基于多级特征融合的运动意向识别方法研究  

Research on Motion Intention Recognition Based on Multilevel Feature Fusion

作　　者：孙文斌 白雪梅[1] 刘一凡 胡延飞 SUN Wen-bin;BAI Xue-mei;LIU Yi-fan;HU Yan-fei(School of Electronics and Information Engineering,Changchun University of Science and Technology,Changchun 130022)

机构地区：[1]长春理工大学电子信息工程学院,长春130022

出　　处：《长春理工大学学报（自然科学版）》2021年第5期63-70,共8页Journal of Changchun University of Science and Technology(Natural Science Edition)

基　　金：吉林省科技厅项目(20200404216YY)。

摘　　要：针对提取的多通道EEG(脑电信号)的CSP(共空间模式)算法在空间滤波上具有很好的性能,但是CSP算法无法体现EEG信号的时域和频率特性,导致分类器的准确率不高和泛化性能不好的问题。提出了一种新的加权正则化的DTCWT(双树复小波变换)与CSP的特征提取方法(R-DTCWT-CSP),充分结合脑电信号的时-频-空域特征信息,减少冗余通道的无用的信息。在国际竞赛BCI Competitions 2008的四分类数据集上,采用多种分类器对比得到分类准确率最好的分类器,网络搜索算法进一步优化参数,最后选取准确率较好的四个分类器组合成一种投票机制的分类器对脑电信号进行分类。识别准确率最高达到94.23%,这表明方法是可行有效的。CSP(Common Space Mode)algorithm for extracting multi-channel EEG(Electroencephalogram)has good performance in spatial filtering,however,CSP algorithm can’t reflect the time domain and frequency characteristics of EEG signal,which leads to the problem of low accuracy and poor generalization performance of the classifier.In this paper,a new weighted regularized feature extraction method of DTCWT(Dual Complex Wavelet Transform)and CSP(R-DTCWT-CSP)is proposed,which fully combines the time-frequency-spatial feature information of EEG signals to reduce the useless information of redundant channels.In the international competition BCI Competitions 2008 four classification data sets,the multiple classifiers can be compared to get the best classifier classification accuracy,the network search algorithm to optimize parameters,then,better accuracy in the first four classifier combination into a voting mechanism of classifier to classify EEG signals.In the end,the best recognition accuracy is up to 94.23%,which indicates that the method of this paper is feasible and effective.

关 键 词：脑电信号 共空间模式 集成分类器 双树复小波分解 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]