Research On Emotion Classification Approaches Base On EEG Signals

Emotions are closely related to people's mental health.If people stay in an abnormal mood for a long time,their mental health will be greatly hurt and even suffer from serious illness.Therefore,accurate emotion recognition and classification contribute to disease prevention and diagnosis.Because of emotions are related to brain activity and EEG acquisition technology is maturing.This article will be based on EEG to achieve emotional recognition and classification.EEG-Based emotion classification consists of EEG acquisition,artifacts removal,feature extraction and emotion classification model construction.This paper will focus on the study of EEG artifacts removal and EEG-based emotion classification which includes feature extraction and model construction.Exploring and experimentally verifying the existing problems and possible solutions to these two tasksThere are two main problems with artifacts removal.First,the scope of artifacts removal is narrow.Most of EEG artifacts removal studies focus on ocular artifacts,and others like Muscle artifacts and Cardiac artifacts are less.The second is that the artifacts removal algorithm is less applied to the actual EEG database and the effect is relatively poor.In order to address these two problems,this paper proposes wavelet packet decomposition and artifacts information enhanced independent component analysis algorithm.The wavelet packet decomposition algorithm is used to extract the band information of the artifacts which will help the algorithm to remove the artifacts correlated independent components as well as eliminates various kinds of EEG artifacts.Therefore,more effective brain electrical information can be retained and better artifacts removal effect can be achieved.Emotional classification based on EEG also has two problems.Firstly,the amount of data in the EEG database is insufficient,resulting in deficient training of the emotional classification model and reducing the classification accuracy.Secondly,it is difficult to study the form and regularity of emotion in EEG,which make emotional recognition and classification is difficult to improve.This article innovatively proposes the concept of emotional patches,obtained by dividing the EEG feature map with time domain information,can increase the number of samples in the database and capture emotions in EEG and learn their regularity over time.Combining with deep learning,this paper proposes Emotional Patches based Deep Belief Networks which achieve the high-precision emotion recognition and classification.This article respectively carries out experimental verification of the above two tasks.For EEG artifacts removal,this article uses the VR EEG database collected by the laboratory to perform experiments.Experimental results show that the stability and effects of the algorithm proposed in this paper are the best.Efforts are also being made to study fear emotions in clean EEG to lay the foundation for emotional classification.For EEG-based emotion classification,this article utilizes the Shanghai Jiao Tong University's public EEG database to extract emotional patches and analyze its ability to capture the law of emotions in the time domain.Finally,we construct a classification model to achieve three classifications of emotions which accuracy reaches 93.15%.Experimental results demonstrate that our method performs better than other emotion classification methods.