Offline Optimization of Filter Bank Canonical Correlation Analysis for SSVEP Classification

Within the field of EEGs, one of the most popular types of signal for classification is the Sinusoidal Steady State Visually Evoked Potential, or SSVEP signal. It is favored for its high signal to noise ratio and its relative ease of use compared to other types. One method for classifying it, the Canonical Corre-lation Analysis, is favored for its accuracy and speed. One way of further improving this accuracy is through the use of a filter bank to isolate the signal into multiple sub bands and performing CCA onto the sub bands. These sub bands then have weights applied to them and are summed together before the classification is determined. This experiment compares the accuracy of both the standard CCA(Canonical Correlation Analysis) method and the FBCCA(Filter Bank Canonical Correlation Analysis) in a series of offline experiments where different parameters are tested and the maximum classification accuracy is found. The overall results of the experiment found that the greatest accuracy for these conditions were yielded under 4 harmonics in the reference signals for the standard CCA experiment. The filter bank CCA experiment resulted in a maximum classification accuracy of around 99 percent for four sub bands and weight coefficients of 2 and 0.75.