Emotion Recognition Based On EEG And Peripheral Physiological Signals

With the continuous development of artificial intelligence in recent years,the research of physiological signals has gradually expanded from the field of medical to the field of emotional computing and human-computer interaction.As physiological signals such as EEG are directly acquired from individuals,they can not be disguised.Therefore,the emotional research of physiological signals has attracted more and more attention.In addition,the research of emotional physiological signals should be more diversified.In this context,we investigate and analyze in detail the physiological signals such as EEG in recent years,and propose our own models from two aspects of signal processing and emotional feature extraction.The main work of this thesis is as follows:(1)By analyzing the stability of baseline and emotional signals in different channels under different characteristics,baseline strategy is summarized.The power spectral density is selected as the pre-feature,and different post-feature combinations and different classifiers are used to verify the experiment.The experimental results show that the baseline strategy can effectively improve the representation ability of emotional features under different feature combinations and classification algorithms.(2)By analyzing the spatial location relationship of brain electrodes in the twodimensional electrodes map,the spatial correlation model of 3D convolution network was summarized and constructed.The model can extract spatial correlation features of different EEG channels.In order to verify the effectiveness of the model,this thesis chooses a variety of feature combinations and classifiers for comparative experiments.The results show that the spatial correlation 3D convolution network are better for multichannel emotional characteristics.On this basis,by analyzing the emotional characteristics of EEG and peripheral physiological signals,the fusion experiments of EEG and peripheral physiological signals were carried out at the feature level and decision level respectively.In feature level fusion,the frequency domain and time domain features of EEG and peripheral physiological signals are extracted respectively.In decision level fusion,four groups of better classification results of EEG and peripheral physiological signals are grouped by weighted voting method.The experimental results show that decision level fusion improves the final classification accuracy,and the best classification results under two emotional labels are 83.2% and 81.7% respectively.