Research On Space-frequency Pattern Optimization And Feature Fusion For Emotion Classification Of Physiological Signals

Emotion classification helps machines better understand human intentions,and enhance the harmony and efficiency in the Human-Machine Interaction.The emotion classification based on physiological signals has become an important research branch because of its objectivity and reliability.However,current research focuses on emotional classification on a single modality of EEG signals or peripheral physiological signals.There a re problems with singularity and one-sidedness and cannot provide comprehensive and accurate emotional information.Emotion classification research on EEG signals also pays more attention to the improvement of classification accuracy,while ignoring the physiological background of EEG signals,and also deficiencies in the interpretability of the model.In this paper,we study these issues,establish feature extraction and classification models from the spatial array characteristics and frequency characteristics of EEG signals,and on this basis,we will further construct a bimodal multi-physiological signal emotion of EEG signals with peripheral physiological signals based on the feature fusion theory.The classification model makes full use of the complementary information between different signals and breaks the single-modality limitation of physiological information.EEG signals under different emotions have obvious spatial frequency characteristics.In this dissertation,the spatial characteristics of EEG signals under multiple leads are utilized to enhance the differences in EEG spatial patterns of different emotions through matrix diagonalization,so that the processed signals are more conducive to discriminant analysis.Experiments show that the propose d method can increase the recognition rate of valence and arousal by 1.67 and 1.94 percent respectively,which verify the validity of spatial information.Based on this,this article further digs out the information of emotions in the frequency pattern of EEG signals,proposes an emotion classification model optimized for spatial-frequency pattern synchronization,and uses the inverse distance weighting method to improve the loss of information in labeling,the estimation of the center point of the sample is realized more accurately,thereby improving the model.Finally,the classification accuracy rates of 81.42% and 82.02% were achieved in terms of valence and arousal,which increased by 4.64 and 4.84 percent respectively.And compared with other research results,the proposed model also has advantages in accuracy and interpretability.For the research on emotion classification of multiple physiological signals,this article uses the induction paradigm of audiovisual stimuli to collect and build a multi-person and multiple physiological signal emotion database to provide data support for the research.A recursive feature elimination algorithm based on mutual information is proposed for feature selection,and the effect of feature fusion strategy is verified on the Augsburg database and self-built database.Experiments show that the recognition rate has increased by 5.47 percent on the Augsburg database,and the valence and arousal rate has increased by 3.73 and4.19 percent on the self-built database,respectively.At the same time,the importance of the proposed features is sorted to verify the superiority of EEG signals in emotion classification.It should be selected first under limited collection conditions.