摘要
由于合成孔径雷达(SAR)和光学图像两种模态之间存在显著的几何和辐射差异,传统方法难以实现SAR与光学图像的精准匹配。文中提出了一种基于端到端网络机制的跨域稀疏SAR与光学图像精准匹配方法。该方法首先预测每幅图像中最适合匹配的区域,然后通过多尺度特征空间互相关运算生成匹配热图,最后将匹配热图分为正匹配和负匹配来消除异常值,实现SAR与光学图像匹配的精确匹配。实验结果表明,所提方法性能指标优于已往SAR与光学图像匹配方法,可用于大规模场景的精确匹配,利于提升光学卫星图像的地理定位精度。
Due to the significant geometric and radiation differences between synthetic aperture radar(SAR) and optical images, it is difficult for traditional methods to achieve accurate matching between SAR and optical images. In this paper, an accurate matching method between cross domain sparse SAR and optical image based on end-to-end network framework is proposed. The method first predicts the region that is considered to be the most suitable for matching in each image, and then generates the matching heatmap through multi-scale feature space cross-correlation operator. Finally, the matching heatmap is classified into positive and negative matching to eliminate outliers and achieve accurate matching between SAR and optical images. The experimental results show that that the proposed approach provides a substantial improvement over previous methods for SAR-optical image matching and can be used to register even large-scale scenes, which is conducive to improve the accuracy of geographic positioning of optical satellite images.
作者
黄柏圣
HUANG Baisheng(School of Electronics and Information Engineering,Nanjing University of Information Science and Technology,Nanjing 210044,China)
出处
《现代雷达》
CSCD
北大核心
2022年第12期106-111,共6页
Modern Radar
基金
江苏省自然科学基金资助项目(20190772)。
关键词
图像配准
特征检测
深度学习
合成孔径雷达
光学图像
image correspondence
feature detection
deep learning
synthetic aperture radar(SAR)
optical image
