Optimized CNN for Enhanced Cassava Leaf Disease Classification and Yield Improvement

Sub-Saharan Africa depends on cassava as its essential staple crop and carbohydrate foundation for its thousands of citizens. The cultivation of cassava faces severe risks because bacterial and viral infections cause yield decreases of as much as 70% within one season, which threatens the food security of smallholder farmers extensively. The research develops a modern deep learning-based diagnostic platform to demonstrate accurate disease classification of cassava leaf samples. ResNet50 and MobileNetV2 networks backed by boosting methods, including AdaBoost and Gradient Boosting, serve to improve classification outcomes from the proposed approach. A transfer learning approach and attention mechanisms enhance feature extraction within the system, which processes the 21,367 labeled images containing cassava leaf samples both in their healthy and diseased states using real-world image noise reduction methods. The proposed model produces classification outcomes with more than 99% precision for various disease categories, which surpasses conventional manual inspections and spreadsheet-based approaches. The conducted research established a mobile diagnosis tool that enables quick disease detection for cassava, thus empowering farmers to apply timely actions to promote sustainable agricultural growth and increase food security.