中文摘要：

研究了基于辅助训练思想的半监督线性支持向量机方法在脑机接口EEG分类中的应用。首先采用高斯混合模型、Parzen窗、K。一近邻估计三种方法估计概率密度，得到未标记样本的密度信息，选择部分概率较高样本，使用比较置零法避免错分。其次采用线性支持向量机作为判别分类器得到已选样本的边界信息，通过距离判别条件选出高置信度的样本，使用方向判别条件避免错分。结合密度和边界信息完成高置信度未标记样本选择的方法称为辅助训练半监督支持向量机。本文的实验数据包括茚0c、BCII、BCIII—IV、USPS，分类正确率分别为91．6％，97％，84％，90．4％，运算速度最慢的仅需约3．5S。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、监督SVM两种方法。

英文摘要：

Abstract：In this paper the application of semi-supervised cation for brain-computer interface （BCI） is investigated. linear SVM algorithm using help training in EEG classifi- Firstly, GMM, Parzen window and KN-neighbor estimation methods are adopted to estimate the probability density of the unlabeled data, and the probability density information of the unlabeled data samples is obtained;some data samples with high probabilities are selected. To avoid misclassi- fication, the strategy called compare and return to zero is applied. Secondly, the linear SVM is used as discriminative classifier to obtain the boundary information of the selected data samples. The final unlabeled data samples with high confidence are selected using the criterion containing distance information, and the direction criterion is used to avoid misclassification. This method is called semi-supervise linear support vector machine using help training, which com- bines the information of density and boundary to select the unlabeled data samples with high confidence. This method is applied to four groups datasets of benchmark, which are gS0c, BCI I , BCI lI _ IV, USPS, and their classification correct rates are 91.6% ,97% ,84% and90.4% ,respectively. Moreover,the operation at the slowest speed needs on- ly about 3.5 s. The results indicate that the new method is superior to self training semi-supervised SVM and super- vised SVM methods in the respects of classification correct rate and operation efficiency, which proves that the pro- posed new method is effective.