Emotion detection using EEG signals based on Multivariate Synchrosqueezing Transform and Deep Learning

Emotion recognition from EEG signals has gained a great research interest in brain-computer interface (BCI) studies. As the result of the outstanding success of deep neural networks in the image classification area, deep learning methods have become popular in the subject of emotion classification from EEG signals. In this study, we have used the Alexnet structure for the classification of emotions in Arousal and Valence domains separately. We generate TF images of 32-channel EEG data we collected by using Multivariate Synchrosqueezing Transform (MSST) and then these TF images are used to feed to the AlexNet model. A 3-fold cross-validation strategy was adopted to evaluate the robustness of the models. By training the AlexNet architecture an average accuracy of 71.60% is yielded on Arousal and an average accuracy of 67.93% is yielded on Valence. The results demonstrated that the proposed method achieved promising performance to classify emotions.