文章摘要

文章信息/Info

Title:: The Comparison of EEG Feature Extraction of Motor Imagery between CSP Algorithm and Wavelet Packet Analysis

Author(s):: WU Linyan; LU Hao; GAO Nuo; WANG Tao; College of Information and Electrical Engineering，Shandong University of Architecture，Jinan 250100，China

Keywords:: Wavelet packet analysis; Common space model (CSP); Support vector machines (SVM); Brain computer interface(BCI); Motor imagery(MI)

摘要:: 运动想象脑电特征快速准确提取是脑-机接口技术研究的重要问题。本研究分别讨论了共同空间模式(common spatial pattern, CSP)与小波包分析关于左右手运动想象特征提取的原理，并对两种方法进行了比较。对于GRAZ大学提供的运动想象脑电数据，使用CSP 与支持向量机(support vector machine, SVM)结合的分类正确率最高为85.5%；使用小波包分析与SVM结合的分类正确率最高为99%。同时对于本实验室采用Emotiv epoc+ 系统采集的运动想象脑电数据，利用小波包分析与 SVM结合的分类正确率也保持在98%以上。实验结果表明,相较于CSP算法，小波包分析对于运动想象特征提取的效果更好。

Abstract:: The rapid and accurate extraction of motor imagery feature of EEG signals is an important issue in brain computer interface (BCI) research. This paper discussed the theory of common spatial pattern(CSP) and wavelet packet analysis in feature extraction of two classes motion imagery. For the data provided byGRAZ university, the highest classify accuracy with CSP and support vector machines (SVM) was 85.5%; the wavelet packet analysis classify accuracy was 99%. And the accuracy to classify the data by emotiv epoc+ system by wavelet packet analysis and SVM could reach 98%. Experimental results show that, compared to CSP algorithm, wavelet packet analysis is better for feature extraction.