Green gram (Mongbean) crop health monitoring, leaf segmentation and disease classification using Deep Learning

Abstract

Green gram (Vigna radiate), a widely cultivated annual herbaceous legume from the family Fabaceae, is highly sensitive to various biotic and abiotic stresses, re sulting in significant yield losses. Notably, two key diseases, the Yellow Mosaic Virus and Leaf Tan Spot, pose considerable challenges to the successful cultiva tion of this crop. Timely detection and accurate classification of these diseases are crucial for effective disease management, thus forming the problem statement for this study. This novel study presents an integrated approach to precision agricul ture focusing on Mungbean crop health monitoring through the fusion of remote sensing and state-of-the-art deep-learning techniques. Two unique datasets were assembled: a 15-day temporal multispectral dataset captured via drones and a disease-specific dataset for the segmentation and classification of Yellow Mosaic Virus and Tan Leaf Spot diseases. The research involved the application of ad vanced image segmentation models, YOLOv8 and Mask R-CNN, and classification models, DEIT and MobileNetV2, demonstrating notable success. YOLOv8, par ticularly its Nano variant, achieved an mAP of over 77%, with an accuracy of 94.9% & mIOU of 82.13% on the test set, showcasing its potential for real-time application on edge devices. DEIT outperformed in classification tasks with a 99% accuracy rate. The study further leveraged vegetative indices—NDVI, VARI, and TGI—to provide a comprehensive assessment of crop health, establishing NDVI as the most reliable index for this purpose. The implications of this work are significant, offering scalable solutions for disease mapping and smart pesticide de ployment, contributing to sustainable agricultural practices and aligning with the United Nations Sustainable Development Goals of Zero Hunger and Responsible Production and Consumption. The research paves the way for impactful and scal able future innovation, highlighting the transformative potential of AI and remote sensing in advancing global agricultural practices and ensuring food security.