A STUDY ON ADVANCEMENT OF INTERNAL DEFECT ESTIMATION METHOD FROM RADAR IMAGES USING GAN AND FDTD METHOD

Abstract

This paper presents a study on the effective use of simulation, which enables efficient data acquisition, in the method proposed by the authors for estimating and visualizing cracks inside concrete from radar images using the adversarial generation network (GAN). Specifically, we proposed a method that can use simulated data as training data even if there are some discrepancies between experimental and simulated results by constructing an additional model that converts experimental radar images into pseudo-simulated radar images using pix2pix, which is a kind of applied technique of GAN. As a result of validation using several example problems, it is found that the proposed method improves the estimation accuracy in the case of thin defects where the reflected wave is small compared to the conventional method that uses only experimental data.