PCB Defect Inspection based on YOLOv5-GCBNet Model

Addressing the issues of missed detections, challenges in feature extraction, and low detection accuracy associated with the traditional YOLOv5 model for Printed Circuit Boards (PCB) defect inspection, this study proposes the YOLOv5-GCBNet model algorithm. Firstly, the C3 module in the backbone network of the YOLOv5s model is replaced with the Global Context Network (GCNet) to more effectively capture the spatial and feature information of defect targets. Secondly, to enhance the detection accuracy for small targets, a Bidirectional Feature Pyramid Network (BiFPN) is integrated into the Neck structure to fully exploit multi-scale feature information and mitigate the issue of feature information loss. Lastly, the use of Distance Intersection over Union (DIoU) Loss as the model’s loss function significantly improves regression accuracy. Experimental results demonstrate that, compared to the traditional YOLOv5 model, the proposed method achieves increases in precision, recall, and mean average precision (mAP) by 2.9%, 1.4%, and 2.6%, respectively.