SawsanYousef/uganda-cassava-disease-classifier

# Cassava Leaf Disease Classification Using Transfer Learning (ResNet50) ## Project Overview Cassava is a critical food crop in many African countries, including Uganda. However, viral and bacterial diseases significantly reduce crop yield and threaten food security. This project develops a deep learning-based image classification system to detect cassava leaf diseases using **Transfer Learning with ResNet50**. The objective is to build a robust and generalizable model capable of accurately identifying plant health conditions from leaf images. The project follows a structured experimentation process: - Building a baseline CNN from scratch - Applying Transfer Learning with K-Fold Cross Validation - Refining the final model using callbacks and checkpointing ## Dataset Description The dataset consists of approximately **21,000 cassava leaf images** divided into five classes: - Cassava Bacterial Blight (CBB) - Cassava Brown Streak Disease (CBSD) - Cassava Mosaic Disease (CMD) - Cassava Green Mottle (CGM) - Healthy Leaves ### Class Imbalance Handling The dataset was originally imbalanced, with significantly more samples belonging to common diseases compared to rarer but critical target diseases. To ensure fair learning across all classes, **Random Undersampling** was applied. After balancing, the dataset contained: - **1,523 images per class** - **Total balanced dataset: 7,615 images** This approach ensured equal representation and prevented bias toward dominant classes. ## Data Preprocessing The following preprocessing steps were applied: - Resized all images to **224 × 224** - Converted all images to **RGB format** - Applied **Normalization** - Converted images into **PyTorch tensors** These preprocessing steps ensured compatibility with pretrained CNN architectures such as ResNet50 and improved convergence during training. ## Model Development & Evaluation ### Baseline Model – Custom CNN We first built a Convolutional Neural Network (CNN) from …