Improved CS-SVM And Its Application In Pattern Recognition Of Motor Imagery EEG

Brain-Computer Interface technology is a hotspot in recent years,such as control and neural repair,signal processing,artificial intelligence,pattern recognition and other fields.Motor imagery BCI is the most typical,and it can help patients to carry out rehabilitation training and restore motor function.In the recognition of EEG signals,the most fundamental problem is the need for an effective feature extraction and classification algorithm.Based on the analysis of the characteristics of support vector machine(SVM)and Cuckoo search(CS),this paper presents an improved Cuckoo search algorithm optimizes the support vector identification method(abbreviated as ICS-SVM)and applies it to the pattern recognition of motion imagination.The main work is as follows:(1)To extract the feature of the EEG signal,this paper Combining the wavelet packet decomposition algorithm and the Common Spatial Pattern(CSP)method.According to the characteristics of motor imagery EEG and event-related de-synchroniz-ation/synchronization,the specific frequency components in the relevant time period are extracted,and the energy ratio feature of left and right channel are proposed.In addition,the CSP algorithm is used to extract the feature of spatial domain information implied in multi-channel EEG signal.The combined time-frequencyspace feature is complete and can improve the classification accuracy.(2)Aiming at the support vector machine(SVM)classification algorithm,this paper selected the ant colony algorithm,particle swarm algorithm,fish swarm algorithm and cuckoo search algorithm used to optimize the SVM parameters.Based on the three UCI data,The results show that the cuckoo search algorithm is superior to other algorithms in terms of classification accuracy and convergence speed.(3)This paper proposed the improved cuckoo search algorithm(ICS).Aiming at the problems existing in the cuckoo search algorithm,three improvements are proposed:1)The adaptive improvement of the step factor.Increases the step size updating method of the optimal nest.2)proposed the adaptive variation formula of discovery probability.3)proposed a new search formula combined with global optimal nest guidance.Inthis paper,we use the basic CS algorithm and the improved ICS algorithm to compare experiments on five test functions.The results show that ICS algorithm is superior to CS algorithm in terms of searching accuracy and convergence speed in addition to Rosenbrock test function.(4)In this paper,SVM,CS-SVM and ICS-SVM are used to compare the classification accuracy of three UCI data sets.The results show that the ICS-SVM algorithm is better than the other two algorithms.(5)Based on the feature extraction algorithm and the improved ICS-SVM classification algorithm,the experiments were carried out on five motor imagery EEG datasets in BCI competiton IV(Data sets 1),and the basic SVM and CS-SVM algorithm was selected for comparison experiments,the results show that the highest accuracy of the algorithm is ICS-SVM,verify the effectiveness of this algorithm.