多层次深度特征交换的人体解析方法  

Human Parsing Method for Multi-level Deep Feature Exchange

作　　者：罗文劼 倪鹏 张涵 LUO Wen-jie;NI Peng;ZHANG Han(School of Cyber Security and Computer,Hebei University,Baoding 071002,China)

机构地区：[1]河北大学网络空间安全与计算机学院

出　　处：《小型微型计算机系统》2020年第1期149-154,共6页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(61375075)资助;河北省自然科学基金项目(F2019201451)资助

摘　　要：人体解析因其潜在的应用前景,成为计算机视觉领域重要的研究内容之一.虽然Segnet对全局与局部信息有较好的学习,但该网络只是进行简单的下采样和上采样操作,忽略了特征信息之间的交换学习,本文提出一种基于多层次深度特征交换网络(DFEnet)的人体解析方法.DFEnet网络既兼顾不同分辨率下高维特征学习,又可以满足不同分辨率下的特征交换学习.在DFEnet提取完人体语义特征后,空洞沙漏池化会对提取后的特征进行多尺度学习.在LIP数据集上的实验结果表明本文提出的方法具有更好的解析结果,与SS-JPPnet相比提高了1.4%MIoU,超过Segnet近26.51%MIoU.Human parsing has become one of the important research contents in the field of computer vision due to its potential application prospects.Although Segnet has a good learning of global and local information,the network only performs simple downsampling and upsampling operations,ignoring the exchange learning between feature information.This paper proposes a multi-level deep feature exchange network(DFEnet).Human body analysis method.The DFEnet network not only considers high-dimensional feature learning at different resolutions,but also satisfies feature exchange learning at different resolutions.After DFEnet extracts the human semantic features,the atrous hourglass pooling will perform multi-scale learning on the extracted features.Experimental results on LIP dataset showthat the method proposed in this paper has better results,which improves by 1.4%MIoU compared with SS-JPPnet and surpasses Segnet by nearly 26.51%MIoU.

关 键 词：人体解析 卷积神经网络 特征交换 多尺度 多分辨率 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]