Research On Audio-Visual Evoked Emotion Recognition Based On EEG Signals

Emotion is a mental and physiological state accompanied by cognitive and conscious processes.It plays a very important role in human communication.The analysis and recognition of emotion has broad application prospectsin the areas of human-computer interaction and rehabilitation medical care.Among various types of information sources that can be used for emotion recognition,brain signals are not easy to disguise,reflect sensitively,and have an objective and true recognition result,which is the current research hotspot.At present,there are many studies based on the discrete model,few based on the dimensional model and classification accuracy rate is not high,besides it is mainly for the single type of stimuli,but the emotion in real life is often induced by multiple types of stimul,therefore,it is very necessary to improve the recognition accuracy of the emotional dimension space under multiple types of inducing methods.Based on the research of feature extraction algorithm using multivariate empirical mode decomposition and feature selection algorithm applying sequential floating forward search algorithm,improved algorithms are proposed,This paper completes the recognition and analysis of emotional dimension space under visualauditory evoked mode and designs two typical emotional evoked experiments of music and pictures.Finally,an audio-visual evoked emotion recognition system based on EEG signals is constructed.The main research work is as follows:(1)Aiming at the problem that spatially uniform sampling can not reflect the multichannel signals dynamic characteristics well,based on the MEMD algorithm,a non-uniform sampling MEMD algorithm which adaptively selects the projection direction is proposed.First,non-uniform sampling multivariate empirical mode decomposition of multichannel EEG signals is performed,and the effective intrinsic mode function is used to extract emotional EEG features,evaluate the positive negative valence and high low arousal in the two-dimensional coordinate system of emotional dimension model.The experimental results show that this method can effectively extract the features of EEG signals and improve the classification accuracy of EEG signals.(2)Aiming at the feature redundancy,characteristics of high dimension in the process of feature extraction,the SFFS algorithm is used as the search strategy for selecting and rejecting optimal feature sets,combined filters and wrappers as evaluation criteria for optimal feature subsets,a feature selection algorithm of hybrid SFFS is proposed to optimize the high-dimensional vector space formed by feature extraction.Experimental results show that the algorithm can effectively reduce the feature dimension,eliminate redundant features,and at the same time improve the classification accuracy of EEG signals.(3)Constructed the emotion recognition system based on EEG and completed the audio-visual evoked experiment,realize the online recognition of emotion based on the valence-arousal two dimensions.Two kinds of typical evoking methods,emotional music and pictures,are designed to complete auditory stimulation and visual stimulation to identify emotions,the effectiveness of the feature extraction and selection methods are verified.based on this,an emotion recognition system was constructed.The system receives the EEG data collected by the UE-16 B EEG amplifier through socket communication,and calls MATLAB to perform feature extraction,selection and classification.After the control of the experimental flow and data synchronization,the system finally returns the emotional information in the dimension space to the subject.