Emotion Recognition Based On Hierarchical Fusion Of Wavelet And Deep Features Of Multi-Channel Physiological Signals

Emotion,as a short-term and strong external reaction,plays a very important role in human life and affects human physiological and psychological state.The research of emotion recognition has attracted more and more attention,and it has been applied in the fields of medical care,safe driving,social security and so on.Compared with the external physical signals,physiological signals are directly generated and controlled by the nervous system(central nervous system,peripheral nervous system),which can be more objective.Therefore,the research content of this paper is the emotion recognition based on physiological signals.At present,there are two limitations in the study of emotion recognition based on physiological signals.(1)The limitation of the features.Most of the features extracted in the current study are not comprehensive,which are extracted manually or learned by deep learning methods for specific tasks without excavating the inherent common information in physiological signals and lacking considering the temporal context of physiological signals.(2)The limitation of the fusion methods.Common feature fusion and decision fusion do not make full use of physiological signals,which affects the final recognition accuracy.According to these problems,the following three main tasks are carried out in this paper:(1)wavelet transform based emotion recognition algorithm from multi-channel physiological signals is proposed.In view of the non-stationary and nonlinear characteristics of physiological signals,the wavelet transform is adopted to decompose the signals,and various type of features are extracted from the decomposed signals.Finally,support vector machine(SVM)is applied and achieved the accuracy rate of 92.68% in the happy and sad two-class classification experiments,which verify the effectiveness of wavelet features and the fusion of multi-channel physiological signals.(2)GRU-AE based emotion recognition algorithm from multi-channel physiological signals is proposed.Gated recurrent unit autoencoder(GRU-AE)was used to learn the potential common information of the temporal context of physiological signals in emotional changes.The accuracy rate in the happy and sad two-class classification experiments is reached 91.48%,and the performance is greatly improved compared with the separate GRU and AE classifiers.The experiments show that the proposed GRU-AE can effectively mine the deep information of physiological signals.(3)An algorithm called emotion recognition based on hierarchical fusion of wavelet and deep features of multi-channel physiological signals is proposed.By means of hierarchical fusion,combining the hand-crafted wavelet features and self-learning GRU-AE deep features,the information at different levels within and between physiological signals is effectively utilized and the accuracy rate of 95.30% in the happy and sad two-class classification experiments is achieved.Compared with different fusion methods and related literature algorithms,the algorithm proposed in this paper has better performance and advancement.The algorithm proposed in this paper utilizes the hierarchical fusion of wavelet features and GRU-AE features for emotion recognition,which effectively makes use of different levels of information within and between physiological signals,and provides a new idea for promoting the development of emotion recognition of multi-channel physiological signals.