摘要
提出了一种基于双树复小波变换的运动想象脑电信号特征提取方法。针对传统离散小波抗混叠性差的缺陷,采用双树复小波变换对脑电信号进行分解与重构,得到各子带信号能量并进行归一化处理,选取α、β节律信号的归一化能量作为想象运动的特征进行SVM分类。通过对仿真信号的分析,证实双树复小波变换具有良好的混叠抑制能力和抗噪性。最后选用国际脑机接口竞赛和实验室实测的运动想象数据进行分类识别。实验结果表明,双树复小波变换是一种有效的特征提取方法,其运动想象特征的识别率要优于常用的特征分析方法。
The paper proposed an algorithm of feature extraction of EEG based on Dual-Tree Complex Wavelet Transform. Considering the defect of severe frequency aliasing resulted from Discrete Wavelet Transform, this paper first extracted the sub-band signals of EEG by DTCWT decomposition and reconstruction, and then calculated the energy of each signal and normalized them. Support Vector Machine was applied to recognize the pattern of motor imagery by selecting the normalized rhythm c~ ,/3 as the features. Also, the simulated signals were analysed to confirm that the DTCWT had a satisfying effect on reducing aliasing effects and noise resistance. Finally, international BCI competition signals and the measured motor imagery data were selected for classification. The results showed that the DTCWT was an effective method of feature extraction, which could also obtain a higher recognition rate than the methods in common use.
出处
《传感技术学报》
CAS
CSCD
北大核心
2014年第5期575-580,共6页
Chinese Journal of Sensors and Actuators
基金
国家自然科学基金项目(61172134)
浙江省国际科技合作项目(2013C24016)
