Deploying lightweight YOLO for Real-Time Sunflower Leaf Disease Detection and Severity Classification on Edge Devices

Sunflower, an important oilseed crop, is prone to numerous bacterial, viral, and fungal infections that can severely impact both yield and economic returns. Ensuring early and accurate identification of these diseases is essential to avoid large-scale crop damage and maintain productivity. To facilitate practical, on-field application, this study integrates a trained deep learning model onto the NVIDIA Jetson Nano, enabling real-time disease detection directly on a portable, low-power device. Conventional methods such as manual observation often lack reliability and efficiency, making automated detection a more effective alternative. This work focuses on utilizing object detection algorithms, specifically YOLOv8 and YOLOv11, for recognizing disease symptoms in sunflower leaves. The models were trained using a carefully compiled dataset that includes both healthy and diseased samples, enhanced with data augmentation and segmentation techniques to improve detection accuracy. Among the models, YOLOv11 showed stronger performance across key metrics such as precision, recall, and mean average precision. Deployed on Jetson Nano, the system classifies affected leaf regions into mild, moderate, and severe stages, allowing farmers to take timely preventive measures and reduce crop losses.