Abstract: Aiming at the serious threat to coal mine safety production caused by conveyor belt longitudinal tearing, this paper proposes an anti-longitudinal tearing detection system based on an improved YOLOv8 algorithm. Through lightweight modifications to the YOLOv8 network architecture, incorporating attention mechanisms and multiple optimization strategies, the system achieves high-precision, real-time detection of conveyor belt longitudinal tearing. Experimental results show that the improved YOLOv8 model reaches a mean average precision（mAP） of 88.3%, with the number of parameters reduced by 18.51%, computational load decreased by 20.73%, and model size shrunk by 15.87%. These improvements significantly lower the hardware requirements for edge devices while maintaining the accuracy required for coal mine safety monitoring. By comprehensively utilizing multi-dimensional information such as belt width detection, foreign object detection, and crack detection, along with logical judgment, the system effectively reduces the false alarm rate, providing reliable technical support for the safe operation of coal mine conveyor belts.