corn-gray-leaf-spot-xai

Explainable Deep Learning for Corn Gray Leaf Spot Detection

This repository provides four PyTorch models trained for binary classification of corn (maize) leaf images into the following categories:

• Healthy
• Gray Leaf Spot (Cercospora Leaf Spot)

The models were developed as part of a research study focused on improving the transparency and trustworthiness of deep learning–based plant disease detection systems using explainable artificial intelligence techniques.

---

Models Included

• ResNet50 (trained without random data augmentation)
• ResNet50 (trained with random data augmentation)
• EfficientNet-B0 (trained without random data augmentation)
• EfficientNet-B0 (trained with random data augmentation)

---

Dataset

• Training Dataset:
  PlantVillage dataset (Corn Healthy and Corn Gray Leaf Spot classes)

• External Evaluation Dataset:
  PlantDoc dataset (used to evaluate generalization performance)

---

Explainability

All trained models were analyzed using Gradient-weighted Class Activation Mapping (Grad-CAM) to visualize the spatial regions of the leaf images that most strongly influenced model predictions. This helps verify whether the models focus on biologically meaningful disease symptoms rather than irrelevant background features.

---

Inference Instructions

Users can perform inference on their own corn leaf images using the provided inference.py script.

Step 1: Clone the Repository

git clone https://huggingface.co/PulinduVR/corn-gray-leaf-spot-xai

Step 1: Create and Activate a Virtual Environment

python -m venv venv
source venv/bin/activate        # Linux / macOS
venv\Scripts\activate           # Windows

Step 2: Install the dependencies

pip install -r requirements.txt

Step 3: Run inference on a test image

python inference.py --image ./path/to/image --model_path ./saved_models/path/to/model/you/want --model_type [EfficientNet / ResNet]

Examples

python inference.py --image ./test_images/healthy.jpg --model_path ./saved_models/EfficientNet_basic/model.pth --model_type EfficientNet
python inference.py --image ./test_images/diseased.jpg --model_path ./saved_models/ResNet50_basic/model.pth --model_type ResNet

Sample Test Images

Below are examples of the corn leaf.

Usage

Refer to inference.py for implementation details related to:

• Model loading
• Image preprocessing
• Prediction logic

---

Citation

If you use these models or this repository in your research, please cite our paper:

Explainable Deep Learning for Early Detection of Corn Gray Leaf Spot Using Grad-CAM

---

Notes

• These models are intended for research and educational purposes.
• Performance may vary under real-world field conditions with complex backgrounds.
• Grad-CAM visualizations are provided to support interpretability and trust.

---

License

This project is intended for academic and research use.
Please review the licenses of the PlantVillage and PlantDoc datasets before any commercial deployment.