|本期目录/Table of Contents|

[1]高诺△,赵凯.基于黎曼空间的脑电信号特征提取和分类算法的对比研究*[J].生物医学工程研究,2023,01:8-14.
 GAO Nuo,ZHAO Kai.Comparison of feature extraction and classification algorithms of EEG based on Riemannian space[J].Journal of Biomedical Engineering Research,2023,01:8-14.
点击复制

基于黎曼空间的脑电信号特征提取和分类算法的对比研究*(PDF)

《生物医学工程研究》[ISSN:1006-6977/CN:61-1281/TN]

期数:
2023年01期
页码:
8-14
栏目:
出版日期:
2023-03-25

文章信息/Info

Title:
Comparison of feature extraction and classification algorithms of EEG based on Riemannian space
文章编号:
1672-6278 (2023)01-0008-07
作者:
高诺△赵凯
(山东建筑大学 信息与电气工程学院,济南 250101)
Author(s):
GAO Nuo ZHAO Kai
(College of Information and Electrical Engineering, Shandong Jianzhu University,Jinan 250101,China)
关键词:
运动想象黎曼流形Stein散度K最邻近(KNN)神经网络
Keywords:
Motor imagery Riemannian manifold Stein divergence K-nearest neighbor(KNN) Neural network
分类号:
R318
DOI:
10.19529/j.cnki.1672-6278.2023.01.02
文献标识码:
A
摘要:
目前运动想象脑电信号(motor imagery electroencephalogram,MI-EEG)的分类方法主要分为两种,一种是利用人工设计MI-EEG特征的相似度/非相似度度量进行分类的算法,另一种是利用深度学习自动学习特征完成分类的算法。为探究两种方法的优劣及适用场景,本研究首先基于黎曼空间提出利用Stein散度作为MI-EEG的相似度/非相似度度量,用K最邻近法进行分类的算法;其次,提出利用黎曼流形结构下的卷积神经网络自动提取脑电信号特征进行分类的算法,最后对两种分类算法进行对比研究。为验证两种算法的有效性,在BCI Competition IV-2a公开数据集上进行实验测试。结果证明,两种分类算法均具有较强的稳定性和分类准确率,利用黎曼流形结构的卷积神经网络算法可获得更高的分类准确率,传统机器学习中利用Stein散度作为MI-EEG相似度/非相似度度量的脑电分类算法运行时间更短,更适合MI-EEG的在线解码。
Abstract:
There are two main approaches for classifying motor imagery electroencephalogram (MI-EEG): One is to use similarity/non-similarity measures of manually designed MI-EEG features for classification, and the other is to use the deep learning to automatically learn features for classification. In order to explore the advantages and disadvantages of these two methods and their applicable scenarios, firstly, based on Riemannian space, Stein divergence was used as the similarity/dissimilarity measure of MI-EEG, and K-nearest method was used to classify them. Secondly, an algorithm based on the convolutional neural network under the structure of Riemannian manifold was proposed to automatically extract EEG signal features for classification. Finally, the two classification algorithms were compared. To verify the effectiveness of the algorithm, experimental tests were carried out on the BCI Competition IV-2a public dataset, and the results proved that the two classification algorithms had strong stability and classification accuracy. The convolutional neural network algorithm using Riemannian flow structure can obtain higher classification accuracy, the traditional machine learning EEG classification algorithm using Stein divergence as MI-EEG similarity/non-similarity measure has shorter running time and is more suitable for online decoding of MI-EEG.

参考文献/References

备注/Memo

备注/Memo:
(收稿日期:2023-01-11)山东省自然科学基金资助项目(ZR2022MF309);山东省科技型中小企业创新能力提升工程项目(2022TSGC2554)。△通信作者Email:gaonuo@sdjzu.edu.cn
更新日期/Last Update: 2023-04-28