基于CSSD-DE通道选择算法的ECoG分类研究

CSSD-DE Channel Selection for ECoG Classification

在脑机接口(BCI)的研究中,通道在提取脑电信息的过程中起着十分关键的作用。本研究提出基于共空域子空间分解-微分进化算法(CSSD-DE)的脑机接口通道选择方法,并且使用逻辑线性分类器进行分类。在对皮层脑电信号(ECoG)进行通道选择的过程中取得了使用少数通道就可以达到令人满意的分类效果。当最优通道个数为6,识别正确率达到93%,优于2005年脑机接口竞赛III数据集I的第一名的正确率(91%)。并提出将最大相关最小冗余度(mRMR)和支持向量机回归特征消去(SVM-RFE)算法应用于通道选择进行对比,mRMR算法得出最优通道个数为7,识别正确率为87%,SVM-RFE算法得出的最优通道个数为6,识别正确率为81%。

For brain computer interface （BCI）, channels play a critical role in extracting brain electrical information. In this article, CSSD-DE （common spatial subspace decompositic,n-differential evolution algorithm） method was proposed for channel selection in brain computer interface. The logical linear classifier was used. For channel selection on ECoG （electrocorticographic signals） classification, the satisfactory classification effect was obtained only using a few channels. When the optimal channel number was 6, recognition accuracy could reach to 93%, superior to 91% accuracy of the topping one during the BCI Competition III dataset I. The mRMR （maximum related minimum redundancy） and SVM-RFE （support vector machine regression characteristic elimination） method are proposed for channel selection for comparison. For mRMR, when the optimal channel number was 7, recognition accuracy could reach to 87%. For SVM-RFE, when the optimal channel number was 6, recognition accuracy could reach to 81%.