LeafDisease-AI: Deployable Deep Learning for Cross-Domain Plant Leaf Disease Detection via Ensemble Learning, Knowledge Distillation, and Quantization

🌟 Overview

LeafDisease-AI is the first comprehensive framework for cross-domain tomato leaf disease detection, bridging the gap between laboratory research and real-world agricultural deployment. This repository implements a unified optimization approach integrating ensemble learning, knowledge distillation, and quantization for edge-compatible disease detection.

🎯 Key Contributions

• First Open Cross-Domain Benchmark: Unifying PlantVillage and TomatoVillage datasets into 15 harmonized disease classes
• Unified Optimization Framework: Integrating ensemble learning, knowledge distillation, and quantization
• Edge-Compatible Deployment: Achieving 671× compression (1.46 MB) with 97.46% accuracy
• Real-World Validation: Cross-domain evaluation on field condition datasets
• Explainable AI: Grad-CAM++ and LIME-based interpretability analysis

📊 Performance Highlights

Model	Accuracy	F1-Score	Parameters	Size	Inference Time
Ensemble (Teacher)	99.15%	97.07%	163M	652 MB	12.6ms
ShuffleNetV2 (Student)	98.53%	96.12%	1M	4.2 MB	0.29ms
Quantized INT8	97.46%	95.36%	1M	1.46 MB	0.29ms

🏗️ Repository Structure

LeafDisease-AI/
├── src/                          # Source code
│   ├── configurations/           # Configuration files
│   ├── data_augmentation/        # Data augmentation strategies
│   ├── data_balancing/          # ADASYN-based balancing
│   ├── datasets/                # Dataset loaders and utilities
│   ├── distillation/            # Knowledge distillation
│   ├── ensemble/                # Ensemble learning
│   ├── evaluation/              # Model evaluation scripts
│   ├── explainable_ai/          # Interpretability analysis
│   ├── hyperparameter_tuning/   # Hyperparameter optimization
│   ├── models/                  # Model architectures
│   ├── quantization/            # Model quantization
│   └── utils/                   # Utility functions
├── data/                        # Datasets (available under data/)
│   ├── combined/                # Unified dataset
│   ├── plantvillage/           # PlantVillage dataset (lab conditions)
│   └── tomatovillage/          # TomatoVillage dataset (field conditions)
├── docs/                        # Documentation
├── scripts/                     # Training and evaluation scripts
├── outputs/                     # Model outputs and results
└── checkpoints/                 # Trained model checkpoints

🚀 Quick Start

Prerequisites

# Python 3.8+ required
pip install -r requirements.txt
pip install -r requirements_balancing.txt

Dataset Preparation

1. Download Datasets:

   • PlantVillage: https://www.kaggle.com/datasets/abdallahalidev/plantvillage-dataset
   • TomatoVillage: https://github.com/mamta-joshi-gehlot/Tomato-Village

   Extract datasets to data/plantvillage/ and data/tomatovillage/ respectively.

2. Datasets are also available under data/ directory

Training Pipeline

1. Baseline Training

python scripts/train.py --model densenet121 --dataset combined --epochs 100

2. Ensemble Training

python src/ensemble/train_best_ensemble.py

3. Knowledge Distillation

python src/distillation/train_best_kd.py

4. Quantization

python src/quantization/mobile_quantization_pipeline.py

Evaluation

# Cross-domain evaluation
python src/evaluation/evaluate_kd_on_test_datasets.py

# Interpretability analysis
python src/explainable_ai/interpretability_analysis.py

🔬 Research Methodology

Cross-Domain Benchmark

Our framework addresses three fundamental challenges:

1. Cross-Domain Generalization: Unified PlantVillage (lab conditions) and TomatoVillage (field conditions) datasets
2. Class Imbalance: ADASYN-based balancing for 75:1 imbalance ratio
3. Computational Constraints: Knowledge distillation and quantization for edge deployment

Optimization Pipeline

1. Data Preprocessing: Strategic augmentation and ADASYN balancing
2. Hyperparameter Tuning: Systematic optimization across 24 architectures
3. Ensemble Learning: Four-model soft voting ensemble
4. Knowledge Distillation: Teacher-student framework with temperature scaling
5. Quantization: INT8 quantization for mobile deployment

Explainable AI

• Grad-CAM++: Attention visualization for model interpretability
• LIME: Local interpretable model-agnostic explanations
• Biological Validation: Alignment with plant pathology principles

📈 Experimental Results

Cross-Dataset Performance

Dataset	Accuracy	F1-Score	Precision	Recall
Combined Test	99.15%	97.07%	97.23%	96.91%
PlantVillage	98.87%	96.45%	96.78%	96.12%
TomatoVillage	95.70%	93.62%	94.01%	93.23%

Computational Analysis

Model	Parameters	FLOPs	Memory	Inference Time
DenseNet-121	7.98M	2.87G	32.4 MB	3.2ms
ResNet-101	44.55M	7.83G	179.8 MB	4.1ms
ShuffleNetV2	1.26M	146M	5.1 MB	0.29ms
Quantized INT8	1.26M	146M	1.46 MB	0.29ms

🏆 Disease Classes

The framework detects 15 tomato leaf disease classes:

1. Healthy
2. Bacterial Spot
3. Early Blight
4. Late Blight
5. Leaf Mold
6. Septoria Leaf Spot
7. Spider Mites
8. Target Spot
9. Yellow Leaf Curl Virus
10. Mosaic Virus
11. Powdery Mildew
12. Nutrient Deficiency
13. Pest Damage
14. Environmental Stress
15. Other Diseases

📚 Documentation

Detailed documentation is available in the docs/ directory:

• Installation Guide
• Project Overview
• Repository Structure

🤝 Contributing

We welcome contributions! Please see our Contributing Guidelines:

1. Fork the repository
2. Create a feature branch (git checkout -b feature/amazing-feature)
3. Commit your changes (git commit -m 'Add amazing feature')
4. Push to the branch (git push origin feature/amazing-feature)
5. Open a Pull Request

📄 Citation

If you use this work in your research, please cite: [citation coming soon]

📞 Contact

Mohammad Junayed Hasan

• Email: junayedhasan100@gmail.com
• GitHub: @junayed-hasan
• LinkedIn: LinkedIn Profile

📜 License

This project is licensed under the MIT License - see the LICENSE file for details.

MIT License

Copyright (c) 2024 Mohammad Junayed Hasan

Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.

🙏 Acknowledgments

• PlantVillage dataset contributors
• TomatoVillage dataset contributors
• PyTorch and timm library developers
• Agricultural research community

---

🌱 Bridging AI Research and Agricultural Reality 🌱