Optics and Precision Engineering, Volume. 32, Issue 11, 1788(2024)

Detection of conductive multi-particles based on circular convolutional neural network

Zilong LIU1...2, Chen LUO1,2,*, Yijun ZHOU1,2, and Lei JIA12 |Show fewer author(s)
Author Affiliations
  • 1College of Mechanical Engineering, Southeast University, Nanjing289, China
  • 2Wuxi Shangshi-finevision Technology Co., Ltd, Wuxi14174, China
  • show less
    Figures & Tables(16)
    Architecture of U-MultiNet and ours
    Schematic diagram of three kinds of convolution
    Comparison of process of calculating control parameters
    Initial layout of sampling points in circular convolution
    Labels processed by attention
    Test samples used in experiments
    Qualitative analysis of each module in ablation experiment
    Loss curves of four methods during training
    Results of conductive particle detection
    Conductive particle detection results of easy-detecting group
    Comparison of stability
    Comparison of stability through data graph
    • Table 1. Effect of parameters(chn & size) on circular convolution

      View table
      View in Article

      Table 1. Effect of parameters(chn & size) on circular convolution

      Parameter combinationAccRepeatReproduce
      Deformable convolution0.834 10.793 50.708 5
      Circular convolution(chn=1,size=7)0.810 50.752 60.665 8
      Circular convolution(chn=9,size=3)0.818 60.776 20.672 1
      Circular convolution(chn=9,size=7)0.825 50.803 00.710 4
      Circular convolution(chn=9,size=11)0.822 40.784 90.683 7
      Circular convolution(chn=18,size=7)0.825 90.803 50.706 5
    • Table 2. Results of ablation experiments

      View table
      View in Article

      Table 2. Results of ablation experiments

      MethodAccRepeatReproduce
      U-MultiNet0.805 40.723 50.674 4
      U-MultiNet+Circlular convolution0.814 20.796 20.708 4
      U-MultiNet+Attention0.813 30.775 40.697 1
      Ours0.825 50.803 00.710 4
    • Table 3. Results of comparison experiments

      View table
      View in Article

      Table 3. Results of comparison experiments

      MethodAccRepeatReproduceAverage time/msParameter number
      Lin60.798 60.764 00.652 612.28
      U-Net0.817 30.757 50.677 252.3775 328
      U-MultiNet0.813 20.730 00.680 536.6126 832
      CPNet0.801 70.665 50.687 741.7548 864
      Ours0.834 10.809 20.705 139.9435 888
    • Table 4. Comparison between deformable convolution and circular convolution (ours)

      View table
      View in Article

      Table 4. Comparison between deformable convolution and circular convolution (ours)

      MethodTraining time in 1 epoch/sAverage time per image/msParameter number
      Deformable convolution16247.9362 528
      Circular convolution(ours)11233.1335 888
    Tools

    Get Citation

    Copy Citation Text

    Zilong LIU, Chen LUO, Yijun ZHOU, Lei JIA. Detection of conductive multi-particles based on circular convolutional neural network[J]. Optics and Precision Engineering, 2024, 32(11): 1788

    Download Citation

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

    Category:

    Received: Dec. 25, 2023

    Accepted: --

    Published Online: Aug. 8, 2024

    The Author Email: LUO Chen (chenluo@seu.edu.cn)

    DOI:10.37188/OPE.20243211.1788

    Topics