基于散度分析的脑电信号特征选择

Electroencephalogram Feature Selection Based on Divergence Analysis

为准确选择脑电信号的频率与通道参数,提高样本的分类识别率,提出一种基于散度的脑电信号特征选择方法。利用散度分析算法从样本数据的原始特征中选取散度值较大的k个特征,并对其进行基于共空间模型的特征提取与线性判别分类器的分类识别。使用2005年BCI竞赛提供的IVa数据集5位样本数据进行实验,结果表明,采用散度分析算法得到的测试样本与训练样本平均识别率为95.54%和84.57%,均高于相关系数和互信息选择算法。

In order to select the effective frequency and electrodes components, this paper promotes an Electroencephalogram（ EEG） feature selection method based on divergence analysis to improve classification accuracy. Throughout the five tested samples from Brain-computer Interface （ BCI ） Competition III dataset IVa, it utilizes divergence analysis algorithms to select the maximum value of the k-space from the original data features, then uses feature extraction based on Common Spatial Pattern （ CSP ） aimed at this k-space feature and classifies by Linear Discriminant Analysis（ LDA） . The experiment identification that the average rate of classification accuracy can obtained is 95. 54% under training pattern, while reached 84. 57% under test pattern, higher than the selection algorithm of correlation coefficient and mutual information.