基于高斯原型网络的小样本逆合成孔径雷达目标识别

杨敏佳; 白雪茹; 刘士豪; 曾磊; 周峰

doi:10.11999/JEIT210724

基于高斯原型网络的小样本逆合成孔径雷达目标识别

doi: 10.11999/JEIT210724

1.
西安电子科技大学雷达信号处理国家重点实验室西安 710071
2.
西安电子科技大学电子信息攻防对抗与仿真技术教育部重点实验室西安 710071

基金项目: 国家自然科学基金(62131020, 61971332, 61631019)

详细信息

作者简介:
杨敏佳：男，博士生，研究方向为雷达目标识别

白雪茹：女，教授，研究方向为高分辨雷达成像、雷达目标识别

刘士豪：男，硕士生，研究方向为高分辨雷达成像

曾磊：男，硕士生，研究方向为雷达目标识别

周峰：男，教授，研究方向为电子对抗、雷达成像

通讯作者:
白雪茹　xrbai@xidian.edu.cn

中图分类号: TN957
计量
- 文章访问数: 1184
- HTML全文浏览量: 659
- PDF下载量: 196
- 被引次数: 0
出版历程
- 收稿日期: 2021-07-16
- 修回日期: 2021-11-15
- 录用日期: 2021-11-18
- 网络出版日期: 2021-11-25
- 刊出日期: 2022-10-19

Small-Data Inverse Synthetic Aperture Radar Object Recognition Based on Gaussian Prototypical Network

1.
National Laboratory of Radar Signal Processing, Xidian University, Xi’an 710071, China
2.
Key Laboratory of Electronic Information Countermeasure and Simulation Technology, Ministry of Education, Xidian University, Xi’an 710071, China

Funds: The National Natural Science Foundation of China (62131020, 61971332, 61631019)

摘要

摘要: 针对现有基于深度卷积神经网络(DCNNs)的逆合成孔径雷达(ISAR)目标识别方法在训练样本不足时性能下降甚至失效等问题，该文提出基于高斯原型网络(GPN)的小样本ISAR目标识别方法。该方法通过嵌入网络将ISAR像映射为嵌入向量，进而根据加权嵌入向量构建高斯原型，最终根据测试样本到原型的马氏距离预测目标类别。3类飞机目标实测数据的识别结果表明，该方法在小样本条件下可获得更高的平均识别精度。
- 逆合成孔径雷达 /
- 目标识别 /
- 深度学习 /
- 小样本学习 /
- 高斯原型网络
Abstract: Considering the issue of performance degradation or even failure of the available Inverse Synthetic Aperture Radar (ISAR) object recognition methods based on Deep Convolution Neural Networks (DCNNs) with insufficient training samples, a small- data ISAR object recognition method based on Gaussian Prototypical Network (GPN) is proposed. Firstly, ISAR images are maped into embedding vectors by the embedding network, and then Gaussian prototypes are constructed according to the weighted embedding vectors. Finally, the object category is output according to the Mahalanobis distance from the test samples to all prototypes. Recognition results of the three different types of aircraft show that the proposed method can obtain higher average recognition accuracy under small-data scenarios.
- Inverse Synthetic Aperture Radar(ISAR) /
- Object recognition /
- Deep learning /
- Small-data learning /
- Gaussian Prototypical Network(GPN)

HTML全文

图 1 基于GPN的小样本ISAR目标识别流程图

下载: 全尺寸图片幻灯片

图 2 GPN中的DCNN结构

下载: 全尺寸图片幻灯片

图 3 5类飞机3D模型及对应的电磁仿真数据典型ISAR像（1 GHz带宽、4°积累角）

下载: 全尺寸图片幻灯片

图 4 F16不同积累角成像结果对比

下载: 全尺寸图片幻灯片

图 5 实测飞机光学图及典型ISAR图像

下载: 全尺寸图片幻灯片

图 6 训练集中5类不同型号飞机电磁仿真数据典型ISAR像的特征分布可视化结果与分布检验结果

下载: 全尺寸图片幻灯片

图 7 GPN嵌入向量的t-SNE可视化结果

下载: 全尺寸图片幻灯片

图 8 GPN识别结果统计直方图

下载: 全尺寸图片幻灯片

表 1 PN与GPN识别结果对比

成像积累角	模型	类型	测试准确率（%）			标准差
成像积累角	模型	类型	均值	最大值	最小值	标准差
3°	PN	1-shot	73.55	89.14	55.06	0.0772
	PN	5-shot	89.95	95.69	73.33	0.0299
	GPN	1-shot	74.31	89.51	45.69	0.0894
	GPN	5-shot	92.52	97.65	77.25	0.0219
4°	PN	1-shot	75.74	91.01	49.81	0.0831
	PN	5-shot	90.56	98.43	72.55	0.0423
	GPN	1-shot	77.05	89.89	55.81	0.0836
	GPN	5-shot	92.82	98.43	83.14	0.0274
5°	PN	1-shot	69.52	90.26	47.94	0.0819
	PN	5-shot	87.36	94.51	72.94	0.0343
	GPN	1-shot	70.09	82.77	43.82	0.0801
	GPN	5-shot	91.99	97.25	76.08	0.0295

下载: 导出CSV

表 2 小样本条件下传统DCNN与GPN识别结果对比(%)

模型	1-shot	5-shot
DCNN (Layer=6)	45.69	68.24
GPN (3°)	74.31	92.52
GPN (4°)	77.05	92.82
GPN (5°)	70.09	91.99

下载: 导出CSV

参考文献(21)

[1]	BAI Xueru, ZHOU Xuening, ZHANG Feng, et al. Robust Pol-ISAR target recognition based on ST-MC-DCNN[J]. IEEE Transactions on Geoscience and Remote Sensing, 2019, 57(12): 9912–9927. doi: 10.1109/TGRS.2019.2930112
[2]	KANG Le, LUO Ying, ZHANG Qun, et al. 3-D scattering image sparse reconstruction via radar network[J]. IEEE Transactions on Geoscience and Remote Sensing, To be published.
[3]	ZHANG Shuanghui, LIU Yongxiang, LI Xiang, et al. Enhancing ISAR image efficiently via convolutional reweighted l₁ minimization[J]. IEEE Transactions on Image Processing, 2021, 30: 4291–4304. doi: 10.1109/TIP.2021.3070442
[4]	杨磊, 夏亚波, 毛欣瑶, 等. 基于分层贝叶斯Lasso的稀疏ISAR成像算法[J]. 电子与信息学报, 2021, 43(3): 623–631. doi: 10.11999/JEIT200292 YANG Lei, XIA Yabo, MAO Xinyao, et al. Sparse ISAR imaging algorithm based on Bayesian-lasso[J]. Journal of Electronics &Information Technology, 2021, 43(3): 623–631. doi: 10.11999/JEIT200292
[5]	LEE S J, LEE M J, KIM K T, et al. Classification of ISAR images using variable cross-range resolutions[J]. IEEE Transactions on Aerospace and Electronic Systems, 2018, 54(5): 2291–2303. doi: 10.1109/TAES.2018.2814211
[6]	BENEDEK C and MARTORELLA M. Moving target analysis in ISAR image sequences with a multiframe marked point process model[J]. IEEE Transactions on Geoscience and Remote Sensing, 2014, 52(4): 2234–2246. doi: 10.1109/TGRS.2013.2258927
[7]	PALADINI R, MARTORELLA M, and BERIZZI F. Classification of man-made targets via invariant coherency-matrix eigenvector decomposition of polarimetric SAR/ISAR images[J]. IEEE Transactions on Geoscience and Remote Sensing, 2011, 49(8): 3022–3034. doi: 10.1109/TGRS.2011.2116121
[8]	PARK S H, JUNG J H, KIM S H, et al. Efficient classification of ISAR images using 2D fourier transform and polar mapping[J]. IEEE Transactions on Aerospace and Electronic Systems, 2015, 51(3): 1726–1736. doi: 10.1109/TAES.2015.140184
[9]	MARTORELLA M, GIUSTI E, CAPRIA A, et al. Automatic target recognition by means of polarimetric ISAR images and neural networks[J]. IEEE Transactions on Geoscience and Remote Sensing, 2009, 47(11): 3786–3794. doi: 10.1109/TGRS.2009.2025371
[10]	TOUMI A, HOUSSEINI A E, and KHENCHAF A. Aircrafts recognition using convolutional neurons network[C]. International Conference on Radar Systems, Belfast, UK, 2017: 1–4.
[11]	LIN Zhao, JI Kefeng, KANG Miao, et al. Deep convolutional highway unit network for SAR target classification with limited labeled training data[J]. IEEE Geoscience and Remote Sensing Letters, 2017, 14(7): 1091–1095. doi: 10.1109/LGRS.2017.2698213
[12]	WANG Li, BAI Xueru, GONG Chen, et al. Hybrid inference network for few-shot SAR automatic target recognition[J]. IEEE Transactions on Geoscience and Remote Sensing, 2021, 59(11): 9257–9269. doi: 10.1109/TGRS.2021.3051024
[13]	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.
[14]	KOCH G, ZEMEL R, SALAKHUTDINOV R, et al. Siamese neural networks for one-shot image recognition[C]. 32nd International Conference on Machine Learning, Lille, France, 2015: 1–27.
[15]	VINYALS O, BLUNDELL C, LILLICRAP T, et al. Matching networks for one shot learning[C]. 30th International Conference on Neural Information Processing System, Barcelona, Spain, 2016: 3637–3645.
[16]	SNELL J, SWERSKY K, and ZEMEL R S. Prototypical networks for few-shot learning[C]. 31st International Conference on Neural Information Processing System, Long Beach, USA, 2017: 4080–4090.
[17]	FORT S. Gaussian prototypical networks for few-shot learning on omniglot[C]. Second workshop on Bayesian Deep Learning (NIPS 2017), Long Beach, USA, 2017: 1–10.
[18]	GLOROT X, BORDES A, and BENGIO Y. Deep sparse rectifier neural networks[C]. Fourteenth International Conference on Artificial Intelligence and Statistics, Fort Lauderdale, USA, 2011: 315–323.
[19]	VAN DER MAATEN L and HINTON G. Visualizing data using t-SNE[J]. Journal of Machine Learning Research, 2008, 9(86): 2579–2605.
[20]	WILK M B and GNANADESIKAN R. Probability plotting methods for the analysis of data[J]. Biometrika, 1968, 55(1): 1–17.
[21]	KINGMA D P and BA J L. Adam: A method for stochastic optimization[C]. 3rd International Conference on Learning Representation, San Diego, USA, 2015: 1–15.