Bright-CG/AgriVision-Nigeria-ML

# AI-Driven Crop Disease Detection for Nigerian Agriculture ## Project Overview This project implements a comparative analysis of Artificial Intelligence models to detect diseases in staple Nigerian crops, specifically **Cassava** and **Maize**. By automating disease identification, we aim to bridge the gap between smallholder farmers and expert diagnostic services. ## Key Features - **Dataset**: 12,866 images sourced from the Mendeley Data Repository. - **Comparison**: Evaluated a Deep Learning CNN (MobileNetV2) against a classical Random Forest model. - **Dynamic Preprocessing**: Implemented robust image loaders that handle truncated data streams and complex directory structures. ## Results The Convolutional Neural Network (CNN) significantly outperformed the classical model: * **CNN (MobileNetV2) Accuracy**: 84.53% * **Random Forest Accuracy**: 65.58% The CNN showed exceptional performance in identifying **Cassava Mosaic** and **Maize Grasshopper** damage (F1-Scores > 0.90). ## Technical Stack - **Languages**: Python - **Frameworks**: TensorFlow, Keras, Scikit-learn - **Tools**: Google Colab (Tesla T4 GPU), OpenCV, Pandas, Matplotlib ## File Structure - `src/data_loader.py`: Handles data augmentation and directory mapping. - `src/model_cnn.py`: Implementation of the Transfer Learning architecture. - `src/model_rf.py`: Classical machine learning implementation using deep feature extraction. ## How to Use / Quick Start 1. Clone the repository. 2. Install dependencies: `pip install -r requirements.txt`. 3. Ensure your dataset is structured into `train`, `validation`, and `test` folders. *(Note: The full dataset must be downloaded from the Mendeley link [Insert Mendeley Link Here].)* 4. Run `python main.py` to train the model. 5. **Quick Start**: To test the model, place an image in the project root and run: `python predict.py --image your_image.jpg` *(Note: The pre-trained model `cnn_mobilenetv2_trained_model.h5` is included, so you can run inference immediat …