基于多尺度双重注意力网络的雷达信号调制识别方法

doi:10.12382/bgxb.2024.0447

摘要/Abstract

摘要：

针对低信噪比环境下复杂多类雷达信号调制识别准确率低的问题,提出一种基于时频融合特征与多尺度双重注意力网络的新方法。通过应用平滑伪Wigner-Ville分布、傅里叶同步压缩变换和基于变分模态分解的希尔伯特黄变换3种时频分析方法,并结合去噪预处理技术,将雷达信号转换为三通道时频特征图,显著增强了特征的稳健性与表达力。设计了一种多尺度双重注意力网络,通过多尺度通道注意力机制实现跨尺度信息融合与噪声抑制,同时利用多尺度空间注意力自适应感知雷达信号的时频结构,并通过门控融合与残差连接技术进一步整合信息。实验结果表明,在信噪比为-10dB的条件下,新方法对12类典型雷达信号调制方式的平均识别率达到98.99%,显示出良好的稳健性。

关键词: 雷达信号, 时频分析, 深度学习, 多尺度空间, 注意力机制

Abstract:

Addressing the issue of low recognition rates for complex multi-class radar signal modulation types under low signal-to-noise ratio,a radar signal modulation recognition method based on time-frequency fusion features and multi-scale dual attention network is proposed.By applying three time-frequency analysis methods,namely,smoothed pseudo Wigner-Ville distribution,Fourier synchrosqueezed transform and Hilbert-Huang transform based on variational modal decomposition,and the denoising preprocessing technique,the radar signals are transformed into three-channel time-frequency feature maps,which significantly enhances the robustness and expressive power of the features.A multi-scale dual attention network is designed to realize denoising preprocessing technique the cross-scale information fusion and noise suppression through the multi-scale channel attention mechanism.The time-frequency structure of radar signals is adaptively perceived by utilizing the multi-scale spatial attention,and the information is further integrated through gated fusion and residual connection.The experimental results show that the method achieves an average recognition rate of 98.99% for 12 types of typical radar signal modulation modes under the condition of a signal-to-noise ratio of -10dB,showing good robustness.

Key words: radar signal, time-frequency analysis, deep learning, multi-scale space, attention mechanism

中图分类号:

TN791

廖仁龙, 罗忠涛, 殷水军, 张伟. 基于多尺度双重注意力网络的雷达信号调制识别方法[J]. 兵工学报, 2025, 46(9): 240447-.

LIAO Renlong, LUO Zhongtao, YIN Shuijun, ZHANG Wei. A Radar Signal Modulation Recognition Method Based on Multi-scale Dual Attention Network[J]. Acta Armamentarii, 2025, 46(9): 240447-.

图/表 13

参考文献 28

[1]	RAO G N, SASTRY C V S, DIVAKAR N. Trends in electronic warfare[J]. IETE Technical Review, 2003, 20(2):139-150.
[2]	WU Y X, GUOCE H, LI W. Waveform design for radar and extended target in the environment of electronic warfare[J]. Journal of Systems Engineering and Electronics, 2018, 29(1):48-57. doi: 10.21629/JSEE.2018.01.05
[3]	WU Z L, HUANG X X, DU M, et al. Intra-pulse recognition of radar signals via bicubic interpolation WVD[J]. IEEE Transactions on Aerospace and Electronic Systems, 2023, 59(6):8668-8680.
[4]	LIAO Y P, JIANG F, WANG J L. Intra-pulse modulation recognition of radar signals based on multi-feature random matching fusion network[J]. The Journal of Supercomputing, 2023, 79(6):6422-6451.
[5]	吴礼洋, 呙鹏程, 刘超, 等. 基于注意力机制增强残差网络的雷达信号调制类型识别[J]. 兵工学报, 2023, 44(8):2310-2318. doi: 10.12382/bgxb.2022.0302
	WU L Y, GUO P C, LIU C, et al. Radar signal modulation type recognition based on attention mechanism enhanced residual networks[J]. Acta Armamentarii, 2023, 44(8):2310-2318. (in Chinese) doi: 10.12382/bgxb.2022.0302
[6]	赵凤, 耿苗苗, 刘汉强, 等. 卷积神经网络与视觉 Transformer 联合驱动的跨层多尺度融合网络高光谱图像分类方法[J]. 电子与信息学报, 2024, 46(5):2237-2248.
	ZHAO F, GENG M M, LIU H Q, et al. Convolutional neural network and vision transformer-driven cross-layer multi-scale fusion network for hyperspectral image classification[J]. Journal of Electronics & Information Technology, 2024, 46(5):2237-2248. (in Chinese)
[7]	LEVANON N, MOZESON E. Radar signals[M]. Hoboken,NJ, US: John Wiley & Sons, 2004.
[8]	QUAN D Y, REN F T, WANG X F, et al. WVD-GAN:a wigner-ville distribution enhancement method based on generative adversarial network[J]. IET Radar,Sonar & Navigation, 2024, 18(6):849-865.
[9]	DRAGOMIRETSKIY K, ZOSSO D. Variational mode decomposition[J]. IEEE Transactions on Signal Processing, 2013, 62(3):531-544.
[10]	HUANG H, LI Y, LIU J Y, et al. LPI waveform recognition using adaptive feature construction and convolutional neural networks[J]. IEEE Aerospace and Electronic Systems Magazine, 2023, 38(4):14-26.
[11]	TANG P. A digitalization-based image edge detection algorithm in intelligent recognition of 5G smart grid[J]. Expert Systems with Applications, 2023, 233:120919.
[12]	LECUN Y, BOTTOU L. Gradient-based learning applied to document recognition[J]. Proceedings of the IEEE, 1998, 86(11):2278-2324.
[13]	TAYE M M. Theoretical understanding of convolutional neural network:concepts,architectures,applications,future directions[J]. Computation, 2023, 11(3):52.
[14]	陈乾, 洪征, 司健鹏. 融合SENet和Transformer的应用层协议识别方法[J]. 计算机科学与探索, 2024, 18(3):805-817. doi: 10.3778/j.issn.1673-9418.2304045
	CHEN Q, HONG Z, SI J P. Application layer protocol recognition incorporating SENet and Transformer[J]. Journal of Frontiers of Computer Science and Technology, 2024, 18(3):805-817. (in Chinese)
[15]	黄鑫, 屈文忠, 肖黎. 基于卷积注意力机制的阀门内漏声发射识别方法[J]. 振动与冲击, 2024, 43(9):105-114.
	HUANG X, QU W Z, XIAO L. Acoustic emission recognition method for valve internal leakage based on convolutional attention mechanism[J]. Journal of Vibration and Shock, 2024, 43(9):105-114. (in Chinese)
[16]	ZHANG J, LI Y, YIN J P. Modulation classification method for frequency modulation signals based on the time-frequency distribution and CNN[J]. IET Radar,Sonar & Navigation, 2018, 12(2):244-249.
[17]	QU Z Y, MAO X J, DENG Z A. Radar signal intra-pulse modulation recognition based on convolutional neural network[J]. IEEE Access, 2018, 6:43874-43884.
[18]	QU Z Y, WANG W, HOU C, et al. Radar signal intra-pulse modulation recognition based on convolutional denoising autoencoder and deep convolutional neural network[J]. IEEE Access, 2019, 7:112339-112347.
[19]	KONG S H, KIM M, HOANG L M, et al. Automatic LPI radar waveform recognition using CNN[J]. IEEE Access, 2018, 6:4207-4219.
[20]	呙鹏程, 吴礼洋. 融合卷积特征与判别字典学习的低截获概率雷达信号识别[J]. 兵工学报, 2019, 40(9):1881-1889. doi: 10.3969/j.issn.1000-1093.2019.09.013
	GUO P C, WU L Y. LPI radar signal recognition with convolution feature and discrimination dictionary learning[J]. Acta Armamentarii, 2019, 40(9):1881-1889. (in Chinese) doi: 10.3969/j.issn.1000-1093.2019.09.013
[21]	SI W J, LUO J J, DENG Z A. Radar signal recognition and localization based on multiscale lightweight attention model[J]. Journal of Sensors, 2022, 2022(1):9970879.
[22]	LI D J, YANG R J, LI X B, et al. Radar signal modulation recognition based on deep joint learning[J]. IEEE Access, 2020, 8:48515-48528.
[23]	LIN A N, MA Z Y, HUANG Z, et al. Unknown radar waveform recognition based on transferred deep learning[J]. IEEE Access, 2020, 8:184793-184807.
[24]	MAO Y J, REN W J, YANG Z P. Radar signal modulation recognition based on sep-ResNet[J]. Sensors, 2021, 21(22):7474.
[25]	徐卓君, 杨雯婷, 杨承志, 等. 雷达脉内调制识别的改进残差神经网络算法[J]. 吉林大学学报(工学版), 2021, 51(4):1454-1460.
	XU Z J, YANG W T, YANG C Z, et al. Improved residual neural network algorithm for radar in-pulse modulation recognition[J]. Journal of Jilin University (Engineering and Technology Edition), 2021, 51(4):1454-1460. (in Chinese)
[26]	徐桂光, 王旭东, 汪飞, 等. 基于 SE-ResNeXt 网络的低信噪比 LPI 雷达辐射源信号识别[J]. 系统工程与电子技术, 2022, 44(12):3676-3684. doi: 10.12305/j.issn.1001-506X.2022.12.11
	XU G G, WANG X D, WANG F, et al. LPI radar emitter signals recognition in low SNR based on SE-ResNeXt network[J]. Systems Engineering and Electronics, 2022, 44(12):3676-3684. (in Chinese) doi: 10.12305/j.issn.1001-506X.2022.12.11
[27]	CHEN K Y, ZHU L Z, CHEN S, et al. Deep residual learning in modulation recognition of radar signals using higher-order spectral distribution[J]. Measurement, 2021, 185:109945.
[28]	LI S N, LI T F, SUN C, et al. Multilayer Grad-CAM:an effective tool towards explainable deep neural networks for intelligent fault diagnosis[J]. Journal of Manufacturing Systems, 2023, 69:20-30.

网络层	神经元数量	卷积核参数	输出大小
输入	N/A	N/A	256×256×3
Conv	32	3×3,1,1	256×256×32
ReLU	N/A	N/A	256×256×32
MP	N/A	2×2,2,0	128×128×32
MSDA	N/A	N/A	128×128×32
Conv	64	3×3,2,1	64×64×64
MSDA	N/A	N/A	64×64×64
Conv	64	3×3,2,1	32×32×64
MSDA	N/A	N/A	32×32×64
Conv	64	3×3,2,1	16×16×64
ReLU	N/A	N/A	16×16×64
MP	N/A	2×2,2,0	8×8×64
Flatten	4096	N/A	1×4096
Linear	12	N/A	1×12

网络层	神经元数量	卷积核参数	输出大小
输入	N/A	N/A	256×256×3
Conv	32	3×3,1,1	256×256×32
ReLU	N/A	N/A	256×256×32
MP	N/A	2×2,2,0	128×128×32
MSDA	N/A	N/A	128×128×32
Conv	64	3×3,2,1	64×64×64
MSDA	N/A	N/A	64×64×64
Conv	64	3×3,2,1	32×32×64
MSDA	N/A	N/A	32×32×64
Conv	64	3×3,2,1	16×16×64
ReLU	N/A	N/A	16×16×64
MP	N/A	2×2,2,0	8×8×64
Flatten	4096	N/A	1×4096
Linear	12	N/A	1×12

调制方式	参数	变化范围/MHz
CW	f_c	45~55
LFM	f_c B_s	45~55 15~20
NLFM	f_c B_s	45~55 15~20
BPSK	f_c 13RandCode	45~55 (0,1)
QPSK	f_c 21RandCode	45~55 (0,1/2,1,3/2)
FSK	f_c Δf 13RandCode	45~55 10~20 (0,1)
4FSK	f_c Δf 16RandCode	45~55 5~15 (0,1,2,3)
Frank	f_c	45~55
P1	f_c	45~55
P2	f_c	45~55
P3	f_c	45~55
P4	f_c	45~55

调制方式	参数	变化范围/MHz
CW	f_c	45~55
LFM	f_c B_s	45~55 15~20
NLFM	f_c B_s	45~55 15~20
BPSK	f_c 13RandCode	45~55 (0,1)
QPSK	f_c 21RandCode	45~55 (0,1/2,1,3/2)
FSK	f_c Δf 13RandCode	45~55 10~20 (0,1)
4FSK	f_c Δf 16RandCode	45~55 5~15 (0,1,2,3)
Frank	f_c	45~55
P1	f_c	45~55
P2	f_c	45~55
P3	f_c	45~55
P4	f_c	45~55

网络模型	识别率/%				参数量/10⁶	浮点运算量/10⁹
网络模型	SNR=-14dB	SNR=-12dB	SNR=-10dB	SNR=-8dB	参数量/10⁶	浮点运算量/10⁹
MSDANet	85.12	96.22	98.99	99.99	0.5412	1.6668
CNN^[16]	34.56	41.12	54.87	60.11	0.0007	0.0035
CNN^[17]	55.23	68.20	75.17	80.53	2.5539	0.0334
CNN^[18]	72.11	87.26	94.76	98.98	0.2373	0.0707
CNN^[19]	52.33	73.83	86.21	95.55	12.3113	0.1509
CNN^[20]	32.71	50.42	74.78	90.45	21.0068	0.0371
LeNet^[21]	51.88	64.63	71.86	79.23	0.0622	0.0007
AlexNet^[22-24]	72.73	85.37	92.38	96.54	14.6060	0.3092
ResNet50^[25-28]	74.57	90.17	97.36	99.56	23.5326	4.1317