Título: Interface Cérebro -Máquina: uma abordagem ótima via distância de Riemann por sub -banda
Autores: Leilane de Jesus dos Anjos, Angello Gabriel Alves Gonçalves & Cleison Daniel Silva
Resumo: Brain-Computer Interface systems based on Motor Imagery decode the user’s motor intentions from electroencephalography (EEG) signals, converting them into control commands for devices external to the human body. This paper proposes an enhanced approach for classifying EEG signals related to four motor imagery classes evaluated across binary combinations. The methodology employs symmetric positive definite covariance matrices extracted after applying a filter bank with overlapping sub-bands. Feature extraction is based on the Riemannian distance between mean matrices computed using the MDRM algorithm; the resulting values are normalized and used as input to a Support Vector Machine (SVM) classifier. Bayesian optimization is applied to fine-tune the system’s hyperparameters automatically. Experimental results show superior performance compared to the reference method in the literature, confirming the effectiveness of the proposed approach in adapting to intra- and inter-subject variability in two evaluated scenarios.
Palavras-chave: Riemann Geometry; Bayesian Optimization; Brain-Computer Interface.
Páginas: 7
Código DOI: 10.21528/CBIC2025-1191476
Artigo em PDF: CBIC_2025_paper1191476.pdf
Arquivo BibTeX:
CBIC_2025_1191476.bib