合成孔径雷达图像与可见光图像配准方法综述

Review of synthetic aperture radar and visible image registration methods

合成孔径雷达(SAR)图像与可见光图像之间存在大量互补特性,二者的综合利用能够提供更丰富的信息。由于SAR图像与可见光图像之间存在多种差异,作为信息融合的基础,两种图像的配准问题成为当前亟待解决的难题。通过对SAR图像与可见光图像的配准方法进行梳理和总结,根据配准过程对专家先验信息的依赖程度及是否具有训练、学习过程这两个角度,将SAR图像与可见光图像配准方法分为基于特征的方法和基于深度学习的方法两大类,并对两类方法的关键技术进行分析,重点阐述了基于几何特征的配准和深度学习的配准,然后介绍配准结果的通用评价指标以及常用的成对的SAR图像与可见光图像数据集,最后总结目前SAR图像与可见光图像配准仍然存在的问题并展望未来发展趋势。

Synthetic Aperture Radar(SAR)images and visible images have many complementary characteristics and their comprehensive utilization can provide abundant information.Since there are so many disparities between SAR images and visible images,the registration of SAR images and visible images is an urgent problem that needs to be resolved.The registration methods of SAR images and visible images were divided into feature-based and deep learning-based methods according to whether the registration process relies on expert prior knowledge and whether there is a training or learning process.The key technologies of the two types of registration methods of SAR images and visible images were analyzed,especially the registration based on geometric features and the registration based on deep learning.The general evaluation metrics of the registration outcomes and widely used paired SAR-visible image datasets were introduced.Finally,the remaining problems in the registration of SAR images and visible images were summarized and the future development trends were forecasted.