基于广义微状态的运动想象分类研究

Research of Motor Imagery Classification Based on Generalized Micro Status

该文根据传统微状态的概念,提出了一种广义微状态定义,并基于该定义实现了一种多分类运动想象任务特征提取算法。该文使用实验室2分类数据和2008年国际脑机接口竞赛4分类数据集对提出的特征提取算法进行了测试。对2分类数据特征用最小夹角算法分类,准确率最高达到95.56%;对多分类数据特征用SVM算法分类,效果稳定,其中竞赛数据4分类kappa分析结果接近2008年竞赛获胜者水平,测试结果验证了算法的有效性。

Based on the traditional definition of micro-status, we proposed a generalized definition , and proposed a feature extraction algorithm for multi-class motor imagery based on the generalized definition .We tested the algorithm using a 2-classes dataset from our lab and a 4-classes dataset from the international BCI competition of 2008.And the highest accuracy we got for 2-classes is 95.56%, based on minimum-between-angle principle.For multi-class tasks, SVM is more stable, and the kappa value is close to the winner＆#39;s level.