基于卷积神经网络的雷达人物身份识别研究

打开文本图片集

关键词：卷积神经网络；人体身份识别；微多普勒信号；雷达传感器

中图分类号：TN957 文献标志码：A 文章编号：1003-5168(2025)12-0024-05

DOI:10.19968/j.cnki.hnkj.1003-5168.2025.12.005

Convolutional Neural Network-Based Radar Human Identification Research

HE Shaofeng (Yantai Institute of Science and Technology,Yantai 264Ooo, China)

Abstract: [Purposes] Convolutional neural network is highly significant in achieveing human identification,mainly in terms of efficient feature extractionand adaptability to complexscenes.Therefore,the article proposes aradar-based human identification method based on convolutional neural network.[Methods]First,the walking gait echo signals of 1O subjects are preprocessed.Second,the Short-Time Fourier Transform (STFT) is performed to obtain the micro-Doppler time-frequency spectrogram of the human body walking.Finaly,nine types of convolutional neural networks are used to extract their features,and the extracted features are categorized using Softmax classifier to complete the human identification. [Findings] The test results show that,considering the accuracy,network parameters,test time and other factors,SqueezeNet network is more advantageous than the other 8 network in terms of network performance.In addition, the research method of using micro-Doppler signal separation resulted in a 2% increase in accuracy.[Conclusions] Therefore,both the use of micro-Doppler signal separation method and the selection of SqueezeNet network can improve the accuracy of human identification,providing valuable reference and insights for future research inthe field of identification.

Keywords: convolution neural network; human identification; micro-Doppler signal; radar sensors

0 引言

在各领域中发挥着越来越重要的作用。（剩余6722字）