A Robust Probabilistic Generative Method for Multitemporal Interferometric Phase Reconstruction: RPGSAR

To improve the quality of products obtained by interferometric phase signal time-series analysis of synthetic aperture radar (TS-InSAR) techniques, various phase reconstruction approaches have been developed recently. The commonly used methods assume that synthetic aperture radar (SAR) observations are complex circular Gaussian (CCG) distributed, making them vulnerable to outliers. Noting this, a robust probabilistic generative SAR (RPGSAR) interferometric phase reconstruction method, named RPGSAR, is proposed in this article. The RPGSAR introduces a multivariate complex t-distribution (MCT) to a probabilistic generative model, which establishes a projection matrix depicting statistical behaviors of reconstructed phase. Furthermore, to resolve this model, a targeted expectation maximization (EM) algorithm is deduced and implemented. It is worth noting that the degree of freedom (DOF) can be automatically estimated by the RPGSAR. To validate the RPGSAR, we first apply the phase linking (PL) and principal components analysis (PCA) methods to simulated data. The experimental results demonstrated the RPGSAR’s ability to handle outliers. Furthermore, real Sentinel-1A SAR stacks consisting of different number of images are reconstructed by these methods. The results show that the proposed method can provide more consistent interferometric phases in all stack size cases compared with the PL and PCA methods, thus further confirming its effectiveness.