Bagging RCSP脑电特征提取算法
张毅;罗久飞;蔡军;尹春林
【期刊名称】《自动化学报》 【年(卷),期】2017(043)011
【摘要】正则化共空间模式(Regularized common spatial pattern,RCSP)解决了共空间模式(Common spatial pattern,CSP)对噪声敏感的问题,但它在小样本脑电数据集中的表现并不理想,针对上述问题,本文提出了Bagging RCSP (BRCSP)算法,通过Bagging方法重复选取训练数据来构造一个个包,并提取RCSP特征,再利用线性判别分析(Linear discriminantanalysis,LDA)将特征向量映射到低维空间中,最后采用最近邻(Nearest neighborhood classifier,NNC)算法判定分类结果,线下实验证明,相比较聚合正则化共空间模式(RCSP with aggregation,RCSP-A),BRCSP的平均准确率提高了2.92%,且方差更小,鲁棒性更好.最后,在智能轮椅平台上,10位受试者利用BRCSP算法实现左右手运动想象脑电信号控制轮椅完成\字形路径的实验,证明了该算法在脑电信号特征提取中的有效性.%The regularized common spatial pattern (RCSP) has solved the problem that the common spatial pattern (CSP)is sensitive to noise. However,its performance on small sample of electro encephalon graph(EEG)data set is not ideal. To deal with this problem, a Bagging RCSP (BRCSP) algorithm is proposed, which divides training samples into packets and extracts RCSP features by Bagging to choose training packets. Furthermore, the feature vector is projected into the lower space with linear discriminant analysis(LDA)and a classification