Analysis Of Emotion Classification Based On Deep Boltzmann Machines And EEG Signals

Emotion is essentially the product of human thought and psychological activities,which can affect human behavior.In the current society,there are many patients who can't control their emotions normally.Once they have unhealthy emotions,it will bring negative impact on society and life.Therefore,the research on how to effectively identify emotions is meaningful.With the continuous development of EEG research and AI technology,the combination of machine learning and deep learning algorithm in emotional computing to model and analyze physiological signal data can identify emotions more effectively than traditional signal analysis methods.However,the physiological signal data is easy to be mixed with noise,which is not conducive to the acquisition of important feature information,so there are still some problems such as low classification accuracy in the current research of emotion recognition based on physiological signal.Based on the above background,the work of this paper is as follows:1)A deep network fusion model,MMDBMS SVM model,is proposed to extract and classify the main emotional features in EEG.The MMDBMS model is composed of multiple depth Boltzmann machines(DBMS),and the DBM of each channel contains two restricted Boltzmann machines(RBMs).Compared with the traditional feature extraction methods,MMDBMS is more suitable for multi-channel data in the experiment.It can not only extract the internal features of the original emotional EEG,but also not lose the main features of the original data,which helps to improve the accuracy of classification.The important features extracted will be input into support vector machine(SVM)for classification.2)In this paper,the physiological signal deap data set generated by video induced emotion is used as experimental data.Firstly,the best number of hidden layer nodes is determined by designing hidden layer nodes and accuracy experiments,and used as the optimal parameter input for subsequent experiments.Secondly,a control experiment is designed to compare and analyze the experimental effect of MMDBMS SVM model proposed in this paper and traditional EEG emotion classification model The conclusion can be summarized as follows: the average accuracy rate of potency and wake-up of MMDBMS SVM deep fusion model can reach 83.9% and 84.7% respectively,which is better than the traditional EEG emotion recognition model.3)In order to explore the classification effect of MMDBMS SVM model under different frequency bands and multimodal factors,this paper also carried out related experiments: firstly,the signals in different frequency bands(?,?,low ?,high ?)were identified,the results show that: the EEG signals in high frequency band have more emotional information than those in low frequency band.Secondly,combining EEG and peripheral physiological signals(eye electrical signal,EMG signal)for emotion recognition research,the results show that: combined with peripheral physiological signals for auxiliary classification will help to improve the accuracy of the experiment,about 1% to 2%.