Title: Motor-imagery EEG signal classification using position matching and vector quantisation

Authors: Tae-Ung Jang; Byeong Man Kim; Yeon-Mo Yang; Wansu Lim; Duk-Hwan Oh

Addresses: Department of Software Engineering, 1 Yangho-dong, Gumi-si, Gyeongsangbuk-do, Korea ' Department of Software Engineering, 1 Yangho-dong, Gumi-si, Gyeongsangbuk-do, Korea ' School of Electronic Engineering, 1 Yangho-dong, Gumi-si, Gyeongsangbuk-do, Korea ' School of Electronic Engineering, 1 Yangho-dong, Gumi-si, Gyeongsangbuk-do, Korea ' Department of Software Engineering, 1 Yangho-dong, Gumi-si, Gyeongsangbuk-do, Korea

Abstract: This paper proposes a motor-imagery electroencephalograph (EEG) signal classification method using vector quantisation and position matching. An EEG signal is transformed into a sequence of feature vectors which are extracted from the signal by short-time Fourier transform (STFT). A sequence of feature vectors, in turn, is transformed into code index sequence using the code book which is created by vector quantisation through Linde-Buzo-Gray (LBG) algorithm. Test signals are classified by K-nearest neighbour (KNN) where similarity among vectors is obtained by position matching. Brain computer interface (BCI) competition 2003 dataset III was used in the proposed method to perform the classification test and the maximum performance of 83.57% was obtained.

Keywords: brain computer interface; BCI; short-time Fourier transform; STFT; signal classification; vector quantisation; position matching; inverse document frequency; motor imagery; EEG signals; electroencephalograms; feature extraction; K-nearest neighbour; kNN.

DOI: 10.1504/IJTMCP.2016.078426

International Journal of Telemedicine and Clinical Practices, 2016 Vol.1 No.4, pp.306 - 313

Received: 07 Sep 2015
Accepted: 18 Sep 2015

Published online: 17 Aug 2016 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article