Spaceborne Staring Spotlight SAR Tomography - A First Demonstration With TerraSAR-X

Nan Ge, Fernando Rodríguez González, Yuanyuan Wang, Yilei Shi, Xiao Xiang ZhuIEEE JSTARS 2018 · DOI

TL;DR

This paper demonstrates the first successful application of SAR tomography (TomoSAR) using TerraSAR-X's "staring spotlight" mode. This new mode significantly improves azimuth resolution, leading to denser 3D point clouds and more accurate height estimations, paving the way for enhanced urban mapping and infrastructure monitoring.

What Problem Does This Paper Address?

Traditional spaceborne Synthetic Aperture Radar (SAR) systems, while powerful, have limitations in resolving fine details, especially in complex urban environments. SAR tomography aims to overcome the inherent 2D imaging limitation of SAR by exploiting multiple acquisitions to reconstruct a 3D reflectivity profile, effectively estimating elevation and motion. However, achieving high-resolution 3D information from SAR data, particularly for urban structures, has been challenging. The authors address the gap in demonstrating the potential of TerraSAR-X's novel "staring spotlight" mode for TomoSAR. This mode offers a significantly higher azimuth resolution (0.24m vs. 1.1m in sliding spotlight) compared to previous modes, promising denser point clouds and improved accuracy, but its suitability for TomoSAR was unproven.

Methods

The core of this work lies in adapting existing SAR tomography processing chains to the unique characteristics of the staring spotlight mode. Key adaptations include:

  • Tailored Interferometric and Tomographic Processors: The authors modified their processors to handle the specific Doppler centroid frequency variations and the trade-off between azimuth resolution and scene extent inherent in staring spotlight data.
  • Doppler Centroid Frequency (fDC) Handling: Unlike sliding spotlight, the staring spotlight mode requires a different approach to time conversion for fDC. The paper leverages a 3x3 grid of fDC values provided in TerraSAR-X annotations to accurately account for second-order variations.
  • Sidelobe Detection and APS Compensation: To improve the reliability of TomoSAR results, the processing chain incorporates a step for detecting and mitigating sidelobes, which can corrupt measurements. Atmospheric Phase Screen (APS) is then compensated for pixels deemed reliable.
  • Spectrum Estimation and Model Selection: Both Tikhonov regularization (for general scene estimation) and L1 regularization (for sparsity-driven estimation, particularly in urban areas) are employed to estimate the elevation-motion spectrum. Model selection techniques are used to estimate the number of scatterers per resolution cell.
  • Off-grid Correction and Outlier Rejection: To address potential inaccuracies due to discretizing parameters, an oversampling approach is used to refine scatterer locations. An ensemble coherence metric is used to identify and reject outliers.
  • ADMM-based Solver: A novel, fast solver based on the Alternating Direction Method of Multipliers (ADMM) is developed to efficiently estimate the plane fitting parameters for relative height accuracy assessment, crucial for large datasets.

Datasets

The primary dataset used is from the TerraSAR-X satellite, specifically focusing on acquisitions in Washington, D.C. and Las Vegas.

  • Washington, D.C. Dataset: A stack of 41 staring spotlight images acquired between July 2014 and November 2016 was used for the first practical demonstration of TomoSAR. This dataset allowed for the estimation of topography and deformation parameters for individual infrastructures and the city scale.
  • Las Vegas Dataset: Two small descending interferometric stacks (12 images each) from the TanDEM-X Science Phase (October 2014 - February 2015) were used for a preliminary comparison between sliding and staring spotlight TomoSAR. These datasets were chosen for their similar baseline distributions.

Key Insights and Ideas

  • Staring Spotlight's Superiority for TomoSAR: The enhanced azimuth resolution of staring spotlight mode directly translates to a denser point cloud and improved height accuracy, making it a highly promising mode for TomoSAR applications.
  • Resolving Layover Effects: The increased resolution allows for better separation of scatterers in layover situations (e.g., roof-facade interactions), leading to more accurate individual scatterer characterization.
  • Importance of Accurate Doppler Centroid Handling: The paper highlights that standard time conversion methods for Doppler centroid are insufficient for staring spotlight, necessitating the use of annotated fDC grids for accurate processing.
  • Sidelobe Mitigation is Crucial: The inclusion of sidelobe detection in the processing chain is vital for improving the reliability of TomoSAR results, especially in complex urban scenes.
  • ADMM for Efficient Height Accuracy Assessment: The development of an ADMM-based solver significantly speeds up the process of calculating relative height accuracy, enabling the analysis of large datasets.

Key Results

  • Point Cloud Density: The staring spotlight mode resulted in a point cloud density approximately 5.1-5.5 times higher than the sliding spotlight mode in the tested urban areas.
  • Relative Height Accuracy: The relative height accuracy of the staring spotlight point cloud was approximately 1.7 times higher than that of the sliding spotlight mode.
  • Improved SCR: Staring spotlight mode leads to higher Signal-to-Clutter Ratio (SCR), making point-like targets more distinct and improving the contrast in intensity maps.
  • Detailed Infrastructure Monitoring: The study successfully visualized and interpreted updated topography and deformation (linear and periodical) for various urban structures in Washington, D.C., including bridges, hotels, and towers.
  • Layover Resolution: The paper demonstrates the ability to resolve layover effects, such as roof-facade interactions, with high fidelity.

Limitations and Future Work

  • Dataset Availability: The authors acknowledge that a lack of datasets might have hindered previous TomoSAR applications with staring spotlight data.
  • Preliminary Comparison: The comparison between sliding and staring spotlight TomoSAR was based on a limited number of acquisitions and small areas, suggesting the need for more extensive studies.
  • Future Work: The authors suggest that multiple-snapshot TomoSAR approaches, such as those using adaptive neighborhood searches or incorporating building footprint information, could further benefit from the enhanced resolution of the staring spotlight mode, leading to even larger numbers of snapshots and improved estimation accuracy.

Summary updated

BibTeX
@article{ge2018spaceborne,
  author = {Nan Ge and Fernando Rodríguez González and Yuanyuan Wang and Yilei Shi and Xiao Xiang Zhu},
  title = {{Spaceborne Staring Spotlight SAR Tomography - A First Demonstration With TerraSAR-X}},
  journal = {IEEE JSTARS},
  year = {2018},
  doi = {10.1109/jstars.2018.2851752},
}

Loading related content...

Figure