| 引用本文: | 张云，穆慧琳，姜义成，丁畅. 基于深度学习的雷达成像技术研究进展[J]. 雷达科学与技术, 2021, 19(5): 467-478.[点击复制] |
| ZHANG Yun, MU Huilin, JIANG Yicheng, DING Chang. Overview of Radar Imaging Techniques Based on Deep Learning[J]. Radar Science and Technology, 2021, 19(5): 467-478.[点击复制] |
|
| 摘要: |
| 成像雷达具有全天时、全天候、远距离、高分辨对地观测的能力，使得雷达系统具有对观测区域进行成像和解译的能力。利用先进信号处理技术实现实时高分辨成像以满足图像解译的需求是雷达成像技术研究的重要目的和意义。随着深度学习的迅速兴起，深度学习网络在逆问题求解中得到广泛应用，也为提升成像质量和成像效率提供新的求解思路。本文基于雷达成像数学模型将雷达成像问题建模为成像逆问题，从逆问题求解的角度分析了基于深度学习的雷达成像方法的可行性。并综述了近年来雷达深度学习技术在合成孔径雷达（Synthetic Aperture Radar，SAR）、逆合成孔径雷达（Inverse Synthetic Aperture Radar，ISAR）、SAR运动目标成像等雷达成像领域的研究现状，在此基础上探讨了目前面临的亟待解决的问题，并对未来发展方向进行了展望。 |
| 关键词: 深度学习 雷达成像 逆问题 卷积神经网络 复数域卷积神经网络 |
| DOI：DOI:10.3969/j.issn.1672-2337.2021.05.001 |
| 分类号:TN958.3 |
| 基金项目:国家自然科学基金面上项目（No.61971163） |
|
| Overview of Radar Imaging Techniques Based on Deep Learning |
|
ZHANG Yun, MU Huilin, JIANG Yicheng, DING Chang
|
|
School of Electronics and Information Engineering, Harbin Institute of Technology, Harbin 150000, China
|
| Abstract: |
| The imaging radar has advantages of all-day, all-weather, long-distance, and high-resolution imaging for the observation scene, which makes it capable of imaging and interpreting the observation scene. The important purpose and significance of the radar imaging technology research is to utilize the advanced signal processing technology to achieve real-time high-resolution imaging for the requirement of image interpretation. With the rapid development of deep learning, deep learning networks have been widely used for inverse problems. And they also provide new solutions for improving imaging quality and efficiency. In this paper, the radar imaging problem is modeled as an inverse imaging problem based on the radar imaging mathematical model. Then the feasibility of the radar imaging method based on deep learning is analyzed from the perspective of solving the inverse problem. Moreover, the state-of-the-art radar deep learning technologies are reviewed in the field of synthetic aperture radar (SAR), inverse synthetic aperture radar (ISAR), and SAR moving target imaging in recent years. The future perspectives are finally discussed according to the existed challenges. |
| Key words: deep learning radar imaging inverse problem convolutional neural network complex-valued convolutional neural network |
