A Genetic Algorithm-Based Multi-Parameter Method for Asynchronous Working State Classification Under SSVEP

Working state classification, which is a critical part of an asynchronous control system based on a brain-computer interface (BCI). The traditional algorithms for the working state classification are based on the threshold of power spectral density (PSD) or the canonical correlation analysis (CCA) parameters, which usually suffer insufficient robustness in case of using one threshold. To improve the robustness of working state classification, a multi-parameter asynchronous working state classification method combined with a genetic algorithm (GA-PSD-CCA) is proposed based on SSVEP. GA-PSD-CCA obtains the CCA parameter and the PSD parameter from the processed electroencephalogram (EEG) signals. Then GA-PSD-CCA iteratively searches the optimization of the PSD and CCA parameters and finds the threshold value by the genetic algorithm for the working state classification to recognize the working state. An experiment is designed to compare the application effect of GA-PSD-CCA with the traditional algorithm based on the EEG signals of ten subjects. The results show that the average accuracy with GA-PSD-CCA achieved 89.35% ± 8.39% and has an improvement of more than 7% compared with the traditional algorithm which has the highest accuracy. GA-PSD-CCA has a large improvement in asynchronous working state classification accuracy and achieves high target recognition accuracy. The BCI systems with the GA-PSD-CCA will have lower false recognition of commands and higher reliability and smoothness of state recognition processes.