Wavelet Transform Based Multichannel Emotion Detection through EEG Signal

Abstract

- Artificial emotion intelligence is one of the fast growing research areas now a day. These emotion
identification and classification schemes are foundation on Brain / Human Computer Interface and can be
used in various applications like Medical, Entertainment, Education and Gaming. The EEG signals from
SEED database are considered for emotion recognition. In this paper we use time frequency analysis of
wavelet transform (WT) for extracting features from five frequency bands of EEG signal. Wavelet
coefficients contain temporal information of EEG signal which gives lot of information about emotion.
Linear classifiers are used to classify emotions (Positive, Negative and Neutral). Various wavelet functions
like “db4”, “db6”, “db8”, “sym6”, “coif5” are used in this paper, to extract various features form brain
signal to categorize various emotions. We compare classification accuracy with 12 channels, 9 channels, 6
channels and 4 channels using SVM and KNN classifiers. Overall accuracy of SVM classifier is 65.41% for
energy feature, while for Entropy feature it gives accuracy of 69.23% using KNN classifier. Accordingly,
we bring to end that higher frequency bands like beta and gamma gives important information related to
emotional state as compared with lower frequency bands.