基于YOLOv5网络的复杂雷达脉内调制类型快速识别方法  

Fast recognition method of intra-pulse modulation type in complex radar based on YOLOv5 network

作　　者：沈永健 李育恒 张鹏宇 蔡敏康 李景文[1] SHEN Yongjian;LI Yuheng;ZHANG Pengyu;CAI Minkang;LI Jingwen(School of Electronic and Information Engineering,Beihang University,Beijng 100191,China;Beijing Research Institute of Telemetry,Beijng 100094,China)

机构地区：[1]北京航空航天大学电子信息工程学院,北京100191 [2]北京遥测技术研究所,北京100094

出　　处：《应用科技》2023年第1期118-126,共9页Applied Science and Technology

摘　　要：近年来雷达系统的抗截获能力快速发展,信号体制变得非常复杂,给空间态势感知信号处理带来困难。针对实际场景中极可能出现的多信号时频域交叠难以识别的问题,提出基于时频分析和深度神经网络的脉内特征识别的方法,利用时频分析手段将不同类型雷达信号转换为时频图像,基于深度学习的YOLOv5网络对不同混合交叠的雷达脉冲信号时频图开展研究。结果表明:在不同信噪比下,实现对两调相信号、三调频信号及调相与调频信号3种交叠情况的混合雷达信号脉内调制方式的检测与识别,信号调制体制识别准确率均达到95%以上。本文分析结果可为空间态势感知信号处理提供参考。In recent years,with the rapid development of anti-interception ability of radar system,the signal system has become very complex,which brings difficulties to the signal processing of space situation awareness.Aiming at the problem that it is very difficult to recognize the overlapping of multiple signals in time-frequency domain,which may very likely occur in actual scenes,we propose a method of intra-pulse feature recognition based on time-frequency analysis and depth neural network.Different types of radar signals are converted into time-frequency images by time-frequency analysis,and time-frequency images of different mixed overlapping radar pulse signals are studied by YOLOv5 network based on deep learning.The results show that under different signal-to-noise ratios,the detection and recognition of the mixed radar signals in pulse modulation mode with two-phase modulation signals,three-frequency modulation signals and three overlapping situations of phase modulation and frequency modulation signals is realized,with the recognition accuracy of the signal modulation system more than 95%.The results of this paper can provide reference for signal processing of space situation awareness.

关 键 词：雷达脉冲信号 时频分析 脉内特征识别 态势感知 YOLOv5 深度学习 特征向量 自动分类识别 

分 类 号：TN98[电子电信—信息与通信工程]