PPGINF PÓS-GRADUAÇÃO EM ENGENHARIA DA INFORMAÇÃO FUNDAÇÃO UNIVERSIDADE FEDERAL DO ABC Phone: (11) 4996-8289 http://propg.ufabc.edu.br/ppginfo

Banca de QUALIFICAÇÃO: ELIANA MARIA DOS SANTOS

Uma banca de QUALIFICAÇÃO de DOUTORADO foi cadastrada pelo programa.
DISCENTE : ELIANA MARIA DOS SANTOS
DATA : 06/02/2019
HORA: 10:00
LOCAL: Sala 306, Bloco B, Campus Santo André da UFABC
TÍTULO:

Source Localization of Electroencephalogram Signals for Brain-Computer Interface Applications


PÁGINAS: 93
GRANDE ÁREA: Engenharias
ÁREA: Engenharia Biomédica
SUBÁREA: Bioengenharia
ESPECIALIDADE: Processamento de Sinais Biológicos
RESUMO:

EEG signals have been studied since the beginning of the 20th century, when researchers began to realize that it was possible to pick up electrical signals through the cranial surface without the need for surgeries. This acquisition is made with electrodes and amplifiers that, when in contact with the scalp, capture variations of electric currents according to the individual's brain activity. Correct identification of brain activity associated with motor imagery (MI) is necessary for the development of BCI. Thus, many studies compare the effects of real movement on the activities of alpha rhythms to the effects of the imagined movement, with the objective of assisting in the development of BCI's. Other studies, however, are concerned with improving the accuracy of motor imagery tasks to assist in the development of more reliable BCI's. For this, several techniques of signal preprocessing are applied, seeking to improve the selection of characteristics to obtain the best results in pattern recognition. In this work, we will cover the following signal preprocessing techniques: location of signal sources, amplitude modulation and Common Spatial Patterns (CSP) with Tikhonov regularization. Study 1 was developed using the location of signal sources applied to 2 BCI Competition databases. It is a pilot study where, using the software LORETA, we applied the technique of locating sources in a little precise way, which resulted in a poor performance of the IM tasks. In study 2, we implemented the amplitude modulation technique as the initial stage of analysis to define filtering regions in the frequency domain of modulation and the CSP with Tikhonov regularization, to decompose the signals into spatial patterns extracted from two classes, maximizing the characteristics . We have noticed that when we define different regions for each subject, it is possible to improve the performance of IM tasks. The secondary study, so named for not addressing the classification of tasks, was inserted in this work because it was developed from the UFABC database, acquired in the context of the research project "Location of electroencephalography signal sources for use in Brain-Computer Interfaces. " This study addresses the laterality of motor tasks, now performed, now imagined, differentiating them from the signal analysis using the event related to spectral disturbance (ERSP). The LDA classifier was used in studies 1 and 2, and in study 2, it was used as the first classification layer, generating binary outputs that served as inputs to the Naive Bayes classifier implemented for multiclasses. Finally, we detail the future performance of the study 3.


MEMBROS DA BANCA:
Presidente - 1545987 - FRANCISCO JOSE FRAGA DA SILVA
Interno - 2334927 - ANDRE KAZUO TAKAHATA
Interno - 1761107 - RICARDO SUYAMA
Notícia cadastrada em: 19/12/2018 11:00
SIGAA | UFABC - Núcleo de Tecnologia da Informação - ||||| | Copyright © 2006-2019 - UFRN - sig.ufabc.edu.br."Melchior"