Interface Cérebro-Computador Baseada em EEG Utilizando Redes Neurais Auto-Organizadas

Abstract

This Doctoral Thesis presents the development of a Brain Computer Inter-face (BCI) system using Electroencephalography (EEG) signals and Self OrganizingMaps (SOM) artificial neural networks as classifier. In this Thesis the problems ofa BCI are analyzed and the classification results of the system is presented. Thissystem uses a clinic acquisition equipment for EEG signal acquisition and a personalcomputer to process the data, using the energy of the frequency components of theEEG signal as characteristics and a classifier based on a Self Organizing Map asclassifier. The great challenge in using SOM as a classifier is the interpretation ofthe outputs of the map, as it has as many outputs as it has neurons in the map.The contribution of this Thesis is in the interpretation method of the outputs ofthe map, which is done by means of the use of a set of masks that represents theprobability of the activation of a neuron in the map representing a specific class.The algorithms used on this Doctoral Thesis can be easily adapted to be executed inembedded systems with less processing power, like Digital Signal Processors (DSP)or microcontrollers. The Brain Computer Interface developed in this Doctoral The-sis was tested and validated off–line, with an external database, and with data fromvolunteers, presenting satisfactory results in both cases, according to similar resultsfrom the literature.