Emotion Recognition Based on EEG and DE-CNN-RNN

Convolutional neural networks (CNNs) can extract local spatial features from data well, but it cannot capture long-term dependencies contained in serial data. Recurrent neural networks (RNNs) can compensate for this defect. In the study, according to the differential entropy (DE) feature of electroencephalogram (EEG), DE-CNN-RNN model is proposed to explore emotion recognition based on EEG signals. Experimental tests were conducted with the SEED emotion dataset to compare the accuracy of emotion classification of DE-CNN, DE-RNN, CNN-RNN, and DE-CNN-RNN models. The proposed DE-CNN-RNN model could improve the accuracy of emotion classification by 10% and effectively identify emotion types.