Houssam-123-ship-it/Crop-Disease-in-Uganda

--- # 🌿 Crop Disease Classification in Uganda — WQU Work Simulation ## 📘 Project Overview This project is part of the **WorldQuant University (WQU) Data Science Work Simulation** and is inspired by real-world challenges in **agriculture and computer vision**. The goal is to build a **deep learning model** that classifies images of **cassava plants in Uganda** into **five categories** — identifying whether a crop is healthy or affected by a specific disease. Throughout the project, I worked with **Convolutional Neural Networks (CNNs)** and **Transfer Learning** techniques to improve accuracy and efficiency while preventing model overfitting. --- ## 🎯 Learning Objectives ### 🧩 Part 1 — Data Exploration & Preparation * Explore the dataset of crop disease images. * Check image properties, data balance, and class distribution. * Normalize pixel values and **balance unbalanced classes** through undersampling. * Understand the impact of **data imbalance** on model performance. **New Terms:** `Unbalanced classes`, `Undersampling` --- ### 🧠 Part 2 — Building a CNN from Scratch * Convert images from grayscale to RGB and **resize them for uniformity**. * Create a transformation pipeline to prepare data for training. * Build a **Convolutional Neural Network** to classify crop diseases into five classes. * Train and evaluate the model using **learning curves** to detect overfitting. **New Terms:** `Overfitting`, `Learning Curve` --- ### ⚙️ Part 3 — Transfer Learning & Callbacks * Load and preprocess cassava plant images for training. * Use **Transfer Learning** by adapting a **pre-trained image classification model** (e.g., ResNet or MobileNet). * Apply **Callbacks** to enhance model optimization: * **Learning Rate Scheduling** — dynamically adjust learning rate during training. * **Model Checkpointing** — save the best-performing model automatically. * **Early Stopping** — stop training when validation accuracy stops improving. * Evaluate model performance usi …