摘 要：针对目前智慧安监领域对于安全帽佩戴的检测存在尺度多样化、检测难度大、中小目标漏检率高的问题，提出了一种基于改进的 YOLOv4 的安全帽佩戴检测算法。首先，改进 K-means 算法重新选择锚框，然后在网络中引入 CBAM 注意力模块来增强安全帽佩戴信息的特征表达，最后对模型进行加速剪枝。实验结果表明，提出的算法在检测中 mAP@0.5 值提升了6.7%，检测速度提升了 35%，模型参数量减少了 48%，改进后的模型更适用于实际场景中对安全帽佩戴行为的识别。
关键词：安全帽佩戴检测；YOLOv4 网络；改进 K-means；CBAM；剪枝
DOI:10.19850/j.cnki.2096-4706.2021.22.044
基金项目：国家自然科学基金项目（661472258）
中图分类号：TP391.4 文献标识码：A 文章编号：2096-4706（2021）22-0156-06
Safety Helmet Wearing Detection Algorithm Based on Improved YOLOv4
WANG Yuchen, XU Mingkun
(Beijing University of Posts and Telecommunications, Beijing 100876, China)
Abstract: Aiming at the current problems in the detection of safety helmet wearing in the field of intelligent safety supervision of diversified scales, difficult detection and high missed detection rate of small and medium-sized targets, a safety helmet wearing detection algorithm based on improved YOLOv4 is proposed. Firstly, the improved k-means algorithm reselects the anchor box, then introduces the CBAM attention module into the network to enhance the feature expression of safety helmet wearing information, and finally speeds up the pruning of the model. The experimental results show that the proposed algorithm improves the mAP@0.5 value by 6.7%, improves the detection speed by 35%, the amount of model parameters is reduced by 48%. The improved model is more suitable for the identification of safety helmet wearing behavior in the actual scene.
Keywords: safety helmet wearing detection; YOLOv4 network; improved K-means; CBAM; pruning
参考文献：
[1] 郝忠，魏延晓 .448 例建筑施工工伤事故的标准统计特征分析 [J]. 中国标准化，2017（2）：245-247+249.
[2] CHEN S B，TANG W H，JI T Y，et al. Detection of Safety Helmet Wearing Based on Improved Faster R-CNN [C]//2020 International Joint Conference on Neural Networks （IJCNN）. Glasgow：IEEE，2020：1-7.
[3] FANG Q，LI H，LUO X C，et al. Detecting non-hardhatuse by a deep learning method from far-field surveillance videos [J]. Automation in Construction，2018，85：1-9.
[4] GIRSHICK R，DONAHUE J，DARRELL T， et al. Rich Feature Hierarchies for Accurate Object Detection and Semantic Segmentation [C]//2014 IEEE Conference on Computer Vision and Pattern Recognition.Columbus：IEEE，2014：580-587.
[5] GIRSHICK R. Fast R-CNN [J/OL].arXiv：1504.08083 [cs. CV].[2021-09-25].https://arxiv.org/abs/1504.08083.
[6] RENS Q，HEK M，GIRSHICK R，et al． Faster R-CNN：Towards Real-Time Object Detection with Region Proposal Networks ［J］．IEEE Transactions on Pattern Analysis and Machine Intelligence，2017，39（6）：1137-1149．
[7] REDMON J，DIVVALA S，GIRSHICK R，et al. You Only Look Once：Unified，Real-Time Object Detection [C]//2016 IEEE Conference on Computer Vision and Pattern Recognition（CVPR）. LasVegas：IEEE，2016：779-788.
[8] LIU W，ANGUELOV D，ERHAN D，et al. SSD： Single Shot MultiBox Detector [C]//Computer Vision–ECCV 2016. Amsterdam：Springer，Cham，2016：21-37.
[9] FANG Q，LI H，LUO X，et al. Detecting non-hardhatuse by a deep learning method from far-field urveillance videos [J]. Automation in Construction，2018，85：1-9.
[10] HUANG J，RATHOD V，SUN C，et al. Speed/Accuracy Trade-Offs for Modern Convolutional Object Detectors [C]//2017 IEEE Conference on Computer Vision and Pattern Recognition（CVPR）. Honolulu：IEEE，2017：3296-3297.
[11] 王兵，李文璟，唐欢 . 改进 YOLO v3 算法及其在安全帽检测中的应用 [J]. 计算机工程与应用，2020，56（9）：33-40.
[12] 施辉，陈先桥，杨英 . 改进 YOLOv3 的安全帽佩戴检测方法 [J]. 计算机工程与应用，2019，55（11）：213-220.
[13] WANG C Y，LIAO H Y M，WU Y H，et al. CSPNet： A new backbone that can enhance learning capability of CNN [C]//2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops（CVPRW）.Seattle：IEEE，2020：1571-1580.
[14] REDMON J，FARHADI A. Yolov3：YOLOv3：An Incremental Improvement [J/OL].arXiv：1804.02767 [cs.CV].[2021-10- 25].https://arxiv.org/abs/1804.02767.
[15] LIN T Y，DOLLÁR P，GIRSHICK R，et al. Feature Pyramid Networks for Object Detection [C]//2017 IEEE Conference on Computer Vision and Pattern Recognition （CVPR）.Honolulu： IEEE，2017：936-944.
[16] 蔡畅 . 基于改进 K-means 的 K 近邻算法在电影推荐系统中的应用 [J]. 电子技术与软件工程，2020（18）：182-183.
[17] HU J，SHEN L，ALBANIE S，et al. Squeeze-andExcitation Networks [J].IEEE Transactions on Pattern Analysis and Machine Intelligence，2020，42（8）：2011-2023.
[18] WOO S，PARK J，LEE J Y，et al. CBAM：Convolutional Block Attention Module [C]//Computer Vision–ECCV 2018.Munich： Springer，Cham.2018：3-19.
[19] ZHANG P Y，ZHONG Y X，LI X Q. SlimYOLOv3： Narrower，Faster and Better for Real-Time UAV Applications [C]//2019 IEEE/CVF International Conference on Computer Vision Workshop （ICCVW）.Seoul：IEEE，2019：37-45.
[20] 周志华，机器学习 [M]，北京：清华大学出版社，2016： 23-28.
[21] KAMBOJ A，POWAR N. SAFETY HELMET DETECTION IN INDUSTRIAL ENVIRONMENT USING DEEP LEARNING [EB/OL]. [2021-09-25].https://aircconline.com/csit/papers/vol10/csit100518.pdf.
作者简介：王雨晨（1997—），女，汉族，河南洛阳人，硕士研究生在读，研究方向：大数据技术及智能信息处理；徐明昆 （1963—），男，汉族，北京人，硕士研究生导师，高级工程师， 硕士研究生，研究方向：软件理论与技术。