Optimized Deep Forest Emotional Awareness Recognition Based on EEG Rhythm Characteristics

Emotion is the psychological and physiological response of human to external things, and occupies an important place in the study of human-computer interaction. Electroencephalogram (EEG) is a physiological signal that is widely used in the field of emotion awareness recognition. EEG signals can be divided into 5 basic rhythms according to frequency, and most of the existing research on emotion awareness recognition based on EEG signals directly processes the rhythms and extracts the corresponding features. This method is easy to lose the rhythm information and cannot give full play to its function. Therefore, in this study, we propose an optimized deep forest emotion awareness recognition method based on EEG rhythm characteristics to investigate the effect of rhythms on emotion awareness recognition performance. Firstly, we classify EEG rhythms into 3 types of rhythm combinations: single rhythm, compound rhythm and full rhythm according to the principle of adjacent combination, and fully consider various combination forms of rhythms; Secondly, we construct a two-dimensional input model to retain the spatial information of multi-channel EEG signals; Finally, we use the gcForest classification model for emotion recognition, which does not require feature extraction and maximizes the retention of rhythm information. We conducted extensive experiments on the DEAP dataset, and the experimental results show that the frequency band and number of rhythms affect the performance of emotion recognition, and the high frequency band rhythms have better emotion classification performance compared with the low frequency band rhythms, among which the $\beta$ rhythms are higher than the Y rhythms in validity and arousal dimension, and their classification accuracy is 96.711% and 96.633%; the classification accuracy of compound rhythms was higher than that of single rhythms, but a higher number of compound rhythms does not necessarily lead to better emotion classification performance, where in the arousal dimension, compound rhythms $((\mathrm{x}+\beta+\gamma)$ have better awareness emotion recognition performance compared to full rhythms $((+\mathrm{t}\mathrm{K}+\beta+\gamma)$, with 97% classification accuracy.