EEG Feature Extraction And Classification Method For The Assessment Of Spatial Cognition Of The Elderly In The Community

With the rise of artificial intelligence,brain science has become a hot topic in the field of cognitive science.Regarding the development of brain science cognition,the research headed by spatial cognition training and evaluation has become the focus of researchers at home and abroad.At present,great progress has been made in the recognition of EEG data of spatial cognitive training,mainly includes the extraction of coupling features of corresponding EEG data and diversified data classification.However,the current task state EEG feature extraction of spatial cognitive training has some shortcomings.when calculating the coupling characteristics of two EEG channels,the influence of other channels and the spatial position of the EEG channels were less considered.In addition,it is also a hard problem in the research of EEG data,that is,whether a classifier with better generalization performance and higher classification performance can be found.Therefore,this paper proposed a new EEG coupling feature extraction method and classification method based on the current domestic and foreign research materials,combined with deep learning and other technologies.Firstly,an EEG coupling feature extraction method based on multivariate permutation conditional mutual information common spatial pattern(MPCMICSP)is proposed,which breaks the strict linear assumptions of traditonal co-space model.In MPCMICSP method,the covariance matrix in the traditional co-space algorithm is replaced by the multidimensional sorting coditional mutual information coupling coefficient martix,which not only considers the influence of other EEG channels on the coupling characteristics,but also establishes the corresponding spatial filter according to the linear correlation degree of EEG signal,so as to obtain the corresponding EEG spatial characteristics,which further improves the previous brain electric feature extraction method.Then,the convolutional parallel fusion strategy is introduced on the basis of the convolutional neural network,and a multi-scale convolution feature parallel fusion classification model is proposed.This is due to the fact that the single input size of the traditional convolutional neural network,it is unable to obtain diversified image information,so multi-scale convolution kernel is used to improvethe image quality;In addition,in order to achieve the complementarity between the high and low level features,the parallel fusion strategy in the Res Net network is introduced and combined The gradient automatic stochastic descent method is used to optimize the model to improve the stability of the network structure.Finally,in order to verify the performance of the above two algorithms,this paper selected two test data before and after training as the data set,and compared the past feature extraction methods and classification methods with the new algorithm proposed in this paper.According to the results,the new algorithm proposed in this paper has higher accuracy and better robustness than the previous algorithm,and has a certain improvement on the recognition of EEG signals.