Novel SSVEP Processing Method Based on Correlation and Feedforward Neural Network for Embedded Brain Computer Interface

Abstract

Steady State Visually Evoked Potential (SSVEP) is a successful strategy in electroencephalographic (EEG) processing applied to spellers, games, rehabilitation, prosthesis, etc. There are many algorithms proposed in literature to detect the SSVEP frequency, however, most of them must to be implemented in high processing computers because SSVEP methods require many EEG input channels and the algorithms are computationally complex. Then, this paper proposes a low computational cost method for SSVEP embedded processing (EP-SSVEP) whose input is one EEG channel and is based on Canonical Correlation and a Feedforward Neural Network. Additionally, this paper also proposes an embedded system to implement EP-SSVEP and a dataset composed with the EEG signals from eight subjects. According to the results, EP-SSVEP is one of the best methods in literature to SSVEP embedded processing because it reports an accuracy of 96.09% with the proposed dataset and the EEG input is acquired with one channel.