小样本下雷达复合干扰半监督迁移学习识别方法

doi:10.11959/j.issn.1000-0801.2023182

电信科学 ›› 2023, Vol. 39 ›› Issue (10): 15-28.doi: 10.11959/j.issn.1000-0801.2023182

• 研究与开发 • 上一篇

小样本下雷达复合干扰半监督迁移学习识别方法

王金强¹, 孙闽红¹, 唐向宏¹, 仇兆炀¹, 曾德国²

¹ 杭州电子科技大学通信工程学院，浙江杭州 310018
² 中国航天科工集团八五一一研究所，江苏南京 210007

修回日期:2023-09-06 出版日期:2023-10-01 发布日期:2023-10-01
作者简介:王金强（1998- ），男，杭州电子科技大学硕士生，主要研究方向为雷达信号处理及抗干扰
孙闽红（1974- ），男，博士，杭州电子科技大学教授、博士生导师，主要研究方向为信号处理、信息对抗、雷达系统与成像技术
唐向宏（1962- ），男，博士，杭州电子科技大学教授、博士生导师，主要研究方向为图像处理与传输、通信与信息系统、信息安全
仇兆炀（1987- ），男，博士，杭州电子科技大学讲师、硕士生导师，主要研究方向为信号采样、复杂信号处理和宽带接收
曾德国（1985- ），男，博士，中国航天科工集团八五一一研究所研究员，主要研究方向为信号识别与新型接收机结构
基金资助:
国家自然科学基金资助项目(61901149);国防特色学科发展项目(JCKY2019415D002)

A semi-supervised transfer learning recognition method for radar compound jamming under small samples

Jinqiang WANG¹, Minhong SUN¹, Xianghong TANG¹, Zhaoyang QIU¹, Deguo ZENG²

¹ School of Communication Engineering, Hangzhou Dianzi University, Hangzhou 310018, China
² China Aerospace Science and Technology Group 8511 Research Institute, Nanjing 210007, China

Revised:2023-09-06 Online:2023-10-01 Published:2023-10-01
Supported by:
The National Natural Science Foundation of China(61901149);The National Defense Characteristic Discipline Development Project(JCKY2019415D002)

摘要/Abstract

摘要：

针对雷达复合干扰信号种类越来越多以及训练样本过少难以令深度学习模型达到最优状态的问题，提出一种在小样本下雷达复合干扰半监督迁移学习识别的方法，通过未带标签样本来解决标签样本难以获取而导致网络训练精度不高的问题，将在单一干扰数据集预训练后得到的特征提取器和分类器迁移到小规模复合干扰数据集上，并利用权重印记和半监督学习对模型进行微调，通过所提出的最近邻相关性损失（nearest neighbor correlation loss，NNCL）来优化模型参数。实验结果表明，在干噪比为10 dB、新类复合干扰信号带标签样本仅有5个时，模型可达93.20%的识别准确率。

关键词: 雷达抗干扰, 复合干扰识别, 迁移学习, 权重印记, 半监督学习

Abstract:

Aiming at the problem that more and more kinds of radar compound jamming signals and too few training samples were difficult to make the deep learning model reach the optimal state, a semi-supervised transfer learning recognition method for radar compound jamming under small samples was proposed, which solved the problem of low network training accuracy caused by the difficulty in obtaining labeled samples through unlabeled samples.The feature extractor and classifier obtained after pre-training of single jamming data set were transferred to small-scale compound jamming data set, and the model was fine-tuning by using weight imprinting and semi-supervised learning.The model parameters were optimized by the proposed nearest neighbor correlation loss nearest neighbor correlation loss (NNCL).The experimental results show that the recognition accuracy of the model can reach 93.20% when the jamming-to-noise ratio is 10 dB and there are only 5 labeled samples of the new class of compound jamming signals.

Key words: radar anti-jamming, compound jamming recognition, transfer learning, weight imprinting, semi-supervised learning

中图分类号:

TN975

王金强, 孙闽红, 唐向宏, 仇兆炀, 曾德国. 小样本下雷达复合干扰半监督迁移学习识别方法[J]. 电信科学, 2023, 39(10): 15-28.

Jinqiang WANG, Minhong SUN, Xianghong TANG, Zhaoyang QIU, Deguo ZENG. A semi-supervised transfer learning recognition method for radar compound jamming under small samples[J]. Telecommunications Science, 2023, 39(10): 15-28.

图/表 14

图1

图2

图3

图4

表1

WRN-28-10网络模型"

层名	输出维度	ResNet块类型B(3,3)
conv1	32×32	$[3 \times 3 ， 16]$
conv2	32×32	$[\begin{matrix} 3 \times 3 ， 16 \times 10 \\ 3 \times 3 ， 16 \times 10 \end{matrix}] \times 3$
conv3	16×16	$[\begin{matrix} 3 \times 3 ， 32 \times 10 \\ 3 \times 3 ， 32 \times 10 \end{matrix}] \times 4$
conv4	8×8	$[\begin{matrix} 3 \times 3 ， 64 \times 10 \\ 3 \times 3 ， 64 \times 10 \end{matrix}] \times 6$
avg-pool	1×1	$[8 \times 8]$

表1

表2

表3

图5

图6

表4

表5

表6

图7

表7

参考文献 23

[1]	ADAMY D . EW101:电子战基础[M]. 王燕,朱松,译. 北京: 电子工业出版社, 2009.
	ADAMY D . EW101:fundamentals of electronic warfare[M]. Translated by WANG Y,ZHU S. Beijing: Publishing House of Electronics Industry, 2009.
[2]	周红平, 王子伟, 郭忠义 . 雷达有源干扰识别算法综述[J]. 数据采集与处理, 2022,37(1): 1-20.
	ZHOU H P , WANG Z W , GUO Z Y . Overview on recognition algorithms of radar active jamming[J]. Journal of Data Acquisition and Processing, 2022,37(1): 1-20.
[3]	FU R R . Compound jamming signal recognition based on neural networks[C]// Proceedings of the Sixth International Conference on Instrumentation ＆ Measurement,Computer,Communication and Control (IMCCC). Piscataway:IEEE Press, 2016: 737-740.
[4]	ZHU M , LI Y , PAN Z ,et al. Automatic modulation recognition of compound signals using a deep multi-label classifier:a case study with radar jamming signals[J]. Signal Processing, 2019,169(9): 107393.
[5]	ZHANG J X , LIANG Z N , ZHOU C ,et al. Radar compound jamming cognition based on a deep object detection network[J]. IEEE Transactions on Aerospace and Electronic Systems, 2023,59(3): 3251-3263.
[6]	李浩 . 基于多标签学习的雷达复合干扰识别研究[D]. 成都:电子科技大学, 2021.
	LI H . Radar compound jamming recognition based on multi-label learning[D]. Chengdu:University of Electronic Science and Technology, 2021.
[7]	QU Q , WEI S , LIU S ,et al. JRNet:jamming recognition networks for radar compound suppression jamming signals[J]. IEEE Transactions on Vehicular Technology, 2020,69(12): 15035-15045.
[8]	庞伊琼, 许华, 张悦 ,等. 基于迁移元学习的调制识别算法[J]. 兵工学报, 2022.doi:10.12382/bgxb.2022.0583.
	PANG Y Q , XU H , ZHANG Y ,et al. Modulation recognition algorithm based on transfer meta-learning[J]. Acta Armamentarii, 2022.doi:10.12382/bgxb.2022.0583.
[9]	季鼎承, 蒋亦樟, 王士同 . 基于域与样例平衡的多源迁移学习方法[J]. 电子学报, 2019,47(3): 692-699.
	JI D C , JIANG Y Z , WANG S T . Multi-source transfer learning method by balancing both the do?mains and instances[J]. Acta Electronica Sinica, 2019,47(3): 692-699.
[10]	SHAO G,CHENY , WEI Y . Convolutional neural network-based radar jamming signal classification with sufficient and limited samples[J]. IEEE Access, 2020(8): 80588-80598.
[11]	LYU Q Z , QUAN Y H , FENG W ,et al. Radar Deception jamming recognition based on weighted ensemble CNN with transfer learning[J]. IEEE Transactions on Geoscience and Remote Sensing, 2022(60): 1-11.
[12]	CHUNG H , LEE J . Iterative semi-supervised learning using softmax probability[J]. Materials ＆ Continua, 2022,72(3): 5607-5628.
[13]	BAE J , LEE M , KIM S B . Safe semi-supervised learning using a bayesian neural network[J]. Information Sciences, 2022,6(12): 453-464.
[14]	PASSALIS N , IOSIFIDIS A , GABBOUJ M ,et al. Hypersphere-based weight imprinting for few-shot learning on embedded devices[J]. IEEE Transactions on Neural Networks and Learning Systems, 2021,32(2): 925-930.
[15]	ZHUANG F , QI Z , DUAN K ,et al. A comprehensive survey on transfer learning[J]. Proceedings of the IEEE, 2020,109(1): 43-76.
[16]	BERTHELOT D , CARLINI N , GOODFELLOW I ,et al. MixMatch:a holistic approach to semi-supervised learning[J]. Advances in Neural Information Processing Systems, 2019,32(454): 5050-5060.
[17]	LI P , ZHAO G , XU X . Coarse-to-fine few-shot classification with deep metric learning[J]. Information Sciences, 2022(610): 592-604.
[18]	QIAO S Y , LIU C X , SHEN W ,et al. Few-shot image recognition by predicting parameters from activations[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition(CVPR). Piscataway:IEEE Press, 2018: 7229-7238.
[19]	HE K , ZHANG X , REN S ,et al. Deep residual learning for image recognition[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition(CVPR). Piscataway:IEEE Press, 2016: 770-778.
[20]	HAFIZ A M , BHAT R A , HASSABALLAH M . Image classification using convolutional neural network tree ensembles[J]. Multimedia Tools and Applications, 2023,82(5): 6867-6884.
[21]	LI T W , LEE G C . Performance analysis of fine-tune transferred deep learning[C]// Proceedings of the 2021 3rd Eurasia Conference on Communication and Engineering. Piscataway:IEEE Press, 2021: 315-319.
[22]	HIGUCHI Y , MORITZ N , ROUX J L ,et al. Momentum pseudo-labeling:semi-supervised ASR with continuously improving pseudo-labels[J]. IEEE Journal of Selected Topics in Signal Processing, 2022,16(6): 1424-1438.
[23]	CHEN W Y , LIU Y C , KIRA Z ,et al. A closer look at few-shot classification[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway:IEEE Press, 2019:1904.04232.

信号参数	参数值
调制类型	LFM
脉宽/μs	20
采样频率/MHz	40
RD时延/μs	2～4
SMSP采样倍数	3～5
AMN/FMN带宽/MHz	20～40
载频/MHz	20
信号带宽/MHz	10
脉冲重复间隔/μs	100
欺骗速度/(m·s^-1)	100～300
COMB谱个数	4～6
FMN调频斜率/(MHz·V^-1)	2

未带标签样本数	5-way 1-shot	5-way 5-shot
5	74.21%±0.77%	82.15%±0.86%
10	77.82%±0.75%	85.69%±0.78%
15	80.12%±0.61%	88.51%±0.72%
20	81.56%±0.46%	88.74%±0.53%
30	81.32%±0.39%	89.06%±0.32%

复合干扰类型		JNR比值
复合干扰类型	1:1	1:5	1:10
RD+AMN	92.62%±0.51%	88.63%±0.61%	81.55%±0.32%
RD+FMN	92.07%±0.16%	88.33%±0.56%	81.16%±0.82%
SMSP+AMN	93.14%±0.35%	89.20%±0.24%	82.87%±0.48%
SMSP+FMN	93.67%±0.64%	89.52%±0.37%	82.64%±0.94%
R-VD+AMN	92.85%±0.72%	87.79%±0.58%	80.86%±0.16%
R-VD+FMN	92.04%±0.64%	87.06%±0.35%	81.21%±0.42%

方法	5-way 1-shot	5-way 5-shot
WI^[14]	69.84%±0.86%	78.92%±0.62%
MixMatch^[16]	64.62%±0.77%	81.65%±0.74%
MixMatch+NNCL	71.68%±0.79%	88.36%±0.48%
WI+MixMatch	77.63%±0.91%	89.86%±0.42%
WI+MixMatch+NNCL	84.31%±0.68%	93.20%±0.55%

算法模型	特征提取网络	5-way 1-shot	5-way 5-shot
Fine-tuning^[21]	ResNet 34	54.41%±0.89%	63.09%±0.72%
	WRN-28-10	58.63%±0.75%	66.21%±0.69%
Pseudo-Labeling^[22]	ResNet 34	73.33%±0.96%	84.30%±0.91%
	WRN-28-10	75.16%±0.63%	85.21%±0.80%
Baseline++^[23]	ResNet 34	67.54%±0.82%	74.68%±0.89%
	WRN-28-10	69.66%±0.74%	75.54%±0.52%
本文算法	ResNet 34	82.75%±0.71%	91.02%±0.65%
	WRN-28-10	84.31%±0.68%	93.20%±0.55%

小样本下雷达复合干扰半监督迁移学习识别方法

A semi-supervised transfer learning recognition method for radar compound jamming under small samples

在线阅读

PDF下载

可视化

摘要/Abstract

引用本文

使用本文

图/表 14

参考文献 23

相关文章 6

Metrics

推荐阅读 0

算法模型	单次训练平均时间/s	单次识别平均时间/s
Fine-tuning^[21]	0.330	0.208
Pseudo-Labeling^[22]	0.289	0.126
Baseline++^[23]	0.355	0.061
本文算法	0.409	0.059

[1]	刘昊双, 张永, 曹莹波. 基于K-means聚类的子结构相关适配迁移学习方法[J]. 电信科学, 2023, 39(3): 124-134.
[2]	祝文军, 王思宁, 高晓欣, 郑倩. 基于知识流和迁移学习的负荷预测[J]. 电信科学, 2022, 38(5): 114-123.
[3]	余立, 李哲, 高飞, 袁向阳, 杨永. 改进自训练模型在业务质差用户识别中的应用[J]. 电信科学, 2021, 37(10): 136-142.
[4]	朱应钊. 异构迁移学习研究综述[J]. 电信科学, 2020, 36(3): 100-110.
[5]	刘参,尚俊娜,李蕊江,岳克强. 基于迁移学习的室内动态环境定位算法[J]. 电信科学, 2018, 34(8): 98-108.
[6]	陈康,向勇,喻超. 大数据时代机器学习的新趋势[J]. 电信科学, 2012, 28(12): 77-85.