On the Use of Hybrid Neuro-fuzzy Systems for the Solution of Inverse Radiative Transfer Problems

Abstract. Inverse problems related to the interaction of radiation with participating media have been attracting the attention of various researchers for many years due to the relevant applications not only in engineering, but also in astrophysics, physical oceanography (hydrologic optics), remote sensing and atmosphere/hydrosphere optics among many others. In this work the determination the single scattering albedo, optical thickness and diffuse reflectivities in one-dimensional homogeneous participating media is investigated using neuro-fuzzy networks, and a hybrid solution of neuro-fuzzy networks with artificial neural networks. Such neural networks are massively distributed and parallel structures, inspired upon the functioning of the human brain, able to implemented in software and hardware, and they try to reproduce the dynamics of biological networks. As they are developed through algorithmic numerals, the knowledge which is symbolic remains represented by the network in numerical form. Inverse problems are very susceptible to the error which is always present in experimental data, being therefore an essential part of its formulations and solutions. The fuzzy logic, which is based upon the ability of human beings to deal with inexact, imprecise and vague information, gives us a tool that may be helpful to handle with the experimental uncertainty. Here we use hybrid neuro-fuzzy systems for the solution of inverse problems in radiative transfer.