Strip Pickling Defect Detection Based on Improved YOLOv5s Algorithm

In the strip steel production industry, strip pickling is the main method to remove surface oxides. However, pickling defects will occur during strip pickling. Various steel manufacturers are trying to find effective ways to solve this problem. A strip steel pickling defect detection algorithm based on improved YOLOv5s algorithm is proposed. Through this algorithm, the image information of the actual working condition of strip steel after pickling is extracted, and then different kinds of strip steel pickling defects can be distinguished, then realize the process control of strip steel pickling defects. In order to improve the recognition accuracy of the algorithm, Soft NMS is used instead of NMS to prevent the situation that targets overlap and are ignored; CBAM attention mechanism is integrated into the Neck module to improve the recognition accuracy of specific targets; Use SIoU Loss instead of GIoU Loss as the improvement of loss function. The experimental simulation shows that the average recognition accuracy of the algorithm before optimization is 75.5%, and the recognition accuracy of the algorithm after optimization is 79.2%, and the size of the algorithm model is also reduced.