Abstract

Electroencephalogram (EEG) signals can intuitively reflect the slight variations in human emotions. Consequently, they are the first choice for emotion recognition media. However, EEG signals at different time steps have different emotion representing abilities. By filtering out EEG signals with low representing abilities, the efficacy of extracted EEG features will increase. Thus emotion recognition accuracy can be improved. Therefore, a new feature extraction method called Selective Gated Recurrent Unit (SGRU) is proposed in this paper. From SGRU, we design a new method for emotion recognition. Firstly, SGRU is constructed to extract features from EEG signals. Secondly, a Fully Connected Neural Network (FCNN) is built to classify emotions with the features obtained by SGRU. Finally, the experiment results on DEAP dataset indicate that the method proposed can achieve better performance on emotion recognition compared with other similar methods.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call