Abstract: Coal mine conveyor belt is one of the important equipment for underground coal transportation. During the operation，due to the mixing of large coal，gangue，anchor rod，channel steel and other foreign matters，it is easy to lead to a belt tearing failure，which seriously affects the safe production of coal mine and even threatens the miners’ life. In order to realize the automatic， rapid and accurate detection of large foreign bodies on conveyor belt in coal mine，a detection method of large foreign bodies based on computer vision technology was designed. Aiming at the problem that the foreign object image in the conveyor belt is difficult to be accurately detected by the traditional im⁃ age detection algorithm due to the interference of conveyor belt coal dust，high⁃speed movement of conveyor belt and uneven illumination，a YOLOv5 target detection algorithm based on the convolutional block attention model was pro⁃ posed，denoted as CBAM-YOLOv5. First，the adaptive histogram equalization algorithm is used to enhance the contrast of the coal mine underground conveyor belt image and reduce coal dust interference. Then，for the high⁃speed move⁃ ment of the conveyor belt，the image of the target to be detected is likely to be blurred and the target is difficult to be accurately detected. Under the framework of the YOLOv5 algorithm，the detection speed of the network is improved by introducing a deep separable convolution，and the detection accuracy of the entire network is improved by optimi⁃ zing the loss function of the detection network. Secondly，in view of the problem that the foreign object is difficult to be accurately detected due to the uneven illumination， the convolutional block attention model is introduced in the YOLOv5 detection network to increase the saliency of foreign objects in the image，enhance the feature expression abil⁃ ity of foreign objects in the detection network，and thereby improve the detection accuracy of foreign objects. Finally，the monitoring video data of a coal mine underground conveyor belt is used to prepare training samples and test samples，and the proposed algorithm is compared with four classic target detection algorithms. The experimen⁃ tal results show that the proposed detection algorithm can better solve the influence of coal dust interference，high⁃ speed movement of the conveyor belt and uneven illumination on the target detection accuracy. The average detection accuracy can reach 94.7% for the image with resolution of 1 280×720，and the detection speed is 31 fps.