The Prior Model-Guided Network for Bearing Surface Defect Detection

Surface defect detection is a key task in industrial production processes. However, the existing methods still suffer from low detective accuracy to pit and small defects. To solve those problems, we establish dataset of defects propose prior model-guided network for detection. This composed segmentation with weight label, classification network, pyramid feature fusion module. The as model can improve network. label center distance transformation reduce cost module adapt different scales avoid information loss comparison experiments are implemented identify performance proposed Ablation designed specify effectiveness every Finally, performed on public verify its robustness. Experimental results reveal that method effectively reach 99.3%, which increased by 2~5% compared other methods, revealing excellent applicability automatic quality inspection production.