An Improved Yolov8 for Enhanced Disease Detection and Localization in Cotton Plants

Abstract

Enhancing disease detection and localization in cotton plants using an upgraded version of YOLOv8, a cutting-edge object detection model noted for its efficiency and accuracy. Cotton plants are subject to a variety of diseases, including bacterial blight, armyworm, and powdery mildew, which reduce agricultural production. The study focuses on using advanced deep learning architectures, notably ResNet50 combined with YOLOv8 and Faster R-CNN, to improve disease identification and localization in cotton plants. CVAT was used to annotate a rigorously curated dataset of 2100 photos with bounding boxes, allowing for model training. To guarantee accurate performance assessment, the dataset was divided into three sets: training, validation, and test. ResNet50 was incorporated into the YOLOv8 and Faster R-CNN architectures, which were specifically designed to identify and localize bacterial blight, armyworm, and powdery mildew, as well as healthy cotton plants. Fine-tuning hyperparameters and optimizing model configurations were used in the experiments to obtain high illness classification accuracy and exact localization. The results obtained were commendable, with both models achieving an object detection IoU of approximately 0.95. It is worth noting that ResNet50-YOLOv8 exhibited superior classification accuracy at 98%, while ResNet50-FRCNN achieved a respectable 90%, thereby illustrating nuanced performance variations. In terms of precision, recall, and F1-score, ResNet50-YOLOv8 presented impressive values of 0.989, 0.986, and 0.988, respectively. Contrastingly, ResNet50-FRCNN displayed values of 0.858, 0.905, and 0.88, indicating variations in precision and recall