支持向量机后验概率方法在多任务脑机接口中的应用  

SVM with Posteriori Probabilistic Output Applied in Multi-class Brain Computer Interface

作　　者：杨平[1] 刘铁军[1] 尧德中[1] 

机构地区：[1]电子科技大学生命科学与技术学院知觉与运动实验室,成都610054

出　　处：《中国生物医学工程学报》2009年第2期171-176,共6页Chinese Journal of Biomedical Engineering

基　　金：国家自然科学基金资助项目(30525030);国家自然科学基金资助项目(60736029)

摘　　要：基于运动想象的脑机接口是把使用者的运动意图转变成控制外部设备的信号,它包括脑电数据采集,特征提取和模式分类等几个基本环节。本研究发展了用支持向量机后验概率输出进行分类,并用分类结果中具有大概率的测试样本扩充训练集的模式分类与特征更新方法,并把此方法应用于4类任务运动想象脑机接口实验。使用BCI CompetitionⅢ的数据Ⅲa,运用一对一共空间模式扩展方法进行特征提取,用支持向量机后验概率方法进行分类和训练样本扩充。结果表明:概率信息能提高BCI的性能;应用概率信息选取样本扩充训练集能增加分类器的稳健性。A brain computer interface （BCI） based on motor imagery translates the user＇s motor intention into a control signal for peripheral equipments. The translation process includes data acquisition, feature extract and pattern recognition. In this article, we have developed SVM with posteriori probabilistic output for patterns recognition and expanded the training sets by adding test samples with great probability output. We applied this method to dataset Ⅲ a in BCI CompetitionⅢ 2005 which contained 4 motor image tasks. One-Versus-One Common Spatial Patterns （CSP） algorithm was adopted to extract feature vectors. Support vector machine （SVM） with posteriori probabilistic output was used for patterns recognition and expanding the training sets. The results showed that probabilistic information could improve the performance of BCI and the application of probabilistic information was available in enlarging the training dataset by adding test samples with big probability, thus to get an even more robust classifier.

关 键 词：脑机接口 样本扩充 后验概率 共空间模式 Kappa系数 

分 类 号：R318[医药卫生—生物医学工程]