A Robust Super-Resolution Gridless Imaging Framework for UAV-Borne SAR Tomography

Silin Gao, Wenlong Wang, Muhan Wang, Zhe Zhang, Zai Yang, Xiaolan Qiu, Bingchen Zhang, Yirong WuIEEE TGRS 2024 · DOI

TL;DR

This paper introduces a novel gridless 3D imaging framework for UAV-borne SAR tomography, called EMPAST. It addresses the "off-grid" effect in traditional methods by leveraging pulse repetition frequency (PRF) redundancy in UAVs to create multiple measurement vectors (MMVs) and employing an atomic norm soft thresholding algorithm for super-resolution and noise immunity.

What Problem Does This Paper Address?

Synthetic Aperture Radar (SAR) tomography (TomoSAR) is a powerful technique for creating 3D images by combining multiple 2D SAR images, effectively solving the layover problem common in urban mapping. Unmanned Aerial Vehicles (UAVs) are increasingly used as TomoSAR platforms due to their cost-effectiveness and flexibility. However, existing TomoSAR methods, especially those using compressed sensing (CS) with grid-based approaches, suffer from discretization errors that degrade reconstruction accuracy, known as the "off-grid" effect. While gridless CS methods exist, they often struggle with partial observations and multiple measurement vectors (MMVs) in the context of UAV-borne SAR, which has unique characteristics like slow flight speeds and sparse array elements. This paper aims to develop a robust, super-resolution, and noise-immune gridless 3D imaging framework specifically for UAV-borne TomoSAR.

Methods

The proposed framework, EMPAST, introduces several key innovations:

  • Extracted-MMV (EMMV) Construction: EMPAST leverages the pulse repetition frequency (PRF) redundancy inherent in slow UAV platforms. By extracting multiple measurement vectors (MMVs) from the oversampled data, the framework enhances noise immunity without sacrificing azimuth-range resolution. This is achieved by constructing sub-PRF measurements, effectively creating more data from the same acquisition.
  • Atomic Norm Soft Thresholding for Partially Observed MMV: A novel atomic norm soft thresholding algorithm is developed to handle the MMV scenario with partial observations. This algorithm is designed to overcome the "off-grid" effect by working in a continuous domain and is tailored for the sparse arrangement of array elements often found in UAV platforms.
  • ADMM-based Optimization: To address the computational complexity of solving large-scale semi-definite programming (SDP) problems associated with atomic norm minimization, an efficient alternative optimization solver using the Alternating Direction Method of Multipliers (ADMM) is proposed.

Datasets

The paper uses both simulated data and real data from the Microwave-Vision 3-D SAR (MV3DSAR) experimental system. The MV3DSAR system is a small UAV-borne array InSAR system. The flight experiment involved two 15-story buildings.

Key Insights and Ideas

  • PRF Redundancy is an Opportunity: Instead of a limitation, the slow speed of UAVs and the resulting PRF redundancy are exploited as an opportunity to generate multiple measurement vectors (MMVs), significantly improving noise resilience.
  • Gridless is Crucial for Off-Grid Targets: The "off-grid" effect, where targets don't align with discrete grid points, is a significant problem for traditional methods. The paper demonstrates that gridless approaches, particularly those based on atomic norms, are essential for accurate reconstruction of these targets.
  • MMV Enhances Robustness: By constructing multiple measurement vectors, the framework can better distinguish between signal and noise, leading to more robust reconstructions, especially in challenging low SNR conditions.
  • ADMM for Practical Efficiency: The computational burden of advanced optimization techniques like SDP is a bottleneck. Using ADMM provides a significant speedup, making the proposed framework more practical for real-world applications.

Key Results

  • Super-Resolution: EMPAST demonstrates superior super-resolution capabilities, enabling the separation of closely spaced scatterers that are indistinguishable by other methods, especially at lower SNRs. For example, EMPAST (L=8) achieved a super-resolution factor of 16.0 at 20 dB SNR, significantly outperforming other methods.
  • Accuracy and Robustness: EMPAST consistently outperforms SVD-Wiener, GBCS, and PAST in terms of normalized elevation estimation error (σs) and probability of detection (PD), particularly under low SNR conditions. For instance, at 10 dB SNR, EMPAST achieved a significantly lower σs compared to other methods.
  • 3D Point Cloud Quality: In practical demonstrations with real data, EMPAST produced clearer and more concise 3D point clouds with fewer false targets and outliers compared to other methods, accurately depicting building outlines and structures.
  • Efficiency: The ADMM-based solver for EMPAST showed significantly faster runtimes compared to SDP-based solvers, especially as the problem size (N+L) increased.

Limitations and Future Work

  • Parameter Selection: The paper notes that the selection of the regularization parameter (τ) and the number of EMMV (L) is crucial and depends on noise characteristics and system parameters. While guidance is provided, optimal selection might require problem-specific tuning.
  • High Accuracy at Low SNR: The paper mentions that achieving very high accuracy at low SNR might require a significant increase in iteration count for ADMM, potentially reducing efficiency. This suggests a trade-off that needs careful consideration.
  • Future Work: The authors suggest that further research could focus on optimizing ADMM parameters for a better trade-off between efficiency and accuracy, especially in challenging low SNR scenarios. The framework's applicability to other complex scenarios and its potential for real-time processing could also be explored.

Summary updated

BibTeX
@article{gao2024robust,
  author = {Silin Gao and Wenlong Wang and Muhan Wang and Zhe Zhang and Zai Yang and Xiaolan Qiu and Bingchen Zhang and Yirong Wu},
  title = {{A Robust Super-Resolution Gridless Imaging Framework for UAV-Borne SAR Tomography}},
  journal = {IEEE TGRS},
  year = {2024},
  doi = {10.1109/tgrs.2024.3393972},
}
Figure