Study On The Application Of Evolutionary Algorithms In Channel Selection Of Motor Imagery Brain-Computer Interface

The brain-computer interface(BCI) is a communication system that a man communicates with a machine(eg a computer) without relying on the body's neuron and muscle, and the extraction of brain signal is directly transmitted to the outside world to control the external equipments. With the 40 years development of BCI, it has gradually had some application into life from the laboratory's study. The common spatial pattern(CSP) is a algorithm that can do some well in spatial filter and EEG feature extraction in BCI system, which based on motor imagery, but it uses a strict line model relationship during recording brain signals. Thus the CSP is severly subjected to the subject's original record parameters such as the EEG period, the filter band, the number of electrode, and therefore it can't be accurately and effectively presented the brain features.In order to match and adapt the subject's original record parameters for extracting EEG feature signals to improve the performance of CSP, two optimization algorithms are used to select the best set of channels in this thesis. One is backtracking search optimization algorithm(BSA), the other is binary particle swarm optimization(BPSO). Besides, two classifiers are also described in this thesis. One is support vector machine(SVM), the other is fisher linear discriminant analysis(FLDA). The number of the selection channels are determined by the BSA and BPSO in process of the channel optimization and the third and fourth BCI competition datasets are used for selection experiments. It shows, compared with all channels, the channels used two optimization algorithms are drammtically reduced and the classification accuracy achieves better.