Emotion Recognition Research Based On EEG Signal

Emotion recognition is an important part of artificial intelligence,and it has been widely used in medical care,education,product design,transportation,military and other fields.Emotion recognition based on peripheral physiological signals,speech,and facial expressions is achieved through human subjective judgments and cannot guarantee the accuracy of the emotion recognition results.Emotion recognition based on EEG signals can objectively and accurately identify emotions.EEG signal is a kind of non-stationary and non-linear random signal,which needs to be analyzed by combining the characteristics of time domain and frequency domain.The empirical mode decomposition(EMD)and wavelet transform are the two most widely used time-frequency analysis methods.At present,when EMD deals with EEG signals,there is still an undershoot phenomenon after envelope fitting.Moreover,the use of approximate entropy and permutation of entropy of permutation entropy for accurate recognition of emotions is not high.Therefore,this thesis adopts the envelope fitting in EMD,the characteristics of EEG signals,and the emotion recognition of EEG signals.The main contributions are as follows:(1)An EMD algorithm for reducing undershoot is proposed.By introducing "pseudo-extreme points",the algorithm increases the number of extreme points to form a new extreme sequence,and then uses the new extreme value sequence interpolation to obtain a new envelope.Experimental results show that this algorithm can effectively reduce the number of undershooting points.The fitted envelope is closer to the original signal and has better smoothness.(2)A new combination of EEG emotion features is proposed.Combining wavelet transform and empirical mode decomposition to analyze EEG signals,the approximate entropy,permutation entropy,and energy moment of EEG signals are extracted and fused as new emotion EEG features,and then support vector machine(SVM)is used.Identify the classification.The experimental results show that the classification and recognition result based on the new EEG emotion feature combination is superior to the classification recognition result based on a single feature parameter.