Rice Leaf Disease Detection with ResNet-50

This project demonstrates the application of deep learning to classify rice leaf diseases using the ResNet-50 architecture. It employs transfer learning and data augmentation techniques to achieve high accuracy in detecting diseases across four classes.

---

Table of Contents

1. Project Overview
2. Dataset Description
3. Model Architecture
4. Training Process
5. Evaluation Metrics
6. Results
7. Use Case
8. How to Use
9. Dependencies
10. Acknowledgments

---

Project Overview

Rice is a staple crop, and its diseases significantly impact food security worldwide. This project addresses the challenge of rice leaf disease classification using a Convolutional Neural Network (CNN). By leveraging the ResNet-50 model, this project classifies images into one of four disease categories:

1. Bacterial Blight
2. Blast
3. Brown Spot
4. Tungro

---

Dataset Description

The dataset consists of images of rice leaves categorized into four classes of diseases. The dataset is split as follows:

• Training Set: 3,082 images
• Validation Set: 2,052 images (split from the training set)
• Testing Set: 800 images

Image dimensions are resized to 224x224 pixels to fit the ResNet-50 input requirements.

---

Model Architecture

This project uses a pre-trained ResNet-50 model as the feature extractor, followed by a custom classification head. The detailed architecture includes:

1. Input Layer: (224, 224, 3)
2. ResNet-50 Feature Extractor: Pre-trained on ImageNet.
3. Global Average Pooling: Reduces dimensionality.
4. Dropout Layer: Prevents overfitting.
5. Dense Layers:
   • Fully connected layer with 128 neurons (ReLU activation).
   • Output layer with 4 neurons (linear activation for multi-class classification).

Optimizer: Stochastic Gradient Descent (SGD)
Loss Function: Squared Hinge Loss

---

Training Process

Key aspects of the training process include:

1. Image Augmentation:
   • Rotation up to 90 degrees.
   • Horizontal and vertical flipping.
   • Nearest neighbor fill for augmented pixels.
2. Validation Split: 40% of the training set is used for validation.
3. Callback: Training stops early if the model achieves >90% accuracy and validation accuracy.

---

Evaluation Metrics

The model is evaluated using the following metrics:

• Accuracy: Overall prediction correctness.
• Precision: Correctly identified positives.
• Recall: Coverage of actual positives.
• F1-Score: Harmonic mean of precision and recall.
• Confusion Matrix: Visualizes classification results.

---

Results

Training and Validation Loss

Training and Validation Accuracy

Confusion Matrix

Final Metrics

• Training Accuracy: 100%
• Validation Accuracy: 95.86%
• Test Accuracy: 98.37%

Classification Report

Class	Precision	Recall	F1-Score	Support
Bacterial Blight	1.0000	0.9450	0.9717	200
Blast	0.9851	0.9950	0.9900	200
Brown Spot	0.9950	0.9950	0.9950	200
Tungro	0.9569	1.0000	0.9780	200
Overall	0.9838	0.9838	0.9837	800

---

Use Case

This repository is designed for researchers, agronomists, and developers looking for a robust solution to detect and classify rice leaf diseases, contributing to better crop management and food security.

---

How to Use

1. Clone this repository:

   git clone https://github.com/filzarahma/padiku-app.git
   

2. Install the required dependencies:

   pip install tensorflow==2.8.0 keras==2.8.0 scikit-learn matplotlib seaborn
   

3. Upload the dataset into the appropriate directories:
   • train/
   • test/
4. Run the Jupyter Notebook:

   jupyter notebook rice_leaf_disease_detection.ipynb
   

5. Train the model or use the pre-trained weights model_CS.h5.

---

Dependencies

The project was built and tested using the following dependencies:

• TensorFlow 2.8.0
• Keras 2.8.0
• scikit-learn
• matplotlib
• seaborn

Ensure you have Python 3.7 or higher installed.

---

Acknowledgments

This project is made possible with:

• The Rice Leaf Disease Images Dataset.
• The ResNet-50 architecture for its robust feature extraction capabilities.

If you use this project, please cite or reference appropriately.