特征筛选对脑-机接口信号单次提取精度的影响

Selection of features in single-trial BCI signal estimation

脑-之机接口的核心问题之一是通信载体信号的单次提取．在构建脑控拼写器的过程中，通过“模拟自然阅读”诱发模式产生的视觉诱发电位作为人脑与计算机之间的通信载体，采用支持向量机方法进行特征信号的单次识别．为提高识别精度，详细研究了信号时程、时段的选择对模式识别精度的影响．结果表明，信号时程越长分类精度越高，时程达到300ms时，分类精度就可达到最大值（且趋于饱和）；信号时段的选择对分类精度亦有较大影响，最佳时段在靶刺激出现后约250～350ms作为起始处．这一结果为提高系统的整体速度与精度打下了基础．

One of the key issues in constructing a brain-computer interface is the single-trial estimation of its communication carriers. The support vector machine was exploited to do the estimation of single-trial VEP evoked by imitating-natural-reading. In order to get a better accuracy of classification, we investigated how the classification accuracy was affected by the selection of signal length and interval. The results show that the longer the length of signals, the better the accuracy of classification. The accuracy approaches to maximum value when the length of signal is up to 300 ms. The selection of signal interval also played a key role in classification. The best accuracy appeared at the interval about 250～350 ms after the target stimulus onset. The works are essential for boosting up the speed of whole BCI system.