Minimum Spanning Tree Co-registration Approach for Time-Series Sentinel-1 TOPS Data

Abstract

Synthetic aperture radar (SAR) image co-registration is one of the essential steps for interferometric SAR processing and the following multitemporal analysis. All acquisitions from time-series images with the Sentinel-1 Terrain Observation by Progressive Scans (TOPS) model should be co-registrated to a common reference geometry with an accuracy of approximately 0.001 pixels. Such a high accuracy can be achieved by correcting the residual azimuth mis-registration using the Enhanced Spectral Diversity (ESD) technique. However, the performance of the ESD depends on the coherence of image pairs, especially over fast decorrelation areas. To improve the coherence between acquisitions, we develop a joint co-registration procedure based on the minimum-spanning-tree algorithm and iterative reweighted least squares. In order to further improve the accuracy of the azimuth mis-registration over low-coherence scenes, we present a coherence estimator by combining two consecutive bursts of SLC samples to reduce both bias and variance. The proposed method is tested over two low-coherence scenes and compared to the network-based ESD approach. The results from both synthetic and real data demonstrated the advantages of our method.