基于WiFi信号和深度学习的身份识别技术研究  

Research on Identity Recognition Technology Based on WiFi Signal and Deep Learning

作　　者：吴哲夫[1] 肖新宇 林超 龚树凤[1] 方路平[1] WU Zhefu;XIAO Xinyu;LIN Chao;GONG Shufeng;FANG Luping(College of Information Engineering,Zhejiang University of Technology,Hangzhou Zhejiang 310023,China)

机构地区：[1]浙江工业大学信息工程学院,浙江杭州310023

出　　处：《传感技术学报》2023年第1期78-84,共7页Chinese Journal of Sensors and Actuators

基　　金：浙江省自然科学基金项目(LZ22F010005);浙江省教育厅科研项目(Y201839636)。

摘　　要：随着价格低廉WiFi设备的广泛部署,无处不在的WiFi信号在人体感知和身份识别方面得到了应用。现有基于WiFi的人体身份识别大多依赖人的步态特征,需要人在WiFi收发设备间来回走动,这种方法限制了识别的速度、规模和应用场景。针对这一不足,提出了一种静态、非接触式快速人体身份识别方法,首先基于人体生物特征影响的射频信号生成特有的信道状态信息(CSI)指纹,这种静态的特征可以提高多人识别的效率;然后对信号进行数据增强和主成分分析(PCA)以减少训练时间和存储空间;最后将预处理后的数据进行多层深度卷积神经网络(DCNN)处理,提取出辨别性特征并进行身份识别。实验结果表明,所提方法可以在多达35人场景下进行快速识别,平均识别精度为95%,优于现有的方法。With the universal deployment of economical WiFi devices,ubiquitous WiFi signals have been applied to WiFi-based human sensing and individual identification.However,traditional methods rely on the individual’s gait characters,which need several minutes of walking to extract the wireless signature,limiting the speed and size of the identification,and finally decrease the applicability.A device-free fast person identification approach is proposed to address the challenges.Firstly,RF-based biometric characters are used to derive the personal Channel State Information(CSI)fingerprint,such static information could effectively enable quick identification of more people.Then data augment and Principal Component Analysis(PCA)are performed on the signal to greatly reduce training time and storage resources.Finally,a multi-layer deep convolutional neural network(DCNN)is designed to take the preprocessed data to extract distinctive features and identify individuals.Experimental results show that the proposed system can identify more people more quickly and accurately.In particular,the average identification accuracy can reach 95%with a total of 35 volunteers.

关 键 词：人体身份识别 信道状态信息 WIFI 深度卷积神经网络 

分 类 号：TN929[电子电信—通信与信息系统]