基于压缩残差网络的雷达辐射源识别方法研究  

Radar radiation source recognition method based on compressed residual network

作　　者：郭恩泽 刘正堂 崔博 刘国彬 史航宇 蒋旭 Guo Enze;Liu Zhengtang;Cui Bo;Liu Guobin;Shi Hangyu;Jiang Xu(Unit 63893 of the PLA,Luoyang 471003,China;Unit 63896 of the PLA,Luoyang 471003,China)

机构地区：[1]中国人民解放军63893部队,河南洛阳471003 [2]中国人民解放军63896部队,河南洛阳471003

出　　处：《强激光与粒子束》2024年第4期113-122,共10页High Power Laser and Particle Beams

基　　金：国家自然科学基金项目(61571043)。

摘　　要：针对低信噪比条件下,现有的雷达辐射源信号识别方法存在识别正确率低、时效性差的问题,提出了一种基于压缩残差网络的雷达辐射源信号识别方法。首先,利用Choi-Williams分布的时频分析方法将时域信号转换为二维时频图像;然后,根据应用场景特点,选择卷积神经网络(Convolutional Neural Networks,CNN)“压缩”范围;最后,构建压缩残差网络来自动提取图像特征并完成分类。仿真实验结果表明,在同等体量的设计下,与当前较为常用的标准CNN以及ResNet模型相比,所提模型能够降低信号识别运行时间约88%,在信噪比为-14 dB条件下对14种雷达辐射源信号的平均识别率高约5%。提供了一种高效的雷达辐射源信号智能识别方法,具有潜在的工程应用前景。Aiming at the problems of low recognition accuracy and poor timeliness of existing radar emitter signal recognition methods under the condition of low SNR,this paper proposes a radar emitter signal recognition method based on compressed residual network.Using Choi-Williams distribution for reference,the time-domain signal is converted into a two-dimensional time-frequency image,which improves the effectiveness of signal essential feature extraction.According to the characteristics of the application scenario,it selects the“compression”range of convolutional neural networks(CNN),and builds a compression residual network to automatically extract image features and identify.The simulation results show that compared with other advanced models,the proposed method can reduce the running time of signal recognition by about 88%,and the average recognition rate of 14 radar emitter signals is at least 5%higher when the signal-to-noise ratio is-14 dB.This paper provides an efficient intelligent recognition method of radar emitter signal,which has potential engineering application prospects.

关 键 词：压缩残差网络 时频分析 雷达辐射源识别 深度学习 扩张卷积 

分 类 号：TN971[电子电信—信号与信息处理]