Neural Network Inverse Method for Dielectric Structures by Multifrequency Signals

Abstract

Application of neural network ideology for layered dielectric structure inverse problem solution is under consideration. The 3-layered feed-forward with activation function as a sigmoid (S-shaped) was chosen. The real part of reflection coefficient at many frequencies is used as input signal. Levenberg-Marquardt algorithm was used for training of the network. Some results of numerical simulation for two-layered structure in presence of Gaussian additive noise are demonstrated