基于改进YOLOv7的防护用品穿戴检测  

Detection of Wearing Personal Protective Equipment Based on Improved YOLOv7

作　　者：杨晓帆 韦少钗 YANG Xiaofan;WEI Shaochai(Guangdong Chaozhou Health Vocational College,Chaozhou 521000,Guangdong,China)

机构地区：[1]广东潮州卫生健康职业学院,广东潮州521000

出　　处：《汕头大学学报（自然科学版）》2024年第3期61-72,共12页Journal of Shantou University：Natural Science Edition

摘　　要：为缓解YOLOv7在检测个人防护用品时面临标签重写、标签分配不平衡和特征耦合等问题,提出一种基于改进YOLOv7的检测方法.首先去除YOLOv7的大尺度和中尺度输出层,以降低标签重写率,且保证输出层得到充分训练;其次将输出层的定位和分类解耦,避免不同任务的特征表示互相影响,并选择在边界框级别检测防护服,在关键点级别检测防护帽和防护手套;最后引入部分卷积,实现实时检测.为验证该方法的有效性,使用实验人员穿戴防护用品的图像数据对所提方法进行验证.结果表明,相比YOLOv7,该方法的精确率和召回率分别提高了4.1和4.5个百分点,FPS(Frames Per Second)提升了1.3帧,可满足实验室场景下的个人防护用品穿戴检测需求.To alleviate the problems of label rewriting,unbalanced label assignment and feature coupling faced by YOLOv7 in detecting personal protective equipment,an improved YOLOv7 detection method is proposed.Firstly,the large-scale and medium-scale output layers of YOLOv7 are removed to reduce the label rewrite rate and to ensure that the output layer is adequately trained;secondly,the localization and classification of the output layer are decoupled to avoid that the feature representations of the different tasks affect each other and choose to detect the protective clothing at the bounding-box level,and the protective cap and protective gloves at the key-point level;Finally,a partial convolution is introduced to achieve real-time detection.In order to verify the effectiveness of the method,the proposed method is validated using image data of experimenters wearing protective equipment.The results show that compared with YOLOv7,the method improves the precision and recall by 4.1 and 4.5 percentage points,respectively,and the FPS is improved by 1.3 frames,which can satisfy the needs of personal protective equipment wearing detection in laboratory scenarios.

关 键 词：个人防护用品 穿戴检测 YOLOv7 单尺度输出 特征解耦 部分卷积 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]