Chinese Journal of Lasers, Volume. 48, Issue 22, 2202011(2021)

Automatic Weld Tracking Based on Convolution Neural Network and Correlation Filter

Guowei Yang, Nan Zhou, Min Yang, Yongshuai Zhang, and Yizhong Wang*
Author Affiliations
  • College of Electronic Information and Automation, Tianjin University of Science & Technology, Tianjin 300222, China
  • show less
    Figures & Tables(12)
    Network structure diagram of weld feature point extraction
    Schematics of prior box generation process. (a) Weld image; (b) feature map in network
    Schematics of marking weld feature points. (a) Weld image; (b) marking graph
    Extracted weld feature points under different noise interference conditions
    Extracted feature points of weld with different grooves
    Tracking results of weld feature points under different levels of noise
    Enlarged display of weld feature point tracking results under different levels of noise
    Feature points tracking of weld with different groove types
    Enlarged display of feature points tracking of weld with different groove types
    Weld feature points tracking error graph
    Error analysis diagram
    • Table 1. Convolution kernel parameters

      View table

      Table 1. Convolution kernel parameters

      LayerKernel(width×height×number)StrideReceptive field size
      Conv 111×11×32111×11
      Pool 14×4414×14
      Conv 27×7×64138×38
      Pool 22×2242×42
      Conv 35×5×128174×74
      Pool 32×2282×82
      Conv 43×3×2561114×114
      Pool 42×22130×130
    Tools

    Get Citation

    Copy Citation Text

    Guowei Yang, Nan Zhou, Min Yang, Yongshuai Zhang, Yizhong Wang. Automatic Weld Tracking Based on Convolution Neural Network and Correlation Filter[J]. Chinese Journal of Lasers, 2021, 48(22): 2202011

    Download Citation

    EndNote(RIS)BibTexPlain Text
    Save article for my favorites
    Paper Information

    Category: laser manufacturing

    Received: May. 8, 2021

    Accepted: Jun. 15, 2021

    Published Online: Oct. 28, 2021

    The Author Email: Wang Yizhong (yzwang@tust.edu.cn)

    DOI:10.3788/CJL202148.2202011

    Topics