13 VII July 2025
[Link]
International Journal for Research in Applied Science & Engineering Technology (IJRASET)
ISSN: 2321-9653; IC Value: 45.98; SJ Impact Factor: 7.538
Volume 13 Issue VII July 2025- Available at [Link]
Plant Disease Detection Using Deep Learning
Jaya Kethan Reddy Dulla1, M. Dhanalakshmi2
1
Post-Graduate Student, Department of Information Technology, Data Science, Jawaharlal Nehru Technological University,
Hyderabad, India
2
Professor of Information Technology, Jawaharlal Nehru Technological University, Hyderabad, India
Abstract: The early and accurate detection of plant diseases plays a pivotal role in enhancing crop health and ensuring food
security in modern agriculture. Traditional disease diagnosis techniques often rely on visual inspection by experts, which may be
subjective, time-consuming, and inaccessible to farmers in remote regions. To overcome these limitations, this project introduces
an intelligent plant disease detection system that leverages deep learning and image analysis to identify symptoms directly from
leaf images. The proposed methodology begins with the collection of a curated dataset comprising various plant species
exhibiting both healthy and diseased conditions. The images undergo preprocessing to enhance quality and ensure consistency,
followed by feature extraction using Convolutional Neural Networks (CNNs). Transfer learning is applied to improve model
generalization and reduce the training time by utilizing pre-trained models. The system is integrated into a Flask-based web
application, enabling users to upload leaf images and receive instant disease diagnoses along with treatment suggestions and
suitable fertilizers. Evaluation of the model has shown high classification accuracy across multiple disease classes, affirming its
potential to support precision agriculture. The solution is designed to be lightweight, user-friendly, and deployable in real-world
agricultural settings, aiming to assist farmers with timely and informed interventions to mitigate crop loss and promote
sustainable farming practices.
Keywords: CNN, Plant Disease Detection, Deep Learning, Image Processing, Flask Web App, Transfer Learning, Agriculture,
Real-time Diagnosis.
I. INTRODUCTION
Agriculture plays a crucial role in India's economic development, with nearly 70% of the population depending on it either directly
or indirectly. Given its significance, any decline in crop productivity due to plant diseases can lead to considerable economic
setbacks. Since plant leaves often exhibit the earliest signs of disease, they serve as key indicators for early diagnosis. Continuous
monitoring of crops from their germination stage to harvest is essential to mitigate the impact of diseases. Traditionally, disease
detection relied on visual inspection by experts—a manual, time-consuming, and labour-intensive method that is not scalable for
large-scale farming. To address this challenge, our project introduces a deep learning-based system that leverages Convolutional
Neural Networks (CNNs) to detect diseases in plant leaves, particularly focusing on crops like cotton. This automated solution
facilitates early diagnosis of bacterial infections, reducing the need for farmers to be physically present in the field for prolonged
periods. Early and accurate disease identification is crucial, as it supports timely treatment and minimizes damage to both crop yield
and quality. Additionally, the system empowers farmers by providing reliable diagnoses without requiring advanced technical
expertise. In recent years, intelligent plant disease detection techniques have gained traction, offering faster, cost-effective, and
more accurate alternatives to traditional methods. These AI-powered solutions eliminate the dependency on manual observation and
reduce the likelihood of human error in disease recognition. Our model is trained to recognize the subtle visual symptoms that often
go unnoticed by the naked eye, such as changes in leaf color, texture, and shape. By automating the diagnosis process, this system
promotes more sustainable agricultural practices and supports food security by minimizing yield loss due to undetected plant
diseases.
A. Objective
1) To build an automated system that takes audio input and turns it into a complete, high-quality video, helping reduce manual
work and speed up the content creation process.
2) To make use of powerful AI tools—like Whisper for converting speech to text, Groq for generating meaningful scripts, Pexels
API for finding suitable images, and ESRGAN for improving video quality—to create visually engaging and informative
content.
3) To ensure the final video feels immersive by properly syncing audio with relevant visuals, offering a smooth and impactful
viewing experience.
©IJRASET: All Rights are Reserved | SJ Impact Factor 7.538 | ISRA Journal Impact Factor 7.894 | 979
International Journal for Research in Applied Science & Engineering Technology (IJRASET)
ISSN: 2321-9653; IC Value: 45.98; SJ Impact Factor: 7.538
Volume 13 Issue VII July 2025- Available at [Link]
4) To provide an easy and innovative way for creators, educators, and media professionals to generate videos directly from audio,
making the content creation process faster and more accessible.
II. LITERATURE SURVEY
1) Brahimi et al. presented a CNN-based system for identifying plant leaf diseases. The model worked well for common crops, but
its performance was constrained by the size and quality of the training dataset. Their work showed promising accuracy but
lacked scalability across diverse plant types.
2) Sladojevic et al. implemented a deep learning-based approach using CNN for feature extraction. The system demonstrated high
accuracy in disease classification but was resource-intensive, requiring longer training times and access to powerful hardware
like GPUs.
3) Mohanty et al. used transfer learning by applying models like AlexNet and VGG trained on large image datasets. This method
significantly improved efficiency and reduced training time, though the system’s success heavily depended on having well-
labelled and clean image data.
4) Kamilaris and Prenafeta-Boldú developed a real-time, mobile-compatible deep learning model that allowed users to detect plant
diseases on-site using a smartphone. While convenient and scalable, their app’s accuracy dropped under poor lighting or image
quality conditions.
5) Ferentinos trained CNNs on a very large dataset containing more than 80,000 images across 58 different classes of plant
diseases and healthy crops. This approach achieved very high accuracy but faced challenges when applied to new or uncommon
plant diseases do not present in the training set.
III. METHODOLOGY OF THE PROPOSED SYSTEM
A. Proposed System
The proposed system follows a systematic pipeline that includes data collection, preprocessing, feature extraction, classification,
and web-based deployment. Each step is crucial in building a robust and accurate disease detection model using deep learning
techniques. The methodology is designed to ensure high performance in both model accuracy and real-time usability.
1) Data Collection
A high-quality image dataset is critical for training a deep learning model effectively. This project uses the PlantVillage dataset, a
well-known open-source repository that contains labeled images of healthy and diseased plant leaves. The dataset includes multiple
classes covering crops like tomato, potato, and bell pepper with various disease categories. Each image is labelled for supervised
learning and represents diverse real-world conditions.
2) Image Preprocessing
To ensure consistency and improve learning performance, preprocessing steps are applied to the raw input images:
Resizing: All images are resized to a standard dimension of 256x256 pixels.
Normalization: Pixel values are scaled between 0 and 1 to ensure numerical stability during training.
Augmentation: Techniques like rotation, flipping, and zooming are applied to artificially expand the dataset and improve
generalization.
Color Standardization: Images are converted to RGB format for consistent channel interpretation.
Feature Extraction using CNN: Convolutional Neural Network (CNN) is used to extract hierarchical image features. CNN is
well-suited for spatial data like images and excels in detecting edges, textures, and patterns which are essential for identifying
symptoms of plant diseases. The model architecture includes:
Convolutional Layers to extract low-level features
ReLU Activation to introduce non-linearity
Pooling Layers to reduce spatial dimensions
Fully Connected Layers for classification
Transfer learning is applied using a pre-trained model (e.g., ResNet50 or VGG16) to reduce training time and improve performance.
©IJRASET: All Rights are Reserved | SJ Impact Factor 7.538 | ISRA Journal Impact Factor 7.894 | 980
International Journal for Research in Applied Science & Engineering Technology (IJRASET)
ISSN: 2321-9653; IC Value: 45.98; SJ Impact Factor: 7.538
Volume 13 Issue VII July 2025- Available at [Link]
3) Model Training
The model is trained using the pre-processed image data with categorical cross-entropy loss and the Adam optimizer. The training
process involves:
Dividing the dataset into training and testing subsets.
Monitoring performance through metrics like accuracy, precision, recall, and F1-score.
Using early stopping to prevent overfitting and save the best performing model.
Real-Time Prediction via Flask Web Interface
To make the model accessible to end-users, a Flask-based web application is developed. This allows users to upload a leaf image
and receive instant disease diagnosis. The pipeline includes:
Accepting the uploaded image.
Preprocessing it similarly to the training pipeline.
Predicting the class using the trained CNN.
Displaying disease name, plant type, and treatment recommendations.
B. System Architecture
The system architecture defines the overall structure and workflow of the plant disease detection model. It outlines how various
components interact to process input leaf images, extract meaningful features, and classify them based on health conditions. This
design integrates both feature compression and deep learning techniques to ensure accurate and efficient disease identification. By
combining a Convolutional Autoencoder (CAE) for dimensionality reduction and a Convolutional Neural Network (CNN) for
classification, the system achieves a balance between performance and computational efficiency. The architecture also supports a
structured flow for both training and testing phases, ensuring consistent results across different image inputs.
Fig: System Architecture
1) Data Collection: The initial step involves gathering a comprehensive set of plant leaf images from various crops. This dataset
contains examples of both healthy and infected leaves, each displaying different visual symptoms. A diverse and well-balanced
dataset is essential for training a reliable model, as it allows the system to learn from a wide range of disease conditions and
improves its ability to handle unseen test images.
2) Image Preprocessing: Before any analysis begins, the raw images undergo preprocessing to prepare them for model input. This
includes resizing each image to a fixed dimension (commonly 256×256 pixels), normalizing pixel values, converting color
formats, and enhancing contrast where needed. These steps ensure consistency in input format and eliminate noise or distortions,
resulting in better training performance.
©IJRASET: All Rights are Reserved | SJ Impact Factor 7.538 | ISRA Journal Impact Factor 7.894 | 981
International Journal for Research in Applied Science & Engineering Technology (IJRASET)
ISSN: 2321-9653; IC Value: 45.98; SJ Impact Factor: 7.538
Volume 13 Issue VII July 2025- Available at [Link]
3) Feature Reduction Using Autoencoders: After preprocessing, the images are passed through a Convolutional Autoencoder
(CAE). This specialized neural network compresses each image into a lower-dimensional feature vector while retaining
important visual characteristics. The encoder part of the CAE captures only the most meaningful data, helping reduce storage
and processing demands without sacrificing accuracy.
4) CNN Model Training: The compressed outputs from the encoder are then used to train a Convolutional Neural Network (CNN).
The CNN is responsible for learning to differentiate between various plant diseases and healthy leaves. Since the input data is
already simplified through compression, the CNN can train more efficiently and focus on recognizing disease-specific textures
and patterns.
5) Test Image Input via Web Interface: In the practical application phase, users such as farmers upload new leaf images through a
user-friendly web interface. The system then follows the same pipeline that was used during training to analyze the new image.
6) Preprocessing of Test Images: The uploaded image is subjected to the same preprocessing operations applied during training.
This ensures the image has the correct format, resolution, and quality before feature extraction begins, which is crucial for
consistent results.
7) Feature Extraction from Test Images: Using the pre-trained encoder component of the CAE, the system generates a compact set
of features from the test image. These features summarize the image's content—such as shape, edges, and color patterns—that
are most relevant for detecting signs of disease.
8) Classification using CNN: Finally, the extracted features are passed to the trained CNN, which predicts whether the leaf is
healthy or affected by a disease. If a disease is detected, the system may further specify the type of infection. The classification
decision is based on the network’s prior learning and pattern recognition capabilities.
C. Algorithm
The Convolutional Neural Network (CNN) algorithm used for plant disease detection follows a systematic sequence of operations to
process input images and classify them accurately. The process begins with the input layer, where an image of a plant leaf is taken
and resized to a consistent dimension, typically 256×256 pixels. This image is then passed into the convolutional layers, where
multiple filters slide over the image and extract essential features such as color changes, spots, edges, or textures that are commonly
associated with plant diseases. After each convolution operation, the ReLU (Rectified Linear Unit) activation function is applied to
introduce non-linearity, ensuring the model can capture complex patterns in the data. Next, the data is passed through a pooling
layer, usually max pooling, which reduces the spatial size of the feature maps. This helps in decreasing computation time and
making the model less sensitive to small distortions in the input. The output of the pooling layer is then flattened into a one-
dimensional vector to prepare it for the dense layers. This flattened feature vector is fed into one or more fully connected layers,
where the model combines all the extracted features and learns to classify the image based on disease characteristics. Finally, the
data reaches the output layer, which applies a softmax activation function to generate a probability distribution over all possible
classes. The class with the highest probability is selected as the prediction — for example, “Potato – Late Blight” or “Tomato –
Healthy.” This structured approach allows CNNs to effectively analyze plant leaf images and identify diseases with high accuracy.
IV. IMPLEMENTATION AND RESULTS
A. Implementation Steps
1) Image Upload via Web Interface: The process begins when the user—typically a farmer or agronomist—accesses the Flask-
based web application. The interface allows the user to upload an image of a plant leaf, which could either be healthy or
affected by a disease. This feature ensures ease of use and accessibility across devices
2) Image Storage and Path Retrieval: Once the image is uploaded, it is securely stored in a predefined local directory on the server
(e.g., /uploads). The system then retrieves the full file path of the image, which will be used in subsequent processing stages.
3) Image Resizing and Normalization (Preprocessing): To make the uploaded image compatible with the CNN model’s expected
input format, it is resized to 256x256 pixels. The resizing ensures that the model can process different image dimensions
consistently. The pixel values are also scaled (typically divided by 255) so that the data falls within a normalized range (0 to 1),
which improves model performance and accuracy.
4) Real-Time Prediction Pipeline: The preprocessed image is then passed into the pre-trained Convolutional Neural Network
(CNN). This model has already been trained on a labeled dataset of plant leaf images—both healthy and diseased. It analyzes
the features in the uploaded image and classifies it into one of the predefined categories.
©IJRASET: All Rights are Reserved | SJ Impact Factor 7.538 | ISRA Journal Impact Factor 7.894 | 982
International Journal for Research in Applied Science & Engineering Technology (IJRASET)
ISSN: 2321-9653; IC Value: 45.98; SJ Impact Factor: 7.538
Volume 13 Issue VII July 2025- Available at [Link]
5) Diagnosis and Feature Matching: The CNN processes the image through multiple internal layers (like convolution, activation,
pooling, and fully connected layers) to extract and match features such as color changes, texture anomalies, or spots that
indicate disease. It then predicts the plant name and identifies the disease, if any.
6) Intelligent Output Generation: Based on the classification result (using the model’s highest-probability output), the system
dynamically generates a diagnosis report. This includes:
The name of the plant.
The detected disease or an indication that the plant is healthy.
Suggested treatment methods.
Recommended fertilizers or fungicides (if needed).
7) Result Rendering and Display: All this information is passed to the result page ([Link]) and displayed to the user in an
easily readable format. The entire flow—from image upload to output rendering—is executed in real time, typically taking just
a few seconds.
8) Model Maintenance and Updates: The underlying model can be periodically updated with new data collected from real-world
usage. By retraining with an expanded dataset, the system improves its ability to detect a wider range of diseases and adapt to
different environmental conditions.
B. Results
It highlights how effectively the system identifies various plant diseases using deep learning techniques, particularly Convolutional
Neural Networks (CNN). The evaluation is based on test images, comparing the system’s predictions with actual classifications to
measure accuracy and reliability. The results demonstrate the practicality and efficiency of the model in real-time disease
identification, providing valuable insights into its effectiveness and potential for real-world agricultural use.
Fig: User Interface
It shows the User Interface to choose and upload the image from your device.
Fig: Uploading images
©IJRASET: All Rights are Reserved | SJ Impact Factor 7.538 | ISRA Journal Impact Factor 7.894 | 983
International Journal for Research in Applied Science & Engineering Technology (IJRASET)
ISSN: 2321-9653; IC Value: 45.98; SJ Impact Factor: 7.538
Volume 13 Issue VII July 2025- Available at [Link]
In fig: We must choose one image. By clicking on predict the output will be displayed.
Fig: Detecting Disease
V. LIMITATIONS AND FUTURE SCOPE
Although the proposed plant disease detection system using deep learning shows promising accuracy and efficiency, it does come
with a few limitations. The performance of the model is heavily dependent on the quality and size of the dataset. In real-world
environments, varying lighting conditions, complex backgrounds, and overlapping leaves can impact the accuracy of predictions.
Additionally, the model might struggle with identifying diseases in their very early stages if symptoms are not visually prominent.
Another limitation is that the model is typically trained on a fixed set of plant species and diseases; thus, its ability to generalize to
new or rare cases is limited. Moreover, reliance on high computational resources may restrict deployment on devices with lower
processing capabilities such as mobile phones without optimization. Looking ahead, the system can be expanded to include real-
time mobile applications using optimized lightweight models like MobileNet or TensorFlow Lite. Increasing the dataset size with
diverse environmental conditions and disease stages can help improve robustness. Integration with GPS and IoT devices can make it
more useful for smart farming applications. Moreover, by incorporating multi-modal data like soil condition, humidity, and
temperature, the model can offer more accurate and context-aware predictions. Future versions can also support automated
treatment suggestions, notifications to local agricultural officers, or integration with drones for large-scale farm monitoring. These
improvements would significantly enhance the utility and scalability of the system for modern agriculture.
VI. CONCLUSION
The proposed system effectively demonstrates how deep learning techniques, particularly Convolutional Neural Networks (CNNs),
can be applied to identify plant diseases with high accuracy using leaf images. By automating the detection process, the model helps
in early diagnosis, which is crucial in minimizing crop damage and improving agricultural productivity. Through systematic stages
such as data collection, preprocessing, feature extraction, and classification, the system is capable of distinguishing between healthy
and diseased plants and even identifying specific types of diseases. The integration of a Flask-based web application further
enhances the usability of the system by offering real-time predictions to end-users like farmers or agronomists. Although certain
limitations exist—such as environmental noise or limited training data—the project lays a solid foundation for building intelligent,
scalable, and practical tools for plant health monitoring. Overall, this work highlights the potential of AI-driven solutions in
advancing smart agriculture and promoting sustainable farming practices.
VII. ACKNOWLEDGMENT
I sincerely thank Sri M. Dhanalakshmi madam for her continual support in helping me carry out this project. Also extend my
heartfelt gratitude to the authors of the research papers reviewed in this study.
REFERENCES
[1] M. H. Saleem, J. Potgieter, and K. M. Arif, “Plant Disease Detection and Classification by Deep Learning,” Plants, vol. 8, no. 11, Art. 468, Nov. 2019.
PubMed+11MDPI+11PMC+11
[2] S. Mohanty, D. P. Hughes, and M. Salathé, “Using Deep Learning for Image-Based Plant Disease Detection,” Frontiers in Plant Science, vol. 7, p. 1419, 2016.
arXiv+1Wikipedia+1
[3] L. Li, S. Zhang, and B. Wang, “Plant Disease Detection and Classification by Deep Learning—A Review,” IEEE Access, vol. 2021, Art. No. 3069646, 2021.
Astrophysics Data System+1ResearchGate+1
[4] Z. Zainab et al., “Plant Disease Detection Using Deep Learning Techniques,” IECE Journal of Image Analysis and Processing, vol. 1, no. 1, pp. 36–44,
Mar. 2025.
[5] B. S. Hamed, M. M. Hussein, and A. M. Mousa, “Plant Disease Detection Using Deep Learning,” International Journal of Intelligent Systems and
Applications (IJISA), vol. 15, no. 6, pp. 38–50, Dec. 2023
©IJRASET: All Rights are Reserved | SJ Impact Factor 7.538 | ISRA Journal Impact Factor 7.894 | 984