Research On Emotion Recognition Method Of Brain Signal Based On Deep Learning

Emotions affect everyone’s life.Positive emotions can improve human health and work efficiency,while negative emotions can cause health problems.Emotion recognition research has positive help in various fields such as safe driving,mental health monitoring,and social safety.With the rapid development of wearable EEG devices,simple,portable and accurate EEG signal acquisition equipment and emotion recognition technology have become research hotspots in recent years.Because of its low acquisition cost and high time resolution,EEG signals have attracted more and more researchers to invest in EEG-based emotion recognition research.Most of the existing researches mainly start from the aspects of EEG signal feature extraction and model optimization,and there are problems such as low accuracy of emotion recognition and complex models.In recent years,with the improvement of computer computing power and the development of deep learning technology.Deep learning is gradually applied in EEG emotion recognition tasks,and has achieved good results.Using deep learning technology to perform emotion recognition tasks on EEG signals has become a current research hotspot.First of all,this thesis is based on the deep learning method for EEG emotion recognition task,extracts the differential entropy of EEG signal as the characteristic of EEG signal,at the same time,preserves the original spatial information of EEG signal,and constructs a three-dimensional feature input model.Emotion recognition in the process.Secondly,this thesis proposes an EEG emotion recognition model base on deep residual network and attention mechanism,using attention mechanism to extract emotion-related features in EEG signals,dig out important information,and suppress redundant information,noise,etc.Through experiments on the two public data sets of DEAP and SEED,the experimental results and analysis show that the method proposed in this thesis obtains better recognition accuracy than traditional methods and most deep learning methods.Finally,this thesis attempts to improve on the SEED data set based on the shortcomings of the existing model,and uses the graph attention network to filter the edges in the original PLV correlation matrix,and retains its important edges to better represent the electrode channels.The new PLV correlation matrix obtained is then used as the adjacency matrix of the graph convolutional network for emotion recognition tasks.Experiments show that the method proposed in this thesis provides a new idea for EEG emotion recognition.