Chernof f加权分类器框架在运动想象脑-机接口中的应用

针对现有脑机接口(BCI)分类器与大脑认知过程结合不够紧密的问题，该文提出一种基于Chernoff加权的分类器集成框架方法，并用于同步运动想象脑机接口中。通过对训练数据进行统计分析，获得各时刻脑电信号(EEG)的统计特性，并建立基于大脑认知过程的高斯概率模型。然后利用Chernoff边界特性得到该概率模型的最小误差，并以此确定该时刻分类器的权重，通过对各时刻分类器的加权，实现同步脑机接口的信号分类。以脑机接口竞赛数据作为测试，并与线性判决分析、支持向量机和极限学习方法分别结合构成新的集成方法。由实验结果可知，加权集成框架方法的分类性能比原独立分类方法有显著提高。

Applying Chernoff Weighted Classification Frame Method to MotorImagery Brain Computer Interface

For the problem that the classifier is less considered to be combined with the brain's cognitive process in the Brain-Computer Interface (BCI) system, a Chernoff-weighted based classifier integrated frame method is proposed and used in synchronous motor imagery BCI. In the method, the statistic characteristics of ElectroEncephaloGraphy (EEG) are obtained by analyzing in each time point of synchronous BCI, and then the probability model is established to compute the Chernoff error bound, which is adopted as the weight of common classifier to take the discriminant process. The test experiments are based on the datasets from BCI competitions, and the proposed frame method is employed to compose with LDA, SVM, ELM respectively. The experimental results demonstrate that the proposed frame method shows competitive performance compared with other methods.