快速多变量自回归模型的意识任务的特征提取与分类  

Feature Extraction and Classification of EEG during Mental Tasks Based on Fast Multivariate Autoregressive Models

作　　者：薛建中[1] 郑崇勋[1] 闫相国[1] 

机构地区：[1]西安交通大学生命科学与技术学院,西安710049

出　　处：《西安交通大学学报》2003年第8期861-864,共4页Journal of Xi'an Jiaotong University

基　　金：国家自然科学基金资助项目(30170257).

摘　　要：给出了一种有效的多变量自回归(MultivariateAutoregressive,MVAR)模型参数和阶数估计的快速算法,通过QR分解技术减少了多维矩阵求逆的运算量,提高了模型估计的速度.利用该算法对4个实验对象的3种不同意识任务的脑电(EEG)信号进行特征提取,并将得到的特征向量作为径向基函数神经网络的输入进行训练和分类测试.分类结果表明,利用该方法进行EEG信号的特征提取,分类的正确率明显高于单变量自回归模型,而且算法运算速度快,可以用于不同意识任务EEG信号的在线特征提取与分类系统.A fast estimation method of parameters and order of multivariate autoregressive (MVAR) models is presented, where QR factorization is adopted to reduce the inverse of a multidimensional matrix, and speeds up the estimation of the MVARmodel parameters. With this method, feature vectors are able to be extracted from sixchannel electroencephalogram (EEG) of four subjects during three mental tasks. These vectors are considered as the inputs of radial basis function (RBF) network to test classification accuracies for three task pairs. The optimal model order can be determined as 3 according to Schwarz's Bayesian Criterion. The classification results are significantly better than those of sixorder scale AR model with an obviously higher operation rate than traditional YuleWalker method. Experiment indicates that the presented method is efficient to extract features of mental tasks from the spontaneous EEG signals online.

关 键 词：多变量自回归模型 意识任务 脑电 特征提取 径向基函数神经网络 

分 类 号：R318[医药卫生—生物医学工程]