Emotional State Recognition Based On Visually-induced Bimodal Bioelectricity

Now the pace of life is getting faster and faster,people’s pressure is everywhere,China has become the country with the largest number of depression patients.With the rapid development of brain-computer interface(BCI)and sensor technology,the research on emotion state analysis based on bioelectrical signal has attracted extensive attention.In recent years,it has been gradually used to treat patients with depression.Based on EEG and EMG dual-mode signals combined with width learning network and D-S evidence theory,visual evoked potentials were analyzed for emotion recognition.The main research work is as follows:First,the stimulus videos were selected and divided into positive,neutral and negative categories according to the James emotion model.Classification criteria were established according to the signal mode of video induced EEG.Twelve subjects were arranged to conduct adaptation experiment to select stimulus materials to obtain accurate bioelectrical signals.Then the experimental paradigm was designed to collect bimodal bioelectrical signals using Emotiv EPOC and NeuSen WN.Then the band stop filter and Busvoton filter were used to remove the high and low frequency noise of the EEG signal,and the pure signal was obtained.Emg signals use synchronous compression wavelet transform to remove noise.Second,Based on the nonlinear characteristics,this paper uses the concept of entropy to obtain the differential entropy of three different emotions,and then uses the power spectrum method to extract their frequency domain characteristics,and analyzes the brain topographic maps of each frequency band data in different emotions.The characteristic wave forms of δ,θ,α,β and γ bands were obtained by extracting the time and frequency characteristics of EEG signals with wavelet packet transform.Then,the characteristic data of EMG were extracted by RMS of time domain feature and mean frequency of frequency domain feature.Finally,the BLS classification method is introduced,and the BLS incremental learning algorithm is analyzed.Experimental results show that the differential entropy extraction algorithm is better than the wavelet packet feature and power spectral density extraction algorithm,reaching78.3%,77.6% and 77.3% in the case of positive and negative,positive and neutral,neutral and negative binary classification.The experimental results show that the classification accuracy of BLS incremental learning algorithm is higher than that of the original BLS method,and the classification effect of EMG and EEG fusion can reach more than 81.4%.In order to further improve the accuracy of classification,d-S evidence theory is introduced to optimize the classification network structure.Finally,D-S evidence theory is used to make decision analysis on the known classification data,and the fusion results of EEG and EMG evidence theory are analyzed.Experiments show that the accuracy of emotion recognition model based on D-S evidence theory fusion is higher than that of width learning classification model,and the classification accuracy of dual-mode bioelectrical signals fusion is more than 83.8%.