A Method for SSVEP Recognition Based on Weighted Canonical Correlation Analysis
Minghui Shi,Chang Liu,Tingting Chen,Min Jiang,Fei Chao,Meihong Wu,Enze Wang,Yuan Shi,Tongrui Li,Jianting Liu,Jingkang Lin,Zilong Li
DOI: https://doi.org/10.1109/ssci44817.2019.9003022
2019-01-01
Abstract:As a multivariate statistical method, canonical correlation analysis (CCA) has been one of the most common methods for recognizing steady-state visual evoked potential (SSVEP) in the field of brain-computer interface (BCI). Normally CCA-based methods do not distinguish the different visual feeling of the subject made by the stimuli with different frequencies. To address this issue, this paper proposed a novel method called weighted canonical correlation analysis (WCCA). To recognize the SSVEP frequency, WCCA first employs the standard CCA to obtain its canonical correlation coefficients with each of the reference signals, which correspond to each stimulus frequency. Then each coefficient is modified with a weight, which is associated with each stimulus frequency and obtained by a training procedure. Different from the SSVEP recognition method based on the standard CCA, WCCA uses the weighted coefficients, rather than the original coefficients, to recognize the SSVEP frequency of the testing EEG data. WCCA may be regarded as a generalized CCA, since WCCA is just CCA when each weight value is one. Therefore, by optimizing the weight values, WCCA may outperform CCA. The effectiveness of WCCA was verified by the experiment results.