Hemispheric Asymmetry Measurement Network for Emotion Classification

Ruofan Yan, Na Lu, Xu Niu, Yuxuan Yan

Abstract:

Abstract. Electroencephalogram (EEG) based emotion recognition has received
considerable attention from many researchers. Methods based on deep learning
have made significant progress. However, most of the existing solutions still need
to use manually extracted features as the input to train the network model. Neuroscience studies suggest that emotion reveals asymmetric differences between
the left and right hemispheres of the brain. Inspired by this fact, we proposed a
hemispheric asymmetry measurement network (HAMNet) to learn discriminant
features for emotion classification tasks. Our network is end-to-end and reaches the
average accuracy of 96.45%, which achieves the state-of-the-art (SOTA) performance. Moreover, the visualization and analysis of the learned features provides
a possibility for neuroscience to study the mechanism of emotion.