Deep learning algorithm to generate real radar images

Abstract

This report discusses a novel way to generate GPR straight scans using generative adversarial networks (GANs). Two pix2pix GANs were developed to generate simulated B-scans from 2D trunk images and realistic B-scans from simulated B-scans. Simulated B-scans were obtained through gprMax models and real B-scans were obtained from measurement campaigns on real trunks. This method was shown to produce realistic images, similar to that of simulated and real measurements.