Scholarlink

<p>In the industrial production of printed circuit board assembly (PCBA) products, accurate defect detection is crucial. Automated optical inspection has been the mainstream method using for detection in industry. However, traditional methods still face numerous challenges in terms of effectiveness and accuracy. In this study, we propose a novel adversarial domain transformation and abnormality decomposition network (ADTAN) for defect detection in PCBA products. Based on the assumption that normal background features and abnormal defect features can be decomposed, and that all background features from normal and abnormal images share a common latent space, our proposed ADTAN effectively detects defects in PCBA products. Through training, ADTAN not only learns to accurately reconstruct the background of defect regions, but also achieves precise defect segmentation in an end-to-end manner. We conducted a series of experiments on mainstream defect datasets and on-site collected PCBA samples. The results demonstrate that our method achieves an AUROC score of 97<i>.</i>8% for mainstream public datasets, and various defects were accurately detected for industrial PCBA products.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

An adversarial network based on anomaly domain decomposition and transformation for industrial PCBA defect inspection

  • 期刊论文

  • 发表日期：

    2025年4月9日

查看全文
  • Qianfeng Pan,
  • Teng Liu,
  • Yue Hou,
  • Jingyun Chang,
  • Hua Yang

摘要

In the industrial production of printed circuit board assembly (PCBA) products, accurate defect detection is crucial. Automated optical inspection has been the mainstream method using for detection in industry. However, traditional methods still face numerous challenges in terms of effectiveness and accuracy. In this study, we propose a novel adversarial domain transformation and abnormality decomposition network (ADTAN) for defect detection in PCBA products. Based on the assumption that normal background features and abnormal defect features can be decomposed, and that all background features from normal and abnormal images share a common latent space, our proposed ADTAN effectively detects defects in PCBA products. Through training, ADTAN not only learns to accurately reconstruct the background of defect regions, but also achieves precise defect segmentation in an end-to-end manner. We conducted a series of experiments on mainstream defect datasets and on-site collected PCBA samples. The results demonstrate that our method achieves an AUROC score of 97.8% for mainstream public datasets, and various defects were accurately detected for industrial PCBA products.