关键词: 关键词: 运动想象, 脑电信号, 跨被试, 黎曼切空间特征, 迁移核学习

Abstract: Motor Imagery ElectroEncephaloGraph (MI-EEG) has gained widespread attention in the development of non-invasive brain-computer interfaces for clinical-assisted rehabilitation. Limited by the differences in the distribution of MI-EEG samples from different subjects, cross-subject MI-EEG feature learning has become the focus of research. However, existing methods have problems such as weak domain-invariant feature expression capabilities and high time complexity, and cannot be directly applied to online brain-computer interfaces. To address this issue, this paper presents an efficient cross-subject MI-EEG classification algorithm based on Riemannian Tangent Space (RTS) features and Transfer Kernel Learning (TKL) method. Our TKRTS method first projects MI-EEG covariance matrices into the Riemannian space and aligns the covariance matrices of different subjects by centering them at their centroids while extracting tangent space features. Subsequently, it learns a domain-invariant kernel matrix on the tangent space feature set, achieving a comprehensive representation of cross-subject MI-EEG features. This matrix is then used to train a kernel support vector machine for classification. To validate the feasibility and effectiveness of our TKRTS method, experiments were conducted on three publicly available datasets, including multi-source domain to single-target domain and single-source domain to single-target domain scenarios, and the average classification accuracy increased by 1.32 percentage points and 0.13 percentage points respectively. Experimental results demonstrate that, compared to state-of-the-art methods, TKRTS improves the average classification accuracy while maintaining similar time complexity. Furthermore, ablation experiments confirm the completeness of TKRTS in cross-subject feature expression and its parameter insensitivity, making it suitable for constructing online brain-computer interfaces.

Key words: Keywords: motor imagery, EEG Signals, cross-subject, Riemannian tangent space features, transfer kernel learning