摘 要：近年来，基于数据驱动的 SAR 自动目标识别研究技术取得了很大进展。虽然这类方法识别性能较好，但是在实际应用场景中很难采集到足够的真实 SAR 数据用于训练。文章通过引入仿真 SAR 数据来扩充训练数据集，弥补真实 SAR 数据的不足。为了解决仿真和真实 SAR 数据之间差异导致的跨域和跨任务迁移问题，提出了一种基于跨域元学习机制的知识迁移算法。利用特征变换和数据增强方法来解决跨领域迁移，利用元学习机制来解决跨任务迁移。实验证明算法可以在有限的训练数据下取得良好的识别性能。
关键词：合成孔径雷达；知识迁移；元学习
DOI:10.19850/j.cnki.2096-4706.2023.04.015
基金项目：江苏省电子信息职业教育研究常规课题（JSDX2021-51）；江苏电子信息职业学院校级科研基金项目（JSEIYY2020003）
中图分类号：TP18；TN957.52 文献标识码：A 文章编号：2096-4706（2023）04-0057-04
Research on SAR Automatic Target Recognition Based on Cross-Domain Meta-Learning
WANG Ke, QIAO Qi
(Jiangsu Vocational College of Electronics and Information, Huai'an 223003, China)
Abstract: In recent years, SAR automatic target recognition research technology based on data driven has made great progress. Although the recognition performance of such methods is better, it is difficult to collect enough real SAR data for training. This paper expands the training data set by introducing simulation SAR data to make up for the shortage of real SAR data. To solve cross-domain and cross-task transfer problems caused by the differences between simulation and real SAR data, a knowledge transfer algorithm based on cross-domain Meta-Learning mechanism is proposed. Feature transformation and data enhancement are used to solve cross-domain transfer, and Meta-Learning mechanism is used to solve cross-task transfer. Experiments demonstrate that the algorithm can achieve good recognition performance with limited training data.
Keywords: synthetic aperture radar; knowledge transfer; Meta-Learning
参考文献：
[1] 保铮，刑孟道，王彤 . 雷达成像技术 [M]. 北京：电子工业出版社，2005.
[2] CHEN S Z，WANG H P，XU F，et al. Target classification using the deep convolutional networks for SAR images [J].IEEE Transactions on Geoscience and Remote Sensing，2016，54（8）：4806-4817.
[3] ZHAO P，LIU K，ZOU H，et al. Multi-stream convolutional neural network for SAR automatic target recognition [J].Remote Sensing，2018，10（9）：1473.
[4] MIN R，LAN H，CAO Z J，et al. A gradually distilled CNN for SAR target recognition [J].IEEE Access，2019，7：42190-42200.
[5] DING B Y，WEN G J. A region matching approach based on 3-D scattering center model with application to SAR target recognition [J]. IEEE Sensors Journal，2018，18（11）：4623-4632.
[6] DING J，CHEN B，LIU H W，et al. Convolutional neural network with data augmentation for SAR target recognition [J].IEEE Geoscience and Remote Sensing Letters，2016，13（3）：364-368.
[7] LIU M，CHEN S C，WU J，et al. Configuration recognition via class-dependent structure preserving projections with application to targets in SAR images [J]，IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing，2018，1（11）：2134–2146.
[8] YU M T，ZHANG S Q，DONG G，et al. Target recognition in SAR image based on robust locality discriminant projection [J]，IET Radar，Sonar & Navigation，2018，12（11）：1285–1293.
[9] MALMGREN-HANSEN D，KUSK A，DALL J，et al. Improving SAR automatic target recognition models with transfer learning from simulated data [J].IEEE Geoscience and Remote Sensing Letters，2017，14（9）：1484-1488.
[10] INKAWHICH N，INKAWHICH M J，DAVIS E K，et al. Bridging a gap in SAR-ATR: Training on fully synthetic and testing on measured data [J].IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing，2021，14：2942-2955.
[11] ZHANG H，CISSE M，DAUPHIN Y N，et al. mixup: Beyond empirical risk minimization [J/OL].arXiv:1710.09412 [cs.LG]. [2022-09-13]. https://arxiv.org/abs/1710.09412.
[12] TSENG H Y，LEE H Y，HUANG J B，et al. Crossdomain few-shot classification via learned feature-wise transformation [J/OL].arXiv:2001.08735 [cs.CV].[2022-09-13].https://arxiv.org/ abs/2001.08735v1.
作者简介：王可（1981—），男，汉族，江苏淮安人，讲师，博士，研究方向：人工智能和大数据技术；乔琪（1982—），男，汉族，江苏淮安人，副教授，硕士，研究方向：通信技术和大数据技术。