Made available in DSpace on 2018-08-02T00:01:59Z (GMT). No. of bitstreams: 1
tese_3870_Tese_Alessandro_Botti_Benevides_600dpi.pdf: 13357880 bytes, checksum: d7eb3ecdca23180cb3af92d0ea795d0e (MD5)
Previous issue date: 2013-08-30 / This present research proposes a Brain-Computer Interface (BCI) architecture
adapted to motor mental tasks and music imagery. For that purpose the statistical
properties of the electroencephalographic signal (EEG) were studied, such as its
probability distribution function, stationarity, correlation and signal-to-noise ratio
(SNR), in order to obtain a minimal empirical and well-founded parameter system for
online classification. Stationarity tests were used to estimate the length of the time
windows and a minimum length of 1.28 s was obtained. Four algorithms for artifact
reduction were tested: threshold analysis, EEG filtering and two Independent
Component Analysis (ICA) algorithms. This analysis concluded that the algorithm
fastICA is suitable for online artifact removal. The feature extraction used the Power
Spectral Density (PSD) and three methods were tested for automatic selection of
features in order to have a training step independent of the mental task paradigm, with
the best performance obtained with the Kullback-Leibler symmetric divergence method.
For the classification, the Linear Discriminant Analysis (LDA) was used and a step of
reclassification is suggested. A study of four motor mental tasks and a non-motor
related mental task is performed by comparing their periodograms, Event-Related
desynchronization/synchronization (ERD/ERS) and SNR. The mental tasks are the
imagination of either movement of right and left hands, both feet, rotation of a cube and
sound imagery. The EEG SNR was estimated by a comparison with the correlation
between the ongoing average and the final ERD/ERS curve, in which we concluded that
the mental task of sound imagery would need approximately five times more epochs
than the motor-related mental tasks. The ERD/ERS could be measured even for
frequencies near 100 Hz, but in absolute amplitudes, the energy variation at 100 Hz was
one thousand times smaller than for 10 Hz, which implies that there is a small
probability of online detection for BCI applications in high frequency. Thus, most of the
usable information for online processing and BCIs corresponds to the α/µ band (low
frequency). Finally, the ERD/ERS scalp maps show that the main difference between
the sound imagery task and the motor-related mentaltasks is the absence of ERD at the
µ band, in the central electrodes, and the presence of ERD at the αband in the temporal
and lateral-frontal electrodes, which correspond tothe auditory cortex, the Wernickes
area and the Brocas area.
| Identifer | oai:union.ndltd.org:IBICT/oai:dspace2.ufes.br:10/9709 |
| Date | 30 August 2013 |
| Creators | BENEVIDES, A. B. |
| Contributors | SARCINELLI FILHO, M., Anselmo Frizera Neto, FERREIRA, A., ALVES, E. C., CONCI, A., BASTOS FILHO, T. F. |
| Publisher | Universidade Federal do Espírito Santo, Doutorado em Engenharia Elétrica, Programa de Pós-Graduação em Engenharia Elétrica, UFES, BR |
| Source Sets | IBICT Brazilian ETDs |
| Detected Language | English |
| Type | info:eu-repo/semantics/publishedVersion, info:eu-repo/semantics/doctoralThesis |
| Format | application/pdf |
| Source | reponame:Repositório Institucional da UFES, instname:Universidade Federal do Espírito Santo, instacron:UFES |
| Rights | info:eu-repo/semantics/openAccess |
Page generated in 0.0016 seconds