Coal Foreign Object Detection Algorithm Based on Improved YOLOv8

As the primary equipment for coal transportation in mines, belt conveyors frequently encounter foreign objects such as anchor rods, plastic bottles, and large pieces of gangue mixed within the transported material. Failure to promptly remove these contaminants can severely compromise coal production safety. Addressing challenges in conveyor belt foreign object detection—including difficulties in extracting small target features, partial object occlusion, and low detection accuracy—this study proposes YOLOv8-CCR, an enhanced YOLOv8-based algorithm. By introducing the novel Receptive-Field Attention Convolution (RFAConv) mechanism, the method enables the model to effectively capture and utilise image information. It replaces the C2f module with the novel CSPPC architecture to fully exploit feature map redundancy, thereby reducing computational load and memory access. The lightweight CARAFE upsampling operator is introduced to enhance detection performance for small objects. Experimental results demonstrate that the YOLOv8-CCR model outperforms the baseline model by 5.7%, 5%, and 2.3% in precision, recall, and Map50 respectively, with a 24% reduction in parameters and a 21% optimisation in computational complexity (Giga floating-point operators per second). Through these model enhancements, this research significantly improves the accuracy and deployment efficiency of conveyor belt foreign object detection, providing technical support for the high-quality development of intelligent mines.