EEG-based emotion recognition via capsule network with channel-wise attention and LSTM models

• Published in: CCF transactions on pervasive computing and interaction (Online) Vol. 3; no. 4; pp. 425 - 435
• Main Authors: Deng, Lina, Wang, Xiaoliang, Jiang, Frank, Doss, Robin
• Format: Journal Article
• Language: English
• Published: Singapore Springer Singapore 01.12.2021; Springer Nature B.V
• Subjects: Accuracy; Computer Science; Computing and Networking; Data collection; Datasets; Deep learning; Electroencephalography; Emotion recognition; Emotions; Experiments; Man-machine interfaces; Mobile Intelligence: Sensing; Physiology; Regular Paper; User Interfaces and Human Computer Interaction; Emotion recognition; Channel-wise attention; LSTM; Capsule network; EEG
• ISSN: 2524-521X; 2524-5228
• DOI: 10.1007/s42486-021-00078-y

Emotion is a kind of psychological and physical state that people produce to objective things. Accurate recognition of emotions is very important in the field of human-machine interface. It is still challenging to extract discriminative features used for electroencephalogram (EEG) emotion recognition which contain the subtle spatial feature of EEG signal and the temporal representations of EEG signal. In order to overcome these challenges, the paper proposes a model based on the capsule network for multi-channel EEG emotion recognition, which combines the attention mechanism and the LSTM network. First, the channel-wise attention mechanism is used to adaptively assign different weights to each channel, then the CapsNet is used to extract the spatial features of the EEG channel, and LSTM is used to extract temporal features of the EEG sequences. The paper proposed method achieves average accuracy of 97.17%, 97.34% and 96.50% on the valence, arouse and dominance of the DEAP dataset, respectively. These results show that the proposed method is better than the current state-of-the-art method.