Classification of thought evoked potentials for navigation and communication using multilayer neural network

Abstract

ABSTRACT In this paper, an intelligent classification system has been developed to command a robot chair by means of direct brain activity, aided by amplification. The intelligent system classifies seven fundamental tasks based on measuring ElectroEncephaloGraphic (EEG) brain activity. The seven tasks were used to control a robot chair and also to interact with others. In this analysis, a simple protocol for the EEG data acquisition procedure has been proposed to perform seven tasks based on thought evoked potentials (TEP’s). The evoked potentials were converted into control signals to navigate the robot chair and also to choose words/letters in an oddball paradigm for communication. In the EEG acquiring process, five volunteers participated and brain activities related to navigational movements (Forward, Left, and Right) and communication (Yes, No, and Help) were recorded from the volunteers to form the database. The acquired EEG signals are visually validated upon recording each trial and pre-processed to eliminate the noise contents. The pre-processed signals were segmented into six frequency bands to extract spectral band energy and spectral band centroid features. The extracted features were then formed to classify the tasks using a feed-forward Multilayer Neural Network algorithm to exhibit customized (subject wise) features. The trained models of the neural networks were compared to validate the classification results. From the results, it is observed that the Spectral centroid features have the highest classification rate of 98.50%.