Feature Analysis Of Motor EEG Signal Based On The Fusion Of Brain Complex Network And Machine Learning

Human brain is the most complex system in human structure and function,it is a great challenge to study human brain.Speeding up brain science research not only improves the diagnosis and treatment of brain science,but also dominates the leading role of scientific research in the future,promotes the revitalization of global industries,and brings economic and social benefits.Therefore,Brain-Computer Interface(BCI)has been developed rapidly with its unique advantages and technical attributes.Through BCI technology,we can help patients with movement disorders to regain their movement ability,and use electroencephalogram to realize the operation of related auxiliary equipment,which is significant for the patients to recover to the normal living level.At the same time,an important branch of current BCI research is the analysis of exerciserelated electroencephalogram to explore the medical and non-medical uses of exerciserelated electroencephalogram.The research on the functional connection of the brain region during exercise can better explain the internal connection between brain regions,thus serving the field of brain-computer interface.Based on this,the main work of this paper is as follows:(1)Based on complex network method and transfer entropy theory,a dynamic transfer entropy network construction method is proposed.This method combines timefrequency analysis with transfer entropy to analyze the special rhythms of brain regions such as default network under exercise stimulation,so as to reveal the extraction of stable brain response signals under exercise stimulation.The research process of this paper is as follows: By extracting the original signal,removing the artifacts such as eye movement and power frequency interference in the data set,using the fast Fourier transform and sliding window method to divide the EEG data into bands,and constructing a directed network to transmit entropy for EEG in different rhythms,so as to measure the information transmission intensity between the cortex during sports-related tasks.The results show that default network and Motor cortex,default network and prefrontal are significant in sports-related tasks,and the results are universal in different sample spaces.(2)Propose a classification method of motor EEG signals based on machine learning fusion transfer entropy,which combines nonlinear analysis with time-frequency analysis.Firstly,the original signals are preprocessed,and the interference such as eye movement artifacts in EEG data is removed by baseline sampling.Then,the disturbed EEG signals are divided into frequency bands by FFT,and the selected EEG data is reduced to a onedimensional sequence by weighted average.The maximum,minimum,median,variance,standard deviation and other features in the data are extracted as feature vectors.Finally,these feature vectors are used as the input of the classifier and classified by logistic regression,decision tree and XGBoost.The results show that the construction of transfer entropy network can effectively improve the classification accuracy when classifying sports-related tasks.(3)Aiming at the problems of simple paradigm and low complexity of the current sports-related tasks,the EEG signal acquisition experiment under multi-sports tasks is carried out,and the features of EEG are extracted through functional networks.By constructing the graph convolutional neural network(GCNs-Net)model,the EEG signal classification under multi-sports tasks is completed.The model comprehensively evaluates the classification accuracy through EEG and behavioral indicators.The final experimental results show that the framework has good classification accuracy when classifying sports tasks,which also proves the effectiveness of multi-sports task classification under this framework,and provides feasible ideas and methods for the stability and accuracy of sports EEG identification in brain-computer control.