Robust Interferometric Phase Estimation in InSAR via Joint Subspace Projection

Hai Li, Renbiao WuInTech 2012 · DOI · OpenAlex

Abstract

Synthetic aperture radar interferometry (InSAR) is an important remote sensing technique to retrieve the terrain digital elevation model (DEM). Image coregistration, InSAR interferometric phase estimation (or noise filtering) and interferometric phase unwrapping are three key processing procedures of InSAR. It is well known that the performance of interferometric phase estimation suffers seriously from poor image coregistration. This paper introduces a robust method for estimating interferometric phase in InSAR that remains accurate even when satellite images are poorly aligned.

Introduction

In Synthetic Aperture Radar Interferometry (InSAR), terrain height is calculated by comparing two radar images of the same area taken from slightly different positions. For this process to be effective, the images must undergo coregistration—a precise alignment of pixels. Even a misalignment of half a pixel can cause the correlation between images to drop significantly, resulting in an extremely noisy interferogram.

Conventional noise-reduction filters, such as pivoting mean or median filters, often fail when coregistration is poor. This research proposes a "joint subspace projection" technique to bridge this gap, enabling accurate phase estimation even with coregistration errors as large as one full pixel, a scenario where traditional filters typically yield unusable results.

Methodology: Joint Subspace Projection

The proposed method shifts from analyzing individual pixel pairs to a Joint Subspace Projection approach that captures spatial correlation information more effectively.

Joint Data Vector Formulation

Instead of processing pixels in isolation, the algorithm constructs a joint data vector from a neighborhood of pixels across both satellite images (Sat1 and Sat2). This "joint" approach allows the algorithm to statistically compensate for misalignment rather than requiring perfect physical alignment beforehand.

Illustration of the joint data vector construction from neighboring pixels in two satellite images.

Subspace Decomposition and Invariance

The core of the method involves performing an eigendecomposition of the covariance matrix to separate the "signal subspace" from the "noise subspace." A key innovation is the specific formulation of the joint data vector, which ensures that the noise subspace dimension remains constant regardless of the alignment error. By exploiting the fact that the eigenspectrum is invariant to the permutation of elements in the data vector, the algorithm's settings (specifically the noise subspace dimension) become independent of the coregistration error.

Experimental Validation

The method was validated using both simulated and real-world datasets:

  • Simulated InSAR Data: Modeled after a cartwheel formation of two satellites with a 281.46m baseline and an 18dB Signal-to-Noise Ratio (SNR).
  • Real-World Data: Data from the ERS-1/ERS-2 tandem satellites were used to test the algorithm's performance against actual Earth terrain complexities, providing a "ground truth" test for fringe clarity.

Key Results and Discussion

The proposed method demonstrates remarkable stability compared to traditional filtering techniques. At a 1-pixel coregistration error, while standard filters produce "salt and pepper" noise, this method maintains clean and distinct phase fringes.

Eigenspectrum analysis showing the behavior of eigenvalues under different registration error conditions (0, 0.5, and 1 pixel).

As shown in the eigenspectrum analysis, the distribution of eigenvalues remains predictable across various error levels. For a 20x20 joint covariance matrix, the authors proved the noise subspace dimension remains exactly 4 across all tested error ranges (0 to 1 pixel). This invariance allows for a simplified implementation where the algorithm's parameters do not need to be adjusted for different levels of misalignment. When applied to real ERS-1/2 data, the method significantly improved fringe contrast compared to conventional processing.

Conclusion and Future Work

The joint subspace projection method provides a robust alternative to fragile traditional filters, effectively compensating for coregistration errors statistically. While the current approach assumes "locally flat" terrain, future work could explore its application in more rugged environments or integrate it directly into automated coregistration software to perform alignment and phase estimation simultaneously in a single robust step.

Summary updated

Abstract

Synthetic aperture radar interferometry (InSAR) is an important remote sensing technique to retrieve the terrain digital elevation model (DEM)[1][2]. Image coregistration, InSAR interferometric phase estimation (or noise filtering) and interferometric phase unwrapping[3][4][5][6] are three key processing procedures of InSAR. It is well known that the performance of interferometric phase estimation suffers seriously from poor image coregistration.
BibTeX
@inproceedings{li2012robust,
  author = {Hai Li and Renbiao Wu},
  title = {{Robust Interferometric Phase Estimation in InSAR via Joint Subspace Projection}},
  booktitle = {InTech},
  year = {2012},
  doi = {10.5772/36820},
}
Figure