点击次数：

影响因子：6.6

DOI码：10.1016/j.asoc.2025.114072

发表刊物：Applied Soft Computing

关键字：AI in engineering; Defect detection; Spatial attention mechanism; Deep learning; Computer vision; Transformer

摘要：The integrity of Printed Circuit Boards (PCBs) is critical to the performance and reliability of electronic devices. However, existing deep learning-based detection methods often struggle to accurately identify small and complex defects in challenging industrial environments, primarily due to their inability to effectively model irregular defect morphologies, sensitivity to background noise, insufficient multi-scale feature fusion, and difficulties in achieving efficient detection under limited computational resources. To address these issues, we propose a novel object detection framework based on YOLOv8, which integrates a dual-branch feature extraction module, a deformable attention mechanism, and an enhanced spatial attention head. Specifically, we design a Dual-Transformer downsampling module that effectively captures both global context and local details of PCB defects. We also introduce a deformable attention mechanism into the C2f module to handle irregular defect shapes adaptively. Furthermore, we propose a lightweight detection head that employs multi-scale spatial attention and depthwise separable convolutions to enhance feature representation while reducing computational cost. To improve the localization accuracy of small defects, we introduce a new loss function that combines Normalized Wasserstein Distance with Wise-IoUv3. Extensive experiments on the publicly available PKU-Market-PCB dataset demonstrate that YOLO-DTS achieves a precision of 88.4%, a recall of 69.9%, and an mAP50 of 77.5%, outperforming the baseline YOLOv8 by 4.9%, 7.5%, and 7.5%, respectively. The parameters used have been reduced by 13.4% compared to the baseline model. Additional experiments on DeepPCB and aluminum profile defect datasets further validate the strong generalization capability of our method. The results indicate that YOLO-DTS is a robust and efficient solution for PCB defect detection in real-world industrial scenarios.

论文类型：期刊论文

学科门类：工学

文献类型：J

卷号：186

期号：A

页面范围：114072

ISSN号：1568-4946

是否译文：否

发表时间：2025-10-21

收录刊物：SCI

发布期刊链接：https://www.sciencedirect.com/science/article/pii/S1568494625013857

第一作者：Yufeng Ou

通讯作者：Chia-Hung Wang