Improving CCA-Based Methods for SSVEP Classification using a Common Source Graph
Nastaran Noori, Sepideh Hajipour Sardouie · 2024
Steady State Visual Evoked Potential (SSVEP)-based Brain Computer Interfaces (BCIs) is one of the most prevalent methods in BCIs. The main purpose of these systems is to detect the frequency of SSVEP to understand which visual stimulus a person is looking at. Various methods have been proposed to detect the frequency of SSVEP. One of the most widely used methods is the Canonical Correlation Analysis (CCA) algorithm. Despite the simplicity and efficiency of the CCA algorithm, none of the methods based on CCA have incorporated the graph structure presented in the EEG and reference signals into the CCA cost function. In this paper, we introduce a novel approach that considers the graph structure between time samples of our signals. This additional graph information can be exploited to estimate more meaningful canonical coefficients. Based on this approach, we propose three algorithms: graph canonical correlation analysis (GCCA), fusing graph canonical correlation analysis (FoGCCA), and individual-template graph canonical correlation analysis (ITGCCA). Evaluating their performance on the SSVEP dataset, we achieve an average accuracy of 82.08% for the FoGCCA algorithm in a 1.5-second time window, which suggests the effectiveness of our methods.