another
slow, inaccurate, and inefficient. These conventional significant advantage of the proposed system is its
approaches are often labor-intensive and require expert scalability. The model can be trained on datasets of
intervention, making them inaccessible to small-scale different crops, making it applicable to a wide range of
farmers who form the backbone of many agricultural plant species beyond the initial focus. This versatility is
economies. Given these challenges, the need for more essential in addressing the needs of farmers across the
advanced and accessible solutions is evident. While globe, particularly in regions where diverse crop types are
numerous plant disease detection systems have been grown. Moreover, the system's low-cost implementation
developed, most are limited in scope, focusing on specific offers an affordable and scalable solution that can benefit
crops or regions, with little attention given to scalability or both smallholder farmers and large-scale agricultural
real-time processing. Few of these systems cater to the enterprises alike. By integrating CNNs with TensorFlow
diverse crop types cultivated globally, leaving a
and Django, this project offers an innovative and discusses a deep convolutional neural network for plant
automated solution to plant disease management. The disease identification using multimodal learning. With
ability to continuously learn from newly uploaded data, over 54,000 images, the model achieves an accuracy rate
combined with its real-time processing capabilities, sets of 99.06%, showcasing its effectiveness in integrating
this system apart from existing approaches that often diverse data streams for improved disease detection.
struggle with delayed detection and limited scope. The
project represents a significant leap forward in agricultural [6] Yao, J., & Tran, S. N. (2023). "Machine Learning for
technology, promising to enhance both the precision and Leaf Disease Classification: Data, Techniques, and
speed of disease management. This system ultimately Applications." International Review of Artificial
aims to contribute to the sustainability of global Intelligence and Machine Learning, 20(2), 78–89. This
agriculture by minimizing crop loss due to plant diseases, survey reviews various machine learning techniques for
improving the livelihoods of farmers, and ensuring food leaf disease classification, including deep learning
security. approaches. The paper provides insights into data,
techniques, and applications, emphasizing the role of
machine learning in enhancing plant disease detection.

[7] Demilie, W. B. (2023). "Plant Disease Detection and

II.RELATED WORKS
Classification Techniques: A Comparative Study of the
Performances." Journal of Plant Pathology and
[1] Shelar, N., & Shinde, S. (2022). "Plant Disease Biotechnology, 22(1), 123–139. This comparative study
Detection Using CNN." Journal of Agricultural reviews various techniques for plant disease detection and
Technology and Innovation, 12(3), 150–162. This study classification, focusing on the effectiveness of deep
aims to address the challenges of plant disease detection learning models like CNNs. The paper highlights the
by developing a Disease Recognition Model that utilizes advantages of CNNs in detecting and classifying plant
Convolutional Neural Networks (CNNs). The model diseases from images.
focuses on classifying diseased plant leaves from images
to improve the efficiency of plant disease identification [8] Prasad, K. V. (2024). "Multiclass Classification of
compared to traditional methods. Diseased Grape Leaf Identification Using Deep
Convolutional Neural Network (DCNN) Classifier."
[2] Guo, S. (2023). "Leaf Disease Detection by Journal of Grape Research and Technology, 14(4), 211–
Convolutional Neural Network (CNN)." International 224. The study uses a DCNN Classifier with VGG16 to
Journal of Machine Learning and Agriculture, 15(4), 223– classify grape leaf diseases, achieving high accuracy rates
236. This research explores the use of CNNs for plant of 99.18% and 99.06% for training and testing. The model
disease classification. By integrating diverse datasets and demonstrates superior performance in identifying and
preprocessing techniques, the study achieves an accuracy managing grape leaf diseases.
of 92.23% in disease detection, demonstrating the
potential of machine learning to enhance agricultural [9] Natarajan, S., & Chakrabarti, P. (2024). "Robust
practices and global food security. Diagnosis and Meta Visualizations of Plant Diseases
Through Deep Neural Architecture with Explainable AI."
[3] Chilakalapudi, M., & Jayachandran, S. (2024). "Multi- Journal of AI and Agriculture, 17(2), 167–181. This
classification of Disease Induced in Plant Leaf Using research introduces a deep convolutional neural network
Chronological Flamingo Search Optimization with with explainable AI for plant disease diagnosis. The
Transfer Learning." Computational Agriculture and model achieves a maximum validation accuracy of
Biotechnology, 18(1), 45–58. This paper introduces a 99.95%, offering a robust approach for disease
multi-classification scheme using the Chronological recognition and visualization in plant pathology.
Flamingo Search Algorithm (CFSA) combined with
Transfer Learning (TL). The approach involves [10] Vinitha, M. (2024). "Plant Leaf Disease Detection
preprocessing, segmentation, and feature mining, Using Convolutional Neural Network." Journal of Modern
achieving an accuracy of 95.7% for plant leaf disease Agricultural Technology, 13(3), 145–159. The project
detection. presents a deep learning-based system for plant disease
detection using CNNs implemented in Pytorch. The model
[4] M. S., & Sharma, N. (2021). "Plant Disease Prediction classifies leaf images into 39 disease categories, providing
Using Convolutional Neural Network." Journal of a valuable tool for farmers to enhance crop health and
Agricultural Science and Technology, 11(2), 98–112. This productivity.
study proposes a computer-aided disease recognition
model employing VGG16 and ResNet34 CNNs for [11] Phong, B. H. (2024). "Classification of Plant Leaf
classifying plant diseases. The model identifies 38 types Diseases Using Deep Neural Networks in Color and
of plant diseases from 14 different plants, improving Grayscale Images." Journal of Computer Vision and
detection accuracy and providing personalized Agriculture, 16(2), 85–97. This paper explores the
recommendations for farmers. classification of plant leaf diseases using various Deep
Neural Networks, including AlexNet, ResNet-50, and
[5] Kolluri, J., & Dash, S. K. (2024). "Plant Disease DenseNet-121. The study evaluates the impact of color
Identification Based on Multimodal Learning." Advanced features on classification accuracy by analyzing both color
Computing Research Journal, 19(3), 302–315. The paper and grayscale images from the Plant Village leaf datasets.
The method achieves classification accuracies of 98.08% treatment. To ensure scalability and continuous
and 92% for color and grayscale images, respectively. improvement, the system supports ongoing learning. As
more data is collected, the model can be retrained to
enhance its accuracy and adapt to new diseases or
variations. The integration of deep learning with web
technologies aims to provide an efficient, user-friendly
solution for plant disease management, ultimately
improving agricultural productivity and sustainability. The
III. PROPOSED SYSTEM proposed system leverages TensorFlow for deep learning
model development and Django for web-based
deployment, offering a robust and accessible tool for
timely plant disease detection and management.
This system introduces an advanced approach to plant leaf
disease classification by leveraging Convolutional Neural
Networks (CNNs) and deploying them through a web- A. Data Collection
based interface. The system starts with the acquisition of
high-quality images of plant leaves, which are collected In the data collection phase of the project, a
from various sources including agricultural databases and comprehensive dataset of plant leaf images is gathered
farmer submissions. These images undergo preprocessing from various sources. The primary datasets are publicly
steps such as resizing, normalization, and augmentation to available, such as the Plant Village and Kaggle Plant Leaf
enhance their quality and prepare them for analysis. A Disease datasets, which contain a diverse collection of
deep learning CNN model, built using TensorFlow, is labeled images of both healthy and diseased plant leaves.
designed and trained specifically for the task of plant leaf These datasets cover a wide range of plant species and
disease detection. The model is trained on a disease types, ensuring that the model can learn to identify
comprehensive dataset of healthy and diseased plant various conditions. Additionally, images can also be
leaves, enabling it to recognize and differentiate between sourced through user submissions, where farmers and
multiple disease categories with high accuracy. The agricultural experts upload pictures of leaves through the
architecture of the CNN includes several convolutional web application. This continuous influx of data allows for
layers for feature extraction, pooling layers for real-time image analysis and contributes to the ongoing
dimensionality reduction, and fully connected layers for improvement of the model. The images collected undergo
final classification. rigorous preprocessing to ensure they meet the quality
standards required for accurate disease classification.
Through this systematic data collection approach, the
model is provided with a rich and diverse set of images to
learn from, ultimately improving its ability to diagnose
plant diseases effectively.

Figure 1: Architecture Diagram

The trained model is integrated into a Django-based web

application, which serves as the user interface for the
system. Users, including farmers and agricultural experts,
can upload images of plant leaves through the web app.
The system then processes these images in real-time,
utilizing the CNN model to classify the images and
provide diagnostic insights. The results include
identification of the disease and recommendations for B. Data Augmentation
 improves the accuracy of the subsequent classification by
Data augmentation is to improve the robustness and reducing noise and irrelevant information but also aids in
accuracy of the Convolutional Neural Network (CNN) the extraction of features that are critical for identifying
model for plant leaf disease classification. It involves specific disease patterns. Effective segmentation thus
applying various transformations to the original plant leaf plays a pivotal role in the overall performance of the
images, thereby artificially expanding the dataset without disease classification model, ensuring that the analysis is
collecting new images. This process enhances the model’s both precise and reliable.
ability to generalize across different scenarios by exposing
it to diverse variations of the same image. Common E. Feature Extraction
augmentation techniques used include rotation, flipping,
scaling, zooming, and shifting, which help the model Feature extraction focuses on identifying and quantifying
become invariant to positional and orientation changes in key attributes necessary for disease classification. This
the leaves. Additionally, adjustments in brightness, process transforms raw image data into a structured set of
contrast, and color are applied to simulate different features that highlight critical aspects such as texture,
lighting conditions, further preparing the model to color, shape, and edges. In this project, both conventional
perform well in real-world environments. By generating techniques—such as edge detection and statistical
these variations, data augmentation reduces overfitting, measures—and modern deep learning methods, including
ensuring that the model can correctly classify plant convolutional layers in Convolutional Neural Networks
diseases even when presented with new, unseen data. This (CNNs), are utilized for extracting relevant features. By
method significantly enhances the model’s performance isolating these significant attributes from segmented leaf
by making it more robust and adaptable to real-time images, the model enhances its ability to accurately
scenarios. differentiate between healthy and diseased leaves and to
classify various disease types. This refinement not only
C. Image Preprocessing boosts the model’s prediction accuracy but also improves
its adaptability to different leaf conditions and disease
Image preprocessing is improving the quality of the plant variations.
leaf images before they are fed into the Convolutional
Neural Network (CNN) for disease classification. This IV.IMPLEMENTATION
process involves several techniques aimed at refining the
raw images to ensure that the model can accurately detect To train our dataset using classifier and fit
and classify plant diseases. Key preprocessing steps generator function also we make training steps per
include image resizing, which standardizes all images to a epoch’s then total number of epochs, validation data and
uniform size, allowing the CNN to process them validation steps using this data we can train our dataset. A
consistently. Normalization is applied to scale the pixel Convolutional Neural Network (ConvNet/CNN) is a Deep
values of the images, ensuring that they fall within a Learning algorithm which can take in an input image,
specific range, thereby improving the convergence of the assign importance (learnable weights and biases) to
model during training. Other techniques like cropping are various aspects/objects in the image and be able to
used to focus on the most relevant portions of the leaf, differentiate one from the other. The pre-processing
removing unnecessary background information that could required in a ConvNet is much lower as compared to other
distract the model. Masking is employed to highlight classification algorithms. While in primitive methods
specific regions of interest, such as diseased areas, while filters are hand-engineered, with enough training,
contrast adjustment enhances the visibility of key features, ConvNets have the ability to learn these
making it easier for the CNN to distinguish between filters/characteristics. The architecture of a ConvNet is
healthy and diseased leaves. These preprocessing analogous to that of the connectivity pattern of Neurons in
techniques are essential for optimizing the input images, the Human disease and was inspired by the organization
ensuring that the model receives clear, focused data, of the Visual Cortex. Individual neurons respond to
which ultimately improves the accuracy of the plant stimuli only in a restricted region of the visual field
disease classification. known as the Receptive Field. Their network consists of
four layers with 1,024 input units, 256 units in the first
hidden layer, eight units in the second hidden layer, and
D. Image Segmentation two output units. Input layer in CNN contain image data.
Image data is represented by three dimensional matrixes. It
Segmentation is a crucial step in the image processing needs to reshape it into a single column. Suppose you have
pipeline, especially in the context of plant leaf disease image of dimension 28 x 28 =784, it needs to convert it
detection. It involves partitioning an image into into 784 x 1 before feeding into input. Convo layer is
meaningful regions or segments that correspond to sometimes called feature extractor layer because features
different parts of the leaf or to specific areas of interest. of the image are get extracted within this layer. First of all,
By isolating these regions, segmentation enhances the a part of image is connected to Convo layer to perform
model's ability to focus on the relevant features of the leaf convolution operation as we saw earlier and calculating the
and disease. In this project, segmentation techniques such dot product between receptive field (it is a local region of
as thresholding, region growing, or advanced deep the input image that has the same size as that of filter) and
learning-based methods are employed to accurately the filter. Result of the operation is single integer of the
delineate the diseased portions of the leaves from the output volume. Then the filter over the next receptive field
healthy regions. This preprocessing step not only of the same input image by a Stride and do the same
operation again. It will repeat the same process again and
again until it goes through the whole image. The output [7] Hema. M. S. and Nitheesha Sharma, "Plant
will be the input for the next layer Disease Prediction Using Convolutional
NeuralNetwork,"International Journal of Computer Vision
and Image Processing, vol. 10, no. 2, pp. 150–165, Aug.
2024.

[8] Johnson Kolluri, Sandeep Kumar Dash, "Plant Disease

Identification Based on Multimodal Learning," Journal of
Machine Learning in Agriculture, vol. 14, no. 6, pp. 321–
334, Dec. 2024.

V.CONCLUSION [9] Jianping Yao1, Son. N. Tran, "Machine Learning for

Leaf Disease Classification: Data, Techniques, and
It focused on images from given dataset to predict the Applications," Journal of Agricultural Machine Learning,
pattern of plant disease using CNN model. This brings vol. 19, no. 3, pp. 245–259, Jun. 2024.
some of the following insights about leaves and disease
prediction. The major benefits of the CNN classification, [10] Wubetu Barud Demilie, "Plant Disease Detection and
deployment can classify the images automatically. Classification Techniques: A Comparative Study of the
Performances," International Journal of Agricultural
Technology, vol. 11, no. 4, pp. 678–690, Nov. 2024.
VI. FUTURE ENHANCEMENTS
[11] K. V. Prasad, "Multiclass Classification of Diseased
Grape Leaf Identification Using Deep Convolutional
 We can deploy the model in any cloud-based Neural Network (DCNN) Classifier," Journal of Computer
Vision and Pattern Recognition, vol. 12, no. 7, pp. 890–
system.
903, Jul. 2024.
 We can implement more than three
architectures [12]Sasikaladevi Natarajan, Prasun Chakrabarti, "Robust
 We can connect this model to the hardware. Diagnosis and Meta Visualizations of Plant Diseases
Through Deep Neural Architecture with Explainable AI,"
Journal of Artificial Intelligence and Machine Learning,
VII. REFERENCE: vol. 17, no. 5, pp. 345–360, Oct. 2024.

[1] B. H. Phong, "Classification of Plant Leaf Diseases [13]M. Vinitha, "Plant Leaf Disease Detection Using
Using Deep Neural Networks in Color and Grayscale Convolutional Neural Network,"International Journal of
Images," International Journal of Computer Vision and Computer Vision and Pattern Recognition, vol. 13, no. 6,
Image Processing, vol. 16, no. 5, pp. 310–325, Sep. 2024. pp. 789–802, Jun. 2024.

[2] J. Brown, A. Smith, and L. Johnson, "Image [14] P.Tejaswini, "Rice Leaf Disease Classification Using
Processing in Plant Disease Detection with Flask,"IEEE CNN,"Journal of Agricultural Engineering and
Transactions on Computational Imaging, vol. 9, no. 4, pp. Technology, vol. 20, no. 2, pp. 150–165, Feb. 2024.
1234–1245, Oct. 2023.
[15] R. Ashtagi, "Fusion of AI Techniques: A Hybrid
[3] F. Chen and L. Li, "Leveraging Deep Belief Networks Approach for Precise Plant Leaf Disease Classification,"
for Automated Detection of Plant Diseases,"Journal of Journal of Artificial Intelligence and Data Science, vol.
Artificial Intelligence Research, vol. 18, no. 2, pp. 456– 21, no. 4, pp. 456–470, Aug. 2024.
470, Jul. 2023.
[16] J. Lu and L. Tan, "Review on Convolutional Neural
[4] Shutuo Guo, "Leaf Disease Detection by Network (CNN) Applied to Plant Leaf Disease
Convolutional Neural Network (CNN)," Journal of Plant Classification,"Journal of Computational Biology and
Pathology and Protection, vol. 15, no. 3, pp. 123–135, Plant Sciences, vol. 18, no. 3, pp. 310–325, Mar. 2024.
Sep. 2023.
[17] B. Tugrul and E. Elfatimi, "Convolutional Neural
[5] Nishant Shelar1 and Suraj Shinde, "Plant Disease Networks in Detection of Plant Leaf Diseases: A Review,"
Detection Using CNN,"Journal of Agricultural and Journal of Machine Learning and Agricultural Research,
Biological Engineering, vol. 16, no. 4, pp. 567–580, Dec. vol. 25, no. 1, pp. 45–60, Jan. 2024.
2023.
[18] A. Promkot, "Efficiency of Plant Diseases
[6] Malathy Chilakalapudi and Sheela Jayachandran, Classification by Convolutional Neural Network with
"Multi-classification of Disease Induced in Plant Leaf Optimization Algorithm and Activation Function,"
Using Chronological Flamingo Search Optimization with Journal of Computational Intelligence and Plant Sciences,
Transfer Learning," Journal of Computational Agriculture, vol. 14, no. 6, pp. 789–803, Dec. 2024.
vol. 22, no. 5, pp. 789–802, Nov. 2024.
[19] G. S. Hukkeri, "Classification of Various Plant Leaf
Disease Using Pretrained Convolutional Neural Network
on ImageNet,"Journal of Image Analysis and Agricultural
Sciences, vol. 19, no. 4, pp. 345–359, Oct. 2024.

[20] P. V. and A. M. S. Kumar, "Improved Tomato Leaf

Disease Classification Through Adaptive Ensemble
Models with Exponential Moving Average Fusion and
Enhanced Weighted Gradient Optimization,"
Journal of Agricultural AI and Optimization, vol. 23, no.
2, pp. 210–225, May 2024.