A Study on Application of Deep Learning to Emotion Recognition by Single Electrode EEG

Abstract

With the development of Human-Machine Interaction technology, Human Computer Interaction (HCI) becomes a new research field. As a crucial part of the HCI research framework, in order to develop friendly and natural human-machine interface, EEG-based emotion recognition which aims to automatically discriminate different emotional states by using physiological signals acquired from human will be effective. Comparing to MRI, EEG has undisputed advantages both in terms of the simplicity and the cost of information collection. But in practice, it is very difficult to collect physiological data by connecting multiple electrodes because of users’ burden. In this study, in order to lighten the burden of wearing equipment, we used the EEG from less electrode to recognize the emotion of human. In the experiment, we employed DEAP dataset, a public available dataset which uses music video. Five IMF bands were extracted from the raw EEG by Empirical Mode Decomposition (EMD), and the sample entropy (SE) of each band were computed as the feature in order to train Bi-LSTM model as the binary classifier. The result showed the accuracy of 67.0% (Arousal) and 67.7% (Valence) in binary classification.