Research On Recognition Method Of Brain Cognitive State Based On Transfer Learning

Brain computer interface(BCI)technology constructs a new connection between brain and external quipment,which has been used in many fields,such as medical treatment,transportation,entertainment and so on.BCI technology based on electroencephalogram(EEG)due to its non-invasive and high time resolution,it is widely used in BCI system.However,in the current BCI technology,the time and economic cost of EEG data annotation are large,and the EEG signal has time-varying and individual differences,which leads to the long calibration time of BCI system and poor generalization ability of recognition methods.In view of the above problems,this paper carries out the following research.Firstly,a sample labeling method in source domain based on semi-supervised learning is proposed.By using a small number of labeled samples in the source domain,unlabeled samples are selected and labeled.The collaborative training method,sample confidence measurement method,and dynamic adjustment mechanism are proposed to improve the labeling accuracy of unlabeled samples,in order to provide a large number of reliable labeled training samples for the target domain.Secondly,a brain cognitive state recognition method for multi-class motor imagery based on semi-supervised active transfer learning is proposed.First of all,through the similarity calculation,the source domain with similar distribution to the target domain is selected,and the samples with higher information entropy are selected from these source domains for transferring.Then,multiple classifiers are trained by the labeled samples in the target domain and the transferred samples in multiple source domains,and the results of these classifiers are weighted and fused to identify the target domain samples.Finally,the samples with large amount of information are selected from the unlabeled samples in the target domain,and the labeled samples are added to increase the number of labeled samples in the target domain.Finally,by experiments based on the source domain of the semi-supervisor learning,the selection of source domains,transfer learning and brain cognitive status identification,the experimental results show that the brain cognitive state recognition method for multiclass motor imagery based on a semi-supervised active transfer learning,the average identification accuracy of the target domain has reached 70.52%,which has better identification performance and training efficiency compared to the methods of the current application.