论文部分内容阅读
脑机接口(BCI)脑电图(EEG)分类能实现人脑直接与外部环境的信息交互。提出了基于辅助训练思想的半监督稀疏表示分类器方法在BCI EEG分类中的应用。首先采用稀疏表示分类器从未标记样本中选择部分相关度较高的样本。其次采用Fisher线性分类器作为判别分类器得到已选样本的边界信息。通过距离大小和方向判别条件进一步选出高置信度样本。本文对三组基准数据集BCIⅠ、BCIⅡ_Ⅳ和USPS分别进行仿真实验,分类正确率分别为97%、82%和84.7%,运算速度最快的仅需约0.2s。在分类正确率和运算效率两个方面,均优于自训练半监督SVM、有导师SVM两种方法。
Brain Machine Interface (BCI) EEG classification enables direct interaction of the human brain with information from the external environment. The application of semi-supervised sparse representation classifier based on auxiliary training ideology in BCI EEG classification is proposed. Firstly, a sparse representation classifier is used to select some samples with high correlation from unlabeled samples. Secondly, the Fisher linear classifier is used as the discriminant classifier to get the boundary information of the selected samples. A further example is to select the high-confidence samples by the condition of distance size and direction. In this paper, three groups of benchmark datasets BCIⅠ, BCIⅡ_Ⅳ and USPS were simulated respectively. The correct rates were 97%, 82% and 84.7%, respectively. The fastest computation speed was only about 0.2s. In terms of classification accuracy and computational efficiency, both are superior to self-training SVM and instructor SVM.