A robust neural network tool for the identification of buried cylinders by a subsurface radar system

This paper is concerned with the application of a neural network based algorithm to face the electromagnetic inverse scattering problem of reconstructing dielectric cylinders with time-domain data, as those that can he available at the receiving terminals of a short-pulse GPR (Ground Penetrating Radar) [l] system. The algorithm can he suitably applied when some a-priori information on the problem are available. The exploitable data are the transient electromagnetic fields scattered by the buried target and collected by a receiver antenna. Some numerical results will he presented concerning the localization and the reconstruction of a circular cylinder in a 2D scenario. Moreover, the neural network robustness will he tested against noise and corruptions in the a-priori information.