基于SRNN+Attention+CNN的雷达辐射源信号识别方法

doi:10.12305/j.issn.1001-506X.2021.12.11

摘要/Abstract

摘要：

针对低信噪比条件下雷达辐射源信号特征提取困难、识别准确率低的问题, 提出一种基于切片循环神经网络(sliced recurrent neural networks, SRNN)、注意力机制和卷积神经网络(convolutional neural networks, CNN)的雷达辐射源信号识别方法, 并在CNN中引入批归一化层, 进一步提升网络的识别能力。模型以雷达辐射源信号幅度序列作为输入, 自动提取信号特征, 输出识别结果。实验结果表明, SRNN相比于门控循环单元(gated recurrent unit, GRU)训练速度大大提升, 注意力机制和批归一化层能有效提高识别准确率; 在采用8种常见雷达辐射源信号进行的实验中, 所提方法在低信噪比条件下仍有较高的识别准确率。

关键词: 辐射源信号识别, 切片循环神经网络, 卷积神经网络, 注意力机制, 批归一化, 时间序列

Abstract:

Aiming at solving the problem of difficulty in extracting features of radar emitter signals and low recognition accuracy under the condition of low signal to noise ratio, a radar emitter signal based on sliced recurrent neural networks (SRNN), attention mechanism and convolutional neural networks (CNN) is proposed. Batch normalization layer is introduced into CNN to further improve the recognition ability of the network.Taking the amplitude sequence of radar emitter signal as input, the signal characteristic is extracted automatically and the recognition result of radar emitter signal is output. Compared with gated recurrent unit (GRU), the experimental results show that the training speed of SRNN is greatly improved, and the attention mechanism and batch normalization layer can effectively improve the recognition accuracy. In the experiments with eight common radar emitter signals, the proposed method still has a high recognition accuracy under the condition of low signal to noise ratio.

Key words: emitter signal recognition, sliced recurrent neural networks (SRNN), convolutional neural networks (CNN), attention mechanism, batch normalization, time series

中图分类号:

TN971.1

高诗飏, 董会旭, 田润澜, 张歆东. 基于SRNN+Attention+CNN的雷达辐射源信号识别方法[J]. 系统工程与电子技术, 2021, 43(12): 3502-3509.

Shiyang GAO, Huixu DONG, Runlan TIAN, Xindong ZHANG. Radar emitter signal recognition method based on SRNN+Attention+CNN[J]. Systems Engineering and Electronics, 2021, 43(12): 3502-3509.

图/表 15

图1

图2

图3

图4

表1

表2

表3

表4

表5

图5

表6

图6

图7

表7

图8

参考文献 31

1	赵国庆. 雷达对抗原理[M]. 西安: 西安电子科技大学出版社, 2012.
	ZHAO G Q . Principle of radar countermeasure[M]. Xi'an: Xidian University Press, 2012.
2	DONG J , WU G W , YANG T T , et al. Battlefield situation awareness and networking based on agent distributed computing[J]. Physical Communication, 2019, 33, 178- 186. doi: 10.1016/j.phycom.2019.01.002
3	ZAKUAN F R A , ZAMZURI H , RAHMAN M A A , et al. Threat assessment algorithm for active blind spot assist system using short range radar sensor[J]. Asian Research Publishing Network (ARPN) Journal of Engineering and Applied Sciences, 2017, 12 (4): 4270- 4275.
4	PARK B, AHN J M. Intra-pulse modulation recognition using pulse description words and complex waveforms[C]//Proc. of the IEEE International Conference on Information and Communication Technology Convergence, 2017: 555-560.
5	ZHOU J L , ZHENG S L , YU X B , et al. Low probability of intercept communication based on structured radio beams using machine learning[J]. IEEE Access, 2019, 7, 169946- 169952. doi: 10.1109/ACCESS.2019.2955509
6	KAWALEC A, OWCZAREK R. Radar emitter recognition using intrapulse data[C]//Proc. of the IEEE 15th International Conference on Microwaves, Radar and Wireless Communications, 2004, 2: 435-438.
7	李昆, 朱卫纲. 基于机器学习的雷达辐射源识别综述[J]. 电子测量技术, 2019, 42 (18): 69- 75.
	LI K , ZHU W G . Overview of radar emitter recognition based on machine learning[J]. Electronic Measurement Technology, 2019, 42 (18): 69- 75.
8	SINGH S P, KUMAR A, DARBARI H, et al. Machine translation using deep learning: an overview[C]//Proc. of the IEEE International Conference on Computer, Communications and Electronics, 2017: 162-167.
9	XIONG C M, ZHONG V, SOCHER R. Dynamic coattention networks for question answering[EB/OL]. [2020-10-20]. https://arxiv.org/abs/1611.01604.
10	LI S T , SONG W W , FANG L Y , et al. Deep learning for hyperspectral image classification: an overview[J]. IEEE Trans.on Geoscience and Remote Sensing, 2019, 57 (9): 6690- 6709. doi: 10.1109/TGRS.2019.2907932
11	ZHANG Z , GEIGER J , POHJALAINEN J , et al. Deep learning for environmentally robust speech recognition: an overview of recent developments[J]. ACM Trans.on Intelligent Systems and Technology, 2018, 9 (5): 1- 28.
12	LI J Y, ZHAO R, HU H, et al. Improving RNN transducer modeling for end-to-end speech recognition[C]//Proc. of the IEEE Automatic Speech Recognition and Understanding Workshop, 2019: 114-121.
13	WANG D, GONG J B, SONG Y X. W-RNN: news text classification based on a weighted RNN[EB/OL]. [2020-10-20]. https://arxiv.org/abs/1909.13077.
14	WAN J , YU X , GUO Q . LPI radar waveform recognition based on CNN and TPOT[J]. Symmetry, 2019, 11 (5): 725- 739. doi: 10.3390/sym11050725
15	ZHANG M , DIAO M , GAO L Q , et al. Neural networks for radar waveform recognition[J]. Symmetry, 2017, 9 (5): 75- 94. doi: 10.3390/sym9050075
16	GUO Q , YU X , RUAN G Q . LPI radar waveform recognition based on deep convolutional neural network transfer learning[J]. Symmetry, 2019, 11 (4): 540- 553. doi: 10.3390/sym11040540
17	MEDSKER L R , JAIN L C . Recurrent neural networks: design and applications[M]. Florida: CRC Press, 1999.
18	陈森森. 基于RNN的雷达辐射源分类识别算法研究[D]. 西安: 西安电子科技大学, 2019.
	CHEN S S. Research on classification and recognition algorithm of radar signal based on RNN[D]. Xi'an: Xidian University, 2019.
19	YU Z P, LIU G S. Sliced recurrent neural networks[EB/OL]. [2020-10-20]. https://arxiv.org/abs/1807.02291.
20	VASWANI A, SHAZEER N, PARMAR N, et al. Attention is all you need[C]//Proc. of the Advances in Neural Information Processing Systems, 2017: 5998-6008.
21	HUANG C W, NARAYANAN S S. Deep convolutional recurrent neural network with attention mechanism for robust speech emotion recognition[C]//Proc. of the IEEE International Conference on Multimedia and Expo, 2017: 583-588.
22	BAHULEYAN H, MOU L, VECHTOMOVA O, et al. Varia-tional attention for sequence-to-sequence models[EB/OL]. [2020-10-20]. https://arxiv.org/abs/1712.08207.
23	苟先太, 吴南方. 基于GRU-Attention神经网络的空中群组态势识别方法[J]. 计算机与现代化, 2019, (10): 11- 16, 33.
	GOU X T , WU N F . Air group situation recognition method based on GRU-attention neural networks[J]. Computer and Modernization, 2019, (10): 11- 16, 33.
24	IOFFE S, SZEGEDY C. Batch normalization: accelerating deep network training by reducing internal covariate shift[EB/OL]. [2020-10-20]. https://arxiv.org/abs/1502.03167.
25	BJORCK N, GOMES C P, SELMAN B, et al. Understanding batch normalization[EB/OL]. [2020-10-20]. https://arxiv.org/abs/1806.02375.
26	SANTURKAR S, TSIPRAS D, ⅡYAS A, et al. How does batch normalization help optimization[C]//Proc. of the 32nd International Conference on Neural Information Processing Systems, 2018: 2483-2493.
27	HASANI M, KHOTANLOU H. An empirical study on position of the batch normalization layer in convolutional neural networks[C]//Proc. of the 5th Iranian Conference on Signal Processing and Intelligent Systems, 2019.
28	CAIN L, CLARK J, PAULS E, et al. Convolutional neural networks for radar emitter classification[C]//Proc. of the IEEE 8th Annual Computing and Communication Workshop and Conference, 2018: 79-83.
29	GU J X , WANG Z H , KUEN J , et al. Recent advances in convolutional neural networks[J]. Pattern Recognition, 2018, 77 (5): 354- 377.
30	NIEPERT M, AHMED M, KUTZKOV K. Learning convolutional neural networks for graphs[C]//Proc. of the International Conference on Machine Learning, 2016: 2014-2023.
31	HAQ I U, GONDAL I, VAMPLEW P, et al. Categorical features transformation with compact one-hot encoder for fraud detection in distributed environment[C]//Proc. of the Australa-sian Conference on Data Mining, 2018: 69-80.

模型	参数	值
SRNN	循环单元	GRU
	输出单元数	32
	激活函数	tanh
Conv1d	卷积核数量	32
	时域窗长度	4
	激活函数	ReLU
MaxPooling	池化窗口大小	2
Dense	激活函数	softmax
Dense	输出维度	8

实验环境	配置
CPU	Intel Core i7-10875H
显卡	NVIDIA GeForce RTX2060
内存	16 GB
操作系统	Windows 10 2004
学习框架	Keras 2.3.1
底层	TensorFlow 2.1.0
平台	PyCharm 2020.1.3×64
编译器	Python 3.7

信号类型	载频kHz	采样频率kHz	参数	取值
BPSK	1~1.2	7	巴克码位数	{7, 11, 13}
Costas	1~1.2	{15, 17}	频率	{[4 7 1 6 5 2 3], [2 6 3 8 7 5 1]}
FMCW	1~1.2	7	调制周期/ms	{50, 25, 35}
FMCW	1~1.2	7	调制带宽/kHz	{0.25, 0.35, 0.5}
Frank	1~1.2	7	码位数	{36, 64}
P1~ P4	1~1.2	7	码位数	{36, 64}

模型	测试集损失	识别准确率	训练轮数	训练总时间/s	单轮平均训练时间/s	测试时间/s
GRU	0.463 4	0.832 6	16.0	7 904	494.00	32.0
SRNN	0.130 6	0.954 2	38.0	1 228	32.32	3.0
SRNN+CNN	0.129 4	0.956 2	29.6	1 073	36.25	3.0

模型	测试集损失	识别准确率	训练轮数	训练总时间/s	测试时间/s
SRNN+CNN	0.129 4	0.956 2	29.6	1 073	3.0
SRNN+Attention+CNN	0.127 9	0.957 7	29.0	987	3.0