| As an important research direction in the field of artificial intelligence,emotion recognition has become a hot topic in current research.Human external behavior may not be able to objectively express human true emotions,because both speech signals and facial expression signals have certain camouflage,and the generation of emotions does not necessarily change people's external behavior.The Electroencephalogram(EEG)signals are directly generated by the central nervous system of the human body,and the central nervous system is closely related to human emotions.Therefore,the EEG signal can directly reflect the emotional state of human beings in real time and objectively.Therefore,more and more researchers use EEG signal-based research methods for emotion recognition research.In the traditional EEG signal-based emotion recognition method,a single EEG feature is extracted from the aspects of time domain,frequency domain or information entropy,resulting in poor recognition effect.On the other hand,since emotion recognition requires more channels of acquisition equipment in the signal acquisition process,it is not conducive to portable wear in actual use.Therefore,it is necessary to study emotion recognition based on less channel EEG signals.This paper has done the following work on the basis of the previous research work on the above two aspects.1.A comprehensive introduction to the research status of EEG features and channel selection research methods used in emotional recognition based on EEG signals has been introduced.2.The time domain multi-feature fusion method is applied to EEG emotion recognition.Six kinds of time domain statistical features are merged into one feature vector to classify emotions.The experimental results show that under the same classification algorithm,compared with the results of each single time domain feature classification,the average recognition accuracy can be increased by up to 12.5%.3.According to the characteristics of information entropy,the characteristics of EEG signal non-stationarity can be better reflected.Combined with sub-band energy characteristics,energy ratio characteristics and root mean characteristics after wavelet transform,high-dimensional EEG combination features are used for emotion recognition.The experimental results show that compared with the previous people's use of only frequency domain features for emotion recognition,the average recognition accuracy rate is improved from 69%to 71.7%,and its classification stability is greatly optimized.4.Starting from practical portable,a PCA-ReliefF algorithm based on EEG channel selection is proposed.Through the algorithm,the weights of all features are sorted,and then the weights of the EEG public channels are sorted.The optimal 6 and 13 EEG emotion channel combinations are selected,and the EEG signals are related to emotions.The four band energy distributions are verified.The experimental results show that the EEG pathways selected for emotion are mainly located in the frontal,parietal and temporal lobe of the brain.These areas are consistent with the physiological principles of emotional production.5.The SVM algorithm is used to verify the channel selection result.The relationship between the recognition accuracy and the number of channels is obtained.When the number of EEG channels in the emotion recognition task is reduced to 13,the average classification accuracy is only reduced by about 1.7%..The emotional classification accuracy of each subject was compared under 6,13,32 channels.The experimental results show that the proposed EEG channel selection method can effectively reduce the number of channels required,which lays a foundation for the development of subsequent wearable devices. |
