A general framework to estimate spatial and spatio-spectral filters for EEG signal classification

Davood Fattahi*, Behrooz Nasihatkon, Reza Boostani

*Corresponding author for this work

    Research output: Contribution to journalArticlepeer-review

    27 Citations (Scopus)

    Abstract

    In this paper, a general framework is proposed for simultaneous design of spatial and spectral filters, which are used to extract discriminant features from EEG signals in Brain Computer Interfacing (BCI) systems. This paper introduces Common Spatial Patterns (CSP) as a step-by-step filter optimization algorithm, and then proposes a generalized type of the CSP which is not limited in a specific optimization constraint. Moreover, it is shown that how this generalization can be extended to a spatio-spectral filter estimation scheme. Then, two specific versions of the generalized CSP are proposed, where a specific target function and optimization constraint are used for estimating the spatial and spectral filters. Unlike the traditional CSP which is not very closely linked to the classification accuracy, the proposed algorithms are able to be more directly aimed at achieving better accuracy and stability. Experimental results obtained from applying the introduced methods on the recorded imagery signals from two datasets, demonstrate considerable improvement in the classification accuracy and stability compared to the standard CSP and other similar methods.

    Original languageEnglish
    Pages (from-to)165-174
    Number of pages10
    JournalNeurocomputing
    Volume119
    DOIs
    Publication statusPublished - 7 Nov 2013

    Fingerprint

    Dive into the research topics of 'A general framework to estimate spatial and spatio-spectral filters for EEG signal classification'. Together they form a unique fingerprint.

    Cite this