Supervised Learning Applied to Microwave Imaging System Calibration

Near-field microwave imaging systems seek a quan-titative reconstruction of the permittivity in a region of interest. These systems measure raw S-parameters which contain the effects of antennas, cables, and other non-ideal aspects of the imaging system. Imaging algorithms use a computational forward model to make images, often assuming that fields are measured at points. In the absence of a sufficiently complex forward model that can model the physical system perfectly, these S-parameters need to be calibrated to facilitate the reconstruction. Herein, using a data set with three known targets over 300 collected S-parameter/electric field pairs, we evaluate a novel calibration method via supervised learning with a U-Net neural network.