基于AlexNet模型的雷达信号调制类型识别

doi:10.13229/j.cnki.jdxbgxb20171056

摘要/Abstract

摘要：

针对现代战场复杂电磁环境，在低信噪比(-6 dB)下传统雷达调制信号采用常规五参数特征的识别方法准确率低的问题,本文采用深度学习中的AlexNet卷积神经网络模型自动提取图像各种特征细节，从而替代手工设计特征的庞大的特征工程以实现信号在低信噪比下的识别。该方法首先利用平滑伪Wigner?Ville时频分析在时频域内生成雷达调制信号的时频图像；然后采用中值和均值滤波结合去噪对时频图像进行预处理；最后使用图像处理器GPU在深度学习架构Tensorflow下搭建AlexNet模型进行训练，对CW、LFM、EQFM、DLFM、BFSK、BPSK以及QPSK 这7种雷达信号进行特征的自动提取和选择，从而实现雷达信号的自动识别。仿真结果表明，在信噪比为-6 dB时，除QPSK信号外其余6种雷达信号的整体识别率均达到90%以上，比采用非深度学习和LeNet5卷积神经网络的识别效果好，从而验证了该方法在低信噪比下雷达信号识别的有效性。

关键词: 信号与信息处理, 平滑伪Wigner?Ville, 中值滤波, 均值滤波, 深度学习

Abstract:

Aiming at the complex electromagnetic environment in modern battlefield, the accuracy of traditional radar modulation signal recognition method based on conventional five?parameter feature is low under low signal?to?noise ratio (-6 dB). In this paper, AlexNet convolution neural network model in deep learning is used to automatically extract various feature details of image, which can replace the huge Feature Engineering designed by hand to realize signal recognition under low signal?to?noise ratio. Firstly, the time?frequency image of radar modulation signal is generated by smoothing pseudo?Wigner?Ville time?frequency analysis in time?frequency domain; then, the time?frequency image is preprocessed by median and mean filtering combined with denoising; finally, the AlexNet model is trained by using image processor GPU under the deep learning framework Tensor flow, and CW, LFM, E are trained. Seven kinds of radar signals, QFM, DLFM, BFSK, BPSK and QPSK, are automatically extracted and selected to realize automatic recognition of radar signals. The simulation results show that the overall recognition rate of the other six radar signals, except QPSK signal, is more than 90% at the signal?to?noise ratio -6 dB, which is better than that of the non?deep learning and LeNet5 convolution neural network, thus verifying the validity of the method for radar signal recognition at low signal?to?noise ratio.

Key words: signal and information processing, smooth pseudo Wigner?Ville, midial filter, mean filter, deep learning

中图分类号:

TN971.1

郭立民,陈鑫,陈涛. 基于AlexNet模型的雷达信号调制类型识别[J]. 吉林大学学报(工学版), 2019, 49(3): 1000-1008.

Li⁃min GUO,Xin CHEN,Tao CHEN. Radar signal modulation type recognition based on AlexNet model[J]. Journal of Jilin University(Engineering and Technology Edition), 2019, 49(3): 1000-1008.

图/表 9

图1

图2

表1

图3

表2

表3

表4

图4

图5

参考文献 22

1	张国柱, 黄可生, 姜文利，等 . 基于信号包络的辐射源细微特征提取方法[J].系统工程与电子技术,2006, 28(6):795⁃797.
	Zhang Guo⁃zhu , Huang Ke⁃sheng , Jiang Wen⁃li ,et al . Emitter feature extract method based on signal envelope[J]. Systems Engineering and Electronics, 2006, 28(6): 795⁃797.
2	Misans P , Terauds M . CW doppler radar based land vehicle speed measurement algorithm using zero crossing and least squares method[C]⫽IEEE Electronics Conference, Tallinn, Estonia, 2012: 161⁃164.
3	王世强，张登福，毕笃彦，等 . 双谱二次特征在雷达信号识别中的应用[J].西安电子科技大学学报,2012,39(2): 127⁃132.
	Wang Shi⁃qiang , Zhang Deng⁃fu , Bi Du⁃yan , et al . Research on recognizing the radar signal using the bispectrum cascade feature[J]. Journal of Xidian University, 2012,39(2): 127⁃132.
4	Teng Xiao⁃yun , Tian Peng⁃wu , Yu Hong⁃yi . Modulation classification based on spectral correlation and SVM [C]⫽IEEE International Conference on Wireless Communications, Dalian, China, 2008: 1⁃4.
5	Li Yi⁃bing , Wang Yan⁃huang , Lin Yun . Recognition of radar signals modulation based on short time fourier transform and reduced fractional Fourier transform[J]. Journal of Information & Computational Science, 2013, 10(16): 5171⁃5178.
6	Gulum T O , Erdogan A Y , Yildirim T ,et al . A parameter extraction technique for FMCW radar signals using Wigner⁃Hough⁃Radon transform[C]⫽IEEE Radar Conference, Atlanta, USA, 2012: 847⁃852.
7	Liu Yong⁃jian , Xiao Peng , Wu Hong⁃chao , et al . LPI radar signal detection based on radial integration of Choi⁃Williams time⁃frequency image[J]. Journal of Systems Engineering and Electronics, 2015, 26(5): 973⁃985.
8	Krizhevsky Alex , Sutskever Ilya , Geoffrey E Hinton . ImageNet classification with deep convolutional neural networks[C]⫽International Conference on Neural Information Processing Systems, Lake Tahoe, USA, 2012: 1097⁃1105.
9	Yu Nai⁃gong , Jiao Pan⁃na , Zheng Yu⁃ling . Handwritten digits recognition base on improved LeNet5[C]⫽IEEE Control & Decision Conference, Qingdao, China, 2015: 4871⁃4875.
10	Xu B Q , Sun L , Xu L , et al . Improvement of the Hilbert method via ESPRIT for detecting rotor fault in induction motors at low slip[J]. IEEE Transactions on Energy Conversion, 2013, 28(1): 225⁃233.
11	付卫红,王璐, 贾坤, 等 . 基于STFT与SPWVD的跳频参数盲估计算法[J]. 华中科技大学学报:自然科学版, 2014, 42(9): 59⁃63.
	Fu Wei⁃hong , Wang Lu , Jia Kun , et al . Blind parameter estimation algorithm for frequency hopping signals based on STFT and SPWVD[J]. Journal of Huazhong University of Science and Technology (Nature Science Edition), 2014, 42(9): 59⁃63.
12	Roy A , Singha J , Manam L , et al . Combination of adaptive vector median filter and weighted mean filter for removal of high⁃density impulse noise from colour images[J]. Iet Image Processing, 2017, 11(6): 352⁃361.
13	Boyat A , Joshi B K . Image denoising using waveket transform and median filtering[C]⫽IEEE Nirma University International Conference on Engineering, Ahmedabad, India, 2014: 1⁃6.
14	Jiang Nan , Wang Luo . Quantum image scaling using nearest neighbor interpolation[J]. Quantum Information Processing, 2015, 14(5): 1559⁃1571.
15	Hu Y C , Su B H , Chen W L , et al . Image zooming for indexed color images based on bilinear interpolation[J]. International Journal of Multimedia and Ubiquitous Engineering, 2012, 7(2): 353⁃358.
16	Zhou D W , Shen X L , Dong W M . Image zooming using directional cubic convolution interpolation[J]. Iet Image Processing, 2012,6(6): 627⁃634.
17	Liu Bin , Zhang Xiao⁃yun , Gao Zhi⁃yong , et al . Weld defect images classification with VGG16⁃Based neural network[C]⫽International Forum on Digital TV and Wireless Multimedia Communications, Shanghai, China, 2017: 215⁃223.
18	Xavier Glorot , Antoine Bordes , Yoshua Bengio . Deep sparse rectifier neural networks[C]⫽International Conference on Artificial Intelligence & Statistics, La Palma, Canary Islands, 2012: 315⁃323.
19	Guo Qiang , Pu⁃long Nan , Zhang Xiao⁃yu ,et al . Recognition of radar emitter signals based on SVD and AF main ridge[J]. Journal of Communications and Networks, 2015, 17(5): 491⁃498.
20	Guo Qiang , Pu⁃long Nan , Wan Jian . Radar signal recognition based on ambiguity function features and cloud model similarity[C]⫽IEEE International Conference on Ultrawideband & Ultrashort Impulse Signals, Odessa, Ukraine, 2016: 128⁃134.
21	陈涛，柳立志，郭立民 . 基于 MWC 压缩采样宽带接收机的雷达信号脉内调制识别[J].电子与信息学报,2018,40(4): 867⁃874.
	Chen Tao , Liu Li⁃zhi , Guo Li⁃min . Intra⁃pulse modulation recognition of radar signals based on MWC compressed sampling wideband receiver[J]. Journal of Electonic & Informatin Technology, 2018, 40(4): 867⁃874.
22	Zhang Ming , Liu Lu⁃tao , Diao Ming . LPI radar waveform recognition based on time⁃frequency distribution [J]. Sensor, 2016,16(10): 1682⁃1688.

Metrics

Viewed

Full text

Abstract

Cited

Shared

Discussed

层数	输入尺寸	卷积核	卷池尺寸	卷池步长	输出特征
1	227×227×3	96	11/3	4/2	27×27×96
2	27×27×256	256	5/3	1/2	13×13×256
3	13×13×384	384	3/0	1/1	13×13×384
4	13×13×384	384	3/0	1/1	13×13*×84
5	13×13×385	256	3/3	1/2	6×6×256

信号类型	载频/MHz	采样频率/MHz	采样点N	调频斜率K
CW	0~30	200	512	-
LFM	10~50	200	512	2~3B/T
EQFM	10~40	200	512	-
DLFM	10~30	200	512	2~3B/T
BFSK	fc1=(10~30) fc2=(10~40)	200	512	-
BPSK	20~30	200	512	-
QPSK	20~30	200	512	-

平台	信噪比 (dB) /时间(s)
平台	-10	-8	-6	-4	-2	0
GPU	1730	1680	1645	1654	1623	1400
CPU	14900	4680	14780	14650	14300	14300

[1]	托乎提努尔,张海龙,王杰,王娜,冶鑫晨,王万琼. 基于图形处理器的高速中值滤波算法[J]. 吉林大学学报(工学版), 2019, 49(3): 979-985.
[2]	李健, 李赫宇, 姚汝婧, 吴林. 基于均值滤波的改进 Canny 算法在核磁共振图像边缘检测中的应用[J]. 吉林大学学报(工学版), 2016, 46(5): 1704-1709.
[3]	李抵非, 田地, 胡雄伟. 基于分布式内存计算的深度学习方法[J]. 吉林大学学报(工学版), 2015, 45(3): 921-925.
[4]	刘光宇，庞永杰. 基于阿尔法均值算法和马氏距离的图像自适应滤波[J]. 吉林大学学报(工学版), 2015, 45(2): 670-674.
[5]	段,孙同景,李振华, 黄长伟, 张光先. 全数字逆变电源IIR Butterworth数字滤波[J]. 吉林大学学报(工学版), 2009, 39(增刊2): 311-0314.