Surface Defect Detection of Steel Strips based on an Improved YOLOv8

To tackle the challenges of false negatives and false alarms in the detection of surface defects on steel strips, this paper introduces a modified YOLOv8 model to improve the performance in surface defect detection tasks. The Wise- IoU (WIoU) bounding box loss function is utilized to refine the localization accuracy of surface defects and enhance the training efficiency against low-quality examples. Furthermore, a dilated convolution residual (DCR) module is proposed to improve the ability of feature extraction by replacing the bottleneck structure in the C2f module. The proposed method was tested and verified on the NEU-DET datasets. And the experimental results showed that our method reached the mean average precision (mAP) of 74.9% which was surpassing the original YOLOv8 by 3.4%. The effectiveness of the improved YOLOv8 was proved on the defect detection tasks on steel strip surface.