PCB Surface Defect Detection Algorithm Based on Multi-Scale Feature Enhancement and Lightweight

With the continuous development of electronic technology, printed circuit boards (PCBs) are evolving toward higher density and precision. However, traditional surface defect detection methods face issues such as low accuracy and poor real-time performance, making it difficult to meet modern production requirements. This paper uses YOLOv8 as the base network and addresses the challenges of model deployment in resource-constrained scenarios and slow detection speeds. It proposes a lightweight PCB defect detection method, PDE-YOLO, based on partial convolution. Additionally, considering the challenges of identifying small defects, insufficient accuracy due to insignificant features, and false positives and false negatives in PCB defect detection, we propose another method, YOLOv8-MPSW, which combines multi-scale feature enhancement and attention mechanisms. Based on the proposed PDE-YOLO and YOLOv8-MPSW algorithms, we have developed a simple and efficient PCB surface defect detection system. Experimental results show that the PDE-YOLO algorithm achieves an mAP of 95.8%, with recall rates improved by 3.2% and 4.76% compared to Faster R-CNN and YOLOv8, respectively, reaching 90.4%, demonstrating significant reliability. The YOLOv8-MPSW algorithm achieved an mAP of 97.79%, which is 3.34% higher than the detection accuracy of YOLOv8s, meeting the high-precision requirements of industrial detection applications.