Research On Optimization Of Medical Data Processing Algorithm For EEG Signals Based On Deep Belief Networks

The brain is the most mysterious and complex organ of human beings.So far,people have not solved the mysterious veil of the brain.EEG signal is the "voice" of brain and external communication,and it is the "medium" for people to explore brain.At present,the research on brain computer interface(BCI)is a hot field of brain function research.In this technology,people’s intention can be judged by EEG signals,so that they can communicate with the outside world without relying on limbs or language organs,which brings good news for patients with language loss and other diseases.The feature extraction and classification of EEG signal is an important part of the brain computer interface technology research,and it is also the key content of this paper.In this thesis the feature extraction methods of EEG signal are studied in depth,mainly including time-domain,frequency-domain and time-frequency-domain extraction methods.Furthermore,the classification methods of EEG signals are studied,mainly including two modes of classification methods: shallow and deep machine learning models.In addition,this paper uses a combination of deep machine learning model points and wavelet packet analysis to extract EEG signal features,and then the off-line EEG data sets are classified.The main research contents and innovations of this article are summarized as follows:(1)In this thesis,we study the theoretical basis of Deep Belief Netwroks(DBN)model,and combine with the wavelet packet analysis method to extract the characteristics of EEG signal,select the most suitable wavelet base,network layers and wavelet packet decomposition layers to extract EEG signals Feature and classify EEG signal.(2)This thesis explores the application of DBN model in the field of EEG signal classification.According to the characteristics of multichannel EEG signal,the multichannel DBN model is applied to the field of EEG signal classification.For the first time,the wavelet packet transform method is combined with the multichannel DBN model to extract data features and classification,and compared with the common single channel DBN model,the superiority of this method is proved(3)In this thesis,the advantages and disadvantages of the training algorithm in DBN model are discussed.For the first time,the Free Energy In Persistent Contrast Divergence(FEPCD)algorithm is applied to the training of multichannel DBN model,and it is used for the first time in EEG signal classification field.The core idea of the FEPCD algorithm is to extract elite samples according to the size of the free energy of the samples,and then use the Persistent Contrastive Divergence(PCD)algorithm to train the DBN model.The advantages of this method are proved by Contrastive Divergence(CD),PCD and FEPCD.