Based on YOLOv11 Steel Surface Defect Detection

Iron and steel is widely used in the field of construction and industry. The traditional steel defect detection technology has problems such as high labor cost, low detection efficiency and limited scope of application, etc. Deep learning defect detection technology has gradually become the mainstream. This paper proposes YOLOv11 steel defect detection algorithm for the case of insufficient detection rate of existing steel defect detection methods based on deep learning. This algorithm uses ODConv and GhostConv to replace the traditional convolution in backbone to reduce the model parameters; uses ASF neck structure to perform multi-dimensional feature fusion to improve the model accuracy and achieve lightweight; replaces the convolution in ASF with the intra-class residual convolution to improve the anti-noise and stability of the model and accelerate the convergence speed. The experiment used Northeastern University public dataset NEU-DET in Windows 11 and Pytorch2.0.1. The results show that the improved YOLOv11 improves the rate with the same defect detection accuracy, and the average accuracy increases by 2.9 %, which effectively improves the performance and efficiency of steel surface defect detection.