Research On Emotion Recognition Based On EEG

With the maturity of artificial intelligence technology and the continuous application of human-computer interaction technology,emotion recognition through EEG signals has become a hot research topic in the field of artificial intelligence.Electroencephalogram(EEG)as a physiological activity of brain nerve cells in the cerebral cortex is an overall response.Due to its characteristics that it is not easy to disguise and reliable,it can objectively and accurately reflect people's internal emotional state.How to construct feature engineering based on EEG emotion signals and ensure the quality of emotion features,and improve the recognition rate and generalization of emotion classification models are still challenges in the field of EEG emotion recognition.In this paper,Deap database is used as the research object to analyze five common emotions: relaxation,happiness,depression,stress and calmness.The main work is as follows :Firstly,db4 wavelet transform is used to preprocess the EEG emotional signals,and the linear feature(wave index)and non-linear dynamic features(sample entropy,approximate entropy,permutation entropy,and Hurst index)in the sub frequency of the EEG signal are extracted.And proposes a weighted feature fusion method based on the principle of feedback.Compared with the traditional simple feature combination,the weighted feature fusion can better highlight the impact of different emotional features on the classification result,so that the more relevant features get larger weights.Finally,the Twin Support Vector Machine(TWSVM),which is faster in computing efficiency,and is used to classify and recognize EEG emotions.The average accuracy rate is 88.2%,which is a certain improvement over the traditional feature combination classification accuracy.Secondly,taking the convolutional neural network(CNN)in deep learning as the basic prerequisite,improving the traditional CNN network structure.A new method of EEG emotion recognition is proposed,which combines the RCNN with the LSTM model.Utilizes the ability of RCNN to automatically extract the abstract features of EEG emotion signals,eliminating the need for artificially designed features and calculations,and single-step segmentation of the original EEG emotion signals.Through the RCNN feature extraction after many times emotional feature fusion.With the advantage of LSTM for classification and recognition of time series,it has a good generalization ability and robustness to the temporal,spatial and frequency domain features of EEG emotional signals.In the end,the average classification recognition rate of emotions obtained by experiments reached 94.63%.The results show that the RCNN-LSTM model can help people effectively recognize EEG emotion signals.Thirdly,based on the traditional residual neural network(Resnet),a multi-scale attention residual neural network(MAResnet)model for EEG emotion signal classification is designed.The attention mechanism is added to the traditional residual learning block,so that it can learn the relevance and importance of the features of different channels of the input,through parallel use in the same space In order to avoid the degradation of the network,the multi-channel output is obtained by convolution kernels of different sizes,so that the multi-scale feature extraction of EEG emotion signal is carried out,and the residual learning is carried out in the network model.The experimental results show that the recognition rate of the original Resnet is 68.96%,and the average accuracy rate of MAResnet is 84.56%,which proves the effectiveness of the method.