摘要
从人体目标雷达回波数据中提取可分性较好的微动特征是实现目标分类的关键。针对传统谱图结构特征无法对体型相似的人体目标精细识别,提出了基于堆栈稀疏自编码器的人体身份认证方法。首先构造堆栈稀疏自编码器网络,利用人体微动数据进行无监督预训练,在不同层提取人体微动特征,然后将得到的特征输入softmax分类器进行有监督训练,用交叉验证调整网络参数,最后用训练好的网络进行人体目标分类。在不同人走路实测数据集上,3人平均识别率达到了83%,优于提取谱图结构特征分类的方法。
Extracting micro-motion features from human radar echo data is a key to human target classification.Aimed at the problem that the traditional spectrum structure is hard to realize the fine recognition of similar body size,a method of human body identity authentication based on stack sparse autoencoder is proposed.First of all,this paper constructs a stack-sparse self-encoder network,performs unsupervised pre-training by using human micro-motion data,and extracts human micro-motion features at different layers.Then the paper inputs the features into the softmax classifier for supervised training,and adjusts the network parameters by cross-validation.Finally,the paper uses the trained network for human target classification.The average recognition rate of 3 people on the measured data set of different people reaches83%,and is better than that by the method of extracting spectral structure feature classification.
作者
袁延鑫
孙莉
张群
YUAN Yanxin;SUN Li;ZHANG Qun(Information and Navigation College,Air Force Engineering University,Xi＇an 710077,China)
出处
《空军工程大学学报（自然科学版）》
CSCD
北大核心
2018年第4期48-53,共6页
Journal of Air Force Engineering University(Natural Science Edition)
基金
国家自然科学基金(61701531)
航空科学基金(20121996016)
陕西省统筹创新工程特色产业创新链项目(2015KTTSGY0406)
关键词
堆栈稀疏自编码器
特征提取
微动特征
身份认证
stack sparse autoencoder
feature extraction
micro-motion feature
identity authentication
