Improving CCA Algorithms on SSVEP Classification with Reinforcement Learning Based Temporal Filtering

Canonical Correlation Analysis (CCA) has been widely used in Steady-State Visually Evoked Potential (SSVEP) analysis, but there are still challenges in this research area, specifically regarding data quality and insufficiency. In contrast to most previous studies that primarily concentrate on the development of spatial or spectral templates for SSVEP data, this paper proposes a novel temporal filtering method based on a reinforcement learning (RL) algorithm for CCA on SSVEP data. The proposed method leverages RL to automatically and precisely detect and filter low-quality segments in the SSVEP data, thereby improving the accuracy of CCA. Additionally, the proposed RL-based Temporal Filtering is algorithm-independent and compatible with various CCA algorithms. The RL-based Temporal Filtering is evaluated using a wearable dataset consisting of 102 subjects. The experimental results demonstrate significant advancements in CCA accuracy, particularly when combined with the extended CCA (ECCA) algorithm. In addition to performance enhancement, the RL-based Temporal Filtering method provides visualizable filters, which can ensure the transparency of the filtering process and the reliability of the obtained results. By addressing data quality and insufficiency concerns, this novel RL-based Temporal Filtering approach demonstrates promise in advancing SSVEP analysis for various applications.