Abstract: Based on the engineering background of belt foreign object recognition of 1305 intelligent working face in a coal mine, in order to solve the problems of belt tearing and even damage and shutdown caused by foreign objects such as waste bolts and large pieces of coal in underground belt conveyors, it is urgent to carry out research on belt foreign object detection. The SE-YOLOv5 belt foreign object detection method is proposed, which is based on the YOLOv5 object detection technology, and the SE channel attention mechanism is added to optimize the model parameters, and the model parameters such as learning rate, image input batch size, and weight attenuation are adjusted, and the constructed dataset is trained and detected, and the detection results are compared with Faster-RCNN, YOLOv3, CenterNet, YOLOv5 and other models. The research results show that the prediction results of the SE-YOLOv5 model are greatly improved, with the prediction accuracy of the anchor rod reaching 0.98, the prediction accuracy of the large coal being 0.88, the recall rate （Recall） being 0.91, the average accuracy of each detection target （mAP） being 0.912, and the single recognition speed being 0.037s. It shows that the optimized SE-YOLOv5 model has high detection accuracy, fast speed and strong robustness, which can meet the requirements of belt foreign object recognition in complex environments. In general, the belts foreign object detection in coal mine is a key component of coal mine safety and production efficiency, and the use of modern detection technology can help improve system production efficiency, ensure equipment safety, and reduce maintenance costs.