基于样本迁移的在线脑电分类方法

Online EEG Classification Based on Instance Transfer

在线分类是脑机接口应用中的一个重要研究方向,传统的在线学习算法需要大量样本来适应脑电信号变化,这增加了计算和内存成本。为此,本文提出一种基于样本迁移的在线脑电分类方法。首先利用源域数据训练离线分类器,然后将目标域数据逐个放入进行在线欧式空间预对齐以减小个体差异性,进而提取CSP特征,最后采用加权组合的在线迁移学习分类器进行标签预测。在BCI竞赛Ⅳ数据集Ⅰ和Ⅱa的跨受试者在线模拟实验中,与4种最先进的算法相比,本文方法表现最好,在线平均分类准确率最高达到了86.02%和75.74%,表明了所提方法的有效性。

Online classification is an important research direction in the application of brain-computer interfaces(BCI).Traditional online learning algorithms require a large number of samples to adapt the changes of Electroencephalogram(EEG)signals,increasing computational and memory costs.Therefore,an Online EEG Classification method based on Instance Transfer(OECIT)algorithm is proposed.Firstly,the source domain data is used to train the offline classifier.Then the target domain data are put into online Euclidean space pre-alignment to reduce individual differences,and then the common spatial pattern(CSP)features are extracted.Finally,the weighted combination of online transfer learning classifier is used for label prediction.In the cross-subject online simulation experiments of the BCI competitionⅣDatasetⅠandⅡa,the OECIT method performed best compared with the four most advanced algorithms,with the highest online average classification accuracy rates reaching 86.02%and 75.74%respectively,which shows the effectiveness of the proposed method.