Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

UnwrapDiff: Conditional Diffusion for Robust InSAR Phase Unwrapping

Yijia Song, Juliet Biggs, Alin Achim, Robert Popescu, Simon Orrego, Nantheera Anantrasirichai

TL;DR

The work tackles the challenge of phase unwrapping in InSAR under noise and decorrelation. It introduces UnwrapDiff, a conditional denoising diffusion model that uses SNAPHU outputs as a global prior to guide reconstruction, enabling both global consistency and local corrections. On a synthetic benchmark with atmospheric effects and multiple noise patterns, it outperforms traditional and learning-based baselines, achieving substantially lower NRMS error and showing robustness in difficult deformations like dyke intrusions. The method demonstrates promising generalization to real InSAR data, suggesting a practical approach for robust deformation monitoring and hazard assessment.

Abstract

Phase unwrapping is a fundamental problem in InSAR data processing, supporting geophysical applications such as deformation monitoring and hazard assessment. Its reliability is limited by noise and decorrelation in radar acquisitions, which makes accurate reconstruction of the deformation signal challenging. We propose a denoising diffusion probabilistic model (DDPM)-based framework for InSAR phase unwrapping, UnwrapDiff, in which the output of the traditional minimum cost flow algorithm (SNAPHU) is incorporated as conditional guidance. To evaluate robustness, we construct a synthetic dataset that incorporates atmospheric effects and diverse noise patterns, representative of realistic InSAR observations. Experiments show that the proposed model leverages the conditional prior while reducing the effect of diverse noise patterns, achieving on average a 10.11\% reduction in NRMSE compared to SNAPHU. It also achieves better reconstruction quality in difficult cases such as dyke intrusions.

UnwrapDiff: Conditional Diffusion for Robust InSAR Phase Unwrapping

TL;DR

The work tackles the challenge of phase unwrapping in InSAR under noise and decorrelation. It introduces UnwrapDiff, a conditional denoising diffusion model that uses SNAPHU outputs as a global prior to guide reconstruction, enabling both global consistency and local corrections. On a synthetic benchmark with atmospheric effects and multiple noise patterns, it outperforms traditional and learning-based baselines, achieving substantially lower NRMS error and showing robustness in difficult deformations like dyke intrusions. The method demonstrates promising generalization to real InSAR data, suggesting a practical approach for robust deformation monitoring and hazard assessment.

Abstract

Phase unwrapping is a fundamental problem in InSAR data processing, supporting geophysical applications such as deformation monitoring and hazard assessment. Its reliability is limited by noise and decorrelation in radar acquisitions, which makes accurate reconstruction of the deformation signal challenging. We propose a denoising diffusion probabilistic model (DDPM)-based framework for InSAR phase unwrapping, UnwrapDiff, in which the output of the traditional minimum cost flow algorithm (SNAPHU) is incorporated as conditional guidance. To evaluate robustness, we construct a synthetic dataset that incorporates atmospheric effects and diverse noise patterns, representative of realistic InSAR observations. Experiments show that the proposed model leverages the conditional prior while reducing the effect of diverse noise patterns, achieving on average a 10.11\% reduction in NRMSE compared to SNAPHU. It also achieves better reconstruction quality in difficult cases such as dyke intrusions.

Paper Structure

This paper contains 11 sections, 5 equations, 4 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. Preliminaries
  4. Conditional Diffusion Framework
  5. Training and Loss Functions
  6. Experimental results
  7. Dataset
  8. Comparative Evaluation
  9. Robustness Analysis
  10. Ablations Study
  11. conclusions

Figures (4)

  • Figure 1: Unwrapping results of a dyke intrusion scene with water body patches, shown as dark blue regions in the wrapped interferogram.
  • Figure 2: Diagram of the proposed UnwrapDiff.
  • Figure 3: Unwrapping results (top row) on a simulated dyke intrusion, with the ground truth shown in (a) and the input wrapped interferogram shown in (h). Corresponding error maps are shown in the bottom row.
  • Figure 4: Unwrapped results from real wrapped interferogram of the 2020 Humahuaca earthquake, Argentina.