Abstract: To address the limitations of traditional longitudinal tear detection systems—including poor illumination resistance, low computational efficiency, and weak generalization capability—this study proposes a novel detection technology based on line structured light.The line structured light is used as the image acquisition system, and then Topk is used to pre-process the image for the purpose of reducing data redundancy and memory usage; dimensionality reduction was applied to the base operators of the YOLOv5 network, effectively reducing both the model parameters and floating-point operations (FLOPs); the detection technology has been successfully ported to an embedded system, resulting in the development of a high-speed and highly accurate intrinsically safe longitudinal tear detection system designed for mining applications.The experimental results show that the operand and parameter quantity of the dimensionally reduction YOLOv5 are lower than the traditional methods, and the F_1score is 0.951 1, which is better than other methods when the resolution of the input map is 640 px×2 592 px; The accuracy of detection P in the simulation experiment is 95.14% and the recall rate R is 92.63%;in industrial trials, the technology successfully detected longitudinal tearing in conveyor belts.