LAUSR

Strip steel surface defect detection is a critical step in the production field of the steel industry and a vital guarantee to improve the quality of strip steel production. However, due to the poor contrast of the strip steel surface defect images, the diversity of defect types, scales, texture structures, and the irregular distribution of defects, it is difficult to achieve rapid and accurate detection of strip steel surface defects with the existing methods. In this article, a rapid detection network for strip steel surface defects based on deformable convolution and attention mechanism (DCAM-Net) is proposed. First, we introduce contrast limited adaptive histogram equalization (CLAHE) as a data augmentation method to improve the contrast of the defect image and highlight the defect feature on the strip steel surface images. Second, we propose a novel enhanced deformation-feature extraction block (EDE-block) for various complex and irregularly distributed strip steel defects. By fusing deformable convolution, the receptive field of the defect feature extraction network is expanded to capture complete and comprehensive defect texture features. Finally, we introduce the coordination attention (CA) module to replace the backbone network’s spatial pyramid pooling (SPP) structure, which further factorizes the pooling operation and effectively improves the network’s ability to locate defects. The experimental results on the NEU-DET dataset showed that the mean Average Precision (mAP@IoU $=0.5$ ) of the proposed algorithm is 82.6%, which is 7.3% higher than the baseline network, and the detection speed is up to 100.2 fps, which effectively improves the detection efficiency of surface defects of strip steel.