Restricted Exhaustive Search for Frequency Band Selection in Motor Imagery Classification

Full text
Author(s):	Bustios, Paul ; Rosa, Joao Luis ; IEEE Total Authors: 3
Document type:	Journal article
Source:	2017 INTERNATIONAL JOINT CONFERENCE ON NEURAL NETWORKS (IJCNN); v. N/A, p. 6-pg., 2017-01-01.
Abstract
Motor imagery is a mental process that produces modulations in the amplitude of ongoing electroencephalogram signals. The patterns present in these modulations can be used to classify this mental process, but the identification of these patterns is not a trivial task, because they are present in frequency bands that are specific for each person. In this paper, we introduce a novel method to select these subject-specific frequency bands in a new configuration for the Filter Bank Common Spatial Pattern approach. Our method uses an exhaustive search to find the best subset of frequency bands containing the most discriminative patterns within a search space restricted to a fixed size for this subset. The size is determined using cross-validation and the Sequential Forward Floating Selection method. As we demonstrate with experiments on the data set 2b of the BCI Competition IV, our method is more accurate than current approaches evaluated on this data set. (AU)

FAPESP's process:	16/02555-8 - Development of algorithms and computational techniques for application in brain-computer interfaces
Grantee:	João Luís Garcia Rosa
Support Opportunities:	Regular Research Grants