Neural networks trained by scattered electromagnetic data for GPR applications

This paper is intended to give a presentation of the first results we have achieved in the field of reconstructing buried objects starting from the data extracted from the time-domain waveform of the scattered electromagnetic field which can be available at the receiving terminals of a GPR equipment. To this aim, neural network methodologies are exploited to face the inverse scattering problem. The main attention has been devoted to identify sets of data that can be "easily" extracted from the transient field. These data, however, represent the "electromagnetic signature" of the investigated scenario and thus they be chosen to be closely related to the geometric as well as the dielectric properties of the target. Preliminary results are shown concerning the localisation of the target and the reconstruction both of its dimension and of its relative dielectric permittivity.