Journal of Applied Optics, Volume. 44, Issue 3, 539(2023)

Real-time segmentation algorithm of crack images based on improved Fast-SCNN

Zheng ZHANG... Qinjian QIAN*, Jiazheng ZHOU, Zipeng KE and Xinyu HU |Show fewer author(s)
Author Affiliations
  • School of Mechanical Engineering, Hubei University of Technology, Wuhan 430068, China
  • show less
    Figures & Tables(11)
    Overall structure diagram of Fast-SCNN and improved Fast-SCNN
    Schematic of bottleneck with different step distances
    Comparison of two multi-scale feature fusion methods
    Schematic of position attention and LPAB insertion method
    Schematic of lightweight position attention module
    Visualization of attention gate response images
    Grad-CAM visualization results for SPP and LFAP
    Comparison results of different up-sampling methods
    Comparison results of different algorithms
    • Table 1. Comparison of experimental results with different up-sampling methods

      View table
      View in Article

      Table 1. Comparison of experimental results with different up-sampling methods

      上采样方式精度 / %召回率 / %F1 score / %mIOU / %
      双线性插值71.5869.6170.5875.53
      反卷积72.4470.8471.1876.23
      LPAB76.6676.8376.7480.31
    • Table 2. Comparison of experimental results with different algorithms

      View table
      View in Article

      Table 2. Comparison of experimental results with different algorithms

      算法精度/%召回率/%F1 score/%mIOU/%FLOPs/G参数量/M推理速度/(f/s)
      FCN[5]51.7662.9256.8069.06185.69134.2727
      U-Net[8]66.9576.8471.5575.92361.1728.9641
      PSP-Net[21]69.2068.1668.6875.65255.2565.5718
      DeeplabV3[22]68.8875.4272.0076.22260.7459.3915
      IC-Net[23]62.4569.0565.5873.815.577.8030
      DFA-Net[24]61.0964.7562.7872.301.772.1546
      Fast-SCNN[13]71.5869.6170.5875.530.871.14166
      Ours76.6676.8376.7480.310.961.20151
    Tools

    Get Citation

    Copy Citation Text

    Zheng ZHANG, Qinjian QIAN, Jiazheng ZHOU, Zipeng KE, Xinyu HU. Real-time segmentation algorithm of crack images based on improved Fast-SCNN[J]. Journal of Applied Optics, 2023, 44(3): 539

    Download Citation

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

    Category: Research Articles

    Received: Jul. 7, 2022

    Accepted: --

    Published Online: Jun. 19, 2023

    The Author Email: QIAN Qinjian (649473676@qq.com)

    DOI:10.5768/JAO202344.0302001

    Topics