Application of statistical models for change detection in SAR imagery

Synthetic Aperture Radar (SAR) imagery is capable of monitoring Earth surface on a massif scale with high resolution. New generation of SAR satellites such as RADARSAT-2, TerraSAR-X, Cosmo-SKymed, etc. with metric resolution and high frequency bands (C-band, Xband) provide the possibility to better characterize surface objects. Moreover, the short revisit time of these satellites allows us to capture time series of images on the same region, which enables the development of change detection techniques and their applications. The Spherically Invariant Random Vector (SIRV) model was designed specifically for the analysis of heterogeneous clutters in high resolution radar images. In this paper, we study four algorithms of change detection based on different criteria including: Gaussian (sample covariance matrix estimator), Gaussian (fixed point estimator), Fisher texture-based and KummerU-based (Fisher distributed texture along with SIRV model). These algorithms are evaluated through simulated dataset and radar images from TerraSAR-X satellite.