仿真数据辅助的雷达HRRP小样本目标识别方法

陈健; 於刚; 杜兰; 董文强; 郭昱辰

doi:10.12000/JR25123

仿真数据辅助的雷达HRRP小样本目标识别方法

DOI: 10.12000/JR25123 CSTR: 32380.14.JR25123

西安电子科技大学雷达信号处理全国重点实验室西安 710071

基金项目: 教育部联合基金(8091B03032401)，国家自然科学基金(U24B20137, U21B2039)，航空科学基金(20230020081006)

详细信息

作者简介:
陈　健，博士，副教授，主要研究方向为雷达目标识别、雷达信号处理、机器学习等

於　刚，硕士生，主要研究方向为雷达自动目标识别

杜　兰，博士，教授，主要研究方向为雷达目标检测识别、雷达信号处理、机器学习等

董文强，硕士，主要研究方向为雷达自动目标识别

郭昱辰，博士，讲师，主要研究方向为SAR目标检测识别

通讯作者:
陈健 jianc@xidian.edu.cn

责任主编：张双辉 Corresponding Editor: ZHANG Shuanghui

中图分类号: TN957.5
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: 2025-07-15

Few-Shot Radar High-Resolution Range Profile: Target Recognition with Simulated Data Assistance

National Key Laboratory of Radar Signal Processing, Xidian University, Xi’an 710071, China

Funds: Joint Fund of the Ministry of Education of China (8091B03032401), The National Natural Science Foundation of China (U24B20137, U21B2039), The Aviation Science Foundation(20230020081006)

More Information

Corresponding author: CHEN Jian, jianc@xidian.edu.cn

摘要

摘要: 雷达高分辨距离像(HRRP)目标识别研究广泛、方法众多，特别是深度学习在雷达HRRP识别领域的应用与发展，为直接利用雷达回波实现高效、精确的目标感知提供了技术支撑。然而，深层识别网络依赖大量训练数据。对于非合作目标，受雷达系统参数、目标快速机动等因素限制，实际很难提前获取姿态完备的HRRP训练样本，深层识别网络面临学习过拟合、泛化能力显著下降的问题。针对上述问题，考虑关注目标的全姿态电磁仿真数据易获取，该文以仿真数据为辅助，从数据扩充和跨域知识迁移学习两方面来缓解小样本问题。数据扩充方面，结合一定姿态角角域范围内仿真、实测HRRP在均值和方差特性两方面的差异分析，对与少量实测HRRP同角域的大量仿真HRRP样本进行线性变换，使其均值、方差满足实测域HRRP特性，实现可表征真实HRRP分布特性的数据扩充。跨域知识迁移学习方面，考虑数据扩充策略仅能实现临近姿态角的样本扩充，对仿真数据知识的利用仍不充分，所提方法利用基于生成对抗约束的域对齐策略和基于对比学习约束的类对齐策略，将具有强可分性与泛化性的仿真域全姿态数据特征和实测域特征按类拉近，进一步辅助实测域数据的学习，实现小样本识别性能的更大提升。基于3类飞机目标以及10类地面车辆目标电磁仿真和实测HRRP数据的实验表明，所提方法相较于现有小样本识别方法具有更优的识别稳健性。
- 雷达目标识别 /
- 高分辨距离像 /
- 小样本 /
- 仿真数据辅助 /
- 数据扩充 /
- 跨域知识迁移学习
Abstract: Research on target recognition using radar high-resolution range profiles (HRRPs) is extensive and diverse in methodology. In particular, the application and development of deep learning to radar HRRP target recognition have enabled efficient, precise target perception directly from radar echoes. However, deep learning-based recognition networks rely on large amounts of training data. For non-cooperative targets, due to limited radar system parameters and rapid target attitude variations, acquiring adequate HRRP training samples that comprehensively cover target attitudes in advance is difficult in practice. Consequently, deep recognition networks are prone to overfitting and exhibit considerably degraded generalization capability. To address these issues, and given the ease of obtaining full-attitude electromagnetic simulation data for the target, this paper leverages simulated data as auxiliary information to mitigate the small-sample-size problem through data augmentation and cross-domain knowledge-transfer learning. For data augmentation, based on the analysis of differences in mean and variance between simulated and measured HRRPs within a given attitude-angle range, a linear transformation is applied to a set of simulated HRRPs spanning the same angular domain as a small set of measured HRRPs. This adjustment ensures that the simulated data’s mean and variance match the characteristics of the measured HRRPs, thereby achieving data augmentation that approximates the true distributional properties of HRRPs. Meanwhile, for cross-domain knowledge transfer learning, the proposed method introduces a domain alignment strategy based on generative adversarial constraints and a class alignment strategy based on contrastive learning constraints. These approaches draw the domain features of full-attitude simulation—strong discriminability and generalizability—closer to the measured domain features on a class-by-class basis, thereby further aiding learning from the measured domain data and leading to substantial improvements in few-shot recognition performance. Experimental results based on electromagnetic simulated and measured HRRP data for three and ten types of aircraft and ground vehicle targets, respectively, demonstrate that the proposed method yields superior recognition robustness compared with existing few-shot recognition methods.
- Radar target recognition /
- High-Resolution Range Profile (HRRP) /
- Few-shot learning /
- Simulated data assistance /
- Data expansion /
- Cross-domain knowledge transfer learning.

HTML全文

图 1 仿真数据辅助的雷达HRRP小样本目标识别方法整体框架图

Figure 1. Overall framework diagram of simulated data-assisted few-shot radar HRRP target recognition method

下载: 全尺寸图片幻灯片

图 2 特征提取模块结构示意图

Figure 2. Schematic diagram of feature extraction module structure

下载: 全尺寸图片幻灯片

图 3 域对齐示意图

Figure 3. Schematic diagram of domain alignment

下载: 全尺寸图片幻灯片

图 4 类对齐示意图

Figure 4. Schematic diagram of class alignment

下载: 全尺寸图片幻灯片

图 5 每类10个实测训练HRRP样本情况下不同类对齐约束方式对仿真HRRP和实测HRRP测试数据的特征对齐可视化

Figure 5. Visualization of feature alignment under different class alignment constraints for simulated HRRPs and measured test HRRPs with 10 HRRP samples per class

下载: 全尺寸图片幻灯片

图 6 3类飞机目标轨迹投影到地平面的展示图

Figure 6. Projection diagram of trajectories of three types of aircraft targets onto the ground plane

下载: 全尺寸图片幻灯片

图 7 SAMPLE数据集场景条件设置及对应的各类样本示例

Figure 7. Scene condition settings for the SAMPLE dataset and corresponding sample examples of various types

下载: 全尺寸图片幻灯片

图 8 每类10个实测训练样本情况下不同方法在实测HRRP测试集上的特征可视化对比(10shot)

Figure 8. Feature visualization comparisons of different methods on the measured HRRP test set in the condition of 10 HRRP training samples per class

下载: 全尺寸图片幻灯片

图 9 雅克-42在不同方位角域内的HRRP数据可视化

Figure 9. HRRP data visualizations of Yak-42 within different angular domains

下载: 全尺寸图片幻灯片

图 10 奖状在不同方位角域内的HRRP数据可视化

Figure 10. HRRP data visualizations of Cessna Citation within different angular domains

下载: 全尺寸图片幻灯片

图 11 安-26在不同方位角域内的HRRP数据可视化

Figure 11. HRRP data visualizations of An-26 within different azimuth domains

下载: 全尺寸图片幻灯片

图 12 每类10个实测训练样本情况下不同方法仿真域特征与实测测试数据的特征分布可视化

Figure 12. Feature visualizations of simulated HRRPs and measured test HRRPs for different methods in the condition of 10 HRRP training samples per class

下载: 全尺寸图片幻灯片

图 13 不同实测训练样本数条件下3类飞机目标HRRP的识别性能随信噪比的变化

Figure 13. The variation of HRRP recognition performance of three types of aircraft targets with the signal-to-noise ratio under different measured training sample numbers

下载: 全尺寸图片幻灯片

图 14 每类10个实测HRRP训练样本情况下不同方法在仿真训练HRRP和实测测试HRRP数据上的特征可视化。不同颜色的圆形表示不同类别的源域训练特征，与圆形同色的黑框菱形表示对应类目标域测试特征。

Figure 14. Visualization of the features of different methods on the simulated training HRRP and the measured test HRRP data under the condition of 10 measured HRRP training samples per class

下载: 全尺寸图片幻灯片

图 15 含杂波/目标有遮挡条件下所提方法在实测HRRP数据上的特征可视化(每类10个实测训练样本)。不同颜色的圆形表示不同类别的源域训练特征，与圆形同色的黑框菱形表示对应类目标域测试特征。

Figure 15. Feature visualization of the proposed method (10 measured training HRRPs per class) on measured HRRP data under conditions with clutter or target occlusion

下载: 全尺寸图片幻灯片

表 1 仿真HRRP信噪比为30 dB、实测HRRP信噪比为30 dB, 20 dB, 10 dB, 5 dB情况下所提方法的识别性能(每类10个实测HRRP训练样本)

Table 1. Performance of proposed method under the SNR of simulated HRRP being 30 dB and the SNR of measured HRRP being 30 dB, 20 dB, 10 dB, 5 dB with 10 measured training HRRPs per class

实测HRRP数据信噪比	识别准确率
30dB	90.22%
20dB	85.24%
10dB	79.58%
5dB	71.93%

下载: 导出CSV

表 2 实验雷达参数

Table 2. Radar parameters in the experiment

参数	数值
中心频率	5520 MHz
脉冲重复频率	400 Hz
Dechirp后采样频率	10 MHz
信号带宽	400 MHz

下载: 导出CSV

表 3 实验飞机参数

Table 3. Aircraft parameters in the experiment

飞机	机长(m)	机宽(m)	机高(m)
安-26	23.80	29.20	9.83
奖状	14.40	15.90	4.57
雅克-42	36.38	34.88	9.83

下载: 导出CSV

表 4 SAMPLE数据集类别和数据量

Table 4. The category and volume of the SAMPLE dataset

类别	仿真	实测训练	实测测试
2S1	174	116	58
BMP2	107	55	52
BTR70	92	43	49
M1	129	78	51
M2	128	75	53
M35	129	76	53
M548	128	75	53
M60	176	116	60
T72	108	56	52
ZSU23	174	116	58

下载: 导出CSV

表 5 实验所用实测与仿真训练样本具体数量

Table 5. The specific quantities of measured and simulated training samples used in the experiment

方法	每类实测/仿真训练样本数
方法	1shot	5shot	10shot
AGC、SVM、FA、CNN、 RNN、LSTM、PN	1/0	5/0	10/0
PN+微调、PNML、MAML、 DAN、DANN、所提方法	1/54000	5/54000	10/54000

下载: 导出CSV

表 6 不同方法小样本识别性能对比(20次实验平均结果)

Table 6. Comparison of few-shot recognition performance among different methods (average results of 20 experiments)

训练数据	方法	1shot	5shot	10shot
	AGC	40.86%±5.23%	49.42%±4.63%	53.42%±4.21%
	SVM	42.81%±5.77%	50.86%±5.08%	54.88%±3.83%
仅使用少	FA	41.58%±5.36%	51.69%±4.89%	57.52%±4.57%
量实测数	CNN	48.62%±3.87%	53.88%±3.42%	57.28%±2.93%
据训练	RNN	47.83%±4.09%	53.62%±3.76%	55.12%±3.25%
	LSTM	49.76%±3.53%	54.12%±3.04%	56.85%±2.73%
	PN	54.27%±3.28%	61.98%±2.59%	65.96%±2.05%
	PN+微调	61.85%±1.84%	70.24%±1.37%	74.21%±1.06%
使用仿真	PNML	63.86%±1.26%	75.56%±1.14%	79.45%±0.87%
数据和少	MAML	64.33%±1.32%	75.85%±1.03%	79.95%±0.97%
量实测数	DAN	57.58%±0.95%	71.75%±0.79%	77.42%±0.72%
据训练	DANN	57.93%±0.93%	72.65%±0.84%	78.65%±0.64%
	所提方法	74.94%±0.68%	85.78%±0.32%	90.22%±0.27%
注：表内加粗数值表示最优指标数值。

下载: 导出CSV

表 7 传统识别方法结合/不结合所提数据扩充策略在每类10个实测HRRP训练样本情况下的识别性能比较

Table 7. Comparison of the recognition performance of traditional recognition methods with/without the proposed data expansion strategy in the case of 10 measured HRRP training samples per class

是否使用所提数据扩充策略	传统方法	识别准确率
否	SVM	54.88%
	CNN	57.28%
	PN+微调	74.21%
	DAN	77.42%
	DANN	78.65%
是	SVM	57.94%
	CNN	63.06%
	PN+微调	78.33%
	DAN	85.78%
	DANN	86.41%

下载: 导出CSV

表 8 每类10个实测训练样本情况下的消融实验结果

Table 8. Ablation experiment results in the condition of 10 HRRP training samples per class

模块	数据扩充	域对齐	类对齐	准确率/%
S2R	-	-	-	54.33
(a)	√			76.19
(b)	√	√		86.47
(c)	√		√	82.06
(d)		√		78.14
(e)			√	67.02
(f)		√	√	79.51
(g)	√	√	√	90.22
注：表内加粗数值表示最优指标数值。

下载: 导出CSV

表 9 不同方法单批次训练计算复杂度和训练时间对比

Table 9. Comparison of computational complexity and training time of single-batch training by different methods

方法	训练计算复杂度(FLOPs)	单批次训练时间(ms)
PN+微调	1.5×10⁹	18.9
PNML	1.4×10⁹	17.6
MAML	2.4×10⁹	30.7
DAN	1.9×10⁹	21.3
DANN	2.0×10⁹	22.5
所提方法	2.1×10⁹	26.9

下载: 导出CSV

表 10 不同方法单个测试HRRP样本计算复杂度和推理时间对比

Table 10. Comparison of computational complexity and inference time for a single test HRRP sample by different methods

方法	测试计算复杂度(FLOPs)	推理时间(ms)
PN+微调	1.5×10⁷	0.197
PNML	1.5×10⁷	0.201
MAML	1.5×10⁷	0.204
DAN	2.3×10⁷	0.296
DANN	2.3×10⁷	0.298
所提方法	2.8×10⁷	0.359

下载: 导出CSV

表 11 不同方法在由标准操作条件下SAMPLE数据集转换成的HRRP数据上的识别性能对比(20次实验平均结果)

Table 11. Comparison of recognition performance of different methods on HRRP data converted from SAMPLE under the standard operating condition (average results of 20 experiments)

方法	1shot	5shot	10shot
PN+微调	60.10%±1.94%	67.89%±1.45%	71.65%±1.12%
DAN	58.14%±1.26%	69.96%±0.97%	75.23%±0.83%
DANN	58.81%±1.17%	70.24%±0.90%	76.39%±0.78%
所提方法	69.52%±0.81%	77.46%±0.52%	82.74%±0.39%
注：表内加粗数值表示最优指标数值。

下载: 导出CSV

表 12 不同方法在含杂波、目标有遮挡条件下SAMPLE数据集转换成的HRRP数据上的识别性能对比(20次实验平均结果)

Table 12. Comparison of recognition performance of different methods on HRRP data converted from SAMPLE under extended operating conditions (average results of 20 experiments)

场景条件	方法	1shot	5shot	10shot
含杂波	PN+微调	57.86%±2.02%	64.02%±1.58%	69.63%±1.24%
	DAN	55.91%±1.59%	65.25%±1.12%	73.04%±0.96%
	DANN	56.48%±1.43%	66.83%±1.04%	74.57%±0.83%
	所提方法	67.36%±0.94%	74.20%±0.61%	80.68%±0.45%
目标有遮挡	PN+微调	51.79%±2.16%	57.92%±1.75%	62.23%±1.46%
	DAN	49.64%±1.72%	60.36%±1.27%	67.42%±1.09%
	DANN	49.37%±1.59%	61.71%±1.13%	69.35%±0.91%
	所提方法	63.83%±1.08%	71.95%±0.79%	78.49%±0.57%
注：表内加粗数值表示最优指标数值。

下载: 导出CSV

参考文献(33)

[1]	LI Liya, LIU Hongwei, JIU Bo, et al. Polarization radar HRRP recognition based on kernel methods[C]. The 6th International Symposium on Neural Networks on Advances in Neural Networks, Wuhan, China, 2009: 304–311. doi: 10.1007/978-3-642-01510-6_35.
[2]	DU Lan, WANG Penghui, LIU Hongwei, et al. Radar HRRP target recognition based on dynamic multi-task hidden Markov model[C]. 2011 IEEE RadarCon (RADAR), Kansas City, USA, 2011: 253–255. doi: 10.1109/RADAR.2011.5960538.
[3]	PAN Mian, DU Lan, WANG Penghui, et al. Noise-robust modification method for Gaussian-based models with application to radar HRRP recognition[J]. IEEE Geoscience and Remote Sensing Letters, 2013, 10(3): 558–562. doi: 10.1109/LGRS.2012.2213234.
[4]	CHEN Bo, LIU Hongwei, and BAO Zheng. PCA and kernel PCA for radar high range resolution profiles recognition[C]. IEEE International Radar Conference, Arlington, USA, 2005: 528–533. doi: 10.1109/RADAR.2005.1435883.
[5]	ZHANG Xianda, SHI Yu, and BAO Zheng. A new feature vector using selected bispectra for signal classification with application in radar target recognition[J]. IEEE Transactions on Signal Processing, 2001, 49(9): 1875–1885. doi: 10.1109/78.942617.
[6]	PEI Bingnan and BAO Zheng. Multi-aspect radar target recognition method based on scattering centers and HMMs classifiers[J]. IEEE Transactions on Aerospace and Electronic Systems, 2005, 41(3): 1067–1074. doi: 10.1109/TAES.2005.1541451.
[7]	DU Lan, LIU Hongwei, BAO Zheng, et al. Radar automatic target recognition using complex high-resolution range profiles[J]. IET Radar, Sonar & Navigation, 2007, 1(1): 18–26. doi: 10.1049/iet-rsn:20050119.
[8]	LIU Hongwei, FENG Bo, CHEN Bo, et al. Radar high-resolution range profiles target recognition based on stable dictionary learning[J]. IET Radar, Sonar & Navigation, 2016, 10(2): 228–237. doi: 10.1049/iet-rsn.2015.0007.
[9]	FENG Bo, DU Lan, LIU Hongwei, et al. Radar HRRP target recognition based on K-SVD algorithm[C]. 2011 IEEE CIE International Conference on Radar, Chengdu, China, 2011: 642–645. doi: 10.1109/CIE-Radar.2011.6159622.
[10]	DU Lan, LIU Hongwei, BAO Zheng, et al. A two-distribution compounded statistical model for radar HRRP target recognition[J]. IEEE Transactions on Signal Processing, 2006, 54(6): 2226–2238. doi: 10.1109/TSP.2006.873534.
[11]	DU Lan, LIU Hongwei, and BAO Zheng. Radar HRRP statistical recognition: Parametric model and model selection[J]. IEEE Transactions on Signal Processing, 2008, 56(5): 1931–1944. doi: 10.1109/TSP.2007.912283.
[12]	BARAT C and DUCOTTET C. String representations and distances in deep Convolutional Neural Networks for image classification[J]. Pattern Recognition, 2016, 54: 104–115. doi: 10.1016/j.patcog.2016.01.007.
[13]	KRIZHEVSKY A, SUTSKEVER I, and HINTON G E. ImageNet classification with deep convolutional neural networks[C]. The 26th International Conference on Neural Information Processing Systems, Lake Tahoe, USA, 2012: 1106–1114.
[14]	ZHU Xiaoxiang, MONTAZERI S, ALI M, et al. Deep learning meets SAR: Concepts, models, pitfalls, and perspectives[J]. IEEE Geoscience and Remote Sensing Magazine, 2021, 9(4): 143–172. doi: 10.1109/MGRS.2020.3046356.
[15]	LUNDÉN J and KOIVUNEN V. Deep learning for HRRP-based target recognition in multistatic radar systems[C]. 2016 IEEE Radar Conference, Philadelphia, USA, 2016: 1–6. doi: 10.1109/RADAR.2016.7485271.
[16]	JITHESH V, SAGAYARAJ M J, and SRINIVASA K G. LSTM recurrent neural networks for high resolution range profile based radar target classification[C]. The 3rd International Conference on Computational Intelligence & Communication Technology, Ghaziabad, India, 2017: 1–6. doi: 10.1109/CIACT.2017.7977298.
[17]	GUO Dandan, CHEN Bo, CHEN Wenchao, et al. Variational temporal deep generative model for radar HRRP target recognition[J]. IEEE Transactions on Signal Processing, 2020, 68: 5795–5809. doi: 10.1109/TSP.2020.3027470.
[18]	JIANG Mengjuan, FAN Jiaqing, HE Jiangzhen, et al. Contrastive prototype network with prototype augmentation for few-shot classification[J]. Information Sciences, 2025, 686: 121372. doi: 10.1016/j.ins.2024.121372.
[19]	宋益恒, 王彦华, 李阳, 等. 基于深度生成网络的雷达HRRP生成技术[J]. 信号处理, 2019, 35(6): 1118–1122. doi: 10.16798/j.issn.1003-0530.2019.06.025. SONG Yiheng, WANG Yanhua, LI Yang, et al. Radar HRRP generation using deep generative networks[J]. Journal of Signal Processing, 2019, 35(6): 1118–1122. doi: 10.16798/j.issn.1003-0530.2019.06.025.
[20]	GAO Xunzhang and PENG Xuan. Robust sequential high-resolution range profile recognition based on the infinite conditional restricted Boltzmann machine[J]. IET Radar, Sonar & Navigation, 2020, 14(1): 71–80. doi: 10.1049/iet-rsn.2019.0012.
[21]	李昱琛, 索继东. 基于FEKO软件的高分辨距离像建模仿真[J]. 弹箭与制导学报, 2020, 40(4): 10–13,19. doi: 10.15892/j.cnki.djzdxb.2020.04.003. LI Yuchen and SUO Jidong. Modeling and simulation of high resolution range profile based on FEKO software[J]. Missiles and Guidance, 2020, 40(4): 10–13,19. doi: 10.15892/j.cnki.djzdxb.2020.04.003.
[22]	REN Zuyu, JIANG Weidong, and ZHANG Xinyu. Few-shot HRRP target recognition method based on Gaussian deep belief network and model-agnostic meta-learning[C]. 2022 7th International Conference on Signal and Image Processing, Suzhou, China, 2022: 260–264. doi: 10.1109/ICSIP55141.2022.9887225.
[23]	LI Jixi, LI Dongying, JIANG Yong, et al. Few-shot radar HRRP recognition based on improved prototypical network[C]. IEEE International Geoscience and Remote Sensing Symposium, Pasadena, USA, 2023: 5277–5280. doi: 10.1109/IGARSS52108.2023.10282950.
[24]	LONG Mingsheng, CAO Yue, WANG Jianmin, et al. Learning transferable features with deep adaptation networks[C]. The 32nd International Conference on Machine Learning, Lille, France, 2015: 97–105.
[25]	CHEN Zhuo, ZHAO Lingjun, HE Qishan, et al. Pixel-level and feature-level domain adaptation for heterogeneous SAR target recognition[J]. IEEE Geoscience and Remote Sensing Letters, 2022, 19: 4515205. doi: 10.1109/LGRS.2022.3214750.
[26]	ZHOU Qiang, WANG Yanhua, ZHANG Xin, et al. Domain-adaptive HRRP generation using two-stage denoising diffusion probability model[J]. IEEE Geoscience and Remote Sensing Letters, 2024, 21: 3504305. doi: 10.1109/LGRS.2024.3379275.
[27]	LI Shuai, LI Wentao, HUANG Pengjun, et al. MTBC: Masked vision transformer and Brownian distance covariance classifier for cross-domain few-shot HRRP recognition[J]. IEEE Sensors Journal, 2025, 25(9): 16440–16454. doi: 10.1109/JSEN.2025.3550584.
[28]	王鹏辉. 基于统计建模的雷达高分辨距离像目标识别方法研究[D]. [博士论文], 西安电子科技大学, 2012. WANG Penghui. Study of radar high resolution range profile target recognition based on statistical modeling[D]. [Ph.D. dissertation], Xidian University, 2012.
[29]	杜兰. 雷达高分辨距离像目标识别方法研究[D]. [博士论文], 西安电子科技大学, 2007. doi: 10.7666/d.Y1137516. DU Lan. Study on radar HRRP target recognition[D]. [Ph.D. dissertation], Xidian University, 2007. doi: 10.7666/d.Y1137516.
[30]	CHEN Jian, DU Lan, GUO Guanbo, et al. Target-attentional CNN for radar automatic target recognition with HRRP[J]. Signal Processing, 2022, 196: 108497. doi: 10.1016/j.sigpro.2022.108497.
[31]	GANIN Y and LEMPITSKY V S. Unsupervised domain adaptation by backpropagation[C]. The 32nd International Conference on Machine Learning, Lille, France, 2015: 1180–1189.
[32]	KONG Yingying, ZHANG Yuxuan, PENG Xiangyang, et al. Few-shot high-resolution range profile ship target recognition based on task-specific meta-learning with mixed training and meta embedding[J]. Remote Sensing, 2023, 15(22): 5301. doi: 10.3390/rs15225301.
[33]	潘勉. 雷达高分辨距离像目标识别技术研究[D]. [博士论文], 西安电子科技大学, 2013. PAN Mian. Study of radar target recognition technology based on high range resolution profile[D]. [Ph.D. dissertation], Xidian University, 2013.