Parameter transfer and Riemannian space coordinate alignment for EEG intention recognition

• Published in: Biomedical signal processing and control Vol. 92; p. 106044
• Main Authors: Li, Xingchen, Tang, Xianlun, Liu, Jieyu, Liu, Qing, Wang, Huiming, Liu, Ke
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.06.2024
• Subjects: EEG intention recognition; Intelligent wheelchair; Parameter transfer; Riemannian manifold; EEG intention recognition; Parameter transfer; Intelligent wheelchair; Riemannian manifold
• ISSN: 1746-8094; 1746-8108
• DOI: 10.1016/j.bspc.2024.106044

•The proposed SPT obtains the correlation between data distribution and classifier parameters with a few labeled data of the target subject.•The proposed RCA defines a translation matrix that is able to align the distribution of data in different domains and preserve its internal features through simple geometric transformations.•RCA does not use any labeled data and align data from multiple domains at the same time, which can effectively promote knowledge transfer and improve learning ability.•An intelligent wheelchair steering system is designed based on the proposed methods. In brain-computer interface (BCI) system, local and temporal information of brain signals are often distorted and different subjects are variable, which leaves the brain-computer interfaces based on motor imagery (MI) challenging in cross-subject variability. This paper proposes a novel transfer learning method and a data alignment algorithm based on Riemannian space to reduce the difference between subjects. Through these methods, the target classification parameters will be close enough to that of the previous subjects which have a similar feature distribution to the target one. Therefore, even when only small amount of subject-specific trials of current target subject are labeled, the classification accuracy will not be compromised. The proposed two algorithms are evaluated using three motor imagery EEG data sets. Furthermore, an intelligent wheelchair steering system is designed based on the proposed algorithms. The experiment results show that in contrast with the subject-specific classifier, the proposed approaches achieve greater cross-subject classification accuracy. Moreover, the improvement is more salient when there are fewer trials available for the current target subject. In addition, the two algorithms proposed in this paper are combined to obtain optimal results.