无监督电子连接器异常检测方法
DOI:
CSTR:
作者:
作者单位:

西安科技大学通信与信息工程学院 西安 710000

作者简介:

通讯作者:

中图分类号:

TN911.73

基金项目:

西安市科技计划科学家工程师队伍建设项目(23KGDW0032-2022)资助


Unsupervised electronic connectors anomaly detection method
Author:
Affiliation:

College of Communication and Information Technology,Xi′an University of Science and Technology,Xi′an 710000, China

Fund Project:

  • 摘要
  • |
  • 图/表
  • |
  • 访问统计
  • |
  • 参考文献
  • |
  • 相似文献
  • |
  • 引证文献
  • |
  • 资源附件
  • |
  • 文章评论
    摘要:

    由于电子连接器异常样本稀缺,使得有监督模型难以捕捉异常样本特征,很大程度上限制了有监督学习方法的检测性能。并且现有无监督模型存在重建图像模糊、缺陷残留的问题,严重影响检测精度。为此,提出一种仅需无异常样本训练的一步去噪单纯形扩散模型电子连接器异常检测方法。由于高斯去噪扩散概率模型在图像重建任务中存在特征投影误差导致重建位置偏差,因此引入单纯形噪声构建单纯形噪声去噪扩散概率模型,并重新制定去噪范式使推理时间降低至0.09 s。此外,研究得到一种图像预处理方法,消除冗余特征干扰,使模型高效学习电子连接器表面特征,提高模型学习效率。实验结果表明,在异常检测标准评估度量AUROC准则下,所提方法显著优于现有无监督模型。图像级检测准确率达99.71%,像素级精度达到93.86%,展现出卓越的异常检测性能。

    Abstract:

    The scarcity of abnormal samples of electronic connectors makes it difficult for supervised models to capture abnormal sample features, which largely limits the detection performance of supervised learning methods. Moreover, the existing unsupervised models have the problems of blurred reconstructed images and defects remaining, which seriously affect the detection accuracy. To this end, a one-step denoising simplex denoising diffusion probabilistic model electronic connector anomaly detection method that requires only anomaly-free sample training is proposed. Since Gaussian denoising diffusion probabilistic model has feature projection error in the image reconstruction task that leads to reconstruction position deviation, simplex noise is introduced to construct a simplex noise denoising diffusion probabilistic model, and the denoising paradigm is reformulated so that the inference time is reduced to 0.09 s. In addition, the research obtains an image preprocessing method that eliminates the interference of redundant features, so that the model learns the surface features of the electronic connector efficiently and improves the model learning efficiency. model learning efficiency. The experimental results show that the proposed method significantly outperforms the existing unsupervised models under the AUROC criterion, a standard evaluation metric for anomaly detection. The image-level detection accuracy reaches 99.71% and the pixel-level accuracy reaches 93.86%, demonstrating excellent anomaly detection performance.

    参考文献
    相似文献
    引证文献
引用本文

唐善成,杨继清,李恒.无监督电子连接器异常检测方法[J].电子测量技术,2025,48(11):24-32

复制
相关视频

分享
文章指标
  • 点击次数:
  • 下载次数:
  • HTML阅读次数:
  • 引用次数:
历史
  • 收稿日期:
  • 最后修改日期:
  • 录用日期:
  • 在线发布日期: 2025-07-07
  • 出版日期:
文章二维码