Enhanced Feature Extraction with Superlet Transformation for EEG Emotion Classification

In the field of emotion recognition, electroencephalography(EEG) technology can accurately capture emotion, and has been widely used in psychology, public safety and other fields. In order to improve the accuracy and efficiency of emotion recognition, a new feature extraction method based on Superlets (SL) is developed in this paper. We designed an emotional experimental paradigm to induce positive, neutral and negative emotions. Further more, we extracted features using Superlets from the F7, F8, T7 and Pz electrode channels and classified three emotion states with SVM and DNN classifiers. In order to validate the performance of the proposed model, the collected EEG data set from our lab and SEED data set were used in this study. With SL features and DNN model, the average recognition accuracy from the 11 subjects was 85.75%, and the average recognition accuracy of SEED data set was 85.67%. The experiment results show that superlet transform could be used as a reliable feature extraction method in emotion recognition from EEG signal.