Research On P300-EEG Transfer Learning Classification Algorithm Based Riemannian Geometry

Brain-computer interface system refers to a technology that directly allows the brain to interact with external systems without relying on the nervous system and muscle tissue.The most common EEG signal in the brain-computer interface is the P300 signal.The P300 EEG signal is an event-related potential and is a signal generated by the brain when it receives a certain external stimulus.In the brain-computer interface technology based on P300 EEG signals,the most common application is the P300 character input device,and the subject can complete the character input using only the brain through a specific experimental paradigm.However,due to the instability of EEG signals and the differences between people,this type of system requires a long time of pre-training before each use,so that the performance of the system is more stable and the classification accuracy is higher.This long-term pre-training is not only time-consuming but also very energy-intensive,especially for some patients with impaired mobility in clinical practice,it is an inhuman experience.Therefore,the research in this paper mainly focuses on how to implement the P300 EEG signal migration learning algorithm,so that the subjects do not have to repeat the long pre-training process every time,but can use the previous training data or other subjects.Training data the main work of this article is as follows:This paper proposes a cross-subject transfer learning algorithm based on XDAWN spatial filter and Riemann geometric classifier.First,use the XDAWN spatial filter to filter and reduce the original P300 EEG signal,and then directly map the EEG signal into the Riemannian manifold space in the form of a symmetric positive definite covariance matrix,and then use the subject’s respective The Riemannian geometric mean performs an affine transformation of its symmetric positive definite covariance matrix on the Riemannian manifold space,making data from different subjects comparable in the Riemannian manifold space,and finally uses the Riemannian geometric classifier for instance transfer learning Experiments,the experimental results show that the method proposed in this paper has a greater improvement than the traditional method,the stability is higher,and the experimental results fluctuate little.This paper proposes a cross-subject feature transfer learning algorithm based on Riemannian manifold tangent space.The feature space of each subject is regarded as a subspace,and the data of each subject is mapped in after the Riemannian manifold space,find the Riemannian geometric mean point of each subject,and construct the Riemannian geometric tangent space at that point,and construct a common feature by combining the Riemannian geometric tangent spaces of all subjects Space,project data into this common feature space,and perform vectorization stitching to obtain new features.Finally,a traditional support vector machine is used to complete a migration learning classification algorithm based on feature migration.The experimental results show that,compared with different frontier processing methods,the feature transfer algorithm of Riemannian manifold space proposed in this paper performs better than other algorithms on average,which verifies the feasibility of the algorithm.