基于多元经验模式分解的SSVEP目标识别研究

Study on Steady State Visual Evoked Potential Target Recognition Based on Multivariate Empirical Mode Decomposition

在脑-机接口(BCI)系统中,对稳态视觉诱发电位(SSVEP)的准确识别在生物医学等领域是至关重要的,而各种伪迹影响了识别准确率。提出一种基于多元经验模式分解的多元同步指数法(MEMD-MSI),首先用白噪声辅助的多元经验模式分解(MEMD)对原信号进行分解,各通道提取出前6个本征模式函数(IMF)分量,提出通过网格搜索法对IMF分量进行加权,从而剔除EEG信号中的伪迹,保留脑电信号中的有效信息,6名受试者的信号数据用来筛选加权系数。接着用多元同步指数法(MSI)对重构信号进行识别。另外,选取了9名受试者的信号数据,对比了MEMD-MSI,MSI及多元经验模态分解典型相关分析(MEMD-CCA)3种算法在不同时窗的准确性。结果表明,MEMD-MSI在三种算法中有着最高的准确率,且在时窗大小为2 s时,其准确率达到了95.24%。证明该算法有效地剔除了伪迹,具有高准确率。

In the brain-computer interface(BCI)system,the accurate recognition of steady-state visual evoked potential(SSVEP)is significant in biomedical and other fields,and various artifacts affect the recognition accuracy.We propose a multiple synchronization index based on multiple empirical mode decomposition(MEMD-MSI).Firstly,the original signal is decomposed by white noise assisted multiple MEMD.The first six intrinsic mode functions(IMF)components are extracted from each channel.A grid search method is proposed to weigh the IMF components so as to eliminate the artifacts in the EEG signal and retain the effective information in the EEG signal.The signal data of the six subjects are used to choose the weighting coefficient.Then,MSI is used to identify the reconstructed signal.Besides,the signal data of 9 subjects are selected to compare the accuracy of the MEMD-MSI,MSI and multiple empirical mode decomposition canonical correlation analysis(MEMD-CCA)in different time windows.The results show that the MEMD-MSI has the highest accuracy among the three algorithms,and the accuracy reaches 95.24% when the time window size is 2 s.It is proved that the proposed algorithm can effectively eliminate the artifacts with high accuracy.