基于PPYOLO的工人着装规范检测  

Detection of Dress Code of Workers Based on PPYOLO

作　　者：韩钰[1] 王紫玉 郑金亮[1] 王磊[1] 王晨旸 蔡培君[1] HAN Yu;WANG Zi-yu;ZHENG Jin-liang;WANG Lei;WANG Chen-yang;CAI Pei-jun(Jianghuai College,Anhui University,Hefei,Anhui 230031,China)

机构地区：[1]安徽大学江淮学院,安徽合肥230031

出　　处：《石家庄学院学报》2024年第3期5-11,30,共8页Journal of Shijiazhuang University

基　　金：安徽大学江淮学院院级科学研究项目(2022KJ0001);安徽省高等学校科学研究重点项目(自然科学类)(2022AH053061)。

摘　　要：针对传统人工检测工人作业时是否规范着装效率低的问题,提出了改进的深度学习检测算法PPYOLO-CA.首先,采用深度可分离卷积将backbone前部分卷积进行替换,深度可分离卷积能够减少参数并且增加网络的非线性度,进而增强其提取特征的能力;其次,通过在backbone的特征提取段添加移动网络注意力机制(CoordAtt)模块,提高图片空间信息提取能力,从而为目标检测提供丰富的特征信息;最后,将空间金字塔池化(SPP)模块改进为SPPF(SPP-Fast)模块,在不影响精度的情况下能有效降低模型的参数量并提高模型的运行速度.为验证该算法,在工人着装数据集上进行仿真实验,实验结果表明:在不同场景下改进模型相比于原始模型的识别效果更好,其中,改进模型平均精度均值(mAP)比原始模型高出1.8%,且在模型参数量上减少了0.14MB.An improved deep learning detection algorithm PPYOLO-CA has been proposed to solve the problem of low efficiency of traditional manual inspection workers'standardized dressing.First,the deep separable convolution is used to replace the former part of the backbone convolution.The deep separable convolution can reduce the parameters and increase the nonlinearity of the network,thus enhancing its ability to extract features;Secondly,by adding CoordAtt module to the feature extraction section of backbone,to improve the ability of image spatial information extraction and provide rich feature information for target detection;Finally,changing SPP module into SPPF module can effectively reduce the parameters of the model and improve the running speed of the model without affecting the accuracy.In order to verify the algorithm,simulation experiments are carried out on the data set of workers'clothing.The experimental results show that the improved model has better recognition effect than the original model in different scenarios.The improved model on mAP is 1.8%higher than that of the original model and 0.14 MB less than that in model parameters.

关 键 词：目标检测 着装规范 PPYOLO CoordAtt 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]