A MEMD Method of Human Emotion Recognition Based on Valence-Arousal Model

Since emotions play an important role in human-machine interaction and EEG based emotion recognition is a challenging study due to nonstationary behavior of the signals, sophisticated signal processing methods are required to extract the hidden patterns in the EEG. In this research, some promising results are presented for classification of emotions induced by watching music videos, a multivariate empirical mode decomposition(MEMD) based feature extraction method is proposed to identify emotional state as high/low arousal and high/low valence. Multi-channel EEG signals are decomposed into intrinsic mode functions (IMFs), the information-bearing IMFs are selected to extract power, volatility index, hjorth parameters and asymmetrical properties on the left and right hemispher. The performances of the proposed MEMD based methods are evaluated using the publicly available DEAP data set. High/low arousal and high/low valence states of participants are recognized using the MEMD based features with KNN and SVM classifier. For the recorded 8-channel EEG signals, extracted features with KNN have accuracy rates of 67.7% and 70.5%, while SVM yields the accuracy of 67.9% and 70.9 % for arousal and valence states.