A NOVEL REINFORCEMENT LEARNING STRATEGY FOR SEQUENTIAL DETECTION OF STEADY-STATE VISUAL EVOKED POTENTIAL-BASED BCI

Steady-state visual evoked potential (SSVEP) had been widely ap-plied for brain computer interfaces (BCI) control. However, low-frequency stimu-lation might induce excessive visual discomfort and photosensitive epilepsy in sub-jects. In this paper, we proposed a novel self-adaptive algorithm combined rein -forcement learning strategy and sequential detection-based canonical correlation analysis (SDCCA) for SSVEP detection. On the independent dataset, our pro-posed method achieved a mean classification accuracy of 85.12% and information transmission rate of 8.86 bpm, which was 2.86%-9.17% and 0.95bpm-3.64bpm higher than those of state-of-the-art algorithms, respectively. It was validated that reinforcement learning strategy was more robust and applicable for sequen-tial detection of non-stationary time-series signals.