Chinese Journal of Liquid Crystals and Displays, Volume. 38, Issue 7, 964(2023)

Improved lightweight helmet wear detection algorithm

Xue-chun LIU1,2, Da-ming LIU1,2、*, and Ruo-chen LIU1,2
Author Affiliations
  • 1School of Electronics and Electrical Engineering,Ningxia University,Yinchuan 750000,China
  • 2Key Laboratory of Intelligent Sensing of Desert Information,Ningxia University,Yinchuan 750000,China
  • show less
    Figures & Tables(16)
    YOLOv5 structure
    GhostNet network structure
    Ghost BottleNeck principle structure
    Quadruple scale detection of improved structures
    CA implementation principle
    Mosaic data enhancement
    Interactive inspection interface
    Results of visual analysis of background occlusion
    Results of visualisation analysis in a rainfall environment
    Results of visualisation analysis in foggy conditions
    • Table 1. YOLOv5-Q backbone extraction network structure

      View table
      View in Article

      Table 1. YOLOv5-Q backbone extraction network structure

      FromNumberParamsModuleArguments
      -11232Conv[3,8,3,2,1]
      -11476GhostBottleneck[8,8,16,3,1]
      -112 280GhostBottleneck[8,16,48,3,2]
      -112 428GhostBottleneck[16,16,72,3,1]
      -116 914GhostBottleneck[16,24,72,3,2]
      -1112 318GhostBottleneck[24,24,120,3,1]
      -1115 684GhostBottleneck[24,40,240,3,2]
      -1110 608GhostBottleneck[40,40,184,3,1]
      -11149 404GhostBottleneck[40,56,480,3,1]
      -11150 452GhostBottleneck[56,56,480,3,1]
      -11295 880GhostBottleneck[56,80,672,3,2]
      -1191 880GhostBottleneck[80,80,960,3,1]
      -11553 880GhostBottleneck[80,80,960,3,1]
      -1191 880GhostBottleneck[80,80,960,3,1]
      -11553 880GhostBottleneck[80,80,960,3,1]
      -1139 360Conv[80,480]
    • Table 2. Comparison results of YOLOv5m and lightweight model

      View table
      View in Article

      Table 2. Comparison results of YOLOv5m and lightweight model

      算法参数量模型大小/MBmAP
      YOLOv5m20 856 97542.170.949
      轻量级模型12 187 95124.930.933
    • Table 3. Correspondence between feature maps and a priori boxes

      View table
      View in Article

      Table 3. Correspondence between feature maps and a priori boxes

      特征图大小感受野先验框大小
      20×20[116,90][156,198][373,326]
      40×40较大[30,61][62,45][59,119]
      80×80101316,30][33,23]
      160×1605679[12,10]
    • Table 4. Experimental operating environment

      View table
      View in Article

      Table 4. Experimental operating environment

      名称环境参数
      操作系统Windows 10 64位
      GPU12th Gen Intel(R)Core(TM)i9 12900 K 3.19 GHz
      内存64 GB
      PythonPython3.9版本
      深度学习框架Pytorch 1.10
    • Table 5. Comparison of experimental results

      View table
      View in Article

      Table 5. Comparison of experimental results

      算法RPmAP参数量模型大小/MB
      SSD0.6210.3340.56726 285 48691.1
      Faster R-CNN0.7610.4430.701137 098 724108
      YOLOv50.9320.9110.94920 856 97542.17
      YOLOv5-Q0.9280.8710.93712 696 64026.47
    • Table 6. Ablation experiments

      View table
      View in Article

      Table 6. Ablation experiments

      改进点1改进点2改进点3PRmAP模型大小/MB参数量
      YOLOv5×××0.9320.9110.94942.1720 856 975
      模型1××0.9250.8750.93324.9312 187 951
      模型2××0.9390.9090.95943.5521 293 508
      模型3××0.9360.9110.95442.3420 933 523
      模型4×0.9180.8790.93926.2912 617 572
      模型5×0.9280.9190.95943.7321 392 304
      模型6×0.9280.8720.93625.112 264 499
      本文算法0.9280.8710.93726.4712 696 640
    Tools

    Get Citation

    Copy Citation Text

    Xue-chun LIU, Da-ming LIU, Ruo-chen LIU. Improved lightweight helmet wear detection algorithm[J]. Chinese Journal of Liquid Crystals and Displays, 2023, 38(7): 964

    Download Citation

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

    Category: Research Articles

    Received: Aug. 13, 2022

    Accepted: --

    Published Online: Jul. 31, 2023

    The Author Email: Da-ming LIU (nxldm@qq.com)

    DOI:10.37188/CJLCD.2022-0268

    Topics