Study On The Effective Channels Of EEG Signals Based On Feature Selection

EEG is an important technology to explore the secret of the human brain.It plays an important role in the application of BCI system and computer-aided diagnosis system for its high time resolution and noninvasive test.In different fields of applications,people collect EEG signals through channels on the scalp surface.However,different applications require different channels.Hence some channels have become redundant or unrelated in specific applications.Therefore,the research on how to select effective channels has become one of the most important problems in the EEG signals processing.In order to ensure the generality of channel selection methods in different fields of applications,more and more attention has been paid to channel selection method based on wrapper technique.In these methods,the channel selection methods based on single objective optimization can not well weigh the number of selected channels against the classification accuracy.While the channel selection methods based on multi-objective optimization investigates the optimal tradeoff between the classification accuracy and the number of selected channels,but it costs too much time.In addition,due to the individual differences on EEG signals have,the combination of channels which is given by multiobjective optimization is only suitable for subjects themselves and cannot be used by other subjects.In this paper,in view of the problem of current channel selection methods,the research on effective channels of the EEG signal is studied.The specific work is as follows:1.The channel selection method based on improved UPS-EMOA is proposed.The improved UPS-EMOA preserves the original UPS-EMOA’s strategy to not restrict population size.It introduces the concept of crowding degree of NSGA-II algorithm,and adopts mutation,crossover operator which are used in differential evolution algorithm.The experimental results show that the method can greatly reduce the cost of time while ensuring high accuracy.In addition,the method has been recommended to use on the EEG data with 64 channels because of its good performance on these datasets.2.The channel selection methods based on three kinds of multi-objective optimization are compared.The advantages and disadvantages of the three methods are analyzed,and their respective applicable conditions are given in the paper.3.A frequent channel combination mining method based on multi granularity and FP-Tree is proprosed.First of all,the EEG signal of subject is translated into subjects’ feature vector.Then,HML algorithm is used on subjects’ feature sets to obtain hierarchy generation,different levels represent different levels of granularity.Finally,the appropriate level is chosen.On this level,FP-Growth algorithm is used to mining frequent channel combination from a cluster,the combination of channel can be extended to other subjects which are in the same cluster.The experimental results show that the method makes full use of the channel combination of other subjects and migrates the channel combination to other subjects.At last,the method realizes the zero training.