An Assisted Method for Multitemporal SAR Image Registration

Abstract

Precise registration of multitemporal synthetic aperture radar (SAR) images is a crucial step in Interferometric SAR (InSAR) data processing and serves as the foundation for high-precision interferometric measurements. Regular SAR image registration methods rely on the coherence between images. However, when faced with decorrelation issues, these methods often fail to yield high-precision registration results, adversely affecting subsequent data processing and interferogram quality. In this article, we propose an assisted method for multitemporal SAR image registration that addresses the challenge. By introducing auxiliary scenes with favorable coherence conditions alongside the primary and secondary images, we establish a mathematical model for the assisted registration method based on geometric relationships. The registration precision of the assisted registration method is evaluated using three indicators: 1) consistency checks; 2) interferogram fringe quality; and 3) coherence coefficient distribution. Sentinel-1 SAR images of the mountainous area in southeastern China were used for the experiment, and results show that the offsets calculated using assisted registration method exhibit greater concentration, and root mean square errors (RMSEs) demonstrate improved accuracy in both range and azimuth directions compared to the regular method, with enhancements of 25.6% and 23.3%, respectively. Additionally, interferograms obtained from the assisted registration show clearer and more complete fringes in regions with low coherence. Notably, the number of samples with coherence coefficients exceeding 0.4 increased significantly by 58.1% in the assisted registration results. While the accuracy of the proposed assisted registration method is comparable to that of regular methods under high-quality conditions, it shows marked advantages in scenarios characterized by severe decorrelation.