Cropper

Cropper: Attention-Guided Ensemble Model for Crop Disease Detection 🌿

Cropper is a lightweight and interpretable deep ensemble framework for detecting crop diseases using leaf images. It integrates multiple convolutional backbones with attention mechanisms to deliver high accuracy, visual interpretability, and edge-device deployability, particularly in data-scarce agricultural environments.

🌟 Key Features

• 🔁 Multi-backbone ensemble (DenseNet121, ResNet18, EfficientNetB0)
• 🧠 Attention-guided feature refinement
• 🖼️ Visualization with class-discriminative attention maps
• 🌾 Supports multiple crops: rice, betel, cabbage, Chinese cabbage, Apple, Pumpkin and Tomato
• ⚙️ Includes benchmark models: PlantDet, Guava(DenseNet169), LeafconvNext, ViT for comparison
• 📉 Ablation-friendly architecture and reproducible evaluation protocol

📁 Repository Structure

File / Folder	Description
Cropper_Rice.py	Main Cropper training + evaluation script for rice dataset
Cropper_Betel.py	Main Cropper training + evaluation script for betel dataset
Cropper_Cabbage.py	Main Cropper training + evaluation script for cabbage dataset
Cropper_ChineseCabbage.py	Main Cropperscript for Chinese cabbage dataset
ViT_Tiny.py	ViT_Tiny baseline implementation
ViT_Base.py	ViT_Base baseline implementation
Guava.py	Guava baseline implementation
LeafNext_Betel.py	LeafNext baseline implementation for betel
LeafNext_Rice.py	LeafNext baseline implementation for rice
PlantDet_Betel.py	PlantDet benchmark model for betel
PlantDet_Rice.py	PlantDet benchmark model for rice
Visualization.py	Script for generating and saving attention visualizations (heatmaps)
dataset_Info.xlsx	Summary of all datasets used (image count, labels, splits, etc.)
Results.xlsx	Evaluation metrics and performance comparison tables
README.md	You are here. Project overview and usage instructions.

---

🧪 Datasets

The model supports 7 plant leaf datasets (publicly available):

• Rice: Multi-class (6 disease types)
• Apple: Multi-class (5 disease types)
• Pumpkin: Multi-class (5 disease types)
• Tomato: Multi-class (10 disease types)
• Betel: Binary classification (healthy vs. unhealthy)
• Cabbage: Binary classification (nutrient deficiency vs. healthy)
• Chinese Cabbage: Binary classification (Botanical Leaf Spot)

The linkings of these datasets can be found in the paper.

Dataset info and splits stats are documented in dataset_Info.xlsx.

📦 Note: Please organize your image folders following the structure:

dataset/
├── rice/
│   ├── Bacterial_leaf_blight/
│   ├── Brown_spot/
│   └── ...
├── betel/
│   ├── Healthy/
│   └── Unhealthy/
...

Environment

Shown in the .yaml file