Improved the Detection Algorithm of Steel Surface Defects Based on YOLOv7

To solve the problem of low accuracy and slow speed of steel surface defect detection, an improved steel surface defect detection algorithm based on YOLOv7 was proposed. To strengthen the fitting ability of the bounding box loss, the bounding box loss function is replaced by Wise-IoU (WIoU). To solve the mismatching problem of each unit during training, improve the sensitivity of each branch to the traget scale, and avoid training trangets on inappropriate units, the scale-aware training strategy is introduced. Omni-Dimensional Dynamic Convolution (ODConv) is added to weight the convolution and improve the detection performance of the output terminal for small targets with defects on the steel surface. Ablation and comparison experiments were carried out on NEU-DET dataset. Compared with the original YOLOv7 algorithm, the Precision of the improved algorithm was increased by 9.8 percentage points, and the mAP was increased by 8.2 percentage points. The results show that the improved algorithm can effectively improve the efficiency of steel surface defect small target detection.