Vision-Based Deep Learning Models for Diagnosing Tropical Crop Diseases and AI Innovations

Tropical agriculture is vital for global food security, yet plant diseases remain a significant threat to crop yields. Traditional methods of disease detection, reliant on human observation, are inefficient for large-scale surveillance. This study investigates the application of vision-based deep learning models for the detection of tropical crop diseases. The research aims to evaluate various deep learning algorithms such as U-Net, YOLO, and Graph Neural Networks (GNN) for classifying, segmenting, and recognizing plant diseases. A systematic literature review (SLR) was conducted following the PRISMA framework, analyzing studies that employed vision-based models for tropical plant disease detection. The results reveal that segmentation models like U-Net are more effective for crops with large leaves, such as bananas and papayas, while object detection algorithms like YOLO are better suited for small-leaf crops like rice and wheat. GNNs prove efficient in diagnosing diseases affecting leaf veins, such as coffee leaf rust. Despite promising results, challenges such as variations in field conditions, lack of annotated datasets, and computational limitations persist. The study highlights the need for lightweight, scalable models that can operate on mobile devices and the importance of multi-modal imaging techniques. It concludes with recommendations for improving disease detection systems, including the development of more diverse datasets, enhanced model interpretability, and the adoption of Explainable AI (XAI) for broader application in tropical agriculture.