改进的半监督聚类在MEG脑机接口中的应用

Application of improved semi-supervised clustering in MEG brain computer interface

脑磁信号(MEG)作为一种新的脑机接口(BCI)输入信号,含有手运动方向的模式信息。鉴于半监督聚类融合了训练数据先验知识的优势,提出一种基于训练中心的半监督模糊聚类算法。该算法分为降维和改进的半监督聚类,采用主成分分析和线性判别分析将高维数据降到低维,改进的半监督聚类在对训练数据进行模糊聚类的基础上,将得到的聚类中心加权到测试数据聚类过程中,以增加测试数据聚类中心的鲁棒性。结果表明,该算法识别率较高,平均识别率达到了55.1%,优于BCI竞赛Ⅳ的最好结果46.9%。

The Magneto-Encephalo-Graphy（MEG） can be used as an input signal for Brain Computer Interface（BCI）,which contains the pattern information of the hand movement direction.In view of the fact that the semi-supervised clustering combines the advantages of training data prior knowledge,a semi-supervies fuzzy clustering algorithm based on training center was put forward.The algorithm was divided into lower-dimensional and improved semi-supervised clustering.Principal component analysis and linear discriminant analysis were used to reduce the data from high-dimension to low-dimension.Improved semi-supervised clustering based on fuzzy clustering for the training data added the training center in proportion to the test data center.The experimental results show that the average recognition rate of the proposed algorithm reaches to 55.1%,higher than that of the winner of the 2008 competition Ⅳ.