摘要
针对传统支持向量机分类方法在脑电信号处理中存在分类正确率低的问题,将聚类思想与二叉树支持向量机结合构造多类SVM分类器。实验以"BCI Competition 2005"中的DatasetⅢa为例,先对采集的4类运动想象脑电信号应用小波变换进行去噪;再在分析小波包频带划分特点的基础上,利用小波包进行分解与重构,获取相应的能量特征;最后应用改进后的支持向量机(SVM)分类方法对特征信号进行分类。结果表明该方法分类正确率较高,可以达到91.12%,并且有效的减少了分类器的个数,最终达到较好的识别效果。
For the disadvantages of the traditional SVM classification in dealing with EEG signal,such as lower accuracy rate in classification,a multi-class SVM classifier is constructed by combining cluster idea with binary tree SVM.Based on data of the DatasetⅢa in the'BCI Competition 2005'.Firstly,fourclass motor imagery EEG data collected is de-noised by the wavelet transform.Secondly,on the basis of analyzing the frequency band feature of wavelet packets,the corresponding energy feature is extracted by using decomposition and reconstruction of wavelet packets.Finally,the classification of the obtained feature signal is completed by using the improved SVM classification method.The simulation results show that the higher accuracy rate in the classification,about 91.12%,can be achieved.The number of classifier can be reduced efficiently and the relatively good identifying effects can be achieved finally.
出处
《常州大学学报（自然科学版）》
CAS
2014年第1期42-46,共5页
Journal of Changzhou University：Natural Science Edition
基金
机器人技术与系统国家重点实验室开放基金重点项目(SKLRS-2010-2D-09)
关键词
脑机接口
4类运动想象
特征提取
聚类思想
支持向量机
brain-computer interface(BCI)
four-class motor imagery
feature extraction
clustering idea
support sector machines(SVM)
