Reconstruction of 3D permittivity profile of a dielectric sample using artificial neural network mathematical model and FDTD simulation

Abstract

The paper presents a new method of determining 3D permittivity profile using electromagnetic measurements in the closed waveguide system. The method is based on the application of artificial neural network as a numerical inverter, and on the approximation of 3D profile with quadratic polynomial function. The neural network is trained with numerical data obtained with FDTD modeling of the electromagnetic system. Special criteria for choice of a number of hidden layer neurons are presented. The results of numerical modeling show possibility of determination of permittivity profile with a relative error less than 10%.