EEG Emotion Recognition Based on Temporal-Spatial Features and Adaptive Attention of EEG Electrodes.

EEG signals are more objective in reflecting human emotions, compared to facial expression and speech which can be disguised. Therefore, EEG emotion recognition is a well-worthy research area. The previous recognition methods fail to fully extract the temporal and spatial information in EEG signals, while ignoring the differences of different EEG channels contributing to emotion recognition. Based on this, a new EEG emotion recognition method is proposed in this paper. The method achieves auto-assignment of EEG channel weights by spatial attention mechanism, extracts spatial and temporal features by continuous convolutional block and LSTM, and achieves feature classification by introducing L2-SVM in the last layer of the model. Finally, we have experimented extensively on the DEAP public dataset, and the experimental results show that our method has an average accuracy of 98.93% and 98.95% in valence and arousal, with a minimum accuracy of 93.37% and a maximum accuracy of 99.88%. Its performance is better than the previous methods under subject-dependent condition, which indicates that the method proposed in this paper provides a feasible solution for the emotion recognition task based on physiological signals.