基于双子空间PCA降维的脑力负荷分类

人类社会至今的飞速发展使得大量体力劳动被机械工程替代，工作者的任务重心也从体力劳动逐渐转变为脑力劳动，对操作者脑力负荷进行实时评估以增强工作效率在当下有着重大意义。目前人类对于脑力负荷评估共有三种方式，有研究表明，采用生物电信号进行脑力负荷分类效果较其余两种方法更客观。但脑电信号经过特征提取后维数极高，所需数据量和运算量巨大，需要对其进行降维。目前降维方面最广泛运用的两种算法为主成分分析（PCA）和线性判别分析（LDA）。针对PCA的非监督性和LDA的特征冗余敏感性，本文提出一种二分类下基于双子空间主成分分析的降维算法，分别对不同类别的训练集数据进行主成分分析，并将所有训练集数据映射到生成的空间中，再次进行PCA-LDA降维，以此提高降维后数据的可分性。实验结果表明，双子空间PCA-LDA降维算法在二分类任务下测试集精度整体高于单子空间PCA-LDA算法，以此为脑力负荷分类领域和高维数据降维领域提供了新思路。

Research on classification of mental workload based on dimension reduction of PCA in two subspaces

With the rapid development of human society so far, a large number of physical labor has been replaced by mechanical engineering, and the task focus of workers has gradually changed from physical labor to mental labor. Real-time evaluation of the operator''s mental load to enhance work efficiency is of great significance at present. At present, there are three ways to assess mental load. Some studies have shown that the effect of using bioelectrical signals to classify mental load is more objective than the other two methods. However, the dimension of EEG signal after feature extraction is very high, and the amount of data and computation required is huge, so it needs to be dimensioned. At present, the two most widely used algorithms in dimension reduction are principal component analysis (PCA) and linear discriminant analysis (LDA). In view of the unsupervised nature of PCA and the feature redundancy sensitivity of LDA, this paper proposes a dimensionality reduction algorithm based on two-subspace principal component analysis under two-classification, which performs principal component analysis on different types of training set data, maps all training set data into the generated space, and then performs PCA-LDA dimensionality reduction again to improve the separability of the data after dimensionality reduction. The experimental results show that the accuracy of the test set of the two-subspace PCA-LDA dimensionality reduction algorithm is higher than that of the single-subspace PCA-LDA algorithm under the two-classification task, which provides a new idea for the field of mental workload classification and high-dimensional data dimensionality reduction.