Lightweight YOLOv8 Model for Weed Detection

In recent years, smart agriculture has emerged as an effective means to enhance crop yields, with weed removal in fields being a critical factor directly impacting production. Traditional methods for weed control often involve the extensive use of herbicides, which can lead to overuse and negatively affect crop health. This paper proposes a lightweight YOLO model tailored for weed detection. Compared to the original YOLO model, this lightweight version maintains comparable recognition accuracy while significantly reducing hardware computational requirements. Specifically, in terms of billion floating-point operations (GFLOPs), the lightweight YOLO model reduces hardware demands by 30.6% compared to the original model. Experimental results demonstrate that the lightweight model exhibits notable advantages in weed detection, effectively broadening its application scope.