A Novel PCB Defect Detection Network Based on the Improved YOLOv8 with Fusion of Hybrid Attention Transformer and Bidirectional Feature Pyramid Network

The quality of the Printed Circuit Board (PCB) directly affects the performance and reliability of the electronic product. However, PCB defect detection still faces challenges in terms of high precision and real-time performance in automated inspection until now. In this article, a novel PCB defect detection network based on the improved YOLOv8 named YOLO-HB is proposed, which integrates Hybrid Attention Transformer (HAT) and Bidirectional Feature Pyramid Network (BiFPN). The proposed PCB defect detection network have the dual advantages of good real-time performance and high detection accuracy in the task of PCB defect detection. The experiments conducted on two well-known PCB defect detection public datasets, PCB_DATASET and DeepPCB, show that the results significantly outperform the original YOLOv8 model and also surpass other classic object detection models.