Cotton care: YOLO-powered cotton leaf disease detection

Abstract

Early detection of plant diseases is an important factor for maintaining crop health and enhancing agricultural productivity. Early observation of symptoms of diseases, followed by accurate treatment in cotton growing, can greatly reduce loss of yield and save unnecessary pesticides. Recent developments in deep learning and computer vision have pushed the automated disease detection based on leaf images to be scalable and applicable in real-world agricultural practices. In this paper, we introduce a cotton disease detection technique with four YOLO (You Only Look Once) based object detection models: YOLOv5, YOLOv6, YOLOv8, and the latest YOLOv11. These models are trained on a generated dataset of annotated cotton leaf images capturing multiple symptoms like leaf enation, sooty mold, various curly forms, as well as healthy leaves. The purpose is to accurately classify and localize the disease-affected patches so as to enable real-time decisions in precision farming. The result showed that among the tested models, YOLOv11 presented the best performance with 98.2% precision, 99.3% recall, 99.6% mAP@0.5, and 0.78 mAP@0.5:0.95. YOLOv8, YOLOv6, and YOLOv5 performed well too. The input devices emphasize the significance of preprocessing for real-world applications to achieve robustness of the model to different lighting conditions. The results indicate that the proposed method can be used as a good and effective tool for automated cotton disease prediction and integrated pest management.