系统工程与电子技术 ›› 2022, Vol. 44 ›› Issue (12): 3676-3684.doi: 10.12305/j.issn.1001-506X.2022.12.11

• 传感器与信号处理 • 上一篇    

基于SE-ResNeXt网络的低信噪比LPI雷达辐射源信号识别

徐桂光1, 王旭东1,*, 汪飞1, 胡国兵2, 高涌荇1, 罗泽虎1   

  1. 1. 南京航空航天大学电子信息工程学院/集成电路学院, 江苏 南京 211106
    2. 金陵科技学院电子信息工程学院, 江苏 南京 211169
  • 收稿日期:2021-09-07 出版日期:2022-11-14 发布日期:2022-11-24
  • 通讯作者: 王旭东
  • 作者简介:徐桂光(1996—), 男, 硕士研究生, 主要研究方向为雷达信号处理|王旭东(1979—), 男, 副教授, 主要研究方向为信号检测、参数估计、FPGA设计应用|汪飞(1976—), 男, 副教授, 主要研究方向为飞机射频隐身技术与微弱信号检测|胡国兵(1978—), 男, 教授, 主要研究方向为信号检测、识别与参数估计|高涌荇(1996—), 男, 硕士研究生, 主要研究方向为气象雷达目标识别|罗泽虎(1997—), 男, 硕士研究生, 主要研究方向为气象雷达目标识别
  • 基金资助:
    江苏省高等学校自然科学研究重大项目(20KJA510008);江苏省基础研究计划(自然科学基金)(BK20161104)

LPI radar emitter signals recognition in low SNR based on SE-ResNeXt network

Guiguang XU1, Xudong WANG1,*, Fei WANG1, Guobing HU2, Yongxing GAO1, Zehu LUO1   

  1. 1. College of Electronic and Information Engineering/College of Integrated Circuits, Nanjing University of Aeronautics and Astronautics, Nanjing 211106, China
    2. Jinling Institute of Technology, Nanjing 211169, China
  • Received:2021-09-07 Online:2022-11-14 Published:2022-11-24
  • Contact: Xudong WANG

摘要:

针对低信噪比(signal to noise ratio, SNR)低截获概率(low probability of intercept, LPI)雷达脉内波形识别准确率低的问题,提出一种基于时频分析、压缩激励(squeeze excitation, SE)和ResNeXt网络的雷达辐射源信号识别方法。首先通过Choi-Williams分布(Choi-Williams distribution, CWD)获得雷达时域信号的二维时频图像(time-frequency image, TFI);然后进行TFI预处理降低噪声干扰和频率维的位置分布差异,以适应深度学习网络输入;最后在ResNeXt基础上加入扩张卷积和SE结构提取TFI特征,实现雷达辐射源分类。实验结果表明,SNR低至-8 dB时,该方法对12类常见LPI雷达波形的整体识别准确率依然能达到98.08%。

关键词: 低截获概率雷达波形, 辐射源信号识别, 残差网络, 压缩激励结构, 时频分析

Abstract:

Aiming at the problem of low signal to noise ratio (SNR) and low probability of intercept (LPI) radar pulse waveform recognition accuracy, a radar emitter signal recognition method based on time-frequency analysis, squeeze-excitation (SE) and ResNeXt network is proposed. Firstly, the radar time domain signal is transformed into a two-dimensional time-frequency image (TFI) by Choi-Williams distribution (CWD); then, the TFI pre-processing is used to reduce the noise interference and the difference in frequency dimension location distribution, adapting to deep learning network input; finally, the TFI features are extracted by adding dilated convolution and SE structure on the basis of ResNeXt to achieve radar emitter classification. The experimental results show that when the SNR is as low as -8 dB, the overall recognition accuracy of the method for 12 types of common LPI radar waveforms can still reach 98.08%.

Key words: low probability of intercept (LPI) radar waveform, emitter signal recognition, residual network, squeeze excitation (SE) structure, time-frequency analysis

中图分类号: