Real-time detection and evaluation of longitudinal tearing of mining conveyor belts based on line laser and YOLO-DM

Abstract

Belt conveyors are important components of mining transportation systems, and ensuring their safe and stable operation is crucial. Traditional detection methods often struggle to provide accurate monitoring in complex environments. This study addresses the challenges of detecting longitudinal tears in mining conveyor belts, including identifying minor damage, inaccurately estimating tear lengths, and a lack of real-time responsiveness. This work proposes a real-time detection approach that combines line laser projection with an enhanced YOLOv11 model. Line lasers project continuous light strips onto the conveyor belt, forming a detection framework based on the geometric deformation of the laser lines at damaged areas. To improve detection accuracy, we introduce a YOLO-DM model, which integrates a Dual Enhanced Squeeze-and-Excitation (DESE) attention mechanism and a Multi-Scale Laser Line Feature Fusion (MSLF) strategy. Additionally, we present a tear length calculation algorithm that uses the duration of tear label positions and belt velocity to achieve high-precision real-time measurements through temporal localization, trajectory tracking, and speed integration across sequential frames. On 3,000 annotated images, YOLO-DM achieved 98.7% mAP at 58.1 FPS, and the length estimator reached 95.8% accuracy. meeting the stringent real-time detection requirements for industrial mining conveyor belt monitoring.