SWM-YOLO: A New PCB Surface Defect Detection Algorithm Based on YOLOv8

Printed circuit boards (PCBs) present significant challenges for defect detection due to their complex structures, small size, and high false positive/negative rates associated with traditional methods. To address these issues, this paper proposes SWM-YOLO, a novel network model based on YOLOv8n. By incorporating Swin Transformer backbone, the model enhances feature extraction capabilities, effectively capturing global contextual information. Furthermore, Mixed Local Channel Attention is integrated into the C2f structure to accurately extract detailed features while preserving overall feature representation. Finally, the Wise-IoU loss function is adopted to improve the model’s ability to detect subtle defects. Experimental results on a PCB surface defect dataset demonstrate a $3.2 \%$ precision improvement over the baseline YOLOv8n, significantly reducing false positives and negatives in PCB surface defect detection.