脑机接口(brain-computer interface, BCI)技术可以实现大脑与外界环境的直接交互，从而辅助脑卒中患者进行主动康复训练。然而，由于脑电信号的非平稳特性，BCI技术在临床应用转化中面临着诸多科学挑战。如何实现脑电信号的快速测量和运动意图的准确解码是当前面临的核心问题。针对该问题，本项目以提升脑卒中患者的BCI性能为导向，分别研究了脑电电极的硬件设计和脑电信号的处理算法。首先，项目设计了一款水凝胶脑电电极，可以实现脑电信号的免胶测量，从而降低BCI应用的准备时间；其次，分析了脑卒中患者的中枢-外周神经信号耦合机制，提出了基于特征自适应的伪迹识别方法，可有效检测代偿伪迹信号；随后，研究了基于用户迁移学习的模式分类算法，实现了BCI解码模型的“零再训练”；最后，设计和制造了脑控外骨骼手功能康复训练系统，实现了脑电测量与信号分析技术的系统集成，为BCI技术的临床推广奠定基础。

Brain-computer Interface (BCI) provides a direct communication channel between the human brain and external environment. It can benefit the stroke patients in active rehabilitation of motor function. However, due to the non-stationarity of brain signals, great challenges are still confronting the clinical application of BCI technique. Two essential requirements are fast acquisition of brain signals and accurate BCI decoding. To address these problems, we focus on the design of EEG recording system and on the developing of BCI decoding algorithm, aiming at improving the BCI performance in stroke patients. Firstly, we designed a gel-free EEG electrode by using hydrogel, which can reduce the preparation time of BCI experiment. Secondly, the relationship between EEG and EMG signals were well analyzed and an adaptive algorithm was accordingly developed for the correction of compensatory artifacts. Then, a zero-retraining algorithm based on user transfer learning was developed for BCI classification. Finally, the EEG measurement and analysis techniques were integrated into the BCI-controlled hand exoskeleton system. These works may facilitate the promotion of BCI rehabilitation in clinic.