先验引导的轻量化噪声鲁棒雷达复合干扰多标签识别网络

赵云飞; 刘梅; 高勋章; 刘烁炜

doi:10.12000/JR25272

先验引导的轻量化噪声鲁棒雷达复合干扰多标签识别网络

DOI: 10.12000/JR25272 CSTR: 32380.14.JR25272

国防科技大学电子科学学院长沙 410073

基金项目: 国家自然科学基金 (61921001)，湖南省研究生创新项目 (CX20250020)

详细信息

作者简介:
赵云飞，硕士生，主要研究方向为雷达信号处理

刘　梅，博士生，主要研究方向为智能感知与目标识别

高勋章，研究员，主要研究方向为雷达目标特性与雷达目标识别

刘烁炜，副研究员，主要研究方向为雷达目标识别

通讯作者:
高勋章 gaoxunzhang@nudt.edu.cn

责任主编：李亚超 Corresponding Editor: LI Yachao

中图分类号: TN958
计量
- 文章访问数:
- HTML全文浏览量:
- PDF下载量:
- 被引次数: 0
出版历程
- 收稿日期: 2025-12-17
- 修回日期: 2026-03-05

Prior-guided Lightweight Noise-robust Multi-label Radar Composite Interference Recognition Network

College of Electronic Science and Technology, National University of Defense Technology, Changsha 410073, China

Funds: The National Natural Science Foundation of China (61921001), The Hunan Provincial Graduate Student Research Innovation Program (CX20250020)

More Information

Corresponding author: GAO Xunzhang, gaoxunzhang@nudt.edu.cn

摘要

摘要: 干扰辨识是提升雷达目标识别系统抗干扰能力的关键环节。相较于单一干扰，复合干扰由于结构复杂、组合形式灵活，其辨识难度显著增加。然而，现有辨识方法大多为数据驱动模型，未引入干扰的先验信息，导致模型在复杂场景下存在性能瓶颈，且可解释性不足；同时，大多方法缺乏噪声抑制机制，在复杂环境下易出现噪声过拟合问题。为此，该文基于不同干扰在时域上的对称性数学先验，提出了一种干扰先验引导的噪声鲁棒雷达复合干扰多标签识别网络。首先，通过由粗到细去噪的策略对复合干扰进行噪声抑制，并优化干扰的先验结构，缓解辨识过程中由噪声引起的过拟合问题；其次，计算自相关对称性得分量化干扰先验强度，并通过对称编码器，将先验强度映射为门控机制，引导干扰特征融合并时序建模；最后，将噪声强度与时序特征联合嵌入到识别网络中，进一步增强模型在不同信噪比条件下的鲁棒性。实验表明，在低信噪比条件下，所提方法对15种间歇采样转发复合干扰和30种复杂复合干扰的平均识别准确率均超过90%，在性能优于最优对比模型的同时，显著降低了模型参数量。
- 雷达复合干扰 /
- 多标签分类 /
- 特征融合 /
- 干扰识别 /
- 小波分解 /
- 噪声抑制
Abstract: Interference identification is a critical component in enhancing the anti-jamming capability of radar target recognition systems. Compared with single-type interference, composite interference poses substantially greater identification challenges due to its structural complexity and flexible combination patterns. However, most existing identification methods are purely data-driven and fail to incorporate interference prior knowledge, resulting in performance bottlenecks in complex scenarios and limited interpretability. Moreover, many approaches lack effective noise suppression mechanisms and are prone to noise overfitting under low Signal-to-Noise Ratio (SNR) conditions. To address these limitations, this study proposes a prior-guided, noise-robust multi-label recognition network for radar composite interference, which exploits time-domain symmetry priors in different interference types. First, a coarse-to-fine denoising strategy is employed to suppress noise while preserving and enhancing their prior structural characteristics, thereby alleviating noise-induced overfitting during the recognition process. Second, an autocorrelation-based symmetry score is introduced to quantify the strength of the interference prior. The score is then mapped into a gating mechanism via a symmetry encoder to guide interference feature fusion and temporal representation learning. Finally, noise intensity and temporal features are jointly embedded into the recognition network, further enhancing the robustness of the proposed method across varying SNR conditions. Experimental results demonstrate that, under low-SNR conditions, the proposed method achieves average recognition accuracies exceeding 90% for 15 types of intermittent sampling repeater composite interference and 30 types of complex composite interference. Moreover, the proposed approach outperforms the strongest baseline model while significantly reducing model parameters.
- Radar composite jamming /
- Multi-label classification /
- Feature fusion /
- Jamming signal recognition /
- Wavelet decomposition /
- Noise Suppression

HTML全文

图 1 8种干扰模型脉压结果图

Figure 1. Eight types of interference model pulse compression results

下载: 全尺寸图片幻灯片

图 2 小波细节系数特征图

Figure 2. Wavelet detail coefficient feature map

下载: 全尺寸图片幻灯片

图 3 本文网络结构示意图

Figure 3. Schematic of the proposed network

下载: 全尺寸图片幻灯片

图 4 噪声抑制与特征优化模块结构图

Figure 4. Architecture of the noise suppression and feature optimization module

下载: 全尺寸图片幻灯片

图 5 数据集分布情况

Figure 5. Dataset distribution

下载: 全尺寸图片幻灯片

图 6 各模型噪声鲁棒性能图

Figure 6. Evaluation of noise robustness for different models

下载: 全尺寸图片幻灯片

图 7 本文与对比网络混淆矩阵结果图

Figure 7. Confusion matrices of our method and comparative networks

下载: 全尺寸图片幻灯片

图 8 复杂干扰数据集识别准确率雷达图

Figure 8. Complex interference dataset classification accuracy radar chart

下载: 全尺寸图片幻灯片

图 9 噪声抑制与特征优化效果图

Figure 9. Visualization of noise suppression and feature optimization effects

下载: 全尺寸图片幻灯片

表 1 噪声感知分类器结构

Table 1. Noise-aware classifier architecture

层	操作	输出形状
输入	F_dec	(B,24+2)
1	线性层 26→64,ReLU	(B,64)
2	Dropout (p=0.3)	(B,64)
3	线性层 64→32, ReLU	(B,32)
4	线性层 32→分类数	(B，分类数)

下载: 导出CSV

表 2 3个数据集的参数配置

Table 2. Parameter configurations for three datasets

数据集	参数	取值范围
1	总样本数	30000
	雷达带宽	10 MHz
	雷达脉宽	20 μs
	雷达波长	0.1911 m
	干扰幅度	0.8～1.2
	SNR	–8～20 dB
	采样占空比	0.16～0.875
	速度	300～1200 m/s
	距离	1500～4900 m
	转发次数	2～6 次
	噪声长度(SNISRJ-1)	0.5～1.5 μs
	噪声长度(SNISRJ-2)	10～30 μs
2	总样本数	40000
	SNR	0～20 dB
	速度	100～600 m/s
	距离	1500～5000 m
	干扰幅度	0.8～1.2
	采样占空比(ISRJ)	0.3～0.5
	谱线数(COMB)	5～15
	频率步长(COMB)	0.5～2.0 MHz
	切片数(SMSP)	2～4
	频率步长(SMSP)	1～3 MHz
	采样宽度(C&I)	0.5～1.5 μs
	采样周期(C&I)	1～6 μs
3	采样占空比(ISRJ)	0.1～0.5
	谱线数(COMB)	5～20
	切片数(SMSP)	2～8
	采样宽度(C&I)	0.8～1.8 μs
	采样周期(C&I)	1.6～7.2 μs

下载: 导出CSV

表 3 各类间歇采样转发干扰在不同网络模型下的识别性能对比

Table 3. Recognition performance comparison of various ISRJ types across different network models

评估类别	干扰类型	模型
评估类别	干扰类型	YOLOv11	FRFTNet	MLACM	MLCNN	SNV2-EHA	CMNet	Proposed
单类干扰识别率(%)	1	90.55	99.50	98.59	99.85	99.90	98.39	99.90
	2	64.95	94.57	95.54	98.15	99.51	98.48	98.21
	3	75.54	91.24	90.62	99.49	96.78	95.54	99.38
	4	84.43	95.14	96.43	95.33	97.62	96.95	98.38
复合干扰识别率(%)	1+2	37.08	82.83	90.33	94.17	96.58	92.50	96.00
	2+4	19.61	49.22	68.83	63.88	78.64	82.04	78.93
	2+3	40.00	68.81	49.25	66.87	81.64	81.27	84.18
	2+3+4	20.15	41.24	34.95	57.68	70.93	71.29	78.66
	1+4	25.26	63.42	56.49	69.39	87.63	80.00	90.96
	1+3	42.41	76.98	63.97	87.16	89.74	90.43	89.22
	1+3+4	12.72	19.47	10.44	52.18	54.95	67.91	69.51
	3+4	34.22	65.62	27.50	68.36	76.56	72.03	83.20
	1+2+4	2.99	30.10	53.97	64.61	74.22	74.07	81.18
	1+2+3	34.26	47.35	78.70	85.37	80.62	84.20	87.78
	1+2+3+4	89.67	89.20	92.30	91.32	93.28	92.83	97.13
总体效果(%)	平均精确率	48.30	65.42	72.38	78.32	81.07	81.37	85.52
	平均召回率	42.12	63.42	63.00	74.61	79.91	79.87	83.29
	平均F1分数	42.78	63.38	63.55	75.98	80.27	80.47	84.24
	子集准确率	54.98	71.58	72.41	82.17	86.60	86.78	90.54
模型复杂度	浮点运算次(10⁹)	0.183	0.166	0.652	0.005	0.037	4.596	0.160
	参数量(10⁰)	1.20	0.66	33.78	0.04	0.65	92.66	0.16
	测试时间(s)	0.78	1.14	0.90	0.61	0.64	3.81	0.74
注：加粗表示最优指标；干扰1-4分别对应ISDJ, ISPJ, SNISRJ-1, SNISRJ-2；为方便布局，表格对模型名称进行了简化。

下载: 导出CSV

表 4 不同模型在数据集2与数据集3上的性能对比

Table 4. Performance comparison on dataset 2 and dataset 3

序号	模型	数据集2准确率 (%)	数据集3(失配)准确率 (%)
1	Proposed	90.66	77.14
2	CrossModality	89.60	76.12
3	ShuffleNetV2-EHA	86.40	75.29
4	MLACM	81.79	70.06
5	MLCNN	78.21	65.68
6	FRFTNet	39.12	37.50
7	YOLOv11	27.36	25.63
注：加粗表示最优指标。

下载: 导出CSV

表 5 模块消融结果

Table 5. Ablation results of modules

实验	噪声抑制	噪声感知	先验引导	小波分支	准确率 (%)
1	√	√	√	√	90.62
2		√	√	√	87.21
3	√		√	√	86.71
4	√	√		√	85.68
5	√	√	√		86.44
6		√		√	85.37
7			√	√	86.68
8		√		√	85.87
9				√	84.43
注：加粗表示最优指标。

下载: 导出CSV

参考文献(33)

[1]	崔国龙, 余显祥, 魏文强, 等. 认知智能雷达抗干扰技术综述与展望[J]. 雷达学报, 2022, 11(6): 974–1002. doi: 10.12000/JR22191. CUI Guolong, YU Xianxiang, WEI Wenqiang, et al. An overview of antijamming methods and future works on cognitive intelligent radar[J]. Journal of Radars, 2022, 11(6): 974–1002. doi: 10.12000/JR22191.
[2]	刘振, 隋金坪, 魏玺章, 等. 雷达有源干扰识别技术研究现状与发展趋势[J]. 信号处理, 2017, 33(12): 1593–1601. doi: 10.16798/j.issn.1003-0530.2017.12.010. LIU Zhen, SUI Jinping, WEI Xizhang, et al. The development and prospect of the radar active jamming recognition[J]. Journal of Signal Processing, 2017, 33(12): 1593–1601. doi: 10.16798/j.issn.1003-0530.2017.12.010.
[3]	周胜文, 沙明辉, 胡小春. 基于梳状谱调制和间歇采样重复转发的复合干扰[J]. 系统工程与电子技术, 2021, 43(12): 3495–3501. doi: 10.12305/j.issn.1001-506X.2021.12.10. ZHOU Shengwen, SHA Minghui, and HU Xiaochun. Composite jamming based on comb spectrum modulation and interrupted sampling repetitive repeater[J]. Systems Engineering and Electronics, 2021, 43(12): 3495–3501. doi: 10.12305/j.issn.1001-506X.2021.12.10.
[4]	JIE Xiao, WEI Xizhang, and SUN Jia. Interrupted-sampling multi-strategy forwarding jamming with amplitude constraints based on simultaneous transmission and reception technology[J]. Digital Signal Processing, 2024, 147: 104416. doi: 10.1016/j.dsp.2024.104416.
[5]	周阳, 毕大平, 沈爱国, 等. 基于运动调制的SAR-GMTI间歇采样遮蔽干扰方法[J]. 雷达学报, 2017, 6(4): 359–367. doi: 10.12000/JR16075. ZHOU Yang, BI Daping, SHEN Aiguo, et al. Intermittent sampling repeater shading jamming method based on motion modulation for SAR-GMTI[J]. Journal of Radars, 2017, 6(4): 359–367. doi: 10.12000/JR16075.
[6]	郭文杰, 吴振华, 曹宜策, 等. 多域浅层特征引导下雷达有源干扰多模态对比识别方法[J]. 雷达学报(中英文), 2024, 13(5): 1004–1018. doi: 10.12000/JR24129. GUO Wenjie, WU Zhenhua, CAO Yice, et al. Multidomain characteristic-guided multimodal contrastive recognition method for active radar jamming[J]. Journal of Radars, 2024, 13(5): 1004–1018. doi: 10.12000/JR24129.
[7]	WANG Zan, GUO Zhengwei, SHU Gaofeng, et al. Radar jamming recognition: Models, methods, and prospects[J]. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2025, 18: 3315–3343. doi: 10.1109/JSTARS.2024.3522951.
[8]	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. doi: 10.16337/j.1004-9037.2022.01.001. ZHOU Hongping, WANG Ziwei, and GUO Zhongyi. Overview on recognition algorithms of radar active jamming[J]. Journal of Data Acquisition and Processing, 2022, 37(1): 1–20. doi: 10.16337/j.1004-9037.2022.01.001.
[9]	HAO Guocheng, BU Laite, LU Mengyuan, et al. Radar jamming signal recognition algorithm based on multi-feature fusion[J]. Digital Signal Processing, 2025, 158: 104950. doi: 10.1016/j.dsp.2024.104950.
[10]	SHI Yuxin, LU Xinjin, NIU Yingtao, et al. Efficient jamming identification in wireless communication: Using small sample data driven naive Bayes classifier[J]. IEEE Wireless Communications Letters, 2021, 10(7): 1375–1379. doi: 10.1109/LWC.2021.3064843.
[11]	刘一兵, 罗强, 胡然, 等. 基于多维小波特征的有源雷达欺骗干扰识别[J]. 现代防御技术, 2024, 52(3): 120–127. doi: 10.3969/j.issn.1009-086x.2024.03.015. LIU Yibing, LUO Qiang, HU Ran, et al. Active radar deception jamming recognition based on multi-dimensional wavelet features[J]. Modern Defence Technology, 2024, 52(3): 120–127. doi: 10.3969/j.issn.1009-086x.2024.03.015.
[12]	王佳祥, 孟进, 李伟, 等. YOLO-S³: 一种轻量化的雷达复合干扰识别网络[J]. 雷达学报(中英文), 待出版. doi: 10.12000/JR25080. WANG Jiaxiang, MENG Jin, LI Wei, et al. YOLO-S³: A lightweight network for radar composite jamming signal recognition[J]. Journal of Radars, in press. doi: 10.12000/JR25080.
[13]	ZHAO Qingyuan, LIU Yang, CAI Linjie, et al. Research on electronic jamming identification based on CNN[C]. 2019 IEEE International Conference on Signal, Information and Data Processing, Chongqing, China, 2019: 1–5. doi: 10.1109/ICSIDP47821.2019.9172911.
[14]	LIN Junjie and FAN Xiaolei. Radar active jamming recognition based on recurrence plot and convolutional neural network[C]. 2021 IEEE 4th Advanced Information Management, Communicates, Electronic and Automation Control Conference, Chongqing, China, 2021: 1511–1515. doi: 10.1109/IMCEC51613.2021.9481990.
[15]	WU Yaojun, DUAN Lining, YANG Liaoming, et al. Fine-grained recognition and suppression of ISRJ based on UNet-A[J]. IEEE Geoscience and Remote Sensing Letters, 2024, 21: 3508805. doi: 10.1109/LGRS.2024.3448611.
[16]	ZHOU Hongping, WANG Ziwei, WU Ruowu, et al. Jamming recognition algorithm based on variational mode decomposition[J]. IEEE Sensors Journal, 2023, 23(15): 17341–17349. doi: 10.1109/JSEN.2023.3283397.
[17]	ZHOU Hongping, WANG Lei, and GUO Zhongyi. Recognition of radar compound jamming based on convolutional neural network[J]. IEEE Transactions on Aerospace and Electronic Systems, 2023, 59(6): 7380–7394. doi: 10.1109/TAES.2023.3288080.
[18]	ZHU Mengtao, LI Yunjie, PAN Zesi, et al. Automatic modulation recognition of compound signals using a deep multi-label classifier: A case study with radar jamming signals[J]. Signal Processing, 2020, 169: 107393. doi: 10.1016/j.sigpro.2019.107393.
[19]	QU Qizhe, WEI Shunjun, LIU Shan, et al. JRNet: Jamming recognition networks for radar compound suppression jamming signals[J]. IEEE Transactions on Vehicular Technology, 2020, 69(12): 15035–15045. doi: 10.1109/TVT.2020.3032197.
[20]	LV Qinzhe, FAN Hanxin, LIU Junliang, et al. Multilabel deep learning-based lightweight radar compound jamming recognition method[J]. IEEE Transactions on Instrumentation and Measurement, 2024, 73: 2521115. doi: 10.1109/TIM.2024.3400337.
[21]	LIU Mei, GAO Xunzhang, QIU Xiangfeng, et al. Deep Gaussian hidden Markov network for robust HRRP sequence modeling and target recognition[J]. IEEE Transactions on Aerospace and Electronic Systems, 2025, 61(6): 19068–19083. doi: 10.1109/TAES.2025.3615896.
[22]	LIU Mei, GAO Xunzhang, and ZHANG Zhiwei. Noise robust HRRP sequence recognition based on a deep unfolded go decomposition network[J]. Signal Processing, 2025, 230: 109876. doi: 10.1016/j.sigpro.2024.109876.
[23]	ZHANG Lingyun, TAN Hui, and WANG Zhili. Interference response prediction of receiver based on wavelet transform and a temporal convolution network[J]. Electronics, 2024, 13(1): 162. doi: 10.3390/electronics13010162.
[24]	YU Yifei, LI Yuanxiang, ZHOU Yunqing, et al. A learnable and explainable wavelet neural network for EEG artifacts detection and classification[J]. IEEE Transactions on Neural Systems and Rehabilitation Engineering, 2024, 32: 3358–3368. doi: 10.1109/TNSRE.2024.3452315.
[25]	寇雯博, 董灏, 邹岷强, 等. 混杂复合材料等效热传导性能预测的小波-机器学习混合方法[J]. 物理学报, 2021, 70(3): 030701. doi: 10.7498/aps.70.20201085. KOU Wenbo, DONG Hao, ZOU Minqiang, et al. Hybrid wavelet-based learning method of predicting effective thermal conductivities of hybrid composite materials[J]. Acta Physica Sinica, 2021, 70(3): 030701. doi: 10.7498/aps.70.20201085.
[26]	SU Hanning, BAO Qinglong, PAN Jiameng, et al. Waveform-domain complementary signal sets for interrupted sampling repeater jamming suppression[J]. IEEE Transactions on Aerospace and Electronic Systems, 2024, 60(5): 7142–7158. doi: 10.1109/TAES.2024.3410952.
[27]	施富强, 周超, 刘泉华. 间歇采样重复转发式干扰特性分析[J]. 信号处理, 2017, 33(12): 1616–1624. doi: 10.16798/j.issn.1003-0530.2017.12.013. SHI Fuqiang, ZHOU Chao, and LIU Quanhua. Characteristics analysis of interrupted-sampling repeater jamming[J]. Journal of Signal Processing, 2017, 33(12): 1616–1624. doi: 10.16798/j.issn.1003-0530.2017.12.013.
[28]	潘小义, 刘晓斌, 陈吉源, 等. 间歇采样转发干扰技术研究述评[J]. 系统工程与电子技术, 2024, 46(9): 2887–2901. doi: 10.12305/j.issn.1001-506X.2024.09.01. PAN Xiaoyi, LIU Xiaobin, CHEN Jiyuan, et al. Overview of intermittent sampling repeater jamming technology[J]. Systems Engineering and Electronics, 2024, 46(9): 2887–2901. doi: 10.12305/j.issn.1001-506X.2024.09.01.
[29]	WU Zhenhua, WANG Tengxin, CAO Yice, et al. Radar active deception jamming recognition based on Siamese squeeze wavelet attention network[J]. IET Radar, Sonar & Navigation, 2023, 17(12): 1886–1898. doi: 10.1049/rsn2.12482.
[30]	HUANG Junjie and DRAGOTTI P L. WINNet: Wavelet-inspired invertible network for image denoising[J]. IEEE Transactions on Image Processing, 2022, 31: 4377–4392. doi: 10.1109/TIP.2022.3184845.
[31]	BATOOL I and IMRAN M. A dual residual dense network for image denoising[J]. Engineering Applications of Artificial Intelligence, 2025, 147: 110275. doi: 10.1016/j.engappai.2025.110275.
[32]	DONG Ganggang, WANG Zixuan, and LIU Hongwei. A cross-modality contrastive learning method for radar jamming recognition[J]. IEEE Transactions on Instrumentation and Measurement, 2025, 74: 2520811. doi: 10.1109/TIM.2025.3554858.
[33]	WANG Yongyao, SUN Haiyang, LUO Kai, et al. A lightweight YOLOv11-based framework for small steel defect detection with a newly enhanced feature fusion module[J]. Scientific Reports, 2025, 15(1): 34322. doi: 10.1038/s41598-025-16619-9.