Research On Emotion Recognition Algorithm Based On EEG Signal

The Electroencephalogram(EEG)signals of human are generated by the central nervous system,and are closely related to human emotional changes.Using EEG signals to analyze human emotional status has become a hotspot in the field of emotional recognition.But EEG signals are non-stationary and non-linear signals,which can be interfered by other physiological signals of the human body and external factors.The conventional method of emotion recognition has some limitations,including but not limited to algorithm with high complexity and recognition with low accuracy.Based on the in-depth analysis of EEG signals,this thesis conducts research on EEG-based emotion recognition from the following aspects:Firstly,in order to solve the problem that EEG signals can be interfered by various noises,a multichannel EEG signals denoising algorithm based on FastICA and dual-tree complex wavelet transform is proposed in this thesis.The algorithm judges the noise components in each independent component by calculating the fractal dimension of each independent component,and uses the dual-tree complex wavelet transform with improved threshold function to further analyze the useful component.Experimental results show that the algorithm can achieve favorable results in the denoising of multichannel EEG signals.Secondly,considering the lack of effective feature selection methods in the current emotion recognition,the binary cuckoo algorithm is adopted to screen the emotion features in this thesis.The multi-domain emotion features of EEG signals are extracted,then the feature set is optimized by using the binary cuckoo algorithm,and the optimal emotional feature subset is obtained in the end.Experimental results show that this algorithm can obtain the optimal emotional feature subset and reduce the feature dimension effectively.Finally,targeting the problem of low accuracy of emotion recognition and long training and testing time,this thesis presents an least squares twin support vector machine emotion recognition algorithm based on improved firefly algorithm.The least squares twin support vector machine is used to carry out the emotion recognition,the improved firefly algorithm is used to further optimize the least squares twin support vector machine and determine the optimal algorithm model.Experimental res ults show that the accuracy of the algorithm for recognitions on four different types of emotion exceeds 80%,and its average recognition accuracy is 11.2%,12.5%,and 8.8% higher than that of LSSVM,SA-IPSO-SVM,and MLSTSVM algorithms,respectively.The training and testing time of the emotion recognition algorithm have been reduced as wel.