EEG-Based Emotion Recognition Using Spatial-Temporal-Connective Features via Multi-Scale CNN

Abstract

Electroencephalography (EEG) is often used for emotion recognition in brain computer interaction (BCI) research. EEG signals from each channel mainly reflect activities of the brain region close to the channel position, and the activities cooperated by various brain regions are response to the emotion-induced stimuli. In this paper, temporal, spatial and connective features are extracted from EEG signals gotten around the head, and used for emotion recognition via a proposed model, spatial-temporal-connective convolutional neural network (STC-CNN). The channel-to-channel connectivity is gotten to describe brain region-to-region cooperation under emotion stimuli. The STC-CNN achieved an average accuracy of 96.79 percent and 96.89 percent after classifying in two emotion dimensions of arousal and valence. Utilization of the method we proposed not only helped to achieve a higher accuracy of emotion recognition, but also explored a new selection of feature for EEG research.