Research on feature fusion for emotion recognition based on discriminative canonical correlation analysis

Abstract

With the rapid development of emotion recognition, emotion recognition based on EEG signals and physiological signals has drawn much attention from researchers. However, due to the consistency of multi-source information in emotional expression, emotion recognition based on single modal information is still unsatisfactory. Therefore, we proposed a feature fusion algorithm based on Discriminative Canonical correlation analysis, two modes are dealt with simultaneously, the correlation between the two classes of samples is taken as a similarity measure, introduced the class information of the sample, Fully consider the correlation between similar samples and the correlation between different samples. We use the DEAP database and use the DCCA method to fuse the physiological signals and the EEG signals, which greatly improves the classification effect. The classification of liking dimension is 68.21%, which is about 10% higher than other methods and about 2% higher than the CCA model.