基于子空间对齐与自适应CSP算法的运动想象脑电信号分类

在脑机接口中,让分类器从一个用户适应到另一个用户是具有挑战性的,但对于减少新用户的训练时间是必要的。但由于每个个体的神经信号存在着差异,常用的特征提取方法训练的分类器,应用于不同的用户时,准确率很低。因此本文提出了一种新的自适应共空间模式的特征提取方法,该算法通过选择合适的候选试验更新协方差矩阵,然后对提取的特征进行子空间对齐,最后用于训练分类器进行分类。由实验结果得出该方法的分类准确率优于传统的CSP算法和传统的自适应CSP算法,最后通过对提取特征的可视化可以看出改进的子空间对齐可以降低源域与目标域的域方差,减小源域与目标域之间的差异。

Classification of motion image EEG signals based on subspace alignment and adaptive CSP algorithm

In EEG signal processing,feature extraction and classification are im portant components,so that the trained classifier can be adapted to each user, this problem is called transfer learning.However,due to the differences in neu ral signals of each individual,the classifier trained by the commonly used feat ure extraction method has low accuracy when applied to different users.Therefor e,this paper proposes a new feature extraction method for adaptive co-space mo d e.This algorithm updates the covariance matrix by selecting appropriate candida te tests,then performs subspace alignment on the extracted features,and finall y uses it to train the classifier for classification.According to the experimen tal results,the classification accuracy of this method is better than that of t he traditional CSP algorithm and the traditional adaptive CSP algorithm.Finally ,through the visualization of the extracted features,it can be seen that the i mproved subspace alignment can reduce the domain variance between the source dom ain and the target domain,and reduce the difference between the source domain a nd the target domain.