A Surface Damage Detection Method for Hoisting Steel Wire Ropes Based on FFD-IDeblurGANv2 and GSEW-YOLO

Visual inspection is critical for detecting surface damage in mine hoisting wire ropes. However, factors, such as wire rope motion, coupled equipment vibrations, and camera focal distance dispersion, can result in blurred inspection images, thereby compromising the effectiveness of surface damage detection. To address these challenges, this study proposes a method for enhancing surface damage detection of steel wire ropes. First, based on MobileNetv3_Small, the DeblurGANv2 network backbone is reshaped, and the patch loss calculation strategy is introduced. In the inference stage, FFD is integrated to propose an efficient deblurring method based on FFD-IDblurGANv2. Second, a slender neck structure based on GSConv was designed, integrating an efficient hybrid attention mechanism (EHAM) and introducing a loss calculation strategy based on Wasserstein distance. The GSEW-You Only Look Once (YOLO) (YOLOv7 enhanced by GSConv-based slender neck, EHAM attention, and Wasserstein Loss) model was proposed. Finally, under strictly consistent experimental conditions, the superiority of the proposed method over YOLOv5s, YOLOX-s, YOLOv7, YOLOv8s, and YOLOv9s models was verified through comparative experiments.