核共空域子空间分解特征提取算法研究

Kernel CSSD-Based Feature Extraction Algorithms

脑-机接口中特征提取算法是脑电信号处理的关键步骤。提出一种基于核方法的核共空域子空间分解特征提取算法,将用于多通道两类别分类的共空域子空间分解算法推广到核空间。应用新算法对BCI竞赛Ⅱ的数据集Ⅳ进行实验仿真。实验中核函数使用的是线性核函数,求解空域滤波器时,为了减小计算的压力,在原空间对每一个试验的训练数据进行层次聚类,训练的分类器为最近邻分类器,实验的测试集结果为84%,与数据集Ⅳ的竞赛胜利者的分类结果相同。

Feature extraction is a key step in EEG signal processing for brain-computer interface system.A new kernel CSSD approach based on kernel method was proposed in this paper.In this approach,conventional CSSD used in multichannel and two class problem was extended to kernel space.We applied the Kernel CSSD approach to dataset IV of BCI competition II by computer simulations.A linear kernel function was used in the experiments.When spatial filter was obtained,a hierarchical clustering method was used in train datasets to solve the complexity problem.After that classification was performed using K-nearest neighbor classifier.The accuracy of the test datasets was 84%,which is same with test accuracy of the winner of dataset IV.