Discrimination of EEG Correlates of EmotionsDiscrimination of EEG Correlates of Emotions
Nadace na podporu rozvoje pokročilých technologií, inovací a technického vzdělávání v České republice

dáváme věci do pohybu ...

  • Projekt je realizován na ČVUT
  • Velikost podpory je 25.000,- Kč
  • Délka projektu činí 9 měsíců

Discrimination of EEG Correlates of Emotions

Řešitel: Ing. Vladana Djordjevic

 

The goal of this project was to make a step forward in the design of an algorithm appropriate to explore regularities related to recognition and classification of EEG correlates of emotions, together with the implementation of its modules.

 

Electroencephalography (EEG) is the non-invasive measurement of brain electrical activity by means of electrodes positioned on the scalp. It has many important applications, not only in medicine but also in cognitive science. The wide list of EEG applications confirms the rich potential for EEG analysis and motivates the need for advanced signal processing techniques to aid in interpretation of this type of complex signal.

 

The processing of EEG signal is a multilevel procedure. Special attention in this project was focused on parts of preprocessing and data representation steps, namely segmentation and feature extraction. The concept of adaptive segmentation together with wavelet transform based features is novel in this field of EEG applications. Applied adaptive segmentation algorithm is based on the principle of two connected windows, sliding along a signal, and a calculation of the predefined differences of the parameters of a signal comprised in windows. This particular algorithm was driven by amplitude and frequency dependent changes in observed signals. Further on, each segment, which was derived from the segmentation process, was represented with a set of features (the feature vector). Among features in time and frequency domain that are typical for EEG signal processing, various wavelet transform based features were extracted, e.g. statistical parameters of obtained wavelet coefficients, zero crossing rate, coefficient of variation, wavelet energy, and energy percent for typical EEG frequency bands. Wavelet transform was also applied to the signals’ first and second derivatives. Implementation was performed in Matlab. Export of feature values to the Weka data mining software was provided in order to enable consequent classification.

 

The data for this project were obtained in a non-clinical institution – at the faculty EEG laboratory at the Department of Cybernetics, FEE. The EEG was recorded during the experiment designed by a psychologist, in which emotions were evoked by presenting pictures with high and low value of valence (pleasant or ugly) and arousal (boring or arousing) in a row. With the use of objective descriptions of emotions together with pictures from an international database (International Affective Picture System database) it is enabled that obtained results can be compared with other studies and that experiments can be repeated in other institutions.