基于张量降维和逻辑张量回归的运动想象分类

Motor imagery EEG classification based on tensor dimensionality reduction and logistic tensor regression

采用向量型分类算法对运动想象脑电信号张量特征进行分类时,往往会破坏张量特征的结构信息和内在相关性。为此,提出一种基于张量降维和逻辑张量回归的运动想象脑电信号分类算法。在特征提取阶段,采用结合多线性主成分分析和高阶判别分析的张量降维算法提取运动想象脑电信号的张量特征;在特征分类阶段,采用L2范数正则化的逻辑张量回归算法对张量特征进行分类。实验结果表明,提出算法的平均分类准确率达到86.22%,相比于基于张量降维和线性判别分析的算法的78.05%分类准确率以及基于张量降维和逻辑回归的算法的84.30%分类准确率,分别提高了8.17%和1.92%。

As the vector type classification algorithm often destroys the structural information and intrinsic correlation of the tensor features when classifying the tensor features of motor imagery EEG signals.We propose a motor imagery EEG classification algorithm based on tensor dimensionality reduction and logistic tensor regression.In the feature extraction stage,the tensor dimensionality reduction algorithms combining multilinear principal component analysis and high-order discriminant analysis is used to extract the tensor features of the motor imagery EEG signals;then,the logistic tensor regression algorithm is used to classify the tensor features in the feature classification stage.Experimental results show that the classification accuracy of the proposed algorithm is 86.22%,which is higher than 78.05%of the algorithm based on tensor dimension reduction and linear discriminant analysis and 84.30%of the algorithm based on tensor dimension reduction and logistic regression.