Research On Deep Learning Modeling Method For Cognitive Function Status Assessment Based On EEG

The cognitive function is the primary function for human beings to understand and feel the world.It is closely related to emotion and affects the expression of emotion.Once the cognitive function is damaged,it will cause great harm to the patient’s brain health,and in severe cases,it will endanger the patient’s life.The evaluation of brain cognitive status through deep learning modeling methods is conducive to the rapid detection of changes or damage to the cognitive status of the brain.Early detection and early treatment are very valuable for improving the brain health of patients and improving their quality of life.Deep learning modeling based on EEG(Electroencephalography),especially Vision Transformer(Vi T)modeling for cognitive function status assessment has attracted widespread attention.However,due to the problems of low signal-to-noise ratio,instability,and significant individual differences in EEG signals,the existing Vi T models still have deficiencies in the joint representation of global and local features of EEG signals,the fusion of features in different frequency bands,and the coupling of spatiotemporal features.It is still very challenging to design an efficient,robust,and generalizable deep learning model based on Vi T for EEG-based cognitive function status assessment.In response to the above challenges,the research work of this thesis is as follows:(1)For the problem that the existing Vi T model is not strong enough to jointly represent the global and local features of EEG signals,this thesis proposes a Vi T-based Domain Adaptation(DA)cognitive function state assessment model called DACVi T.The model combines the self-attention mechanism and the convolution mechanism to simultaneously extract the global and local features of the EEG sample to improve the representation ability of the EEG signal;through the DA method,the distribution difference between the source domain and the target domain data is minimized to improve the ability to assess cognitive functional status across subjects.(2)For the problem that the existing Vi T model is insufficient to extract fusion information of different frequency bands of EEG signals,this thesis proposes a cognitive function state assessment model called DAMBFCVi T based on multi-band fusion features.The model uses a multi-band fusion feature encoder that combines the self-attention mechanism and the convolution mechanism to extract the fusion features between five frequency bands,which effectively solves the problem of insufficient fusion information extraction between different frequency bands.(3)For the insufficient extraction of spatiotemporal coupling features of EEG by existing Vi T models,this thesis proposes a domain-adaptive cognitive function state assessment model called DATSCVi T based on temporal and spatial coupling features.The model combines a multi-level transformer of Transformer i N Transformer(TNT)with a hierarchical spatial learning transformer(HSLT)which has the ability to extract spatial information of the hierarchical brain,a time-spatially coupled EEG Transformer encoder is proposed to extract the spatial-temporal coupling characteristics of EEG signals,so as to effectively solves the problem of insufficient spatiotemporal coupling feature extraction.(4)Based on the above three models,this thesis designs an EEG-based cognitive function status assessment system CFSAS.The system can conveniently collect,analyze,and evaluate the cognitive function status of EEG signals.