摘要
超细铝粉在工业生产中应用广泛,然而它在作业过程中易扩散的特点会造成作业现场粉尘浓度增高,增加粉尘燃爆及人员患职业病的风险。为了保障作业人员的健康,佩戴适宜的个体防护装备至关重要。为了解决超细铝粉称量作业过程中个体防护装备佩戴检测的精度和效率不高的问题,提出一种改进的YOLOv5s算法。首先,针对防护装备的小目标特征,优化原始YOLOv5s算法,引入ConvNeXt网络结构,以更有效地提取和保留小目标防护装备的特征信息;同时,改进非极大值抑制(NMS)的后处理方法,提升复杂作业环境中对防护装备的检测准确率;其次,通过LabelImg标注防护装备图像中的目标特征,形成数据集,并使用该数据集训练和评估改进的YOLOv5s算法;最后,分析并对比实验结果。结果表明,所提算法的平均精度均值(mAP)达到99.30%,参数量为6.95×10^(6),每帧检测时间为4.6 ms,验证了所提算法能够有效识别作业人员是否佩戴防护装备,进而为超细粉尘环境下作业人员的安全防护提供了有力的技术支撑。
Ultra-fine aluminum powder is widely used in industrial production,however,its tendency to disperse during operations increases workplace dust concentration,thereby elevating the risks of dust explosion and people suffering from occupational diseases.The proper use of personal protective equipment is crucial for ensuring worker health.To address low precision and insufficient efficiency in personal protective equipment wearing detection during ultra-fine aluminum powder weighing operations,an improved YOLOv5s algorithm was proposed.Firstly,aiming at the small-object characteristics of protective equipment,the original YOLOv5s algorithm was optimized,the ConvNeXt network structure was introduced,thereby enabling more effective extraction and retention of feature information of small-object protective equipment.At the same time,the post-processing method of Non-Maximum Suppression(NMS)was modified to improve detection accuracy of protective equipment in complex operating environments.Secondly,a dataset was constructed by annotating object features in protective equipment images using LabelImg,and the dataset was utilized to train and evaluate the improved YOLOv5s algorithm.Finally,the experimental results were analyzed and compared.The results demonstrate that the proposed algorithm achieves a mean Average Precision(mAP)of 99.30%,with a parameter number of 6.95×10^(6) and a detection time of 4.6 ms per frame,verifying that the proposed algorithm can identify whether workers are wearing protective equipment effectively,thereby providing strong technical support for safety protection of workers in ultra-fine dust environments.
作者
吴雨佳
杜文霞
南小影
刘仓
邢亚飞
孟荣荣
WU Yujia;DU Wenxia;NAN Xiaoying;LIU Cang;XING Yafei;MENG Rongrong(Institute for Industrial Hygiene of Ordnance Industry,Xi’an Shaanxi 710065,China;Occupational Disease Hazard Engineering Protection Technology Guidance Center of Shaanxi Province,Xi’an Shaanxi 710065,China)
出处
《计算机应用》
北大核心
2025年第S1期271-275,共5页
journal of Computer Applications
基金
国防科工局专项科研项目(GFJ2023-02)
中国兵器工业集团统筹安全生产经费资助项目(JTK2023-01)。
关键词
深度学习
小目标检测
YOLOv5s
铝粉称量
个体防护装备
deep learning
small object detection
YOLOv5s
aluminum powder weighing
personal protective equipment
