Modified statistically homogeneous pixel selection for use with a non-local adaptive multilook approach for small SAR datasets

Abstract

ABSTRACT Multitemporal interferometric synthetic aperture radar (MT-InSAR) technique is an important approach for surface deformation monitoring. Nonlocal adaptive multilooking approach has potential benefits for the filtering and coherence estimation in the data preprocessing steps of MT-InSAR technique. The kernel of nonlocal adaptive multilooking approach lies in the selection of statistically homogeneous pixels (SHPs). Various amplitude-based strategies, such as DespecKS and its variations, have been proposed for selecting SHPs. However, the detection rates of these methods are usually unsatisfactory in the case of small data sets. To overcome this limitation, SHPs are selected based on the adaptive joint data vector, which encompasses both time dimensional samples and spatial information. In this letter, the outliers of the time dimensional samples are removed first based on a revised boxplot method. Additionally, image segmentation and L1 norm are both introduced to adaptively construct joint data vector in the preset window. Experiments on five TSX images are used to verify the reliability and efficiency of the proposed algorithm.