基于微多普勒信号分离和SqueezeNet的人体身份识别

Human Identity Recognition Based on Micro-Doppler Signal Separation and SqueezeNet

针对基于雷达传感器的人体身份识别问题,本文提出一种基于微多普勒信号分离和SqueezeNet的人体身份识别方法。首先利用雷达对人体行走的步态进行探测并收集回波数据,回波数据经过预处理得到微多普勒时频谱图;其次用阈值法对时频谱图进行微多普勒信号分离从而得到四肢的时频谱图;最后将其输入到SqueezeNet网络,采用Softmax分类器来实现人体身份识别。实验结果表明,经过微多普勒信号分离后人体身份识别准确率提高5.25%,SqueezeNet网络与其他网络相比,在网络性能上具有准确率高、网络参数少、测试时间短等优势。

Aiming at the problem of human identity recognition based on radar sensor,this paper proposes a human identity recognition method based on micro-Doppler signal separation and SqueezeNet.Firstly,the radar is used to detect the human walking gait and collect the echo data,which is pre-processed to obtain micro-Doppler time-frequency spectrum.Secondly,the micro-Doppler signal separation is used to obtain the time-frequency spectrum of the limbs by the threshold method.Finally,it is input into the SqueezeNet network and the Softmax classifier is used to achieve the human identity recognition.The experimental results show that the accuracy of human identity recognition is improved by 5.25%after micro-Doppler signal separation.Comparing with other networks,SqueezeNet network has the advantages of high accuracy,less network parameters and shorter testing time.