Adaptive Sequential Phase Estimator Based on Nonconvex Sparsity Regularization and Strain Model

Phase decorrelation hampers the accuracy of distributed scatterer (DS) interferometry (DSI) in high-precision deformation monitoring. While several advanced phase linking (PL) techniques built upon the sample coherence matrix (SCM) have shown effectiveness in enhancing the signal-to-noise ratio (SNR), their performance significantly degrades under suboptimal SCM estimation, particularly in scenarios of fast decorrelation and near-zero coherence levels. This article presents an enhanced sequential phase estimator motivated by the degradation of theoretical accuracy due to pure noise-bearing interferograms in full-stack-exploiting PL schemes. In the proposed estimator, Bayesian ensemble theory is first employed to divide the full-stack data adaptively into ministacks based on the coherence pattern, rather than constant-size ministacks usually determined empirically. Building on this, the estimator introduces a nonconvex regularization-based sparse SCM by constraining the coherence matrix to have potential sparsity and low-rank (LR), which suppresses the influence of noisy interferometric pairs and improves SCM estimation. Moreover, we incorporate the deformation elasticity theory to provide additional spatial constraints on the reconstructed phase time series across adjacent pixels, realized by using the strain model to reduce phase discontinuity and further enhance the SNR. Experiments on the simulated and real Sentinel-1 images over a landslide-prone area in western Guizhou, China, demonstrate the effectiveness and superior performance of the new method.