EEG-based emotional state recognition using deep learning network

Emotion is one of the most important human quality that plays an important role in life. Also, one way of communicating between human and computer is based on emotion recognition. One way of emotion recognition is based on electroencephalographic Signal (EEG). In this paper, according to the non-stationary property of EEG, intrinsic mode functions (IMF) extracted by using empirical mode decomposition (EMD) and, then first 3 IMFs selected. Each IMF converts into smaller pieces with a one-second window and power feature has been extracted from each piece. Then, by using a suitable mapping, the position of the electrodes in the 10-20 system becomes the position of the pixels in the picture. The extracted features are considered as pixel color components. To determine the class of valence, the set of all generated pictures is given as input to a deep learning network and output determine the high or low class of valence. The same process is used to determine the class of arousal. To examining the method, the DEAP dataset is used. By choosing 17 * 17 for the image size, the mean accuracy and standard deviation were obtained of 78.58% and 3.9 for the valence and 78.66% and 3.1 for the arousal which that shows a significant improvement compared to similar tasks.