A new brain-robot interface system based on SVM-PSO classifier

Abstract

This paper presents a new noninvasive brain-robot interface system for control of two degrees of freedom robot through motor imagery EEG signals. Signal classification is based on optimized Support Vector Machine (SVM) by Particle Swarm Optimization (PSO) algorithm. EEG signals of FC3, C3, CP3, FC4, C4 and CP4 Channels that are related to hands movement as well as Cz and FCz channels that are related to feet movement are considered. Radial basis function (RBF) and penalty functions of SVM are optimized through PSO algorithm. For validation of SVM-PSO classifier, the EEG signals are collected from two databases: PhysioNet and BCI Competition III, then features including Power Spectral Density (PSD) and wavelet parameters are used as the input of the classifier. By comparing the results of the SVM and SVM-PSO classifiers, is concluded that performance of classifier in terms of accuracy is increased through PSO algorithm. SVM-PSO classification accuracy for wavelet and PSD features are obtained 81% and 92%, respectively. The best algorithm is used to control a two degrees of freedom (one for left and right hand movements and the other for left and right foot movements) industrial robot experimentally. It shows the applicability and effectiveness of proposed method for high accuracy brain-robot interface systems.