Research On Code Modulation Visual Evoked Potential Brain-computer Interfaces Based On Riemannian Manifold

The brain-computer interface(BCI)system is a communication system that can communicate with the outside world without relying on peripheral nerves,muscles and using external equipment.It converts the EEG signals generated by brain activity into control commands after analysis,so as to realize communication with the outside world by controlling external devices.Based on the Riemannian manifold,this paper studies the code-modulated visual evoked potential(c-VEP)BCI.In Euclidean space,c-VEP BCI usually uses canonical correlation analysis(CCA)algorithm for target recognition,and obtains a high classification recognition rate and information transfer rate.This research uses the algorithm as a Riemann space comparison algorithm,and optimizes the results by processing the original data set through the number of channels,the number of training experiments trials which are used to obtain the template,and the data length.Although the development history of Riemann space in BCI research is relatively short,it enables transfer learning to be successfully applied and greatly improves the performance of the BCI system.This research introduces how to process c-VEP data in Riemann space.In the case of the same preprocessing of Euclidean space,firstly construct a super trial to integrate the template data of all trials with the data that needs to be classified;secondly,covariance estimation;and finally,after using the Riemann metric to calculate the Riemann mean of different targets,use the Minimum Distance to Mean(MDM)classifier for classification and recognition.Dimensionality reduction can not only avoid the disaster of dimensionality,make the data more separable,but also reduce the running time and improve system performance.This study reduces the dimensionality of the data in different spaces: 1)In the Euclidean space,the EEG data is spatially filtered using a spatial filter to reduce the dimensionality of the multi-channel data to single-channel data;2)In the Riemann space,the coordination Project the variance matrix to the tangent space and then reduce its dimensionality;or project the data from a high-dimensional Riemannian manifold to a low-dimensional Riemannian manifold.This paper classifies the test data after dimensionality reduction,and compares and analyzes the classification results with the classification results of traditional methodsThe study found that when the number of channels used for classification and the length of the data were 9 and 3 cycles respectively,the average classification recognition rate of the subjects was the highest;in the Euclidean space,when about 30 training experiments were used,the classification results were the best.In Riemann space,when 22 training experiments are selected,the classification results are the best.