E-EmotiConNet: EEG-based Emotion Recognition with Context Information

Electroencephalogram (EEG) based emotion recognition has received considerable research attention for brain-computer interface (BCI) and health care. The main challenge is to find an effective representation that is invariant to within and inter-subject differences associated with raw EEG data. In this paper, we propose a novel deep learning architecture for subject-independent emotion recognition, which is named E-EmotiConNet. The proposed model is a two-stream CNN architecture that effectively incorporates multi-channel EEG-time series information and spatial interactions. It comprises EEG-Emotion and EEG-Context networks, which can find emotion related features and subject oriented ones from raw EEG data, respectively. By fusing the interaction model with emotional features, the proposed model can improve emotion recognition accuracy. Experiments were performed on two publicly available datasets, DEAP and SEED, and verified the proposed E-EmotiConNet achieved classification accuracy of 93.39% and 93.69% on DEAP and SEED, respectively, compared with current state-of-the-art approach.