Department of Electronic Engineering, Gachon University, Seongnam 13306, Korea.
School of Electronic Engineering, Kumoh National Institute of Technology, Gumi 39177, Korea.
Sensors (Basel). 2022 Aug 12;22(16):6042. doi: 10.3390/s22166042.
A novel whitening technique for motor imagery (MI) classification is proposed to reduce the accuracy variance of brain-computer interfaces (BCIs). This method is intended to improve the electroencephalogram eigenface analysis performance for the MI classification of BCIs. In BCI classification, the variance of the accuracy among subjects is sensitive to the accuracy itself for superior classification results. Hence, with the help of Gram-Schmidt orthogonalization, we propose a BCI channel whitening (BCICW) scheme to minimize the variance among subjects. The newly proposed BCICW method improved the variance of the MI classification in real data. To validate and verify the proposed scheme, we performed an experiment on the BCI competition 3 dataset IIIa (D3D3a) and the BCI competition 4 dataset IIa (D4D2a) using the MATLAB simulation tool. The variance data when using the proposed BCICW method based on Gram-Schmidt orthogonalization was much lower (11.21) than that when using the EFA method (58.33) for D3D3a and decreased from (17.48) to (9.38) for D4D2a. Therefore, the proposed method could be effective for MI classification of BCI applications.
提出了一种新颖的运动想象 (MI) 分类白化技术，以降低脑机接口 (BCI) 的准确性方差。该方法旨在提高脑电特征脸分析在 BCIs 的 MI 分类中的性能。在 BCI 分类中，准确性的方差对分类结果的准确性本身很敏感。因此，借助 Gram-Schmidt 正交化，我们提出了一种 BCI 通道白化 (BCICW) 方案，以最小化受试者之间的方差。新提出的 BCICW 方法提高了真实数据中 MI 分类的方差。为了验证和验证所提出的方案，我们使用 MATLAB 仿真工具在 BCI 竞赛 3 数据集 IIIa (D3D3a) 和 BCI 竞赛 4 数据集 IIa (D4D2a) 上进行了实验。基于 Gram-Schmidt 正交化的所提出的 BCICW 方法的方差数据要低得多 (11.21) 比 EFA 方法 (58.33) 时使用 D3D3a ，并且从 D4D2a 的 (17.48) 降低到 (9.38) 。因此，该方法可有效应用于脑机接口的 MI 分类。
