A Study On Channel Selection And Classification Methods In Multi-Task Brain-Computer Interfaces

The development of brain-computer interface system has attracted lots of researchers at home and abroad, and it has becomes a research focus in neural engineering field in recent years. With the development of brain-computer interface, researchers are now attempting to put current brain-computer interface techniques into practical application. However, the brain-computer interface system based on motor imagery is still not used for real-life application due to the decreasing performance of common spatial pattern algorithm especially when the number of channels is large. In addition, with the increase of the number of channels, multi-channel EEG signals need inconvenient recording preparation and complicated calculation, this will be time-consuming and maybe lead to lower classification accuracy.To address the research question, there are four methods are proposed in my paper:one is binary multi-objective particle swarm optimization (bMOPSO); second is a novel method, named cultural-based MOPSO (cMOPSO); the third is filter banks-based cMOPSO algorithm (fbMOPSO); the last is typical L1-norm algorithm. The bMOPSO algorithm and two cMOPSO algorithms are based on particle swarm optimization (PSO). The PSO algorithm is a heuristic search technique that simulates the movements of a flock of birds, which aim to find food. PSO algorithm, previously used to handle single objective. MOPSO algorithm is the extension of PSO,and it has been widely used to solve multi-objective optimization problems. However, most MOPSO use fixed momentum and acceleration for all particles throughout the evolutionary process. The cMOPSO algorithm different from the bMOPSO algorithm, it introduces a cultural framework to adapt the personalized flight parameters of the mutated particles. The third method is typical L]_norm algorithm,a small number of channels can be picked out in two classes Electroencephalogram data by this method.In addition, three classification methods:Support Vector Machine (SVM)、 k-Nearest Neighbor (k-NN)、Back Propagation (BP)neural network, are used as classifier. Finally, three classification methods and four channel selection algorithms are respectively applied to three-task or four-task data sets recorded during motor imagery based on BCI experiments, the results show that the SVM and k-NN algorithms can achieve a better classification results than BP neural network method. The fbMOPSO algorithm is more effective in selecting a smaller subset of channels while maintaining the classification accuracy unreduced among four channel selection algorithms.