摘要: |
针对缺少合成孔径雷达(Synthetic Aperture Radar,SAR)目标图像数据导致的识别网络难以训练的问题,总结了现有的基于深度学习方法的解决方案。归纳了现阶段生成式对抗网络(Generative Adversarial Network,GAN)的发展情况,以及主要的衍生模型及其特点与优势。综述了GAN在SAR图像生成与风格迁移两方面的应用情况,并合理分析了应用中的技术难点和问题。最后结合深度学习的发展趋势,展望了GAN在SAR智能解译方面的应用。 |
关键词: 生成对抗网络 合成孔径雷达 智能解译 数据扩充 |
DOI: |
|
基金项目: |
|
A survey of generative adversarial network applications for SAR image processing |
LU Qinglin,YE Wei |
(Department of Graduate Management,Space Engineering University,Beijing 101416,China;Department of Astronautics,Space Engineering University,Beijing 101416,China) |
Abstract: |
For the problem that the identification network is difficult to train due to the lack of synthetic aperture radar(SAR) target image,the existing solutions based on deep learning method are summarized.The development of the generative adversarial network(GAN) and its main derivative models are summarized,including their characteristics and advantages.The application of GAN in SAR image generation and style migration is reviewed,and the technical difficulties and problems in the application are analyzed reasonably.Finally,according to the development trend of deep learning,the future application of GAN in SAR intelligent interpretation is discussed. |
Key words: generative adversarial nets(GANs) synthetic aperture radar(SAR) intelligent interpretation data enlargement |