A novel artifact removal method for the SSVEP signal using hybrid CCA-DWT and comparative analysis for feature selection and classification in the P300 signal

Electroencephalogram (EEG) based brain-computer interface speller is a communication rehabilitation tool to help patients suffering from motor disorders. A hybrid EEG signal based on steady-state visual evoke potential (SSVEP) and P300 signals is, although more efficient to have a robust speller application, however, there are some challenges including low signal-to-noise ratio, low information transfer rate, and classification accuracy in a smaller number of trials. To overcome these issues, this study proposes two approaches for both SSVEP and P300. For the former, a novel hybrid denoising approach based on canonical correlation analysis and discrete wavelet transform (DWT) was proposed, which offers a significant improvement in the single-flicker SSVEP signal. Furthermore, four feature selection techniques are applied to a combination of temporal features and DWT features to remove irrelevant and redundant information in the P300 signal based on the rapid serial visual presentation paradigm. Then seven strong popular classification techniques are applied to P300 coefficient detection, where the proposed single-layer discriminative restricted Boltzmann machine has shown more robust results compared with other methods. The average character recognition accuracies among six subjects are 51±5% and 94±4% with the average data rates of 34±5 and 26±2 bit/min, for one and five repetitions, respectively.