连续预测脑机接口的信息积累方法

Accumulate Information for Continuous Prediction in Brain-Computer Interface

设计有效的学习算法快速准确地对脑电信号进行连续预测是脑机接口研究的关键之一。本研究给出了一种基于变分贝叶斯算法的理论框架通过区分度权值进行信息积累,从而对脑电信号分类。此方法将对区分度权值和分类器参数的估计融为一体,使得这两部分在学习的过程中可以互相协调。在两个运动想象数据集上的实验结果表明本方法能够提高BCI系统的性能,具有较好的实用性。

To develop effective learning algorithms for fast and accurate continuous prediction using Electroencephalogram （EEG） signal is a key issue in Brain-Computer Interface （ BCI）. This paper presented a unified framework based on variational Bayesian method to classify EEG trial by accumulating the predictions of segments according to the discriminative powers during a trial. The presented method unified the estimations of discriminative powers and classifier parameters into a whole process, which made the two parts cooperate with each other. The experimental results on two motor imagery datasets have shown that the presented method improves the performance of BCI system and is suitable for online application.