基于元学习的畸变雷达电磁信号识别

颜康; 金炜东; 黄颖坤; 葛鹏; 朱劼昊

doi:10.11999/JEIT210190

基于元学习的畸变雷达电磁信号识别

doi: 10.11999/JEIT210190

1.
西南交通大学电气工程学院成都 610097
2.
电子信息控制重点实验室成都 610036

基金项目: 电子信息控制重点实验室项目开放基金(6142105190312)

详细信息

作者简介:
颜康：男，1988年生，博士生，研究方向为雷达信号处理、机器学习

金炜东：男，1959年生，教授，博士生导师，研究方向为智能信息处理、系统仿真与优化方法

黄颖坤：男，1989年生，博士生，研究方向为雷达信号处理、机器学习

葛鹏：男，1986年生，讲师，研究方向为雷达信号处理、电子对抗

朱劼昊：男，1983年生，高级工程师，研究方向为雷达自动目标识别和目标特性分析

通讯作者:
金炜东　wdjin@home.swjtu.edu.cn

中图分类号: TN95
计量
- 文章访问数: 749
- HTML全文浏览量: 385
- PDF下载量: 102
- 被引次数: 0
出版历程
- 收稿日期: 2021-03-05
- 修回日期: 2021-11-03
- 录用日期: 2021-11-03
- 网络出版日期: 2021-12-19
- 刊出日期: 2022-04-18

Distorted Radar Electromagnetic Signal Recognition Based on Meta-learning

1.
College of Electrical Engineering , Southwest Jiaotong University, Chengdu 610097, China
2.
The Science and Technology on Electronic Information Control Laboratory, Chengdu 610036, China

Funds: The Science and Technology on Electronic Information Control Laboratory (6142105190312)

摘要

摘要: 畸变雷达电磁信号会严重影响雷达侦察装备的探测性能。如何有效地识别畸变信号类型对侦察系统的精确感知具有重要现实意义。针对畸变雷达信号往往存在样本稀缺的问题，该文提出一种基于模型无关元学习的残差网络(MAML-ResNet)。算法首先利用正常雷达信号样本训练元学习器，然后在畸变信号样本进行精调，最后在仅有少量畸变信号样本下，实现畸变信号的识别。实验结果表明该算法有效地提高了在小样本数据下畸变信号的识别准确率。
- 畸变雷达信号识别 /
- 深度学习 /
- 小样本学习 /
- 元学习
Abstract: Distorted radar electromagnetic signals will seriously affect the detection performance of radar reconnaissance equipment. How to identify effectively the type of distorted signal has important practical significance for the accurate perception of radar systems. For distorted radar signals, there is often a problem of sample scarcity. A Residual Network based on Model-Agnostic Meta-Learning (MAML-ResNet) is proposed. The algorithm first uses normal radar signal samples to train the meta-learner, then the meta-learner is fine-tuned in the distorted signal samples. Finally, the distorted signal is recognized with only a small number of distorted signal samples. Experimental results show that the recognition accuracy of distorted signals under small sample data is effectively improved.
- Distorted radar signal recognition /
- Deep learning /
- Few-shot learning /
- Meta learning

HTML全文

图 1 基于模型无关元学习的残差网络结构图

下载: 全尺寸图片幻灯片

图 2 5种畸变雷达信号时频图像

下载: 全尺寸图片幻灯片

图 3 残差模块

下载: 全尺寸图片幻灯片

表 1 算法MAML流程

输入：创建若干个任务$ p(\mathcal{T}) $，步长超参数$ \alpha $,$ \beta $
输出：模型参数${\boldsymbol{\theta}}$
(1) 随机初始化${\boldsymbol{ \theta}}$
(2) 循环1：
(3) 　　随机对若干个任务采样$ {\mathcal{T}_i} \sim p(\mathcal{T}) $
(4) 　　循环2（对所有的任务$ {\mathcal{T}_i} $）：
(5) 　　　　根据$ K $个样本计算梯度${\nabla _{\boldsymbol{\theta}} }{\mathcal{L}_{ {\mathcal{T}_i} } }\left( { {f_{\boldsymbol{\theta}} } } \right)$
(6) 　　　　利用梯度下降法更新参数：${ {\boldsymbol{\theta} } '_i}{\text{ = } }{\boldsymbol{\theta}} - \alpha {\nabla _{\boldsymbol{\theta}} }{\mathcal{L}_{ {\mathcal{T}_i} } }\left( { {f_{\boldsymbol{\theta}} } } \right)$
(7) 　　循环2结束
(8) 　　更新参数$\theta \leftarrow {\boldsymbol{\theta} } - \beta {\nabla _{\boldsymbol{\theta} } }\sum {_{ {\mathcal{T}_i} \sim p(\mathcal{T})} } {\mathcal{L}_{ {\mathcal{T}_i} } }\left( { {f_{ { { {\boldsymbol{\theta } }'}_i} } } } \right)$
(9) 循环1结束

下载: 导出CSV

表 2 ResNet结构参数

层数	输出尺寸/像素	对应卷积层
Conv1	$ {\text{112}} \times {\text{112}} $
Conv2_x	$ {\text{56}} \times {\text{56}} $	$ \left[\begin{array}{cc}\text{3}\times \text{3}，& \text{64}\\ \text{3}\times \text{3}，& \text{64}\end{array}\right]\times \text{2} $
Conv3_x	$ {\text{28}} \times {\text{28}} $	$ \left[\begin{array}{cc}\text{3}\times \text{3}，& \text{128}\\ \text{3}\times \text{3}，& \text{128}\end{array}\right]\times \text{2} $
Conv4_x	$ {\text{14}} \times {\text{14}} $	$ \left[\begin{array}{cc}\text{3}\times \text{3}，& \text{256}\\ \text{3}\times \text{3}，& \text{256}\end{array}\right]\times \text{2} $
Conv5_x	$ {\text{7}} \times {\text{7}} $	$ \left[\begin{array}{cc}\text{3}\times \text{3}，& \text{512}\\ \text{3}\times \text{3}，& \text{512}\end{array}\right]\times \text{2} $
全连接层	$ {\text{1}} \times {\text{1}} $

下载: 导出CSV

表 3 人工特征提取方法特征集合

分析域	人工特征	参考文献	特征维数	描述
时域	复杂度特征，包括多尺度信息熵、信息维数、盒维数和稀疏度	[7]	每种特征1维，总共4维	雷达辐射源信号的变化趋势
频率域	复杂度特征	[7]	每种特征1维，总共4维	雷达辐射源信号的变化趋势
频率域	多尺度信息熵特征	[10]	10维	频率域序列的时间变化趋势
时频域	小波脊频级联特征	[3]	7维	时-频曲线上的7个不同统计特征
其他	模糊函数特征	[8]	3维	表征信号模糊函数主脊能量分布走向、分布重心及其惯性半径的3个特征参数

下载: 导出CSV

表 4 畸变雷达信号识别结果(%)

方法	信噪比0～6 dB		信噪比7～13 dB		信噪比14～20 dB
方法	5 way 5 shot	5 way 10 shot	5 way 5 shot	5 way 10 shot	5 way 5 shot	5 way 10 shot
Decision Tree	60.9±1.5	70.1±0.2	68.0±0.3	78.6±0.3	69.3±0.7	79.5±0.2
K-NN	63.2±0.3	74.6±0.2	77.8±0.2	84.2±0.2	79.0±0.2	85.3±0.1
SVM	58.1±0.7	73.3±0.2	73.6±0.4	85.0±0.1	73.8±0.7	86.8±0.1
ResNet	38.8±1.5	51.1±1.9	59.8±0.4	67.8±0.3	52.3±0.5	56.6±0.3
MAML-ResNet	76.8±0.2	82.3±0.1	86.6±0.1	87.1±0.1	89.9±0.1	94.3±0.1

下载: 导出CSV

表 5 高信噪比元学习器下畸变雷达信号识别结果(%)

方法	信噪比0～6 dB		信噪比7～13 dB
方法	5 way 5 shot	5 way 10 shot	5 way 5 shot	5 way 10 shot
MAML-ResNet	72.7±0.2	78.9±0.1	83.7±0.1	86.3±0.1

下载: 导出CSV

表 6 低信噪比元学习器下畸变雷达信号识别结果(%)

方法	信噪比7～13 dB		信噪比14～20 dB
方法	5 way 5 shot	5 way 10 shot	5 way 5 shot	5 way 10 shot
MAML-ResNet	84.6±0.1	88.3±0.1	90.9±0.1	94.4±0.1

下载: 导出CSV

参考文献(22)

[1]	袁兴鹏, 张金全, 杨立永. 海面雷达信号环境多路径效应建模与仿真技术研究[J]. 舰船电子对抗, 2013, 36(6): 68–72. doi: 10.16426/j.cnki.jcdzdk.2013.06.008 YUAN Xingpeng, ZHANG Jinquan, and Yang Liyong. Research into modeling and simulation technologies of multi-path effect of sea surface radar signal environment[J]. Shipboard Electronic Countermeasure, 2013, 36(6): 68–72. doi: 10.16426/j.cnki.jcdzdk.2013.06.008
[2]	焦培南, 张忠治. 雷达环境与电波传播特性[M]. 北京: 电子工业出版社, 2007: 101–105. JIAO Peinan and ZHANG Zhongzhi. Radar Environment and Radio Wave Propagation Characteristics[M]. Beijing: Publishing House of Electronics Industry, 2007: 101–105.
[3]	余志斌, 金炜东, 陈春霞. 基于小波脊频级联特征的雷达辐射源信号识别[J]. 西南交通大学学报, 2010, 45(2): 290–295. doi: 10.3969/j.issn.0258-2724.2010.02.022 YU Zhibin, JIN Weidong, and CHEN Chunxia. Radar emitter signal recognition based on WRFCCF[J]. Journal of Southwest Jiaotong University, 2010, 45(2): 290–295. doi: 10.3969/j.issn.0258-2724.2010.02.022
[4]	黄颖坤, 金炜东, 余志斌, 等. 基于深度学习和集成学习的辐射源信号识别[J]. 系统工程与电子技术, 2018, 40(11): 2420–2425. doi: 10.3969/j.issn.1001-506X.2018.11.05 HUANG Yingkun, JIN Weidong, YU Zhibin, et al. Radar emitter signal recognition based on deep learning and ensemble learning[J]. Systems Engineering and Electronics, 2018, 40(11): 2420–2425. doi: 10.3969/j.issn.1001-506X.2018.11.05
[5]	周志文, 黄高明, 高俊, 等. 一种深度学习的雷达辐射源识别算法[J]. 西安电子科技大学学报:自然科学版, 2017, 44(3): 77–82. doi: 10.3969/j.issn.1001-2400.2017.03.014 ZHOU Zhiwen, HUANG Gaoming, GAO Jun, et al. Radar emitter identification algorithm based on deep learning[J]. Journal of Xidian University, 2017, 44(3): 77–82. doi: 10.3969/j.issn.1001-2400.2017.03.014
[6]	韩俊, 何明浩, 朱振波, 等. 基于复杂度特征的未知雷达辐射源信号分选[J]. 电子与信息学报, 2009, 31(11): 2552–2556. doi: 10.3724/SP.J.1146.2008.01505 HAN Jun, HE Minghao, ZHU Zhenbo, et al. Sorting unknown radar emitter signal based on the complexity characteristics[J]. Journal of Electronics &Information Technology, 2009, 31(11): 2552–2556. doi: 10.3724/SP.J.1146.2008.01505
[7]	普运伟, 金炜东, 朱明, 等. 雷达辐射源信号模糊函数主脊切面特征提取方法[J]. 红外与毫米波学报, 2008, 27(2): 133–137. doi: 10.3321/j.issn:1001-9014.2008.02.012 PU Yunwei, JIN Weidong, ZHU Ming, et al. Extracting the main ridge slice characteristics of ambiguity function for radar emitter signals[J]. Journal of Infrared and Millimeter Waves, 2008, 27(2): 133–137. doi: 10.3321/j.issn:1001-9014.2008.02.012
[8]	陈韬伟, 金炜东. 雷达辐射源信号符号化脉内特征提取方法[J]. 数据采集与处理, 2008, 23(5): 521–526. doi: 10.3969/j.issn.1004-9037.2008.05.004 CHEN Taowei and JIN Weidong. Intra-pulse feature extraction of radar emitter signals based on symbolization method[J]. Journal of Data Acquisition &Processing, 2008, 23(5): 521–526. doi: 10.3969/j.issn.1004-9037.2008.05.004
[9]	黄颖坤, 金炜东, 葛鹏, 等. 基于多尺度信息熵的雷达辐射源信号识别[J]. 电子与信息学报, 2019, 41(5): 1084–1091. doi: 10.11999/JEIT180535 HUANG Yingkun, JIN Weidong, GE Peng, et al. Radar emitter signal identification based on multi-scale information entropy[J]. Journal of Electronics &Information Technology, 2019, 41(5): 1084–1091. doi: 10.11999/JEIT180535
[10]	张瑜, 李玲玲. 多径条件下雷达到达角的估算及仿真[J]. 电波科学学报, 2004, 19(2): 215–218. doi: 10.3969/j.issn.1005-0388.2004.02.018 ZHANG Yu and LI Lingling. Radar arrived angle estimation and simulation under multi-path condition[J]. Chinese Journal of Radio Science, 2004, 19(2): 215–218. doi: 10.3969/j.issn.1005-0388.2004.02.018
[11]	LECUN Y, BENGIO Y, and HINTON G. Deep learning[J]. Nature, 2015, 521(7553): 436–444. doi: 10.1038/nature14539
[12]	ZHANG Wei, CUI Xiaodong, FINKLER U, et al. Distributed deep learning strategies for automatic speech recognition[C]. ICASSP 2019 – 2019 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), Brighton, UK, 2019: 5706–5710.
[13]	HE Kaiming, ZHANG Xiangyu, REN Shaoqing, et al. Deep residual learning for image recognition[C]. 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Las Vegas, USA, 2016: 770–778.
[14]	WANG Chao, WANG Jian, and ZHANG Xudong. Automatic radar waveform recognition based on time-frequency analysis and convolutional neural network[C]. 2017 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), New Orleans, USA, 2017: 2437–2441.
[15]	周东青, 王玉冰, 王星, 等. 基于深度限制波尔兹曼机的辐射源信号识别[J]. 国防科技大学学报, 2016, 38(6): 136–141. doi: 10.11887/j.cn.201606022 ZHOU Dongqing, WANG Yubing, WANG Xing, et al. Radar emitter signal recognition based on deep restricted Boltzmann machine[J]. Journal of National University of Defense Technology, 2016, 38(6): 136–141. doi: 10.11887/j.cn.201606022
[16]	WANG Xuebao, HUANG Gaoming, ZHOU Zhiwen, et al. Radar emitter recognition based on the short time Fourier transform and convolutional neural networks[C]. 2017 10th International Congress on Image and Signal Processing, BioMedical Engineering and Informatics (CISP-BMEI), Shanghai, China, 2017: 1–5.
[17]	WANG Yaqing, YAO Quanming, KWOK J T, et al. Generalizing from a few examples: A survey on few-shot learning[J]. ACM Computing Surveys, 2020, 53(3): 63. doi: 10.1145/3386252
[18]	JIANG Wen, HUANG Kai, GENG Jie, et al. Multi-scale metric learning for few-shot learning[J]. IEEE Transactions on Circuits and Systems for Video Technology, 2021, 31(3): 1091–1102. doi: 10.1109/TCSVT.2020.2995754
[19]	DAS D and LEE C S G. A two-stage approach to few-shot learning for image recognition[J]. IEEE Transactions on Image Processing, 2019, 29: 3336–3350. doi: 10.1109/TIP.2019.2959254
[20]	LIU Weide, ZHANG Chi, LIN Guosheng, et al. CRNet: Cross-reference networks for few-shot segmentation[C]. 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), Seattle, USA, 2020: 4164–4172.
[21]	FINN C, ABBEEL P, and LEVINE S. Model-agnostic meta-learning for fast adaptation of deep networks[C]. Proceedings of the 34th International Conference on Machine Learning, Sydney, Australia, 2017: 1126–1135.
[22]	HOSPEDALES T M, ANTONIOU A, MICAELLI P, et al. Meta-learning in neural networks: A survey[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, To be published.