Lightweight foreign object detection for mine conveyor belt based on improved YOLOv11n

Addressing safety concerns such as conveyor belt damage and breakage caused by foreign objects during the operation of underground belt conveyors in coal mines, traditional detection methods suffer from high false-negative rates, poor real-time performance, or high costs. This paper proposes an improved foreign object detection algorithm for conveyor belts based on the lightweight object detection model YOLOv11n. It addresses the complex environmental challenges of uneven illumination, dust interference, and diverse foreign object shapes in mining shafts. In the backbone section, the C3k2_Faster_EMA module is introduced, enhancing the extraction of fine-grained foreign object features through heterogeneous convolutions and exponential moving average mechanisms. The Neck section embeds an SE attention module to dynamically amplify effective feature channels while suppressing background interference. A lightweight Detect_Efficient detection head employs separable convolutions to reduce computational overhead. Furthermore, the Inner-CIoU loss function optimizes bounding box regression accuracy for irregularly shaped foreign objects. The experimental results show that the improved model achieved P , R , mAP@0.5, and mAP@0.5–0.95 of 84.1%, 82.5%, 82.8%, and 80.2%, respectively, representing an improvement over the original model. Meanwhile, the number of parameters was reduced to 2.54 M, the FPS dropped to 102 frames, and the computational load was only 4.6 GFLOPs. Compared to mainstream models like YOLOv8n, the proposed model demonstrates superior performance in detection accuracy and lightweight efficiency. It achieves a balanced trade-off between detection precision and computational lightness, providing an effective technical solution for real-time intelligent safety monitoring of mine conveyor belts.