| In today’s era of big data,reliance on the Internet is gradually increasing,so identity verification is everywhere.Identification technology has a wide range of applications in daily life,such as devices unlocking,cell phone payment,system login and so on.Fingerprint recognition requires direct contact between human fingers and sensors,which leads to some safety hazards.Face recognition has certain requirements for environmental lighting to collect information,and there is a risk of privacy leakage of face information.Humanity’s radar echo contains some unique characteristics,which can be used for identification as the information.The information collection is non-contact and is not influenced by light.This paper proposes a human identity recognition technique based on continuous wave radar signals.We analyze the humanity’s radar echo signal and use deep learning to identify individuals.The main research work and contributions of this paper are as follows.(1)Signal pre-processing: this paper proposes a data pre-processing method for multi-channel radar wave signals.By analyzing the waveform characteristics of the radar waves,the method achieves automatic detection of abnormal signals.Based on the detection results,the method can screen out abnormal radar wave signals and retain high-quality radar wave data.Signal fusion is used for this multi-channel radar waves.It can make up for the lack of information of single-channel radar waves.(2)Feature extraction: In this paper,besides using the traditional single-period features,a series of new features are proposed for identity recognition,including the peak point envelope feature,the amplitude change rate feature and the period change rate feature.Combined with the amplitude-frequency mean feature and amplitude-frequency standard deviation feature,multi-feature fusion is used for identification.Compared with the method of only using the single-period features,the added features can provide multi-period correlation information of the signal.(3)Model construction: This paper proposes a identification method based on deep learning and humanity’s radar echo signals.We extracted multiple features of the radar waves in time domain and frequency domain.They are input to the respective convolutional networks.The features are fused by the feature fusion layer,and then go through the neural network module containing residual units.The final identification results can be obtained.The experimental results show that this method for identification achieves a combined accuracy of93.63%,which is 9.34% higher than that of the traditional support vector machine method. |
