Cinderella: Method Generalisation of the Elimination Process to Filter Repeating Patterns

Abstract

Synthetic aperture radar is a popular remote sensing technique to observe the topography of the earth by producing interferograms. After initial corrections, there remains a mixture of different disturbances visible in them. Therefore, an estimation is necessary to distinguish between patterns caused by different sources. Commonly, this is achieved by using Persistent Scatterer Interferometry (PSI) or Small BAseline Subset Algorithm (SBAS) which both utilise large stacks of temporally connected interferograms of the same area. The objective of this paper is to introduce a new technique that can also handle small stacks and temporally unconnected subsets, and does not use the isotropy assumption. The introduced estimation technique, called Cinderella, is related to the elimination process and therefore belongs to the Bionic research area. The proposed technique works as follows. First, the initial images are transferred into the dual tree complex wavelet domain. Second, the Cinderella technique is applied for each complex coefficient and estimates a percentage affiliation of the corresponding wavelet to different causes. Third, the Tikhonov regularisation uses this estimate and provides the resulting assignment. Finally, the synthesis of the dual tree complex domains provides the images of all estimated signals. Verification is achieved by simulating random patterns, mixing them, applying the proposed technique, and comparing the estimates with the initial simulations. In doing so, only six interferograms were required to provide an estimation with a standard deviation below 1 mm.