基于CSP和ICA的多任务脑机接口分类方法比较研究
发布时间：2019-04-26 06:18
【摘要】：特征提取和模式分类是BCI系统最重要的两个环节,直接关系到BCI系统的分类识别率和分类稳健性。本文主要研究内容包括两种特征提取算法——共空间模式(Common Spatial Pattern,CSP)和独立分量分析(Independent Component Analysis,ICA),以及三种分类方法——Fisher判别分析FDA、支持向量机SVM以及KNN近邻法。 在两分类任务的脑电信号特征提取方面,CSP的效果非常好,但是在处理多分类数据时,必须将二进制的CSP算法扩展到多类条件。本文使用基于近似联合对角化的多类CSP方法对脑电数据进行特征提取；为了比较特征提取算法的性能,本文还使用了一种经典的特征提取算法——独立分量分析来提取脑电信号的特征。然后使用三种分类算法,即Fisher判别分析、支持向量机SVM以及KNN近邻法,对提取的脑电特征信号进行分类。 本文使用5个受试者的多任务脑电数据,对这两种特征提取算法与这三种分类方法进行了仿真实验,通过实验比较和分析了它们性能。
[Abstract]:Feature extraction and pattern classification are two of the most important links in BCI system, which are directly related to the classification recognition rate and classification robustness of BCI system. The main contents of this paper include two feature extraction algorithms-(Common Spatial Pattern,CSP (Common Space Model) and (Independent Component Analysis,ICA (Independent component Analysis), as well as three classification methods-FDA, support Vector Machine (SVM) and KNN nearest neighbor method (KNN) based on Fisher discriminant analysis. In the aspect of feature extraction of EEG signals from two classification tasks, CSP has a very good effect, but the binary CSP algorithm must be extended to multi-class conditions when dealing with multi-classification data. In this paper, a multi-class CSP method based on approximate joint diagonalization is used to extract the features of EEG data. In order to compare the performance of the feature extraction algorithm, a classical feature extraction algorithm, Independent component Analysis (ICA), is used to extract the features of EEG signals. Then three classification algorithms, namely Fisher discriminant analysis, support vector machine SVM and KNN nearest neighbor method, are used to classify the extracted EEG feature signals. In this paper, the multi-task EEG data of five subjects are used to simulate the two feature extraction algorithms and these three classification methods, and their performance is compared and analyzed by experiments.
【学位授予单位】：南昌大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：TP334.7
本文编号：2465827
[Abstract]:Feature extraction and pattern classification are two of the most important links in BCI system, which are directly related to the classification recognition rate and classification robustness of BCI system. The main contents of this paper include two feature extraction algorithms-(Common Spatial Pattern,CSP (Common Space Model) and (Independent Component Analysis,ICA (Independent component Analysis), as well as three classification methods-FDA, support Vector Machine (SVM) and KNN nearest neighbor method (KNN) based on Fisher discriminant analysis. In the aspect of feature extraction of EEG signals from two classification tasks, CSP has a very good effect, but the binary CSP algorithm must be extended to multi-class conditions when dealing with multi-classification data. In this paper, a multi-class CSP method based on approximate joint diagonalization is used to extract the features of EEG data. In order to compare the performance of the feature extraction algorithm, a classical feature extraction algorithm, Independent component Analysis (ICA), is used to extract the features of EEG signals. Then three classification algorithms, namely Fisher discriminant analysis, support vector machine SVM and KNN nearest neighbor method, are used to classify the extracted EEG feature signals. In this paper, the multi-task EEG data of five subjects are used to simulate the two feature extraction algorithms and these three classification methods, and their performance is compared and analyzed by experiments.
【学位授予单位】：南昌大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：TP334.7
【参考文献】
相关期刊论文 前5条
1 石海明;刘昱东;;BCI:改变未来战争面貌的脑-机接口技术[J];环球军事;2009年16期
2 高诺;鲁守银;张运楚;姚庆梅;;脑机接口技术的研究现状及发展趋势[J];机器人技术与应用;2008年04期
3 张莉;何传红;何为;;脑-机接口的研究现状与挑战[J];现代科学仪器;2007年02期
4 罗尚清;杨浩;;脑—机接口系统工作原理综述[J];医疗卫生装备;2008年08期
5 高上凯;;无创高通讯速率的实时脑-机接口系统[J];中国基础科学;2007年03期
相关博士学位论文 前1条
1 刘美春;基于运动想象的脑—机接口系统模式识别算法研究[D];华南理工大学;2009年
，本文编号：2465827
本文链接：https://www.wllwen.com/kejilunwen/jisuanjikexuelunwen/2465827.html
