基于自适应邻域相似关系的脑-机接口识别算法

摘要

Focusing on the problem of the unclear data distribution in the Brain-Computer Interface (BCI),a novel approach based on Adaptive Neighborhood Similarity (ANS) is proposed for the Movement Related Potential (MRP).Using the pattern similarity,it searches the optimal direction which maximizes the ratio of the expectation distance between the between-class and the within-class.The similar probabilities are calculated directly with the samples' distances in the mapped space,avoiding the inapplicability caused by using traditional distances measurement in high dimensional space directly.In the application of two datasets,one generated randomly and another from BCI competition,the classification accuracy of ANS is slightly better than the comparison algorithms of Linear Discriminative Analysis (LDA) and Discriminative Spatial Pattern(DSP).The results evaluate the effectiveness of ANS algorithm.%针对脑-机接口(BCI)研究数据分布不确定问题,提出基于自适应邻域相似关系识别算法,以提取脑电信号中的运动相关电位(MRP)特征.采用映射后MRP模式的邻域相似关系,寻找最佳投影方向,使得映射后异类与同类的样本期望距离的比值最大.利用映射后的样本距离确定相似概率,避免在高维空间使用距离方程可能产生的不适应性.在随机生成数据和BCI竞赛公开数据应用中,该算法的识别效果优于线性判别分析算法和共空间模式算法.