Study On Filtering Optimization Algorithm Of Visual Evoked Potentials Brain-Computer Interface Based On Code Modulation

The brain-computer interface(BCI)is a communication system that converts human intentions into control signals to control external devices and communicate with the outside world.The code modulated visual evoked potential(c-VEP)BCI has a high information transmission rate,but has not been studied in depth.In previous studies,spatial filtering was used to integrate multi-channel electroencephalograph(EEG)signals,and canonical correlation analysis(CCA)was used to optimize the spatial filter coefficients.However,due to the visual latency of the human brain cannot be accurately estimated,the quality of the extracted characteristic signals after filtering is not high,which affects the classification performance.From the perspective of signal processing,this paper expands the spatial filter into a spatiotemporal filter,aiming to improve the classification recognition accuracy of the stimulus targets.Several different filtering optimization algorithms are studied,and linear spatiotemporal filtering and nonlinear spatiotemporal filtering are analyzed respectively.Linear spatiotemporal filter is optimized by the least absolute shrinkage and selection operator(LASSO)algorithm,and nonlinear spatiotemporal filter is optimized by neural network algorithm.The results identified by the template matching method indicated that the spatiotemporal filtering algorithm based on LASSO optimization achieved averaged classification accuracy of 95.83%,which is higher than averaged classification accuracy of 94.38% yielded by the CCA-optimized spatial filtering algorithm.The best classification performance can be achieved by selecting data length of a stimulus cycle for target recognition and selecting 100 training experiments.This paper also proposes a classification algorithm that combines multiple different sub-filtering optimization algorithms.Based on the classification results of each sub-algorithm for stimulus target,the classification results of the fusion algorithm are obtained according to the direct method,the voting method and the maximum correlation coefficient method,respectively.CCA-optimized spatial filtering algorithm,LASSO-optimized spatiotemporal filtering algorithm and beamformer-optimized spatial filtering algorithm are used to constitute the fusion algorithm.The experimental results show that the average classification accuracy of the fusion algorithm reaches 97.50%,which is obviously better than the classification accuracy of the sub-algorithm mentioned above.By comparing the correlation coefficients of the sub-algorithms in each target recognition,it is found that the average value of the correlation coefficient of the LASSO-optimized spatiotemporal filtering algorithm is the largest,so its contribution to the fusion algorithm is the largest.