ABSTRACT During the coal mining, large foreign objects may block coal conveyors, leading to a series of safety accidents. The existing models for detecting foreign objects in coal conveyors perform poorly in low‐light environments, resulting in false or missed detections of foreign objects. Additionally, the large size of these models complicates deployment on edge devices. Addressing these issues, this study proposes a lightweight algorithm based on an improved YOLOv7 for detecting foreign objects on coal conveyors. Initially, image enhancement techniques are applied to preprocess low‐light images to enhance the effective feature information of foreign objects on the conveyor belts. Subsequently, the YOLOv7 model is modified using the lightweight ShuffleNetv2 model and the Asymptotic feature pyramid network (AFPN) to reduce its complexity. Furthermore, two squeeze and excitation (SE) attention modules are added following the ShuffleNetv2 module to compensate for the precision loss caused by model lightweighting. Experimental results demonstrate that the enhanced YOLOv7 algorithm significantly reduces Parameters and GFLOPs, increases detection speed by 3.9 ms, and improves mean detection accuracy by 1.1% compared to the original algorithm. When compared to common two‐stage and one‐stage detection models, the improved YOLOv7 model is more lightweight and achieves higher mean detection accuracy.
Zhang et al. (Thu,) studied this question.