Multi-channel EEG time series analysis
Multi-channel EEG time series analysis
Since half a century, the question is discussed, to which extent mental activity is represented in the human electroencephalogram (EEG). In todays EEG recording devices, signals from nurnerous electrodes can be analysed with practically arbitrary sampling rates. The mathematical structure of EEG sig...
Saved in:
Personal Name(s): | Peters, B. O. (Corresponding author) |
---|---|
Contributing Institute: |
Publikationen vor 2000; PRE-2000; Retrocat |
Imprint: |
Jülich
Forschungszentrum Jülich, Zentralbibliothek, Verlag
1999
|
Physical Description: |
IV, 63 p. |
Document Type: |
Report Book |
Research Program: |
ohne Topic |
Series Title: |
Berichte des Forschungszentrums Jülich
3608 |
Link: |
OpenAccess OpenAccess |
Publikationsportal JuSER |
Since half a century, the question is discussed, to which extent mental activity is represented in the human electroencephalogram (EEG). In todays EEG recording devices, signals from nurnerous electrodes can be analysed with practically arbitrary sampling rates. The mathematical structure of EEG signals is, however, not fully understood yet. To develop a method that automatically deals wich such complex, multi-channel information was scope of the thesis. In order to not deal with the mathematically ill-posed problem of source localization, a classification task was chosen to study EEG. Movements of left and right index finger, and right foot were recognized in the EEG of three subjects. EEG data from a memorized delay task was analysed. The EEG data sets were recorded on 56 electrodes positioned around motor-cortical areas on the skull, then scanned manually, and only sets free frorn disturbing artifacts in the signals were considered for later analysis. After passing the signals through some spatial and temporal filters, parametrized estimates of the spectral density were computed. On these compactified data sets, artificial neural nets were trained and combined in a "committee of experts". Correct recognition was achieved in 85% to 98% of EEG data slices recorded just around the movements tank place. In a first chapter, event-related EEG potentials are characterized, the idea of classification as a time series procedure and a possible application - a brain computer interface - are presented. Then, the experimental setup and the filtering and classification algorithms are motivated and introduced. In a third chapter, the classifier is analysed for its general properties. In the last two chapters, the classification results obtained on EEG% recorded from three different subjects are presented and discussed. |