脑-机接口（BCI）系统因能够在大脑和外部设备之间建立一种直接的交流通道而备受关注。近年来，依赖于单一输入信号或模态的传统BCI系统在新技术、新算法的加持下已取得长足进步，但受限于单一信号和模态的固有缺点，系统在性能、适用人群和应用场景等方面仍存在局限性。而将多种脑活动信号或其他生理活动信号相融合所构建的多模态或称混合BCI系统表现出优于单模态BCI系统的实用性、普适性和鲁棒性。基于脑电图（electroencephalography，EEG）技术的稳态视觉诱发电位（SSVEP）BCI系统具有信息传输率高、训练时间少，使用简单等特点，是构建混合BCI系统的常用模态。但目前SSVEP-BCI的研究主要集中在低频刺激，长时间使用容易产生疲劳，而高频刺激的舒适性相对较高。因此本文对结合高频SSVEP的混合BCI系统展开了进一步研究。

针对目前结合运动想象（MI）和SSVEP的并行混合脑-机接口范式中混合任务相关性较差的问题，本文通过引入互调频率的概念，构建了更为自然的MI与SSVEP并行混合范式。进一步，通过对吉洪诺夫正则化共空间模式和共空间谱模式两种MI信号解码算法进行优化结合，提出了吉洪诺夫正则化共空间谱模式；此外，提出一种基于概率分布的融合决策方法，对MI和SSVEP分类器所输出的标签进行融合决策，得到混合系统最终的分类结果。健康受试者和脑卒中患者的在线实验结果验证了混合范式的可行性。

高频SSVEP的识别效率较低，在高频SSVEP-BCI中，通常为了保证分类精度需要在数据长度和目标数量之间进行权衡，因此目前高频SSVEP-BCI系统的信息传输率与中低频SSVEP-BCI相比仍有较大的差距。本文采用联合频率-相位调制方法编码了16个高频目标，以增加各目标之间的可区分性，并首次使用任务判别成分分析算法解码高频SSVEP；同时结合四种手腕动作构建基于肌电图与高频SSVEP的混合BCI系统，将高频SSVEP系统指令集扩展到64个。该混合系统获得了88.07±4.51%的平均正确率和159.12±13.63 bits/min的平均信息传输率，优于目前现有的基于高频SSVEP的BCI系统。

与现有相关混合BCI研究相比，上述两套结合高频SSVEP的混合BCI系统在性能、舒适度等方面均得到提高和改善，实用性和普适性也有所增强。实验结果证明两套系统在健康受试者中的可行性和有效性，且将其应用于病人或残疾人群的运动能力恢复或日常交流等方面的潜力也值得进一步挖掘。

Brain-computer interface (BCI) systems have gained attention for their ability to establish a direct communication channel between the brain and external devices. In recent years, traditional BCI systems that rely on a single input signal or modality have made great progress with the support of new technologies and algorithms, but they are still limited by the inherent characteristics of a single signal and modality. The disadvantage is that the system still has limitations in terms of performance, applicable population, and application scenarios. The multi-modal or hybrid BCI system constructed by fusing a variety of brain activity signals or other physiological activity signals is more practical, universal, and robust than the single-modal BCI system. Steady-state visual evoked potentials (SSVEP)-based BCI with electroencephalography (EEG) technology has the characteristics of high information transmission rate, less training time, and simple use. It is a common modality for constructing hybrid BCI systems. However, the current research on SSVEP-BCI mainly focuses on low-frequency stimulation, which is prone to fatigue when used for a long time, while the comfort of high-frequency stimulation is relatively high. Therefore, this paper further studies the hybrid BCI system combined with high frequency SSVEP.

Aiming at the poor correlation of mixed tasks in the current parallel hybrid brain-computer interface paradigm combining motor imagery (MI) and SSVEP, this paper constructed a more natural parallel hybrid of MI and SSVEP by introducing the concept of intermodulation frequency. Further, by optimizing and combining two MI decoding algorithms, the Tikhonov regularized common spatial pattern and the common spatial-spectral pattern, a Tikhonov regularized common spatial-spectral pattern was proposed. In addition, a probability distribution-based algorithm was proposed. The fusion decision method is to fuse the labels output by the MI and SSVEP classifiers to obtain the final classification result of the hybrid system. Online experimental results in healthy subjects and stroke patients validated the feasibility of the hybrid paradigm.

The recognition efficiency of high-frequency SSVEP is low. In high-frequency SSVEP-BCI, a trade-off between the data length and the number of targets is usually required to ensure the classification accuracy. Therefore, the information transfer rate of the current high-frequency SSVEP-BCI system still has a big gap compared with that of medium- and low-frequency SSVEP. Therefore, this paper introduced the joint frequency-phase modulation method to encode 16 high-frequency targets, and then used task discriminant component analysis (TDCA) to decode high-frequency SSVEP for the first time. At the same time, it combined four wrist movements to construct a hybrid BCI system of electromyography (EMG) and high-frequency SSVEP, expanding the high-frequency SSVEP system instruction set to 64. The hybrid system achieved an average correct rate of 88.07±4.51% and an average information transfer rate in 159.12±13.63 bits/min, outperforming the currently known high-frequency SSVEP-based BCI systems.

Compared with existing related hybrid BCI studies, the above two hybrid BCI systems combined with high-frequency SSVEP have been improved in terms of performance and comfort, and their practicability and universality have also been enhanced. The experimental results demonstrated the feasibility and effectiveness of the two systems in healthy subjects, and the potential of applying them to exercise recovery or daily communication in patients or disabled individuals is also worthy of further exploration.