A Sparsity-Based InSAR Phase Denoising Algorithm Using Nonlocal Wavelet Shrinkage

Dongsheng Fang,Yong Wang,Xiaolei Lv,Xue Lin,Jiang Qian

doi:10.3390/rs8100830

Abstract

An interferometric synthetic aperture radar (InSAR) phase denoising algorithm using the local sparsity of wavelet coefficients and nonlocal similarity of grouped blocks was developed. From the Bayesian perspective, the double- l 1 norm regularization model that enforces the local and nonlocal sparsity constraints was used. Taking advantages of coefficients of the nonlocal similarity between group blocks for the wavelet shrinkage, the proposed algorithm effectively filtered the phase noise. Applying the method to simulated and acquired InSAR data, we obtained satisfactory results. In comparison, the algorithm outperformed several widely-used InSAR phase denoising approaches in terms of the number of residues, root-mean-square errors and other edge preservation indexes.

Highlights

In the data processing of interferometric synthetic aperture radar (InSAR), the quality of the retrieved interferometric phase determines the accuracy of final products such as the estimation of ground deformation and digital elevation model (DEM)
The nonlocal similar blocks of interferogram are clustered by grouping, and the overlapped blocks are shrunk in 2D wavelet domain by the nonlocal wavelet shrinkage function
The finest details shared by the grouped blocks and the essential unique features of each individual block are revealed

Summary

Introduction

In the data processing of interferometric synthetic aperture radar (InSAR), the quality of the retrieved interferometric phase determines the accuracy of final products such as the estimation of ground deformation and digital elevation model (DEM). The phase retrieval is strongly affected by the phase denoising and phase unwrapping. When corrupted random phase noise exists, the result after the phase unwrapping is generally unsatisfactory. The phase denoising using filtering is one of fundamental steps to obtain accurate phase estimation. Numerous filtering approaches in the spatial domain or transformed domain are developed for the denoising. The direct filtering methods [1,2] applied in the spatial domain may not preserve details of fringes the window direction-dependent [3] and size-dependent [4,5,6] methods are able to remedy the preservation difficulty to some degree. With the assumption that the true signal and noise could be separated in the frequency domain after transformation, the denoising is performed by suppressing part of the transformed coefficients. If the coherence is low or fringes are dense, the Goldstein filter [7]

Methods

Results

Discussion

Conclusion