eprintid: 9427
rev_number: 2
eprint_status: archive
userid: 1
dir: disk0/00/00/94/27
datestamp: 2023-11-09 16:21:24
lastmod: 2023-11-09 16:21:24
status_changed: 2023-11-09 16:15:06
type: article
metadata_visibility: show
creators_name: Abdalsalam, E.
creators_name: Yusoff, M.Z.
creators_name: Kamel, N.
creators_name: Malik, A.S.
creators_name: Mahmoud, D.
title: Classification of four class motor imagery for brain computer interface
ispublished: pub
keywords: Biomedical signal processing; Computer vision; Discrete wavelet transforms; Electroencephalography; Extraction; Feature extraction; Image classification; Interfaces (computer); Motion compensation; Multilayer neural networks; Multilayers; Nearest neighbor search; Radial basis function networks; Robotics; Signal reconstruction, Classification accuracy; Electroencephalogram (EEG) datum; Feature investigations; K nearest neighbor (KNN); K-nearest neighbors; Motor imagery classification; Motor imagery tasks; Multi layer perceptron, Brain computer interface
note: cited By 4; Conference of 9th International Conference on Robotic, Vision, Signal Processing and Power Applications, RoViSP 2016 ; Conference Date: 2 February 2016 Through 3 February 2016; Conference Code:184869
abstract: In this paper, four class motor imagery classification has been studied for brain computer interface. Feature investigations were conducted on the Enobio device, firstly with all 8 channels (F3, F4, T7, C3, C4, Cz, T8 and Pz) and subsequently with 3 selected channels (C4 left hand, C3 right hand, C3 and C4 both hand and Cz both feet) in alpha and beta rhythm in order to establish the active networks. Five volunteers were participated, the volunteers were instructed to perform motor imagery tasks, such as to imagine the opening and closing of the left and right hand, both hands, and both feet movement. Electroencephalogram (EEG) data were collected and offline signals processing were performed. Discrete wavelet transform (DWT) was used for feature extraction, while difference classifications methods such as multilayer perceptron (MLP), RBFNetwork, and K-Nearest Neighbors (KNN) were implemented. Best classification of MLP over KNN and RBFNetwork was noticed, whereas the highest accuracy was achieved at sym8 wavelet using DWT based feature extraction. On average over the subjects the selected channel accuracies were in the range of 86.61 . Whereas for all the channels, accuracies were in range of 78.37 . The study has shown that the classification accuracy can significantly improve by using specific channels for the EEG classification rather than using all EEG channels a time. © Springer Science+Business Media Singapore 2017.
date: 2017
publisher: Springer Verlag
official_url: https://www.scopus.com/inward/record.uri?eid=2-s2.0-84992688800&doi=10.1007%2f978-981-10-1721-6_32&partnerID=40&md5=4d0c87ad2644f45c6724e33dc96b2f44
id_number: 10.1007/978-981-10-1721-6₃₂
full_text_status: none
publication: Lecture Notes in Electrical Engineering
volume: 398
pagerange: 297-305
refereed: TRUE
isbn: 9789811017193
issn: 18761100
citation:   Abdalsalam, E. and Yusoff, M.Z. and Kamel, N. and Malik, A.S. and Mahmoud, D.  (2017) Classification of four class motor imagery for brain computer interface.  Lecture Notes in Electrical Engineering, 398.  pp. 297-305.  ISSN 18761100