Presentation Title

A Multiuser EEG Based Imaginary Motion Classification Using Neural Networks

Location

Room 2911

Session Format

Paper Presentation

Research Area Topic:

Engineering and Material Sciences - Electrical

Abstract

Using Electroencephalography (EEG) to detect imaginary motions from brain waves to interface human and computer is a very nascent and challenging field that started developing rapidly in the past few decades. This technique is termed as Brain Computer Interface (BCI). BCI is extremely important in case of people who are incapable of communicating due to spinal cord injury. This technique uses the brain signals to make decisions, control and communicate with the world using brain integration with peripheral devices and systems. In this paper, in order to classify imaginary motions, raw data are used to train a system of neural networks with a majority vote output. The network was trained using the scaled conjugate gradient back propagation algorithm. The novelty of this proposed approach is in using a majority vote system for a network of artificial neural networks (ANNs) that is used to optimally classify imaginary motions performed by multiple subjects. EEG data for 3 subjects are used from the BCI Competition III data set V. Each subject has data represented in three sessions comprising of three different types of imaginary motion tasks in each session. Using an optimized set of electrodes, classification accuracy was optimized for the three users as a group. A cross validation method is applied to improve the reliability of the generated results. The optimization resulted in an electrode structure consisting of 15 electrodes with a relatively high classification accuracy of almost 80%. It was observed that using a separate ANN for every channel coupled with a majority vote system was able to improve the average classification accuracy of such imaginary motion of all three users from a maximum 71% to almost 80% while maintaining a relatively simple ANN structure. In addition, the quality of the EEG signal generated by the users declined with time due to fatigue and loss of concentration. It was also concluded that the classification accuracy is user dependent in nature which limits it optimization for multiple subjects. The proposed method presented is novel in the structure of such classification network and in the optimization of channels for multi user EEG-based BCI system.

Presentation Type and Release Option

Presentation (Open Access)

Start Date

4-16-2016 1:30 PM

End Date

4-16-2016 2:30 PM

This document is currently not available here.

Share

COinS
 
Apr 16th, 1:30 PM Apr 16th, 2:30 PM

A Multiuser EEG Based Imaginary Motion Classification Using Neural Networks

Room 2911

Using Electroencephalography (EEG) to detect imaginary motions from brain waves to interface human and computer is a very nascent and challenging field that started developing rapidly in the past few decades. This technique is termed as Brain Computer Interface (BCI). BCI is extremely important in case of people who are incapable of communicating due to spinal cord injury. This technique uses the brain signals to make decisions, control and communicate with the world using brain integration with peripheral devices and systems. In this paper, in order to classify imaginary motions, raw data are used to train a system of neural networks with a majority vote output. The network was trained using the scaled conjugate gradient back propagation algorithm. The novelty of this proposed approach is in using a majority vote system for a network of artificial neural networks (ANNs) that is used to optimally classify imaginary motions performed by multiple subjects. EEG data for 3 subjects are used from the BCI Competition III data set V. Each subject has data represented in three sessions comprising of three different types of imaginary motion tasks in each session. Using an optimized set of electrodes, classification accuracy was optimized for the three users as a group. A cross validation method is applied to improve the reliability of the generated results. The optimization resulted in an electrode structure consisting of 15 electrodes with a relatively high classification accuracy of almost 80%. It was observed that using a separate ANN for every channel coupled with a majority vote system was able to improve the average classification accuracy of such imaginary motion of all three users from a maximum 71% to almost 80% while maintaining a relatively simple ANN structure. In addition, the quality of the EEG signal generated by the users declined with time due to fatigue and loss of concentration. It was also concluded that the classification accuracy is user dependent in nature which limits it optimization for multiple subjects. The proposed method presented is novel in the structure of such classification network and in the optimization of channels for multi user EEG-based BCI system.