Título: Deep Riemannian Networks with Temporal Approach for Classification in Brain – Computer Interfaces
Autores: Alexandre Herrero Matias, João Guilherme Prado Barbon, Lucas Heck dos Santos & Denis Gustavo Fantinato
Resumo: In the context of electroencephalography (EEG) classification for Motor Imagery (MI) tasks, the use of the framework based on Riemannian Geometry (RG) has shown comparable performances to convolutional neural networks. Its application on EEG data usually considers the extraction of a sample covariance matrix, which captures spatial information (between electrodes). However, the temporal information can be used as well, for instance, considering a set of time-delayed covariance matrices. In that sense, in this work, we propose the use of the temporal information, mainly considering Deep Riemannian Networks (DRN). We also present a modified version of SPDNet in order to encompass a set of covariance matrices as input. Our approach enhances SPDNet’s ability to learn spatio-temporal patterns, resulting in improved classification performance on EEG data.
Palavras-chave: Brain-Computer Interface; Electroencephalography; Riemannian Geometry.
Páginas: 7
Código DOI: 10.21528/CBIC2025-1174180
Artigo em PDF: CBIC_2025_paper1174180.pdf
Arquivo BibTeX:
CBIC_2025_1174180.bib