PSO-Sub-ABLD-Based Parameter Optimization for Motor-Imagery BCI

Feiyu Yin,Yangyang Miao,Xingyu Wang,Jing Jin
DOI: https://doi.org/10.1007/978-981-16-0317-4_23
2021-01-01
Abstract:Common spatial pattern (CSP) is one of effective feature extraction algorithms, which is widely applied to motor imagery (MI)-based brain–computer interface (BCI). However, its performance is susceptible to artifacts and noise. Therefore, some researchers proposed Sub-Alpha-Beta Log-Det Divergences (Sub-ABLD) algorithm to improve the performance of BCI systems. The performance of Sub-ABLD algorithm depends on the values of hyperparameters \(\alpha \), \(\beta \) and \(\eta \). In this study, a strategy named PSO-Sub-ABLD was proposed to select three hyperparameters with particle swarm optimization (PSO). Two public BCI competition datasets were used to validate the effectiveness of the proposed strategy. The results show that compared with CSP and Sub-ABLD with default hyperparameters, PSO-Sub-ABLD method gains better classification accuracy.
What problem does this paper attempt to address?