摘要
在脑机接口系统研究中,基于黎曼流形的协方差矩阵在运动想象特征提取中应用广泛,但维度灾难一直是不可避免的问题。提出一种融合人工智能联合互信息和广义判别分析的特征降维方法称之为SJ-GDA,其对高维向量进行智能降维。SJ-GDA方法首先采用Semi-JMI对特征向量进行特征排序,进而利用GDA对排序后的部分向量智能降维,融合两类向量构造最终特征。将最终特征输入文中提出的DT-KNN决策树框架进行人工智能多类运动人想象识别,结果表明提出的算法在左手、右手、脚和口四类运动想象任务识别中Kappa系数从0. 57提高到了0. 607。
The covariance matrix based on the Riemannian manifold is widely used for feature extraction of motion imagery in the brain-computer interface system,but dimension disaster is inevitable. In this article,a feature dimensionality-reduction method SJ-GDA is worked out,with the combination of artificial-intelligence mutual information and generalized discriminant analysis.Thus,SJ-GDA is adopted for intelligent dimensionality reduction of high-dimensional vectors. SJ-GDA firstly uses Semi-JMI for feature ranking of vectors,and then uses GDA to reduce the dimension of the ranked vectors;as a result,the two types of vectors are integrated,in order to construct the final features. The final features are input into the DT-KNN decision-tree framework for artificial-intelligence multi-class motion-imagery recognition. The results show that this algorithm improves the Kappa coefficient from 0.57 to 0.607 in the four types of motion-imagery recognition tasks: left hand,right hand,foot,and mouth.
作者
林福
关山
LIN Fu;GUAN Shan(School of Information and Manufacturing,Minxi Vocational&Technical College,Longyan 364012;School of Mechanical Engineering,Northeast Electric Power University,Jilin 132012)
出处
《机械设计》
CSCD
北大核心
2020年第7期110-115,共6页
Journal of Machine Design
基金
国家自然科学基金资助项目(51704073)。
关键词
人工智能
运动想象
黎曼流形
特征排序
决策树框架
artificial intelligence
motion imagery
Riemannian manifold
feature ranking
decision-tree framework
