Laser & Optoelectronics Progress, Volume. 60, Issue 12, 1210011(2023)

Improved YOLOv4 Helmet Detection Algorithm Under Complex Scenarios

Guobo Xie1, Jingjing Tang1, Zhiyi Lin1、*, Xiaofeng Zheng1, and Ming Fang2
Author Affiliations
  • 1School of Computer Science and Technology, Guangdong University of Technology, Guangzhou 510006, Guangdong, China
  • 2Transmission Branch of Yunnan Power Grid Co., Ltd., Kunming 650011, Yunnan, China
  • show less
    Figures & Tables(10)
    YOLOv4 network structure
    SMD-YOLOv4 network structure
    SE-Net structure
    Multi-scale information fusion detection network
    DenseASPP network structure
    SMD-YOLOv4 model training situation. (a) Loss; (b) mAP
    Detection effect comparison of different models
    Comparison of detection effect with other deep learning target detection algorithms
    • Table 1. Comparison of the results of ablation experiments

      View table

      Table 1. Comparison of the results of ablation experiments

      YOLOv4SEMulti-scaleDwDenseASPPP /%R /%mAP /%T /ms
      92.3089.7694.1524.7
      93.7288.7894.7434.6
      94.8089.6095.4151.4
      94.5189.4095.2328.5
      95.1993.0097.3447.1
    • Table 2. Performance comparison with other deep learning target detection algorithms

      View table

      Table 2. Performance comparison with other deep learning target detection algorithms

      ModelP /%R /%mAP /%T /ms
      Faster R-CNN57.3175.3170.9340.1
      SSD93.7976.6790.9012.9
      YOLOv594.8186.1694.0923.7
      YOLOx94.5191.3395.8524.6
      SMD-YOLOv495.1993.0097.3447.1
    Tools

    Get Citation

    Copy Citation Text

    Guobo Xie, Jingjing Tang, Zhiyi Lin, Xiaofeng Zheng, Ming Fang. Improved YOLOv4 Helmet Detection Algorithm Under Complex Scenarios[J]. Laser & Optoelectronics Progress, 2023, 60(12): 1210011

    Download Citation

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

    Category: Image Processing

    Received: Apr. 22, 2022

    Accepted: Jun. 16, 2022

    Published Online: May. 23, 2023

    The Author Email: Zhiyi Lin (lzy291@gdut.edu.cn)

    DOI:10.3788/LOP221388

    Topics