Laser & Optoelectronics Progress, Volume. 61, Issue 22, 2212002(2024)

Detection Method of Pavement Cracks in Aerial Images Based on Double Pyramid Network

Mingxing Gao1、*, Zhengfa Jiang1, Lin Zhang2, and Haoyang Wang2
Author Affiliations
  • 1College of Energy and Transportation Engineering, Inner Mongolia Agricultural University, Hohhot 010000, Inner Mongolia , China
  • 2Inner Mongolia Communications Group Co., Ltd. Tongliao Branch, Tongliao 028000, Inner Mongolia , China
  • show less
    Figures & Tables(16)
    Double pyramid network structure
    VPG module diagram matching the decoder 128×128×80 scale
    VPD_top structure
    Scale-aware module
    SAPF module before improvement
    Improved DSPF module
    Different loss function effects
    Combined loss function effect
    Dataset construction process
    Visualize the results of different models
    Visualization of prediction results of CFD datasets
    • Table 1. Improved MobileNetv3_small

      View table

      Table 1. Improved MobileNetv3_small

      Input(H×W×COperational blocks(CKCECCOCSEACTSN
      512×512×3Conv(3, 3, 16, 16)TrueHS22
      256×256×16Bneck(16, 3, 16, 16)FalseRU11
      256×256×16Bneck(16, 3, 64, 24)TrueRU21
      128×128×24Bneck(24, 3, 72, 24)FalseRU11
      128×128×24Bneck(24, 5, 72, 40)TrueRU21
      64×64×40Bneck(40, 5, 120, 40)FalseRU12
      64×64×40Bneck(40, 3, 240, 80)TrueHS21
      32×32×80Bneck(80, 3, 184, 80)FalseHS13
      32×32×80Bneck(80,3,480,112)TrueHS21
      16×16×112Bneck(112, 5, 576, 112)FalseHS11
      16×16×112Conv(112, 3, 288, 288)FalseHS12
      16×16×288Conv(288, 3, 512, 512)TrueRU612
    • Table 2. Model ablation experiments

      View table

      Table 2. Model ablation experiments

      ModelRecall /%F1_score /%MIoU /%Params /106
      CPF83.5685.4777.1632.36
      CPF+VG85.7887.8180.3433.86
      CPF+DS84.9486.9279.1831.70
      M-CPF82.6284.8976.6214.52
      Ours85.3987.5179.8415.63
    • Table 3. Experiments of different lightweight models

      View table

      Table 3. Experiments of different lightweight models

      ModelF1_score /%MioU /%Params /106
      ResNet-3485.4777.1620.80
      MobileNetv281.3272.413.50
      MobileNetv3_small82.1773.382.54
      MobileNetv3_large82.9474.235.48
      MobileNetv3_my84.8976.623.81
    • Table 4. Compare experiments with different models in my dataset

      View table

      Table 4. Compare experiments with different models in my dataset

      ModelF1_score /%MIoU /%Params /106MFLOPs
      U-Net84.8776.5824.89225.84
      FCN80.4471.4333.6185.28
      SegNet85.3477.0629.44160.68
      DeepLabv3+82.5473.7339.76173.40
      CPFNet85.4777.1632.3632.13
      Proposed model87.5279.8515.6316.74
    • Table 5. CFD dataset experimental results

      View table

      Table 5. CFD dataset experimental results

      ModelF1_score /%MIoU /%Params /106MFLOPs
      U-Net74.5865.9024.89M225.84
      FCN71.3263.0333.61M85.28
      SegNet76.9068.0529.44M160.68
      DeepLabv3+75.1966.4239.76M173.40
      CPFNet76.2467.4632.36M32.13
      Proposed model78.1169.2815.63M16.74
    Tools

    Get Citation

    Copy Citation Text

    Mingxing Gao, Zhengfa Jiang, Lin Zhang, Haoyang Wang. Detection Method of Pavement Cracks in Aerial Images Based on Double Pyramid Network[J]. Laser & Optoelectronics Progress, 2024, 61(22): 2212002

    Download Citation

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

    Category: Instrumentation, Measurement and Metrology

    Received: Feb. 29, 2024

    Accepted: Mar. 25, 2024

    Published Online: Nov. 13, 2024

    The Author Email: Mingxing Gao (gaomingxing_2000@126.com)

    DOI:10.3788/LOP240772

    CSTR:32186.14.LOP240772

    Topics