In order to improve the accuracy of classification based on motor imagery brain-computer interface,a feature extraction method based on the combination of the Event Related Desychronization (ERD) feature and accumulated power feature has been proposed,which extracts ERD band power feature from EEG signal and uses Linear Determination Analysis (LDA) classifier to classify band power feature.The LDA classifier''s output D has been taken as confidence level of classification.When it is bigger than the threshold,the motor imagine status has been judged for extraction of accumulated power feature and combining ERD band power feature with accumulated power feature to construct a new vector with combination features.Classification has been conducted with LDA classifier and thus the final results of classification have been achieved.Experiments for verification have been carried out with BCI 2003 Competition''s Data III.The evaluation criteria are classification accuracy and mutual information.A comparison of classification results with teams use the same dataset has been made.The best classification accuracy of proposed method is 90%,and the best mutual information is 0.51.The comparison show that the proposed method is superior to the most of teams used the same dataset and that it is feasible and effective which can act as a reference for design of online BCI system.%为了提高运动想象脑-机接口的分类正确率,提出了一种基于事件相关去同步(ERD)的频带能量特征和累积能量特征相结合的特征提取方法.对脑电信号提取ERD频带能量特征,使用线性判别分析(LDA)分类器进行分类,将LDA分类器的输出D作为分类置信度.当D大于设定的阈值时,判断进入运动想象状态,提取累积能量特征,将ERD频带能量特征与累积能量特征相结合,构建联合特征向量,使用LDA分类器进行了分类,得到最终分类结果.采用BCI 2003竞赛数据集Data III进行了实验.实验结果以分类正确率和互信息(MI)作为评估标准,提出的方法最大分类正确率为90%,最大互信息为0.51,结果优于大部分使用相同数据集的参赛队伍.实验结果验证了所提出方法的可行性、有效性,为设计在线脑-机接口模型提供了参考.
展开▼
