Loading...
| Friend's email | |
| Your name | |
| Your email | |
| enter code | |
This page was sent successfuly
- Type of Document: M.Sc. Thesis
- Language: English
- Document No: 46341 (55)
- University: Sharif University of Technology, International Campus, Kish Island
- Department: Science and Engineering
- Advisor(s): Haj Sadeghi, Khosrow
- Abstract:
- Brain-inspired methods are now widely used to process the data generated by the brain with the aim of improving our understanding of how the brain functions and produces the remarkable intelligence exhibited by humans, which is the source of all realizations, perception and actions. Therefore brain-computer interface (BCI) is one of the most challenging scientific problems which focuses scientists attention, in most cases these systems are based on EEG signals recorded from the surface of the scalp because this method of the brain activity monitoring is noninvasive, easy to use and quit inexpensive. Brain computer interface (BCI) systems analyse the EEG signals and translate person’s intentions into simple commands. It can be used to allow paralyzed as well as healthy individuals to interact with and control the surrounding environment or to communicate simply by the conscious modulation of thought patterns. BCI interfaces make use of several brain potentials such as: P300, SSVEP or ERS/ERD. The most useful case for implementation up to now is BCI based on brain potentials associated with movements (ERD/ERS). The ERD/ERS name is origined from the phenomenon of EEG signal power rise or fall in the frequency bands mu and beta (8-12 Hz and 18-26 Hz), when a subject spontaneously imagines a movement.
Signal processing methods are very important in such systems. Signal processing covers: preprocessing, feature extraction, feature selection and classification. Among these processes a model is trained which can predict the labels of the new data in order to produce a control signal for applying to a BCI system. In this study, training has been done by extracting the suitable feature vector with using spatial patterns obtained by the method of common spatial pattern (CSP). Then three classifiers as Support Vector Machine (SVM), Linear Discriminant Analysis (LDA) and Logistic Regression (LogReg) are trained separately by obtained features and their performance will be evaluated at the end. However, it is a challenging problem and requires the development of an appropriate machine learning and signal processing tools.
Different sets of preprocessing feature extraction and classification methods were implemented to help support and validate the proposed system and to benchmark against the latest developed systems in BCI literature.
This thesis develops an asynchronous multiclass noninvasive EEG - based BCI system based on a novel combination which is done by using CSP for spatial pattern, log variance for feature extraction and LogReg classifier for classifying the features. The developed system showed robust and accurate classification results for all datasets. The proposed system is tested on four separated datasets which is a well-known and publicly available dataset. The asynchronous multiclass BCI problem is particularly important because it closely matches realistic operating conditions (as opposed to synchronous problems) - Keywords:
- Brain Waves ; Processing ; Classification ; Brain-Computer Interface (BCI) ; Signal Processing ; Feature Vector
Digital Object List-
محتواي کتاب - view
Bookmark