Arif-miad/Ghana-Crop-Disease-Detection-Challenge

# Ghana-Crop-Disease-Detection-Challenge --- # Ghana Crop Disease Detection Challenge ### Overview This project is focused on developing a model to detect crop diseases using annotated images. The objective is to identify various diseases present in crops and classify them effectively, contributing to agricultural sustainability and crop health management. ### Dataset - **License**: CC BY 4.0 Attribution 4.0 International - The dataset is divided into: - **Training Set**: Each image in the training set includes bounding box annotations for different types of diseases. Multiple diseases may be present within a single image, each annotated separately. - **Test Set**: Introduces images that may contain new, unannotated diseases, with no bounding boxes provided. This setup challenges the model's adaptability to novel disease types. ```python # 1. Distribution of Class plt.figure(figsize=(10, 6)) sns.countplot(data=df, x='class', palette='viridis') plt.title('Distribution of Crop Disease Classes') plt.xticks(rotation=45) plt.savefig("dist_dis.png") plt.show() ``` ### Approach 1. **Data Preprocessing**: - Processed images to normalize sizes, apply data augmentation techniques for diversity, and optimize bounding box formats for compatibility with object detection models. 2. **Model Selection**: - Selected and fine-tuned a deep learning model capable of multi-class object detection to identify and localize diseases using bounding box annotations. 3. **Training and Evaluation**: - Trained the model on the annotated training set and evaluated its performance using metrics like accuracy, recall, and precision. - Focused on generalization to handle unannotated test images with potential new disease types. 4. **Challenges**: - Adapting the model to identify unseen diseases in the test set, which required implementing techniques to improve generalization and anomaly detection. ### Results Achieved a robust model with high accuracy on annotated images and improved adapt …