Answers are grounded in this paper and cite their sources.
In conditions of extremely low signal-to-noise ratio and complex decorrelation, the interferometric synthetic aperture radar (InSAR) phase is often severely corrupted by dense noise. Existing filtering methods struggle to achieve an effective balance among noise suppression, high-frequency topological detail preservation, and the avoidance of nonphysical spurious fringes, thereby constraining the accuracy of subsequent phase unwrapping. To address this issue, this article proposes a multiscale empirical fringe-manifold adaptive filtering (MEFMA) framework. This algorithm constructs a V-cycle-like phase decomposition and cascaded reconstruction framework, which is primarily driven alternately by two nonlinear operators. The first is a spatial-frequency adaptive operator based on Hoyer’s sparsity, enabling the data-driven extraction of local fringe features. The second is a complex gradient constant phase reconstruction smoothing operator, which effectively filters out broadband noise while preserving high-frequency deformation gradients without relying on heavy subspace eigen-decomposition. Furthermore, the algorithm introduces a coherence-based cascaded weighting and terminal gradient smoothing mechanism to suppress the nonlinear propagation and accumulation of errors. Experiments based on complex simulated data and real spaceborne SAR data from the Myanmar earthquake, the Turkey earthquake, and the Etna volcano demonstrate that the MEFMA algorithm outperforms existing comparative methods in metrics such as structural similarity and root mean square error. The proposed algorithm can successfully reconstruct high-frequency dense fringes in low-coherence regions, significantly reduce the number of residue points, and effectively mitigate unwrapped phase jumps caused by topological discontinuities. Consequently, it provides reliable data support for high-precision InSAR inversions in complex geological disaster environments.
No takes yet. Share an insight, caveat, or question.
Li et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: