A Multifrequency SAR Tomography Method via Mixed-Norm and Deep Unfolding Network

Abstract

Tomographic synthetic aperture radar (TomoSAR) enables 3-D imaging of observed scenes while typically requiring numerous flight passes to achieve high-quality imaging and avoid elevation ambiguities. Recently developed multifrequency TomoSAR (MF-TomoSAR) offers a promising approach to reduce the required flight passes by expanding the effective spatial-frequency sampling in the elevation direction. However, existing MF-TomoSAR approaches fail to preserve frequency-dependent scattering information and are computationally demanding, as they lack explicit per-frequency scattering modeling and require tedious hyperparameter tuning. To address these limitations, this article proposes a mixed-norm multifrequency (MMFreq) TomoSAR imaging framework, and the main contributions are threefold. First, the MMFreq imaging model is proposed, which incorporates a mixed-norm regularization to jointly characterize cross-frequency scattering features while preserving per-frequency scattering information. Second, MMFreq-three-operator splitting (TOS) is derived as an efficient imaging algorithm based on the TOS framework, which provides closed-form update rules for the MMFreq model. Third, MMFreq-Net is developed as a deep unfolding network (DUN) that maps the MMFreq-TOS algorithm into a learnable architecture, achieving faster imaging and automatic parameter optimization. Finally, real-data experiments successfully yield unambiguous 3-D imaging results at each frequency, demonstrating the effectiveness of the proposed MMFreq framework.