基于PSO-CNN的LPI雷达波形识别算法

doi:10.12305/j.issn.1001-506X.2021.12.17

摘要/Abstract

摘要：

低截获概率(low probability of intercept, LPI)雷达作为一种具有强抗干扰能力及低截获特性的新型雷达, 对其精准高效识别已成为雷达对抗一方波形识别的难点。针对该方向主流分类器卷积神经网络(convolution neural network, CNN)的结构智能寻优问题, 提出一种基于粒子群优化(particle swarm optimization, PSO)算法-CNN的波形识别算法。该算法利用PSO的寻优特性, 可实现较大范围内自动搭建不定层数、不定层类别及层内参数的CNN结构并进行迭代寻优; 采用识别精度及网络复杂度相结合的衡量指标, 可根据需求调整两者比重以实现对精度与轻量性的选择。该算法获取的CNN结构实现了比9种经典CNN结构更好的LPI雷达波形识别效果, 同时避免了波形识别时人工选定CNN超参数缺乏智能性、客观性的问题, 提高了选用CNN结构的适配性及高效性。

关键词: 低截获概率雷达, 波形识别, 卷积神经网络优化, 粒子群优化算法

Abstract:

Low probability of intercept (LPI) radar is a new type of radar with strong anti-jamming ability and low interception characteristics. Accurate and efficient recognition of it has become a difficult problem in waveform recognition of radar confrontation. Aiming at the structural intelligent optimization problem of the mainstream classifier convolution neural network (CNN) in this direction, a waveform recognition algorithm based on particle swarm optimization (PSO)-CNN is proposed. The algorithm uses the optimization characteristics of PSO to automatically build CNN structure with indefinite layers, layer types and layer parameters in a large range, evaluate and iteratively optimize it. The algorithm uses a combination of recognition accuracy and network complexity to measure indicators, the proportion of the two can be adjusted according to requirements to achieve the choice of accuracy and lightness. The CNN structure obtained by the algorithm achieves better LPI radar waveform recognition effect than nine classic CNN structures. And the algorithm avoids the lack of intelligence and objectivity of artificially selecting CNN hyper parameters during radar waveform recognition, and improves the adaptability and efficiency of the selecting CNN structure.

Key words: low probability of intercept (LPI) radar, waveform recognition, convolution neural network (CNN) optimization, particle swarm optimization (PSO) algorithm

中图分类号:

TN973

赵帅, 刘松涛, 汪慧阳. 基于PSO-CNN的LPI雷达波形识别算法[J]. 系统工程与电子技术, 2021, 43(12): 3552-3563.

Shuai ZHAO, Songtao LIU, Huiyang WANG. LPI radar waveform recognition algorithm based on PSO-CNN[J]. Systems Engineering and Electronics, 2021, 43(12): 3552-3563.

图/表 20

图1

表

图2

图3

表1

表2

图4

图5

表3

表4

表5

图6

图7

图8

表6

图9

表7

图10

参考文献 21

1	刘松涛, 雷震烁, 温镇铭, 等. 认知电子战研究进展[J]. 探测与控制学报, 2020, 42 (5): 1- 15.
	LIU S T , LEI Z S , WEN Z M , et al. A development review on cognitive electronic warfare[J]. Journal of Detection & Control, 2020, 42 (5): 1- 15.
2	ZHANG M , DIAO M , GUO L M . Convolutional neural networks for automatic cognitive radio waveform recognition[J]. IEEE Access, 2017, 5, 11074- 11082. doi: 10.1109/ACCESS.2017.2716191
3	KONG S H , KIM M , HOANG L M . Automatic LPI radar waveform recognition using CNN[J]. IEEE Access, 2018, 6, 4207- 4219. doi: 10.1109/ACCESS.2017.2788942
4	吴琼. 基于改进CNN的雷达辐射源识别算法研究[D]. 西安: 西安电子科技大学, 2019.
	WU Q. Research on radar emitter recognition algorithm based on improved CNN[D]. Xi'an: Xidian University, 2019.
5	杨洁, 弋佳东. 改进GA优化BP神经网络的雷达信号识别[J]. 西安邮电大学学报, 2019, 24 (6): 11- 15.
	YANG J , YI J D . Radar signal recognition based on BP neural network optimized by improved GA[J]. Journal of Xi'an University of Posts and Telecommunications, 2019, 24 (6): 11- 15.
6	赵毅. 基于改进量子粒子群卷积神经网络雷达信号识别[D]. 镇江: 江苏科技大学, 2019.
	ZHAO Y. Radar signal recognition based on improved quantum particle swarm optimization convolution neural network[D]. Zhenjiang: Jiangsu University of Science and Technology, 2019.
7	高华照. 基于深度卷积神经网络的图像识别算法研究[D]. 长春: 吉林大学, 2018.
	GAO H Z. Research on image recognition algorithm based on deep convolutional neural networks[D]. Changchun: Jilin University, 2018.
8	王金哲, 王泽儒, 王红梅. 基于PSO算法与Dropout的改进CNN算法[J]. 长春工业大学学报, 2019, 40 (1): 26- 30.
	WANG J Z , WANG Z R , WANG H M . An improved CNN algorithm based on PSO and dropout[J]. Journal of Changchun University of Technology, 2019, 40 (1): 26- 30.
9	白燕燕. 基于粒子群算法优化卷积神经网络结构[D]. 呼和浩特: 内蒙古大学, 2019.
	BAI Y Y. Optimization of convolutional neural network structure based on particle swarm optimization[D]. Huhhot: Inner Mongolia University, 2019.
10	JUNIOR F E F , YEN G G . Particle swarm optimization of deep neural networks architectures for image classification[J]. Swarm and Evolutionary Computation, 2019, 49, 62- 74. doi: 10.1016/j.swevo.2019.05.010
11	LECUN Y , BOTTOU L , BENGIO Y , et al. Gradient-based learning applied to document recognition[J]. Proceedings of the IEEE, 1998, 86 (11): 2278- 2324. doi: 10.1109/5.726791
12	KENNEDY J, EBERHART R. Particle swarm optimization[C]// Proc. of the International Conference on Neural Networks, 2002: 1942-1948.
13	SAHU A , PANIGRAHI S K , PATTNAIK S . Fast convergence particle swarm optimization for functions optimization[J]. Procedia Technology, 2012, 4, 319- 324. doi: 10.1016/j.protcy.2012.05.048
14	WANG H , YEN G G . Adaptive multiobjective particle swarm optimization based on parallel cell coordinate system[J]. IEEE Trans.on Evolutionary Computation, 2015, 19 (1): 1- 18. doi: 10.1109/TEVC.2013.2296151
15	FAN X L , KOU K I , LIU M S . Quaternion Wigner-Ville distribution associated with the linear canonical transforms[J]. Signal Processing, 2017, 130, 129- 141. doi: 10.1016/j.sigpro.2016.06.018
16	HARTONO D , HALIM D , WYN R G , et al. Gear fault diagnosis using an improved reassigned smoothed pseudo Wigner-Ville distribution[J]. Cogent Engineering, 2018, 5 (1): 1436928. doi: 10.1080/23311916.2018.1436928
17	YANN L C , BOTTOU L , BENGIO Y , et al. Gradient-based learning applied to document recognition[J]. Proceedings of the IEEE, 1998, 86 (11): 2278- 2324. doi: 10.1109/5.726791
18	KRIZHEVSKY A, SUTSKEVER I, HINTON G. ImageNet classification with deep convolutional neural networks[C]//Proc. of the Conference on Neural Information Processing Systems, 2012.
19	SIMONYAN K, ZISSERMAN A. Very deep convolutional networks for large-scale image recognition[EB/OL]. [2021-01-20]. https://arxiv.org/pdf/1409.1556.pdf.
20	SZEGEDY C, LIU W, JIA Y Q, et al. Going deeper with convolutions[C]//Proc. of the IEEE Conference on Computer Vision and Pattern Recognition, 2015.
21	HE K M, ZHANG X Y, REN S Q, et al. Deep residual learning for image recognition[C]//Proc. of the IEEE Conference on Computer Vision and Pattern Recognition, 2016: 770-778.

雷达波形	参数	取值范围
	f_c	U(f_s/6, f_s/5)
LFM	B	U(f_s/20, f_s/16)
	N	U[500, 1 920]
	L_Bc	{7, 11, 13}
Barker	f_c	U(f_s/6, f_s/5)
	N_cc	U[11,14]
	f_c	U(f_s/6, f_s/5)
Frank & P1	N_cc	{4, 5, 6}
	M	{6, 7, 8}
	f_c	U(f_s/6, f_s/5)
P2	N_cc	{4, 5, 6}
	M	{6, 8}
	f_c	U(f_s/6, f_s/5)
P3 & P4	N_cc	{4, 5, 6}
	m	{36, 49, 64}

信道参数	取值
采样频率/MHz	F_s=100
多径衰减/s	{0, 1.8, 3.4}/F_s
平均路径增益/dB	{0, -2, -10}
K-factor	4
最大多普勒频移/Hz	4
AWGN/dB	SNR=[-6∶6∶30]

设置项	训练集	测试集
SNR范围/dB	[-6, 30]	[-10, 10]
SNR间隔/dB	6	2
单类每SNR下图片数量	1 000/7	200
图片尺寸	80×80	80×80
总量	7 000	15 400

类别	参数项	取值
粒子群相关参数	粒子群迭代次数	10
	粒子群大小	10
	C_tsh	0.5
CNN粒子初始化相关参数	卷积核尺寸范围	[1×1, 7×7]
	卷积层输出通道范围	[3,256]
	全连接层神经元个数范围	[1,300]
	网络层数范围	[3,15]
	卷积层生成概率	0.6
	池化层生成概率	0.3
	全连接层生成概率	0.1
CNN粒子训练相关参数	训练批量	32
	粒子评估前训练epoch	1
	最优粒子测试前训练epoch	10
	Dropout	0.5
	层间是否采用正则化	是

类别	1	2	3	4	5	平均
acc-10	0.978 9	0.968 5	0.990 1	0.983 5	0.989 2	0.982 0
参数数量	320 771	2 757 398	1 469 133	3 420 182	3 477 612	2 289 019
训练时间/s	39 244	98 634	42 687	81 278	67 783	65 925