合成孔径雷达深度学习成像研究综述  

A Survey of Synthetic Aperture Radar Imaging Methods Based on Deep Learning

作　　者：张群[1,3,4] 张宏伟 倪嘉成 罗迎[1,3] ZHANG Qun;ZHANG Hongwei;NI Jiacheng;LUO Ying(College of Information and Navigation,Air Force Engineering University,Xi’an,Shaanxi 710077,China;Graduate School,Aire Force Engineering University,Xi’an,Shaanxi 710077,China;Collaborative Innovation Center of Information Sensing and Understanding,Xi’an,Shaanxi 710077,China;The Key Laboratory for Information Science of Electromagnetic Waves(Ministry of Education),Fudan University,Shanghai 200433,China)

机构地区：[1]空军工程大学信息与导航学院,陕西西安710077 [2]空军工程大学研究生院,陕西西安710077 [3]信息感知技术协同创新中心,陕西西安710077 [4]复旦大学电磁波信息科学教育部重点实验室,上海200433

出　　处：《信号处理》2023年第9期1521-1551,共31页Journal of Signal Processing

基　　金：国家自然科学基金(62131020,62001508,61971434)。

摘　　要：现代合成孔径雷达(SAR)系统工作在日益复杂的电磁环境中,对成像精度、实时性以及算法鲁棒性等要求越来越高。传统的匹配滤波以及压缩感知技术在满足SAR成像的各类高标准要求时局限性较为明显,尤其在成像性能方面。随着机器学习的快速发展,研究人员将深度学习网络与雷达成像算法相结合,提出了学习成像技术,旨在为实现高质量实时成像寻求新的解决方案。本文从数据驱动以及模型驱动同数据驱动相结合的两种思路出发,介绍了用于求解SAR成像逆问题的深度学习网络架构。在此基础上,对SAR静止目标学习成像、SAR运动目标学习成像、SAR三维学习成像以及ISAR学习成像的研究现状进行概述,帮助研究人员和从业人员理解深度学习技术在SAR成像相关问题中的应用。最后,提出该研究方向一些悬而未决的问题,探讨潜在的解决方案和未来趋势。Modern Synthetic Aperture Radar(SAR)systems operate in increasingly complex electromagnetic environ⁃ments,requiring higher imaging accuracy,real-time capability,and algorithm robustness.Traditional matched filtering and compressed sensing technology have apparent limitations in meeting the high-standard requirements of SAR imaging,especially in imaging efficiency and resolution.With the rapid development of machine learning,researchers combine deep learning networks with radar imaging algorithms and propose learning imaging technology,aiming to find new solu⁃tions for high-quality real-time imaging.This paper introduces the deep learning network architecture used to solve the in⁃verse problem of SAR imaging from two perspectives of data-driven and the combination of model-driven and data-driven.On this basis,the development and research status of SAR stationary target learning imaging,SAR moving target learning imaging,3D-SAR learning imaging,and ISAR learning imaging are reviewed to help researchers and practitioners under⁃stand the application of deep learning technology to SAR imaging-related issues.Finally,some unresolved problems are proposed,while discussing potential solutions and future trends.

关 键 词：合成孔径雷达(SAR) 深度学习 SAR成像 数据驱动 模型驱动 

分 类 号：TN957.9[电子电信—信号与信息处理]