Emotion Recognition Research Of EEG Based On Visual Stimulation

Emotional recognition is refers to use the signal processing and analysis to extract and classify the parameters,which come from people’s psychological,physiological or signs of behavior in various emotional state,to confirm the emotional state of a person.Electroencephalogram(EEG)is a kind of physiological signal,which caused by postsynaptic potentials of numerous nerve cells,can reflect the neuronal activity and state of mind Due to the strong functional specificity and high time resolution of EEG,the research of emotion recognition based on EEG has broad application prospects.And it has significant or potential application value in human computer interaction,entertainment development,distance education,and many other fields.Human emotions are complex and diverse,it is difficult and also unrealistic to identify all kinds of emotions.According to the requirement of issue of “Emotion Recognition Research of EEG Based on Visual Stimulation”,this paper starts from The background and current situation of emotional EEG recognition.Dividing the visual stimuli pictures into positive,neutral and negative,and uses these pictures to evoke emotional EEG.Selecting the EEG from Seven leads of frontal lobe to carry on the emotional recognition research.After that,systematically analyzing and processing the de-noising,feature extraction and pattern classification of emotional EEG.The main research contents and innovations are as follows:(1)In order to ensure that the stimulus pictures can correctly induce the corresponding emotional EEG,setting the classification criteria to divide the stimulus pictures into positive,neutral and negative.In this paper,let subjects use self-assessment manikin to get the individual valence and arousal scores of a picture,and averaging the individual scores to get the average score of the picture.Then,according to the classification of emotion and the relationship between pleasure and arousal,the classification standard of pictures is set up.In this standard,the positive picture’s valence degree is high,arousal degree is also high;the neutral picture’s valence degree is flat,arousal degree is also flat;the positive picture’s valence degree is low,but arousal degree is high.subjects divide the picture category according to the classification standard,and removing the picture that average score does not meet thisstandard.So that it can ensure the effectiveness and reliability of the stimulus pictures.(2)To retain more details during the EEG signal de-noising,a de-noising method based on SA4 multi-wavelet is proposed.Firstly,the repeated sample pre-filtering method is applied to pre-process the EEG signal,and the multidimensional multi-wavelet coefficients can be obtained by SA4 multi-wavelet decomposition algorithm.Then,the soft threshold function is used to process the multi-wavelet coefficients of each layer,and the coefficients are reconstructed by multi-wavelet transform to get the de-noised EEG signal.Simulation results show that compared with db4 wavelet algorithm,the better signal-to-noise ratios and root mean square errors of the EEG signal can be achieved by SA4 multi-wavelet algorithm,which can also reduce the detail loss during the EEG signal de-noising.(3)Because the fuzzy entropy can not be used to measure the complexity of different scale factors,this paper chooses to use the multi-scale fuzzy entropy algorithm for feature extraction.And aiming at the problem that the short of the experimental data will limit the scale factor of multi-scale fuzzy entropy algorithm,the coarse grain algorithm is improved.Using the multi-scale fuzzy entropy algorithm,which based on improved coarse-grained,to extracte a eigenvector,then normalizing it.Finally using principal component analysis method to reduce dimensions to pattern classification.The simulation results show that,compared to the fuzzy entropy feature,the multi-scale fuzzy entropy can better improve the average recognition rate of the classification.(4)In order to increase the separability of samples in the feature space,the multi-kernel support vector machine is introduced into the field of emotional classification.The kernel function of radial basis kernel and polynomial mixture is constructed,combining with the radial basis kernel and polynomial kernel’s features.And in order to avoid the global fine search and get more accurate parameters combination,the parameter optimization method based on cross validation and improved grid search method is used to improve the average accuracy of pattern classification.In the case of multi-scale fuzzy entropy feature,this algorithm’s recognition rate of multi kernel support vector machine is 85.8%,it improved by 3.6%and 1.0% compared with the polynomial kernel function support vector machine and the radial basis kernel function support vector machine.