Improved variational Bayesian TomoSAR for height inversion

Abstract

Tomographic synthetic aperture radar (TomoSAR) extends conventional 2D SAR imaging by resolving vertical scattering centers, enabling detailed 3D structural analysis. However, its accuracy declines when multiple targets are located close together in elevation. To overcome this limitation, compressed sensing (CS) techniques have been introduced, which discretize the elevation axis to improve target localization. Building on this approach, we propose a new algorithm called Group-Outlier Variational Bayesian (GOVB). The proposed method incorporates group-based Bayesian inference and outlier detection mechanisms to more effectively manage nonideal conditions commonly found in real-world scenarios. Specifically, it demonstrates strong performance in environments with minor elevation differences among targets, where nonorthogonality and data inconsistencies are more prevalent. Experimental evaluations using real-world data demonstrate that GOVB remains robust and accurate under noise and sparse sampling. These results suggest that GOVB offers a reliable solution for reconstructing complex scenes in advanced SAR systems, especially where precise vertical resolution is essential.