Research On Tranfer Learning Of Motor Imagery EEG Based On Riemannian Space

Brain Computer Interface(BCI)technology provides another way of interaction between brain and machine by identifying brain activity and transforming it into external commands.Sports imagine that brain electrical signals have attracted widespread attention due to their non-invasiveness and easy collection.As an artificial intelligence technology with huge potential,the brain-machine interface system based on sports imagination brain electrical signal has a broad application prospect in many ways such as education life,medical rehabilitation and national defense security.However,due to the characteristics of non-stable and low signal-to-noise ratio,the distribution of the characteristics of the Electrical Electrical Signal between different tests between the tests of the depending on the test of the noise and noise ratio is very different.Construct a precise classification model.The longer training time not only increases the user’s workload,but also weakens the practicality of the brain interface system.Migration learning uses the category model of the current subjects with label samples of other subjects,which can achieve the purpose of reducing and even eliminating the time of calibration.At present,most migration learning algorithms still need to have labels with labels with labels and the accuracy accuracy rates are not high.In response to this issue,this article proposes a motor imagery EEG signal classification algorithm based on Riemannian Space Feature Transfer Learning(RFTL),and is carried out on the public sports imagination of data sets and laboratory self-mining data sets.Verification of the experiment’s content,including the following components:(1)Construct the Riemann alignment framework to operate the coordinated differential matrix with the average value of Riemann and the Riemann differential vector,so as to achieve the rotation of brain electrical data in the Riemannian space.Using the similarity of Electrical Electrical Signals and image signals,the combined distribution adaptation method that performs excellent performance in image processing is referenced to the processing motion imagination of Electrical Electrical signals,and the maximum mean difference distance is used to quantify the distribution differences between different fields.(2)A motion imaging EEG signal feature transfer learning algorithm based on Riemannian space is proposed,and the framework structure and basic process of the algorithm are introduced in detail.Firstly,the source domain and target domain data are distributed aligned in Riemann space,and then the joint distribution adaptation is used to reduce the data distribution differences between different domains,and a domain-invariant classifier model suitable for target domain tasks is constructed.(3)Experimental analysis was carried out on the open competition dataset and the laboratory self-harvesting data set.Experimental results show that the RFTL algorithm can effectively overcome the inconsistency of cross-domain distribution,greatly improve the recognition accuracy of motor imaging EEG signals across objects,and then greatly improve the learning ability of BCI system and improve the universal applicability of classification models.