Dissertation/Thesis Abstract

Performance of an SVM Classifier on Non-invasive EEG for a 2-State BCI Controller
by Samayamantri, Krishna Vijay, M.S., California State University, Los Angeles, 2019, 160; 13880546
Abstract (Summary)

Independent ambulation is a challenging problem for quadriplegics, since most have little functional muscle for an external human-machine interface (HMI), whereas invasive controls are expensive and dangerous. A viable option is the use of surface electroencephalography (s-EEG) based Brain Computer Interface (BCI) that can be achieved using affordable mobile EEG devices like the Emotiv EPOC. This, along with modern machine learning techniques, can allow the implementation of a fully functional BCI to establish an external control system for an assistive device, such as a prosthetic arm or an electric wheelchair.

Multiple methods that were designed for controlling a wheelchair were explored. The main differences observed were in the number of electrodes, their configuration, number of controls, and the methods of feature extraction and classification. The work focused on implementing and comparing two real-time BCI implementations of a 2-state BCI based on linear-support vector machines (linear SVM) classifier. Its performance was analyzed on two s-EEG databases: Physionet EEG Motor Movement/Imagery dataset and EEG data collected in our lab using the Emotiv EPOC BCI device. The Physionet data contains 64 channels, while the Emotiv data contain 14 channels. The accuracy and speed were compared for motor intent and motor action for both the electrode configurations.

A comparison of the differences in covariance matrix of the transformed data between the two input classes revealed that the common spatial patterns feature extraction approach led up to 50\% greater differences in features than that of the common spatial frequency subspace decomposition.

Indexing (document details)
Advisor: Won, Deborah
Commitee: Mondin, Marina, Daneshgaran, Fereydoun
School: California State University, Los Angeles
Department: Electrical and Computer Engineering
School Location: United States -- California
Source: MAI 81/2(E), Masters Abstracts International
Source Type: DISSERTATION
Subjects: Electrical engineering, Biomedical engineering, Engineering
Keywords: Brain computer interface (BCI), Common spatial patterns, Electroencephalography (EEG), Linear support vector machines (Linear-SVM), Machine learning, Wavelet transform
Publication Number: 13880546
ISBN: 9781085599450
Copyright © 2019 ProQuest LLC. All rights reserved. Terms and Conditions Privacy Policy Cookie Policy
ProQuest