摘要
脑—机接口(BCI)是连接大脑和计算机及外部设备的通讯系统,通过连续小波变换(CWT)对采集的脑电信号进行分解,构造由多个尺度对应的方差构成的多维向量,应用支持向量机(SVM)进行分类识别,取得了良好的效果。基于统计学习理论的结构化风险最小化原则,研究了高斯核支持向量机误差惩罚参数C和高斯核参数σ对支持向量机性能的影响,使用仿真实验验证了传统的经验风险最小化原则不能保证良好的推广能力,提出了综合调整参数σ和参数C的方法以优化支持向量机的性能。
Brain-Computer Interface （BCI） is a communication system that connects the brain with computer and peripheral equipments. Continuous Wavelet Transform （CWT） was adopted to decompose the acquired signals of the electroencephalogram （EEG）. A multidimensional vector, composed of the variances corresponding to multi-scales, was constructed. Besides, Support Vector Machine （SVM） was applied to do the classified recognition and achieved good effects. The influences of the error penalty parameter C and the Gaussian kernel parameter σ on the generalization ability of support vector machine were studied on the basis of the structural risk minimization principle of the statistic learning theory. Simulation results show that the traditional experience risk minimization principle cannot guarantee good generalization ability. Therefore, the method of adjusting both the parameter C and the parameter σ was proposed to optimize the performance of support vector machine.
出处
《计算机应用》
CSCD
北大核心
2007年第2期337-339,348,共4页
journal of Computer Applications
基金
国家自然科学基金资助项目(60471055)
教育部博士点基金项目(20040614017)
关键词
脑-机接口
连续小波变换
支持向量机
统计学习理论
模型参数
Brain-Computer Interface （BCI）
Continuous Wavelet Transform （CWT）
Support Vector Machine （SVM）
statistic learning theory
model parameters
