Back-Projection based Image Reconstruction for Time of Flight based Low-Permittivity Radar Tomography

Low-permittivity tomography is a material characterization technique that enables twodimensional imaging of materials, such as gases or industrial foams in spatially confined areas. Due to their favorable performance-to-cost ratio, fully integrated automotive FMCW radar MMICs have been identified as suitable candidates for this application. Building upon this foundation, a low-permittivity tomography system was developed in a prior study, employing regularization-based methods for tomographic reconstruction. However, certain reconstructions obtained from this system exhibit significant artifacts.
This work aims to identify the underlying causes of these artifacts and propose mitigation strategies. To this end, the filtered back-projection (FBP) reconstruction method is implemented. Due to the system’s suboptimal antenna placement, a novel re-binning algorithm is introduced, which enables the application of the FBP reconstruction method for tomographic systems with arbitrary antenna arrangements. This approach facilitates the identification of physically unfeasible measurement values as primary contributors to the observed reconstruction artifacts. To address these issues, an alternative phase unwrapping approach is explored. Furthermore, total variation methods are investigated for sinogram de-noising and outlier removal. Classical sinogram interpolation and filtering techniques are examined to mitigate the impact of erroneous measurement values on reconstruction quality. Additionally, a convolutional neural network-based approach is explored.
The proposed methods are evaluated on four distinct measurement configurations constructed from XPS foam blocks of various shapes