基于HSA注意力模块和关联规则的行人属性识别算法  

Pedestrian Attribute Recognition Algorithm Based on HSA Attention Module and Association Rules

作　　者：孙志伟 师亚涛 马永军[1] 闫潇宁 许能华 SUN Zhiwei;SHI Yatao;MA Yongjun;YAN Xiaoning;XU Nenghua(College of Artificial Intelligence,Tianjin University of Science&Technology,Tianjin 300457,China;Shenzhen Softsz Co.,Ltd.,Shenzhen 518131,China)

机构地区：[1]天津科技大学人工智能学院,天津300457 [2]深圳市安软科技股份有限公司,深圳518131

出　　处：《天津科技大学学报》2023年第1期61-69,共9页Journal of Tianjin University of Science & Technology

基　　金：国家自然科学基金资助项目(61603273);天津市自然科学基金资助项目(18JCQNJC69500)。

摘　　要：针对先前研究工作对行人属性相关性分析不足和行人图像中细粒度属性特征难以捕获的问题,首先基于关联规则寻找属性之间的相关性,然后根据相关性改变网络结构,提升有较强相关性属性的准确率并嵌入改进的HSA(hierarchy split attention)注意力模块,将特征图中的潜在信息进行深度挖掘.HSA注意力模块将特征图进行分组和在子特征图之间增加通道交互操作,整合后的特征图输入挤压激励(squeeze and excitation,SE)模块中,提取图像在通道上的信息.在PA100K、Market-1501和PETA数据集上的实验结果表明,本算法的精确率、召回率、F1值与其他算法相当,但是准确率有较大提升.In view of insufficient correlation analysis of pedestrian attributes in previous research and the difficulty in capturing fine-grained attribute features in pedestrian images,in our present study we first find the correlation between attributes based on association rules,then change the network structure according to the correlation,improve the accuracy of highly correlated attributes,and embed the improved hierarchy split attention(HSA)module to deeply mine the latent information in the feature map.HSA module groups feature maps and adds channel interaction between sub feature maps,inputs the integrated feature map into the squeeze and excitation(SE)module,and extracts the information of the image on the channel.The experimental results on PA100k,Market-1501 and PETA data sets show that this algorithm has little difference with other algorithms in precision,recall and F1,but has a great improvement in accuracy indicators.

关 键 词：图像处理 行人属性识别 注意力模块 协同训练 残差网络 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]