论文部分内容阅读
在基于头皮脑电(EEG)信号的脑-机接口(BCI)研究中,用户个体差异性和背景噪声的复杂性是影响BCI系统稳定性的两个主要因素。因此需要针对不同个体进行BCI系统参数优化,其中包括对时域、空域滤波器参数的优化设计和分类器参数的学习。本文以提高BCI系统的准确性为目标,提出了一种结合独立分量分析空域滤波器(ICA-SF)优化设计和EEG多子带特征的BCI信息处理新方法。基于所提方法,对4位受试者在不同时间采集的三类运动想象EEG(MI-EEG)进行分析。实验结果表明,在同一受试者的自交叉测试和不同受试者数据集之间的互交叉验证中,多子带特征结合方法所得到的平均识别率比仅使用单频带所得的平均识别率普遍提高,识别率最大提升可达6.08%和5.15%。
In brain-computer interface (BCI) studies based on scalp electroencephalography (EEG) signals, the complexity of user individuality and background noise are two major factors that affect the stability of BCI systems. Therefore, it is necessary to optimize BCI system parameters for different individuals, including the optimization of time domain and spatial filter parameters and the learning of classifier parameters. In order to improve the accuracy of BCI system, a new BCI information processing method based on ICA-SFA and EEG multi-subband features is proposed in this paper. Based on the proposed method, three types of motorized imagings EEG (MI-EEG) collected at four different time points from 4 subjects were analyzed. The experimental results show that in the same subject’s self-crossover test and cross-validation between different subject data sets, the average recognition rate obtained by multi-subband feature combination method is higher than the average recognition rate obtained using only single frequency band Generally, the recognition rate can be increased by up to 6.08% and 5.15% respectively.