为解决偏瘫患者在主动康复训练中对上肢外骨骼的控制难题，提出一种基于单次运动想象的脑电信号分类方法，并将其应用于自主研发上肢外骨骼的实时控制中。针对脑电信号信噪比低、个体差异较大的问题，提出一种改进的共同空间模式(Common spatial pattern，CSP)特征提取算法，并结合支持向量机(Support vector machine，SVM)分类器，实现对单次运动想象脑电信号的分类;使用此分类方法对两种不同试验范式建立分类模型，并对其分类表现进行评估;将较好分类表现的分类模型应用于上肢外骨骼的实时控制中，验证方法的可行性。所有被试对上肢外骨骼控制的平均成功率达到87.12%±2.03%。试验结果表明，基于所提出的运动想象分类方法，可以实现上肢外骨骼的准确控制，并为面向康复训练的脑机接口技术提供了理论依据和实践基础。

For solving the problem how the hemiplegic patients control the upper-limb exoskeleton during the active rehabilitation training,this study proposed an EEG classification method based on single-trial motor imagery. And this method in the real-time control of an upper-limb exoskeleton developed is applied. Aiming at the low noise-signal ratio and large individual differences of EEG,an advanced CSP algorithm for feature extraction is proposed. Combining this algorithm with SVM classifier,the single-trial motor imagery EEG is classified. Then,this method to construct classification models in two different paradigms is used,and evaluated the classification performance of two models. The classification model which had a better performance is applied in the real-time control of an upper-limb exoskeleton,to verify the feasibility of this method. The average accuracy is 87.12%±2.03% across all subjects in real-time control. The results demonstrate that the upper-limb exoskeleton can be controlled accurately based on the proposed method,and this study is provided the theory evidence and practical basis for BCI technology used in the rehabilitation training.