| 引用本文: | 孟明,尹旭,高云园,佘青山,罗志增.运动想象脑电的块选择共空间模式特征提取[J].控制理论与应用,2021,38(3):301~308.[点击复制] |
| MENG Ming,YIN Xu,GAO Yun-yuan,SHE Qing-shan,LUO Zhi-zeng.Block-selection based common space pattern feature extraction of motor imagery electroencephalogram[J].Control Theory and Technology,2021,38(3):301~308.[点击复制] |
|
| 运动想象脑电的块选择共空间模式特征提取 |
| Block-selection based common space pattern feature extraction of motor imagery electroencephalogram |
| 摘要点击 2086 全文点击 654 投稿时间：2020-05-25 修订日期：2020-09-24 |
| 查看全文 查看/发表评论 下载PDF阅读器 |
| DOI编号 10.7641/CTA.2020.00290 |
| 2021,38(3):301-308 |
| 中文关键词 脑机接口 运动想象 共空间模式 支持向量机 块选择 |
| 英文关键词 brain-computer interface motor imagery common space pattern support vector machine block-selection |
| 基金项目 国家自然科学基金项目(61671197, 61871427, 61971168), 浙江省自然科学基金项目(LY18F030009)资助. |
|
| 中文摘要 |
| 共空间模式(CSP)作为一种空间滤波方法已在脑电信号(EEG)的特征提取上得到了广泛应用, 而对脑电信
号的通道和频带进行合理选择可以有效改善共空间模式特征在运动想象脑机接口(BCI)中的分类性能. 针对已有选
择方法中未充分考虑通道间差异性的问题, 本文提出一种对通道和频带同时进行选择的块选择共空间模式(BS–
CSP)特征提取方法. 首先针对每个通道进行频带划分从而构建数据块, 然后根据时频特征计算对应的Fisher比表征
每个块的分类能力, 并设置阈值选出一定数量的最优块, 最后用CSP和支持向量机(SVM)分别进行特征提取与分类.
在对BCI Competition III Datesate IVa和BCI Competition IV Datesate I两个二分类运动想象任务的分类实验中, 平均
分类精度达到了90.25%和83.78%, 表明了所提出的特征提取方法的有效性和鲁棒性. |
| 英文摘要 |
| As a space filtering method, common space pattern (CSP) has been widely used in feature extraction of
electroencephalogram (EEG) signal. Reasonable selection of channels and frequency bands of EEG signal can effectively
improve the classification performance of CSP features in brain-computer interface (BCI). To solve the problem that the
difference between channels is not fully considered in the existing selection methods, a block-selection for CSP (BS–CSP)
feature extraction method is proposed in this paper. Firstly, each channel is divided into frequency bands to construct blocks,
then the Fisher ratio is calculated according to the time-frequency characteristics to represent the classification ability of
each block, and a threshold is set to select optimal blocks. Finally, CSP and Support vector machine (SVM) are used for
feature extraction and classification respectively. In the experiment of BCI competition III dataset IVa and BCI competition
IV dataset I, the average classification accuracy is 90.25% and 83.78%, which shows the effectiveness and robustness of
the proposed feature extraction method. |
|
|
|
|
|