METODOLOGIA ADAPTATIVA PARA ATUALIZAÇÃO DOS PARÂMETROS DO CSP E LDA UTILIZANDO UMA ABORDAGEM BASEADA NO FK

Odisley E. Nascimento; Wendy Y. Eras-Herrera; Fabricio J. Erazo-Costa

doi:10.20906/sbai.v1i1.2711

Odisley E. Nascimento Departamento de Engenharia Elétrica, Universidade Federal de Ouro Preto, MG
Wendy Y. Eras-Herrera Departamento de Engenharia Elétrica, Universidade Federal de Ouro Preto, MG
Fabricio J. Erazo-Costa Departamento de Engenharia Elétrica, Universidade Federal de Ouro Preto, MG

DOI: https://doi.org/10.20906/sbai.v1i1.2711

Keywords: brain-computer interface, commom spatial patterns, Kalman filter

Abstract

This article proposes the use of CSP (Common Spatial Patterns) as a feature extraction method for the recognition of patterns present in brain signals from motor imagery. These signals are classified using the LDA (Linear Discriminant Analysis) classifier. In order to improve the classification, an update is proposed, both in the CSP and in the LDA, based on the Kalman filter algorithm (KF). The KF has a prediction and correction structure that allows an improvement in the estimation of states in relation to linear classifiers. The results demonstrate an improvement in classification accuracy when comparing the conventional time-invariant CSP and LDA methods with the adaptive methods.