MATHEMATICAL MODEL OF BRAIN-COMPUTER INTERFACE BASED ON THE ANALYSIS OF P300 EVENT RELATED POTENTIALS

Abstract

The evoked potentials (EP) method consists in recording bioelectric reactions of the brain in response to external stimulation or while performing cognitive tasks. The goal of the work is to develop a mathematical model of the system for detection and classification of evoked potentials on the electroencephalogram (EEG). The main odd of the machine EP detection are artifacts from EEG recordings and the high variability of potentials. EP detection and classification algorithm includes three stages. At the preliminary stage, the frequency-time and spatial signal transformations – a set of Butterworth frequency filters, linear composition and averaging of the recorded signals from different sensors are used to remove noise and uninformative EEG components. The next step is the direct fixation and averaging of the evoked potentials. At the final stage, to reduce the dimension of the problem, the information features vector is formed. The parameterized image is used as input of the binary classifier. The support vector method is used to construct the classifier. During the study, the optimization of the regularization C parameter of the classifier was carried out using the estimation of sliding control. The proposed solution is useful for human-machine interaction and for medical procedures with biofeedback.