Surface Defect Detection Method of Strip Steel Based on Improved YOLOv5

A strip steel surface defect detection method called YOLOv5-ABS is developed in order to address the issues of low detection accuracy, insufficient feature extraction ability, and insufficient feature fusion of YOLOv5. Firstly, in order to enhance the backbone network's capacity for feature extraction, the C3 module is swapped out for the SeC3 module with an attention mechanism. Secondly, in order to improve the network feature fusion capability, the bidirectional weighted feature pyramid network BiFPN is added in the Neck section. Finally, by introducing the SPPFCSPC spatial pyramid pooling structure, speed and accuracy are improved while keeping the receptive field unchanged. According to the experimental findings, the revised YOLOv5-ABS algorithm’s mAP on the NEU-DET dataset is 78.6%, 3.8% larger compared to the initial YOLOv5s algorithm, and the detection speed is 142.8 FPS, enabling the quick and precise identification of strip steel surface defects.