A new EEG feature selection method for self-paced brain-computer interface

In BCI research community, EEG based self-paced brain-computer interfaces (SBCI) have been widely researched in the past several years. SBCI systems allow individuals to control outside device using EEG signals at their own pace. But the performance of current SBCI technology is not suitable for most applications due to the difficult in detection of the non-periodic intentionally brain state changing. In this paper, we propose a new feature selection method based on particle swarm optimization (PSO) for EEG-based motor-imagery (MI) SBCI systems. The method includes the following two steps: (1) an optimization algorithm, i.e. PSO is used to select the EEG features and classifier parameters; and (2) a voting mechanism is introduced to remove the features redundant, which produced by optimization algorithm. We also compare the proposed method with the genetic algorithm (GA) method. Experiment on single-trial MI EEG classification shows the effectiveness of the proposed method.