一种基于参数优化C-SVM的脑电信号分类方法及应用

A parameter optimized C-SVM approach for EEG classification and its application

脑电信号(electroencephalograph,EEG)由于自身信号微弱且容易受到周边环境和大脑内部其他活动的影响,对其进行特征分类并提高分类准确率这一问题一直是脑机接口领域的难点。传统的基于支持向量机(support vector machines,SVM)的脑电信号特征分类方法在选取惩罚参数与核函数参数时大都只是采用经验数据,而忽略了参数优化对提升SVM分类效果重要性,而现有的参数优化方法计算复杂严重影响了分类效率。针对以上问题,提出了一种通过交叉检验和LOO误差上界对C-SVM中的惩罚参数C和核函数参数进行优化的方法,并在理论分析的基础上结合实验证明了参数优化后的分类方法能够有效提高脑电信号分类的准确率且对分类效率影响不大。

The feature classification of EEG signal is one of the key-technology for Brain-Computer Interface technology. The traditional SVM-based EEG feature classification paid more attention on the effectiveness of SVM algorithm but ignored the parameters＂ contribution to increasing the SVM ＂s performance. The existing parameter optimization methods cost too much time in calculating and seriously affect the efficiency of classification. In this paper, a novel parameter optimized C- SVM approach using cross-validation and leave-one-out（LOO） estimation is put forward in order to find the optimal penalty parameter C and kernel function parameters. Experimental results by controlling the intelligent wheelchair via the BCI sys- tem on it verified the theoretical analysis.