Regularized tensor discriminant analysis for single trial EEG classification in BCI

Authors:
Jie Li;Liqing Zhang
Affiliations:
Department of Computer Science and Engineering, Shanghai Jiao Tong Univeristy, Shanghai 200240, PR China;Department of Computer Science and Engineering, Shanghai Jiao Tong Univeristy, Shanghai 200240, PR China
Venue:
Pattern Recognition Letters
Year:
2010

Citing 1
Cited 1

General Tensor Discriminant Analysis and Gabor Features for Gait Recognition

IEEE Transactions on Pattern Analysis and Machine Intelligence

Classifying healthy children and children with attention deficit through features derived from sparse and nonnegative tensor factorization using event-related potential

LVA/ICA'10 Proceedings of the 9th international conference on Latent variable analysis and signal separation

Quantified Score

Hi-index	0.10

Visualization

Abstract

In this paper, a tensor-based scheme is introduced for single trial electroencephalogram (EEG) classification in brain computer interfacing (BCI). Firstly, EEG signals are represented as third order tensors in the spatial-spectral-temporal domain by wavelet transform. Then, a regularized tensor discriminant analysis (RTDA) algorithm is proposed for a multi-way discriminative subspace extraction from tensor-represented EEG data. Unlike the conventional wavelet transform method, the proposed scheme includes the structural information in multi-channel time-varying EEG spectrums endorsed by tensor representation, and improves the performance for EEG classification. Compared with the common spatial pattern (CSP, the most successful algorithm in BCI) in the applications to two classes of datasets, the proposed scheme has the following advantages: (1) an optimal multi-way discriminative subspace can be extracted, obtaining significant spatial-spectral-temporal patterns for EEG classification; (2) the proposed scheme can identify discriminative characteristics robustly, and works well without prior neurophysiologic knowledge. This is a valuable property for developing new paradigms in BCI whose discriminative neural correlates are not known and (3) the proposed scheme is able to find the most significant channels for classification, and can be applied to channel selection in BCI. Computer simulations show that the number of used channels can be reduced to 2in two datasets with very little loss in performance. Therefore, it has great potential for the practical application of BCI.