A statistically homogeneous pixel selection approach for adaptive estimation of multitemporal InSAR covariance matrix

Multitemporal interferometric synthetic aperture radar (InSAR) technology is extensively applied in earth observations. As a critical processing step, the estimation of covariance matrix directly affects the accuracy of its final result. Adaptive multilooking is proven to be an efficient estimator employing statistically homogeneous pixels (SHPs). In this context, the SqueeSAR initially introduced the Kolmogorov-Smirnov test to select SHPs. Furthermore, many test methods were developed based on real-valued data. Therefore, only amplitude is used and complex coherence is typically ignored. In this paper, a new SHP selection algorithm is proposed based on covariance matrix patch, named CMP. In particular, the complex patch is estimated using temporal samples rather than spatial samples. The Kullback-Leibler divergence is employed to evaluate the similarity between two complex patches. Then, the SHP set is initially determined by performing thresholding operation, and an iterative methodology is used to obtain a refined SHP set. Further application of CMP to estimate covariance matrix is investigated. The derived products include the filtered amplitude, interferometric phase and coherence. A series of experiments show the effectiveness of the proposed SHP selection method, especially suitable for cases with only interferometric phase differences. Moreover, due to the high computational cost, this method is generally recommended for small-sized stacks. The experimental results of filtered images also suggest its high performance in both noise suppression and detail preservation.