PolInSAR temporal series analysis with Binary Partition Trees

Abstract

This work is focused on the exploitation of PolSAR and PolInSAR temporal series datasets, in the context of change detection and characterization. Instead of a classical pixel-based approach, a Binary Partition Tree (BPT) data structure is employed in order to perform a region-based processing of the image. The homogeneous regions of the image are obtained, which are characterized by containing pixels following a similar polarimetric temporal evolution. Then the temporal dimension of the data is analyzed firstly to quantify the significance of the polarimetric temporal variation, making possible the detection of scene changes, and secondly to analyze and characterize those changes. In this work, the PolInSAR information will also be exploited for this purpose, in order to achieve a more detailed characterization of change and no-change areas. Finally, the proposed technique is employed to process a real RADARSAT-2 dataset to show its capabilities and potentialities.