🌾 Rice Leaf Disease Analysis Framework

A comprehensive computer vision framework for analyzing rice leaf diseases using multiple approaches:

• 📊 Principal Component Analysis (PCA) for infection pattern discovery
• 🌿 Classical computer vision for disease segmentation and severity estimation
• 🤖 Deep learning models for advanced disease detection

The system processes rice leaf images and produces:

• PCA feature loadings
• spatial infection maps
• automatic leaf segmentation
• disease spot detection
• infection severity estimation
• lesion segmentation using deep learning
• disease classification
• severity heatmaps

The goal is to enhance subtle disease patterns in rice leaves that are difficult to observe in standard RGB images.

---

📌 Project Overview

Plant diseases significantly impact global agricultural productivity, especially in rice farming.

Early disease detection helps:

• reduce crop loss
• improve yield management
• support precision agriculture

This project combines three different computer vision approaches for analyzing rice leaf infections.

1️⃣ PCA Feature Analysis

Unsupervised feature analysis used to detect hidden infection patterns.

2️⃣ Classical Computer Vision Pipeline

Traditional image processing methods for disease detection.

3️⃣ Deep Learning Pipeline

Modern deep learning models for segmentation and classification.

Together these pipelines create a complete research workflow for plant disease detection.

---

🧠 Methodology

The framework contains three main analysis pipelines.

---

📊 Pipeline 1 — PCA Infection Pattern Analysis

This pipeline reveals hidden infection patterns using Principal Component Analysis.

1️⃣ Image Loading

The rice leaf image is loaded using OpenCV and converted from BGR to RGB format.

2️⃣ White Balance Correction

A Gray-World assumption is applied to normalize lighting conditions.

3️⃣ Feature Extraction

Multiple image features are extracted.

Color Channels

• R
• G
• B

Vegetation Indices

• ExG (Excess Green) – vegetation intensity
• ExR (Excess Red) – highlights disease regions
• NDI (Normalized Difference Index) – chlorophyll variation

Normalized RGB

• r_norm
• g_norm
• b_norm

Color Deconvolution (HED)

Separates stain-like components:

• Hematoxylin
• Eosin
• DAB

(optional depending on image compatibility)

Gaussian Blur Features

Two blurred images are generated:

• σ = 1
• σ = 2

These capture texture and spatial information.

---

📊 Principal Component Analysis (PCA)

All extracted features are:

1. Flattened
2. Z-score normalized
3. Processed using PCA

The model extracts:

• PC1
• PC2
• PC3

These components represent the dominant patterns in the leaf image.

---

🌿 Pipeline 2 — Classical Computer Vision Disease Detection

This pipeline performs automatic disease detection using image processing.

Leaf Segmentation

The leaf is isolated using HSV color thresholding.

Disease Spot Detection

Disease regions are detected using vegetation indices such as ExG.

Infection Severity Estimation

Disease severity is calculated as:

severity = diseased_area / leaf_area × 100

Machine Learning Classification

A Random Forest classifier predicts disease type.

Example classes:

• Healthy
• Brown Spot
• Leaf Blast

---

🤖 Pipeline 3 — Deep Learning Disease Detection

A deep learning pipeline is implemented for advanced disease analysis.

Leaf Isolation

Uses DeepLabV3 semantic segmentation to isolate the leaf.

Lesion Segmentation

Uses a U-Net CNN to detect infected regions.

Disease Classification

Uses EfficientNet to classify leaf diseases.

Example categories:

• Healthy
• Brown Spot
• Leaf Blast

Severity Heatmap

A color heatmap visualizes the infection intensity across the leaf.

---

🖼 Generated Outputs

The framework generates several visual outputs.

1️⃣ PCA Loadings Plot

pca_loadings.png

Shows how each feature contributes to principal components.

---

2️⃣ PCA Score Maps

pc1_scoremap.png pc2_scoremap.png pc3_scoremap.png

These maps highlight:

• infection regions
• chlorosis
• texture variation

---

3️⃣ False Color PCA Composite

pc_rgb_composite.png

Mapping:

• Red → PC1
• Green → PC2
• Blue → PC3

This creates a feature-enhanced visualization of the leaf.

---

4️⃣ Leaf Segmentation Map

Binary mask representing detected leaf area.

---

5️⃣ Disease Detection Map

Detected infection regions on the leaf.

---

6️⃣ Severity Heatmap

Color heatmap highlighting disease intensity.

---

⚙️ Installation

Install required dependencies:

pip install scikit-image scikit-learn opencv-python matplotlib torch torchvision segmentation-models-pytorch

Or in Google Colab:

!pip install scikit-image scikit-learn opencv-python matplotlib torch torchvision segmentation-models-pytorch

---

🚀 Usage

Step 1 — Run the script

Run one of the pipelines.

PCA Analysis

python pca_rice_leaf_analysis.py --image rice_leaf.jpg

Classical Detection

python classical_rice_disease_detection.py

Deep Learning Detection

python deep_learning_rice_disease_detection.py

---

📁 Project Structure

rice-leaf-disease-analysis/ │ ├── pca_rice_leaf_analysis.py ├── classical_rice_disease_detection.py ├── deep_learning_rice_disease_detection.py │ ├── README.md │ ├── results/ │ ├── pca_loadings.png │ ├── pc1_scoremap.png │ ├── pc2_scoremap.png │ ├── pc3_scoremap.png │ └── pc_rgb_composite.png │ └── sample_images/ └── rice_leaf_example.jpg

---

🔬 Applications

This framework can be used for:

• 🌾 Rice disease detection
• 🌱 Plant stress analysis
• 🧪 Agricultural research
• 📷 Image-based crop monitoring
• 🤖 Feature engineering for ML models
• 🚜 Precision agriculture

---

🧩 Future Improvements

Possible enhancements:

• real-time disease detection systems
• hyperspectral imaging analysis
• Vision Transformer models
• mobile applications for farmers
• cloud-based crop monitoring platforms

---

🛠 Technologies Used

• Python
• OpenCV
• Scikit-image
• Scikit-learn
• PyTorch
• segmentation-models-pytorch
• Matplotlib
• Google Colab

---

📜 License

This project is released under the MIT License.

---

👨‍💻 Author

Developed for research in plant disease detection using computer vision and deep learning.

If you found this project useful, consider ⭐ starring the repository.