Accurate real-time detection of foreign object during industrial coal mining and processing is crucial to ensuring equipment and personnel safety, as well as maintaining the quality of coal product. The complexity of real industrial coal mining and processing environment poses challenges to vison-based foreign object detection. This work proposes a SHA-DH-YOLOX algorithm to enhance detection accuracy. The proposed algorithm stands out with three improvements. First, Shuffle Attention mechanism (SHA) is integrated into the YOLOX backbone to strengthen the feature extraction of essential information from input images. Second, Dynamic head (Dyhead) is introduced after feature fusion to enhance the detector's representation capability for scale-, spatial-, and task-awareness. Third, the original Intersection over Union (IoU) loss function is replaced by SCYLLA-IoU (SIoU) for more accurate bounding boxes and enhanced training stability. These improvements work collaboratively with YOLOX-M, making the proposed algorithm outperform state-of-the-art baseline models including YOLOv5, YOLOv6, YOLOv7, and original YOLOX series. The developed SHA-DH-YOLOX algorithm improves AP₅₀ by 1.87 - 7.97% compared to baseline models of equivalent size. Robustness tests further demonstrate the stability of SHA-DH-YOLOX model in facing diverse harsh scenarios. This pioneering work provides valuable tools to achieve safe and reliable coal mining and processing.
 

foreign object detection,YOLOX,shuffle attention mechanism,dynamic head,SIoU