Reducing the training time of the brain-controlled music player

Abstract

A brain computer interface (BCI) system is a non-muscular communication link between the user’s brain activities and the surrounding environment. A BCI system detects the executed command by processing the brain signals recorded using electroencephalography (EEG) which is a non-invasive technique used to record the electrical activity present in the brain. The BCI system considered in this project is the brain-controlled music player developed by Zerafa. This is a steady state visual evoked potential (SSVEP)-based BCI system which presents flickering stimuli where each stimuli is flickering at a different frequency representing the different commands. For the user to select a command, the user must attend to one of the stimuli, causing the brain signals in the occipital area to synchronise with the flickering stimulus. For the BCI system to identify the user’s brain signals and assign them to the related command, a training session is required to study the nature of the signals of each individual. The current music player application requires two training sessions; one to find each subject’s optimal parameters and the other to train the classifier to distinguish between the different classes. The current training time required before using the music player amounts to 21 minutes, which is quite long to make this application feasible for everyday use. Therefore this project addresses the issue of training by investigating the possible techniques that can be implemented to reduce the training time of this application. A detailed analysis of the structure, the nature and the length of the training sessions of this application is initially done. This is followed by a review of the methods used for feature extraction and classification in order to get a better understanding on the nature of the features and how they are distinguished. The methods that can be used to reduce the training time are then identified, some of them are derived from the music player application by eliminating one or some parts of the training sessions or by considering the use of common set-up parameters across the subjects. Other methods are derived from the literature review conducted on the methods developed to reduce training time in BCI systems. This leads to the testing a different classifier, mainly through the use of prototypes feature vectors and the use of a generalized classifier. Finally one of the implemented techniques which showed no statistical significance in performance when compared to the current system can be implemented and the performance of the music player in real time can be tested. The tests done in view of this confirmed the offline analysis and therefore this project succeeded in reducing the training time of the brain controlled music player by 8.17 minutes which is equivalent to 38.90 per cent.