Parallel genetic algorithm based common spatial patterns selection on time–frequency decomposed EEG signals for motor imagery brain-computer interface

• Published in: Biomedical signal processing and control Vol. 80; p. 104397
• Main Author: Luo, Tian-jian
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.02.2023
• Subjects: Brain computer interface; Common spatial pattern; Motor imagery; Parallel genetic algorithm; Time segments and frequency bands; Parallel genetic algorithm; Motor imagery; Brain computer interface; Common spatial pattern; Time segments and frequency bands
• ISSN: 1746-8094; 1746-8108
• DOI: 10.1016/j.bspc.2022.104397

•A complete time segments and frequency bands were performed on the EEG signals to extract TSFBCSP features.•A MapReduce framework based genetic algorithm is proposed to select optimal TSFBCSP features.•Experiments on two public benchmark datasets validate the proposed method. Since the nonlinear and non-stationary characteristics of electroencephalogram (EEG) signals, motor imagery based brain-computer interface (MI-BCI) have problems of poor recognition accuracy and robustness across subjects and recording sessions. To address such problems, we extract common spatial pattern features on a detailed decomposition of time segments and frequency bands (TSFBCSP) on EEG signals, and propose a novel encoding approach of genetic algorithm (GA) for features selection. The proposed approach selects the most robust and discriminative CSP features to enhance the accuracy and robustness of EEG signals recognition. To improve the efficiency of TSFBCSP features selection, the parallel GA is implemented on a MapReduce framework (MRPGA). Comparative experiments have been done on two publicly available datasets (2a and 2b) from BCI competition IV. The excellent average recognition accuracy shows that the proposed TSFBCSP-GA/MRPGA has a promising candidate performance improvement on MI-BCI. Classification accuracies based on the optimal EEG segments and sub-bands have achieved 74.92 % and 81.04 % for the dataset 2a and 2b, respectively, and the faster classification accuracies of 74.54 % and 79.78 % has been achieved for the MRPGA architecture. Ablation study further proves the feasibility and effectiveness of the proposed algorithm, and discusses the parameters setting of applying it to actual MI-BCI applications.