面向港区安防的集装箱堆场作业人员与车辆检测算法研究  

Research on detection algorithm of operators and vehicles in container yard for port area security

作　　者：李金阳 李俚[1] 宋海涛 吴丛铭 林勇传[1] LI Jinyang;LI Li;SONG Haitao;WU Congming;LIN Yongchuan(School of Civil Engineering and Architecture,Guangxi University,Nanning Guangxi 530004,China;Guangxi Qingzhou FreeTrade Port Area Shenggang Gulf Port Terminal Co.,Ltd.,Qinzhou Guangxi 535008,China)

机构地区：[1]广西大学机械工程学院,广西南宁530004 [2]广西钦州保税区盛港码头有限公司,广西钦州535008

出　　处：《中国安全生产科学技术》2025年第4期158-164,共7页Journal of Safety Science and Technology

基　　金：国家自然科学基金项目(22272035);广西自然科学基金项目(2020GXNSFAA297214)。

摘　　要：为提高算法在异常天气、背景干扰与数据噪声等因素影响下,对港区集装箱堆场作业人员与车辆的检测精度,以及降低模型参数,满足实时检测需求,提出1种基于改进YOLOv5的面向自动化码头集装箱堆场作业人员与车辆的实时检测算法C3Rep-YOLO。该算法在特征提取网络中使用基于MetaFormer通用体系结构的轻量级ViT代替原来的CSPDarknet53网络,可提高算法的检测速度。在颈部网络中采用改进的RepGFPN结构代替原始的PAN结构,以提升特征融合效果与模型表征能力。由于在码头场景的实际检测过程中会不可避免地混入数据噪声,嵌入CAFM模块,以提高算法的全局特征捕获能力与去噪能力。研究结果表明:改进后的算法将检测速度提高了37.5%,mAP提高了4.8百分点,达到74.8%,可实现识别速度与精度的平衡。研究结果可为港区自动化安防与智慧港口建设提供理论基础和技术参考。In order to improve the detection accuracy of algorithms for the operators and vehicles in the container yard of port areas under the influence of factors such as abnormal weather,background interference,and data noise,etc.,as well as to reduce the model parameters,and meet the real-time detection requirements,a real-time detection algorithm C3Rep-YOLO for the operators and vehicles in the container yard of automated terminal based on improved YOLOv5 was proposed.A lightweight ViT based on the MetaFormer universal architecture was adopted in the feature extraction network instead of the original CSPDarknet53 network,which improved the detection speed of the algorithm.An improved RepGFPN structure was used in the neck network instead of the original PAN structure,so as to enhance the feature fusion effect and model representation ability.Due to the inevitable mixing of data noise in the actual detection process of the dock scene,the CAFM module was embedded to improve the global feature capture and denoising capabilities of the algorithm.The results show that the improved algorithm increases the detection speed by 37.5%,and mAP by 4.8%,reaching 74.8%,achieves a balance between recognition speed and accuracy.The research result can provide theoretical foundation and technical support for the automated security in port areas and the smart port construction.

关 键 词：交通工程 安防 自动化码头 目标检测 轻量化 注意力机制 

分 类 号：X913[环境科学与工程—安全科学]