Mo7amedMok5tar/Crop-Disease-in-Uganda-

# Crop Disease Detection – Kaggle Competition (Uganda Dataset) This project develops an image classification model using PyTorch to identify crop diseases from leaf images. It was built for the Kaggle competition: Cassava Leaf Disease Classification, and incorporates ensembling, cross-validation, and deployment via a Streamlit web app. ## Table of Contents 1. Project Overview 2. Project Structure 3. Data Preparation - Dataset Overview - Handling Imbalanced Data 4. Model Training - Cross-Validation Strategy (K-Fold) - Model Architecture - Callbacks and Training Enhancements 5. Model Evaluation 6. Model Ensembling 7. Deployment - Streamlit Web App 8. How to Run 9. Results Summary --- ## Project Overview This project aims to build a robust crop disease classifier using PyTorch. It tackles class imbalance, leverages K-Fold cross-validation, and ensembles the final predictions for enhanced accuracy. A Streamlit app is also built for easy interaction and real-time prediction. --- ## Project Structure ``` ├── notebooks/ # EDA and model training notebooks ├── src/ │ ├── utils/ # Helper functions (data loading, training, etc.) ├── model/ # Trained model checkpoints │ ├── fold_0/ # Model for Fold 0 │ ├── fold_1/ # Model for Fold 1 │ └── ... # etc. ├── app/ # Streamlit web app code ``` --- ## Data Preparation ### Dataset Overview The dataset includes images of diseased and healthy leaves of cassava crops. The labels represent different disease types. - 📂 **Dataset link**: Cassava Plant Disease – Merged 2019/2020 - 🏆 **Competition link**: Cassava Leaf Disease Classification ### Handling Imbalanced Data The dataset was significantly imbalanced. To address this: - **Under-sampling** of majority classes was applied to ensure fair training across folds. --- ## Model Training ### Cross-Validation Strategy (K-Fold) To ensure model generalization and robustness, a 5-fold **Str …