Surface parameter retrieval using fast learning neural networks

Publisher Website
Google Scholar
Abstract
The application of neural networks to the problem of retrieval of surface electrical and roughness parameters is examined. To illustrate the usefulness of the first training algorithm developed here, simulated data sets based on a surface scattering model were used so that the data may be viewed as taken from a completely known randomly rough surface. A multilayer perceptron (MLP) trained with fast training (the FL network) and back‐propagation (referred to as the BP network) algorithms were tested on the simulated backscattering data. Sensitivity studies were made regarding the number of independent angles and polarizations necessary for the extraction of the surface parameters. The RMS error of training the FL network was found to be less than one half the error of the BP network while requiring one to two orders of magnitude less CPU time for training. When applied to inversion of parameters from a statistically rough surface, the FL network was successful at recovering the layer permittivity, surface correlation length, and the rms surface height while consistently having less RMS retrieval error than the BP network. Further applications of the FL neural network to the inversion of parameters of an inhomogeneous layer above a half space, using available backscatter measurements, are also shown.

This publication has 12 references indexed in Scilit: