An Improved Surface Defect Detection Algorithm for Strip Steel with YOLOv11

Aiming at the existing deep convolutional network in detecting the complex and diverse defects on the surface of strip steel with low accuracy, large number of model parameters and computational volume, an improved strip steel surface defect detection algorithm YOLOv11-WDS is proposed. fusion of the DRFD module and wavelet pooling to design the WaveletRFD, replacing the original network part of the downsampling Conv module; design of lightweight multi-branching module DBBGELAN, replacing the C3k2 module in Neck; designing SEAMHead by using SEAM Attention module to improve Head. Experiments show that compared with the original yoov11, yoov11 WDS’ map is increased by $4.0 \%$, and the amount of calculation, parameters and weight file volume are reduced by $17.2 \%, 13.1 \%$ and $6.0 \%$, respectively. It also maintains a high detection speed and achieves lightweight while maintaining high accuracy, which is suitable for strip surface defect detection in industrial field.