Research On The Classification Algorithm Of Emotional Disorder Tendency Based On Electroencephalogram Signal Processing

With the acceleration of the social life,more and more people are affected by emotional disorders.This intrusion has gradually penetrated into the youth group.Depression,anxiety,mania and other emotional disorders will interfere with the normal learning and life of young people and their physical and mental health.In many cases,emotional disorders do not progress to overt symptoms,but simply manifest as a tendency.Emotional disorder tendencies can also have a negative impact on the teen’s psyche and are at risk of further aggravation.Such tendencies are often difficult to use general methods for diagnosis and analysis.Due to the strong correlation between human emotions and EEG signals,machine learning and deep learning methods can be used to analyze EEG signals to achieve more efficient auxiliary judgments.The main contents of this thesis are as follows:(1)A system process for EEG signal preprocessing is proposed.The process includes three parts: signal filtering,baseline correction and artifact removal.The signal filtering part is mainly used to remove the high and low frequency noise in the EEG signal and the power frequency noise of the electrical equipment.The baseline correction part mainly eliminates the time drift component of the EEG signal over time.The artifact removal part is mainly based on independent component analysis and ADJUST plug-in to remove physiological artifacts contained in EEG signals.The preprocessing process proposed in this thesis has general applicability to various EEG signals.(2)Extract and select EEG features,and use a machine learning classifier to classify emotional disorder tendencies.In this thesis,a variety of features are extracted from the four different sub-bands and the full-band of the EEG signal.In addition,each sub-band and the full-band are spliced to extract features as a new frequency band.The tree model-based feature selection algorithm was used to select 237 EEG features for classification,and the feature selection algorithm could significantly improve the classification accuracy.Two different classification tasks are designed.In the classification task based on EEG segmentation,three classifiers on each frequency band are tested.The classification performance of the three classifiers on the spliced frequency band is better than that on the divided frequency band and the full frequency band,the average classification accuracy rate reached 80%.The K-nearest neighbor classifier performed better in the high frequency band,with a maximum classification accuracy of 95%,and the random forest classifier performs better in the low frequency band,reaching an average classification accuracy of 60%.In the classification task based on the subjects,three classification models were tested on the spliced frequency band.The support vector machine could obtain the best classification effect,with an accuracy rate of 68%.Each test results show the effectiveness of the feature selectionclassification method proposed in this thesis.(3)The EEGNet neural network model is introduced to classify different emotional disorder tendencies based on EEG signals.The EEGNet network structure is built based on the depthwise separable convolution.Ten-fold cross-validation was used to select the model with the highest verification accuracy,and the emotional disorder tendencies were classified at the EEG segmentation and individual level In the classification task based on EEG segmentation,EEGNet can achieve an average classification accuracy of 94% for each emotional disorder tendencies.In the classification task based on the subjects,EEGNet can achieve a classification accuracy of 72%,both excellent on the performance of machine learning classifiers.The test results show that EENGet has a good classification effect on a small sample data set without relying on artificial features,which is of great significance for the classification of emotional disoder tendency.