BCI Competitions III Result
Dataset IVc
Group information:
LIU Yang, HU Dewen, ZHOU zongtan, ZANG Guohua
Department of Automatic Control,
National University of Defense Technology,
Changsha, Hunan, 410073, P.R.C.
gloolar@163.com
Algorithm description:
Preprocessing:
1) filter signals using bandpass filter(2.5Hz to 25Hz).
2) get data of 1s length as trials after the position of every cue for both training and testing data.
3) compute CSP(Common Spatial Pattern) using training data, apply it to both training and testing data, then compute cwt of every trials.
Classifying:
1) compute two correlation matrix A(among training trials) and C(between testing trials and training trials)
2) construct the feature space using every row of matrix A and C as features of corresponding trials.
3) optimize a linear discriminant surface from a initial position decided by the training trials under a energy penalty critera, ie.
min norm(Ax-b)+norm(x)
where x is weight vector of the linear discriminant function, b is vector containing expected positions of every training samples.