Steel surface defect detection based on SSB-YOLO11

Abstract

Due to the low detection accuracy of current steel surface defect detection methods, this paper proposes an SSB-YOLO11-based detection method. Three attention mechanisms, namely SimAM, CBAM, and ECA, are incorporated into the YOLO11 model. Experimental results demonstrate that the SimAM module achieves superior performance in extracting defect region features. Furthermore, in the feature fusion stage, the original Concat operation is replaced with BiFPN to further improve detection accuracy. In addition, the standard convolution (Conv) in the backbone network is replaced with SAConv to enrich the receptive field and enhance multi-scale feature extraction capability. The improved model achieves a 0.5 percentage point increase in mean Average Precision (mAP), with an average detection accuracy reaching 79.7%, demonstrating strong potential for practical applications.