YOLO-DSE: PCB surface defect detection with the improved YOLO

As electronic products advance rapidly, maintaining high-quality standards for printed circuit boards (PCBs) has become more essential than ever. This research introduces an enhanced YOLOv5 algorithm to boost both the accuracy and efficiency of PCB defect detection. Our main contributions include introducing dynamic snake convolution in the C3 structure to enhance feature extraction capabilities, incorporating the SimAM attention mechanism to increase the saliency of feature maps, and refining the feature fusion network for more effective integration of multi-level feature information. As a result of these improvements, we have achieved significant performance gains on PCB defect detection tasks, with a 3.4% increase in mAP (mean average accuracy). Our improved algorithm is superior to traditional methods in terms of accuracy and efficiency, and has good application prospects.