Vision-based surface damage recognition method for elevator hoisting steel belts considering motion blur

Abstract The elevator hoisting steel belt is a critical load-bearing component, and undetected surface damage can pose a serious threat to operational safety. However, due to the small size and complex morphology of hoisting steel belt defects, as well as motion blur caused by high-speed operation, current detection methods remain limited in accuracy, real-time capability, and robustness. To address these challenges, this paper proposes an improved lightweight YOLO11-DED detection framework for vision-based recognition of surface damages on elevator hoisting steel belts. The proposed method achieves lightweight optimization through structural redesign to ensure real-time operation on edge devices, incorporates a multi-scale attention mechanism to enhance the recognition of tiny defects, and integrates a dynamic upsampling strategy to improve robustness under motion blur. To validate the proposed approach, a dedicated elevator belt defect dataset covering multiple damage types was constructed, and systematic experiments were conducted under both static and dynamic operating conditions. Experimental results showed that, compared with the original YOLO11, YOLO11-DED reduces the number of parameters by 24.29%, decreases GFLOPs by 31.75%, increases inference speed by 10.26%, and improves mAP by up to 7.90%. The proposed method effectively addresses the challenges of insufficient accuracy for tiny and complex defects and susceptibility to motion blur in real time, achieving a balance between lightweight design and detection efficiency. It shows considerable promise for elevator safety monitoring and other industrial defect-detection tasks.