Scholarlink

<p>For printed circuit board (PCB) image with spatial deformation, existing detection methods struggle to meet the reliability requirements. In response to this challenge, we propose a template-based spatial deformation perception alignment and feature vector matching network for PCB component detection. The novel network is based on the template-based alignment and matching algorithms. Firstly, it uses a pre-alignment module to eliminate large image offsets. Then, it uses a newly proposed Siamese multi-scale feature extraction network with shared parameters to extract component features synchronously from PCB image and template image. Next, it focuses on the tiny offset and perspective distortion of PCB image through the novel spatial deformation perception alignment module, and achieves high-precision alignment of PCB image with template image through the homography transformation module. Finally, it uses a feature vector matching network to match the aligned PCB image with the template image in terms of similarity, providing information such as the category, missing, and incorrect installation of PCB components. The experimental results demonstrate that the new network achieved an average pixel offset error of 0.82 in image alignment. After alignment, the component classification error rates were 17.38% and 21.09% on the known PCB batches and new PCB batches. The source code is available at <a href="https://github.com/ustl-szh/TSAFM-Net">https://github.com/ustl-szh/TSAFM-Net</a>.</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

TSAFM-Net: template-based spatial deformation perception alignment and feature vector matching network for PCB component detection

  • 期刊论文

  • 发表日期：

    2025年2月22日

查看全文
  • Zhihan Sun,
  • Rongfen Gong,
  • Maoxiang Chu,
  • Guanghu Liu

摘要

For printed circuit board (PCB) image with spatial deformation, existing detection methods struggle to meet the reliability requirements. In response to this challenge, we propose a template-based spatial deformation perception alignment and feature vector matching network for PCB component detection. The novel network is based on the template-based alignment and matching algorithms. Firstly, it uses a pre-alignment module to eliminate large image offsets. Then, it uses a newly proposed Siamese multi-scale feature extraction network with shared parameters to extract component features synchronously from PCB image and template image. Next, it focuses on the tiny offset and perspective distortion of PCB image through the novel spatial deformation perception alignment module, and achieves high-precision alignment of PCB image with template image through the homography transformation module. Finally, it uses a feature vector matching network to match the aligned PCB image with the template image in terms of similarity, providing information such as the category, missing, and incorrect installation of PCB components. The experimental results demonstrate that the new network achieved an average pixel offset error of 0.82 in image alignment. After alignment, the component classification error rates were 17.38% and 21.09% on the known PCB batches and new PCB batches. The source code is available at https://github.com/ustl-szh/TSAFM-Net.