Ablation experimental results under S2S strategy.

In this study, we introduce a novel Field-Agnostic Riemannian-Kernel Alignment (FARKA) method to advance the classification of motion imagination in Brain-Computer Interface (BCI) systems. BCI systems enable direct control of external devices through brain activity, bypassing peripheral nerves and muscles. Among various BCI technologies, electroencephalography (EEG) based on non-intrusive cortical potential signals stands out due to its high temporal resolution and non-invasive nature. EEG-based BCI technology encodes human brain intentions into cortical potentials, which are recorded and decoded into control commands. This technology is crucial for applications in motion rehabilitation, training optimization, and motion control. The proposed FARKA method combines Riemannian Alignment for sample alignment, Riemannian Tangent Space for spatial representation extraction, and Knowledge Kernel Adaptation to learn field-agnostic kernel matrices. Our approach addresses the limitations of current methods by enhancing classification performance and efficiency in inter-individual MI tasks. Experimental results on three public EEG datasets demonstrate the superior performance of FARKA compared to existing methods.

本研究提出一种全新的领域无关黎曼核对齐（Field-Agnostic Riemannian-Kernel Alignment, FARKA）方法，用于提升脑机接口（Brain-Computer Interface, BCI）系统中的运动想象分类任务性能。脑机接口系统可绕过外周神经与肌肉，直接通过脑活动实现对外部设备的操控。在各类脑机接口技术中，基于非侵入式皮层电位信号的脑电图（electroencephalography, EEG）凭借其高时间分辨率与非侵入的特性脱颖而出。基于脑电图的脑机接口技术可将人类的脑意图编码为皮层电位信号，经采集记录后解码为操控指令。该技术在运动康复、训练优化与运动控制等应用场景中具有重要价值。本文提出的FARKA方法整合了用于样本对齐的黎曼对齐、用于空间表征提取的黎曼切空间，以及用于学习领域无关核矩阵的知识核自适应技术。本方法通过提升个体间运动想象（motion imagination, MI）任务的分类性能与运行效率，解决了现有方法存在的局限。在三个公开脑电图数据集上的实验结果表明，相较于现有方法，FARKA展现出更优异的分类性能。