| The problem of attention has always been the most important issue for drivers,pilots,etc.Once their attention is scattered,it will lead to serious traffic accidents.It can reduce the incidence of accidents by monitoring and alerting the driver of attention of different states.The characteristic parameters that can accurately distinguish different attention levels are found,which not only helps to establish a biofeedback system that regulates the attention level of human,but also is beneficial for the diagnosis and treatment of attention-related brain diseases.This thesis aims to achieve an accurate classification of attention.The non-linear features of attention EEG are extracted by using the methods of sample entropy and approximate entropy.However,this method takes longer to compute.The maintenance of attention involves multiple brain regions of the brain,but the above feature extraction methods focus only on the local features of the brain but ignore the interactive features of different functional regions of the brain.The graph-theoretic analysis based on the brain network can obtain information transmission characteristics in different brain regions.This thesis starts from the measures of brain network,and finds the characteristic parameters that represent different levels of attention,and achieves an accurate classification of attention.The main results of this thesis are as follows:(1)Design paradigm and data acquisition.First,experimental paradigm of attention 1(high),attention 2(low)and non attention are designed to obtain EEG data from different states of attention.Then,the experimental data are pretreated.(2)Construct brain network.19 important heads covering the brain are chosen as nodes of the brain network through using sparsity.Then,the edges are established by using the coherence coefficient method and phase locked value method.And then,the attention function brain network is constructed.(3)Analyze feature of attention EEG.It is calculated that network measure of node degree,clustering coefficient and feature path length of attentional brain network based on coherence coefficient(COH)and phase-locked value(PLV).The network measures are analyzed and t-tests are used for significance analysis.Statistical analysis shows that there are significant differences between the three levels of attention at the majority of leads,and the higher the concentration of attention of the subjects,the lower the value of the node degree and the clustering coefficient.This shows that node degree and clustering coefficient of the network can well distinguish different attention levels.(4)Multilevel classification of attention levels.The network measures such as node degree,clustering coefficient and feature path length are extracted as feature parameters,and are classified using support vector machines(SVM).The results of 10-fold cross-validation show that the average classification accuracy of two-level attention level reaches 84.54%,and the average classification accuracy of multi-level attention level reaches 73.43%.This proves that brain network measure based on EEG provids a new way for more accurate assessment of people's attention level. |
