The brain-computer interface (BCI) facilitates a direct communication pathway between the human brain and external devices, bypassing the need for normal motor output pathways. Among various BCI methods, the steady-state visual evoked potential (SSVEP)-based BCI has gained significant attention due to its high signal-to-noise ratio (SNR), short training time, and rapid communication rate. It has been extensively explored in various applications, including assistive technologies, rehabilitation, communication, and entertainment. Despite some progress reported in recent literature, achieving reliable and accurate translations of user intentions in real-world scenarios remains highly challenging. This is mainly attributed to the instability of EEG signals and the disruptions encountered in practical situations. Besides, many existing systems only identify discrete commands, resulting in a gap between a user’s cognitive intentions and the system’s physical actions.
Within the framework of the SSVEP-based BCI system, the crucial significance of reliability and accuracy comes prominently to light. These attributes are pivotal for achieving precise command over external devices, simultaneously enriching the user experience and ensuring safety. Therefore, this thesis focuses on the development of a reliable and accurate BCI based on SSVEP technology, and the work is centered on four aspects: 1) To improve the recognition accuracy of the SSVEP signals; 2) To enhance the reliability of classification by rejecting low-confidence results; 3) To boost recognition performance for a new user through the incorporation of knowledge from existing users; 4) To apply the SSVEP-based BCI for controlling the velocity of the robotic arm according to the user's intention. To accomplish these objectives, the following efforts were undertaken: 1) A multi-objective optimization-based high-pass spatial filtering method was proposed for improving SSVEP recognition accuracy. This approach has the potential to extract target-relevant features, reject target-irrelevant information, and mitigate the impact of volume conduction simultaneously. 2) A Bayesian-based classification confidence estimation method is proposed to improve recognition reliability. This method estimates the probability of correctness for each classification of the recognition system, allowing the identification and rejection of low-confidence results. The BCI system can make high-confidence decisions and mitigate potential errors by incorporating confidence information. 3) An inter-subject transfer learning method was proposed, leveraging SSVEP signals from source subjects to strengthen the recognition performance of a target subject. By transferring knowledge from existing users, this method enhances the adaptability of the BCI system, particularly for the new user with limited training data. 4) In pursuit of more natural and responsive control, a velocity modulation method based on stimulus brightness was integrated into the SSVEP-based BCI system. Unlike conventional approaches with fixed movement direction and speed, this method dynamically adjusts the movement direction and speed based on the subjects' intentions. Users can interactively modulate the robotic arm's movement by focusing on specific high- or low-brightness stimuli, leading to a more natural and intuitive control experience. In conclusion, this thesis contributes to the advancement of SSVEP-based BCI technology by developing a reliable and accurate system while exploring its practical applications.
