Abstract: In the coal mine transportation system, the conveyor belt is often impacted by schist, gangue, bolt and other sharp hard impurities in the coal flow, and it is easy to tear lengthwise. Once the conveyor belt longitudinal tear, if not found in time and shut down will cause great economic losses, and even casualties. In view of the problems existing in the visual detection of longitudinal tearing of conveyor belt, such as low detection accuracy and low automation intelligence of industrial camera or line laser transmitter, a set of vision-based longitudinal tearing inspection system for mining conveyor belt is developed. Firstly, based on the harsh environment of coal mine, the layout scheme of multi-line laser emitter and industrial camera is designed, which highlights the characteristics of longitudinal tear and improves the accuracy of longitudinal tear detection. The optimal layout scheme is determined through experiments. Secondly, an image enhancement algorithm combining piecewise linear transformation and CLAHE is proposed to improve the quality of the collected image. Thirdly, the image stitching algorithm based on SIFT feature extraction and the hat function weighted average fusion algorithm are used to obtain high-quality and complete laser images of the lower surface line of the conveyor belt. Fourthly, this paper proposes a multi-line laser centerline extraction algorithm based on improved Otsu threshold segmentation algorithm, and through experimental comparison, it is proved that the centerline extracted by the algorithm can accurately reflect the linear characteristics of laser lines. Finally, a morphology-based longitudinal tear feature extraction and detection algorithm for mining conveyor belts is proposed, and the number of connected domains is used to determine whether longitudinal tearing occurs in the conveyor belt. In order to verify the superiority of the algorithm, a longitudinal tear detection test bench of mining conveyor belt is built, and experiments are compared and verified in a laboratory environment without dust fog and simulated dust fog. The experimental results show that the accuracy of the proposed method reaches 98.6% under dust-fog-free environment, and the accuracy of the method reaches 97.9% in the dust-fog environment, which proves the advanced effectiveness and practicality of the method.