EEG Emotion Recognition Based On Causal Relationship Feature Of Sparse Group Lasso-granger

Emotion is closely related to human life.It is controlled by the human brain's central nervous system to produce a corresponding emotional state.Emotion recognition is the use of computer technology to perceive changes in human emotional state,so that it can accurately recognize human emotions in different environments.At the same time,it can continuously promote and improve the progress and development of human-computer interaction technology.At present,many scholars use the physiological signals such as respiration,myoelectricity,electrocardiogram,and brain electricity to conduct emotional recognition research.Among them,EEG signals are mostly studied,because EEG can fully reflect the brain's response under different cognitive tasks and functional states.It can express the whole neuron activity of the brain,and it will not be influenced by artificial deliberate concealment.Therefore,emotion recognition research based on EEG can reflect the changes of human emotional state more objectively and fully.The research on emotion recognition based on EEG signals mainly focuses on two key steps of feature extraction and emotion classification.The EEG signal of human cerebral cortex is a non-stationary random signal containing many interference signals,which needs to be corresponding in the research process.The pre-processing extracts valid emotional feature information,and then uses the classification algorithm to classify the extracted emotional features to identify different emotional states of the subject.This paper proposes a new EEG feature and emotion recognition classification method based on the research of EEG signals,the main work is as follows:1.An EEG emotional feature based on the lasso-granger causality of sparse group is proposed.In the experiment,the granger algorithm is used to extract the granger causal eigenvalues of the ?,? and ? EEG bands,and then the causal features are screened by the sparse lasso algorithm to obtain high correlation feature subsets.Finally,the SVM classifier is used to classify the emotions.In addition,in the feature extraction process,the ReliefF algorithm is used to select effective EEG channels to reduce the calculation time.Experiments show that the emotional features of EEG extracted by this method can effectively identify the different emotional states of the subjects,and obtain the average emotion classification accuracy of 87.15% and 86.60% on the Valence-Arousal two-dimensional emotion model.The classification effect is better than the comparison.EEG features.2.An emotion classification method based on CapsNet neural network is proposed.The sparse group lasso-granger causality method is used to extract the granger causality feature of the original EEG signal.The obtained high correlation feature subset is used as the input of the network to realize the final emotion classification.Experiments show that the CapsNet neural network is used to classify EEG signals by adjusting the network structure and model parameters.The average emotion classification accuracy of 88.09% and 87.37% is obtained under the Valence-Arousal emotional dimension,compared with SVM ? CNN and DBN.The system can get better results and significantly improve the performance of EEG emotional classification.