An SSVEP Classification Method Based on a Convolutional Neural Network

Steady-state visually evoked potential (SSVEP) is a typical paradigm of brain-computer interface system (BCIs). How to fully extract and utilize the features of SSVEP to improve the classification accuracy is a crucial task in the development of SSVEP-BCI system. Aiming at the SSVEP-BCI system with little target visual stimulus paradigm, deep learning is applied to the research of SSVEP signal classification in this paper, and a classification method of SSVEP signal based on convolutional neural network (CNN) is proposed. It solves the problem of lower classification accuracy of canonical correlation analysis (CCA) and filter bank canonical correlation analysis (FBCCA) methods in short-time. The data set used for the study was 800 sets of SSVEP data from 20 subjects of Inner Mongolia University of Technology (IMUT). In this study, the effects of data length and number of leads on classification accuracy and information transmission rate (ITR) were analyzed. The experimental results show that the highest classification accuracy of CNN model reaches 95.86%, and the maximum ITR reaches 183.05bit/min. Compared with the existing CCA method and FBCCA method, CNN method can significantly improve the classification accuracy and ITR of SSVEP signal stimulation frequency.