Research On Emotion Recognition Of EEG Based On Deep Learning

Emotion recognition refers to identifying the corresponding emotional state through human behavior and physiological response.Emotion recognition plays an important role in the field of human-computer interaction and artificial intelligence,and has broad application prospects in the fields of education,medical treatment and life.In recent years,with the rapid development of deep learning,deep learning methods are more and more widely used in emotion classification tasks of EEG.Because EEG signals have the characteristics of nonlinearity,complexity and high dimensionality,it is still a great challenge to extract emotion related,sensitive and effective feature components from EEG signals.In this thesis,the deep learning method is used to study EEG emotion recognition from the construction of emotion classification model and the extraction of effective emotion features.The research content mainly includes the following two aspects:(1)Aiming at the problem that traditional machine learning methods need to design features manually,using the advantage that deep learning method can automatically extract features.An EEG emotion recognition model based on CNN is proposed,namely ATCRNN model.The model converts one-dimensional EEG sequence into two-dimensional matrix sequence data through the Transfer Layer,then inputs the matrix sequence into multi-layer CNN to automatically extract spatial features,and then extracts spatio-temporal features through the BILISTM Layer.Finally,subject-dependent verification experiment is carried out on two public datasets.The experimental results show that the average accuracy of Valence and Arousal of the model on DEAP dataset is 91.48% and 91.59%.And Valence on SEED dataset is 94.41%.Compared with the same type method is the best classification results have been achieved.In addition,ablation experiments are carried out on the Transfer Layer and Attention mechanism of the model.The experimental results show that the classification accuracy of the model is greatly improved after adding the Transfer Layer and Attention mechanism.(2)Aiming at the limitation of CNN of scalar neurons in extracting the spatial relationship between brain electrode channels,this paper considers the spatial information,time information and attention information of multi-channel EEG signals at the same time,and proposes an EEG emotion recognition model based on Capsule Network,namely ATCaps LSTM model.The model first integrates the Attention mechanism into the CNN,extracts the differences between channels from EEG signals,then uses the Capsule Network of vector neurons to extract the relative relationship between the local and overall brain electrode channels,and then sequentially connects the LSTM to extract the temporal characteristics of EEG sequence.Finally,subject-dependent verification experiment is carried out on two public datasets.The experimental results show that the average accuracy of Valence and Arousal of the model on DEAP dataset is 97.17% and 97.34%.And Valence on SEED dataset is 98.33%.The experimental results show that the EEG spatial features extracted by Capsule Network are more representative than CNN.