EEG classification based on Small-World neural network for Brain-Computer Interface

Focusing on mental task recognition, a novel Small-World neural network(SWNN) algorithm is proposed for the EEG classification tasks aiming at solving small training sets problem. For making some attempts to discover the agile experimental paradigms, two channel sets having different information-carrying capacities are built to filter the multi-channel EEG data, as two channel-filters. The band-pass filtering preprocessing is performed by IIR Chebyshev I Filter. Common spatial patterns, which can emphasize the greatest distinction among the most outstanding features of different patterns, is used to carry out spatial filtering. Bring in the Small-World neural network, which possesses the complex network structure transformed from the regular network by random rewiring according to the rewiring probability P and the high-dimensional weights adjusting mechanism based on back-propagation. This algorithm was applied to the data set IVa of “BCI Competition iii”, which provides trails for the classes “right hand” and “right foot”, with the classification accuracies of 99.1%~97.7% by 10-fold cross-validation.