The Classification Research Based On Emotional Electroencephalogram Signals

With the rapid development of machine learning and deep learning algorithms in the field of artificial intelligence,people have higher requirements for the application artificial intelligence in our life.Emotion is a kind of spiritual feedback to the outside world and it is unique to the brain of human.The recognition and simulation of human emotions by computers is one of the important signs of computer intelligence.The analysis of Electroencephalogram(EEG)and the classification of emotions by machine learning have become the frontier research directions in the field of artificial intelligence and human-computer interaction.The EEG signal is sampled through multiple channels and its feature vector has high dimensional characteristics.Because emotional EEG signals are not related to all EEG channels,there is a degree of channel redundancy and dimensionality distress in emotional classification through EEG signals.In view of the sparse characteristics of emotional EEG signals in the brain,the following research has been done:Firstly,the emotional data set DEAP based on the multiple physiological modalities generated by music video(MV)induction is studied to complete the temporal and frequency domain features of emotional EEG signals.The delta(0.5-3Hz),theta(4-7Hz),alpha(8-13Hz),beta(14-30Hz),gamma(31-47Hz)wave of emotional EEG signals are extracted by wavelet transform and fast Fourier transform.The four-layer intrinsic mode function(IMF)of EEG signal is calculated by empirical mode decomposition,and its power spectrum value is calculated to construct the feature engineering.Secondly,emotional EEG signals are sparse signals in the human brain.For this characteristic,the orthogonal matching pursuit algorithm(OMP)based on greedy solution and the accelerate proximal gradient(APG)algorithm based on convex optimization are used to solve the sparse coding sparseness.The classification model SRC is used to classify emotions and compare them with the traditional emotional EEG classification algorithmFinally,aiming at the multi-channel sampling of emotional EEG signals and the dimensionality disaster of eigenvectors,this paper uses the sparse group Lasso algorithm as the method of feature dimension reduction and channel selection,and use the sparse representation classification algorithm(SRC)based on the sparse group Lasso as the penalty term to perform four classifications of emotions.The inter-group sparsity and intra-group sparsity of the emotional EEG signal are realized.The experimental results are compared with the traditional emotional EEG classification algorithm and the feature dimension reduction method.It is shown that the sparse representation classification algorithm based on the sparse group Lasso has a good effect on the classification and feature dimension reduction of emotional EEG signals.