摘要：

针对单一种类特征提取方法所得特征信息量不足的问题,通过自回归模型(AR)与小波变换2种方法实现特征提取,在合并特征集后,采用遗传算法进行最优特征集选择。对运动想象脑电信号进行AR建模,将估计得到的参数作为时域特征,并结合小波变换的时频域特征构建特征集,使用基于k最近邻的分类错误率作为适应度函数,实现对特征向量的选择。运用支持向量机等分类方法验证特征选择效果,结果表明,通过遗传算法进行特征选择,可去除冗余的特征信息,分类正确率达到96.43%。

关键词: 运动想象, 脑机接口, 自回归模型, 遗传算法, 特征选择, 支持向量机

Abstract:

Aiming at the insufficient information obtained by a single type of feature extraction method,this paper combines Autoregressive(AR) model and wavelet transform to implement feature extraction.The combined feature set is not directly utilized as input of classifier to identify,while the genetic algorithm is carried out for optimal feature subset selection.The motor imagery electroencephalogram signals are modeled using AR model.The estimated parameters are regarded as time-domain features.Combining time-frequency features of wavelet transform, the final feature set is constructed.Genetic Algorithm(GA) based on k-Nearest Neighbor(kNN) classification error as fitness function is employed feature selection.The Support Vector Machine(SVM) classification method is used to validate the feature selection results.Experimental results show that redundant and non-informative features are removed through the feature selection of genetic algorithm and the classification accuracy reaches 96.43%.

Key words: movement imagery, Brain-computer Interface(BCI), Autoregressive(AR) model, Genetic Algorithm(GA), feature selection, Support Vector Machine(SVM)

中图分类号: 

• TP391