Feature Extraction And Recognition Of Motor Imagery Eeg Signals

One of the challenges of the brain-computer interface(BCI)system research is to reduce the time for user training on the premise of ensuring accurate identification.The BCI system based on the Motor Imagery(MI)paradigm requires training for a period of time to adjust the system to suit each user’s brain,and create a classifier by obtaining the user’s Electroencephalogram(EEG).The performance of this classifier depends on the amount of data used for training.More data can improve the classifier,but it will also increase the training time,which is particularly problematic for some patients.Therefore,this paper proposes to design a BCI system that can shorten the training time of users in an offline state,in order to achieve a feasible EEG recognition accuracy with a small training sample.The main work of this paper is as follows:(1)Aiming at the lack of constraints and feedback in the experimental paradigm of spontaneous EEG,and the low participation of subjects,this paper conceptualizes the life scenario of moving a cup of water as a dynamic and complex task model operating in a virtual environment,which is established by dynamic and static analysis.The model motion equation is programmed to present the task model based on the function library Psych Toolbox in the MATLAB 2016 a programming environment,and the subject’s EEG signals are collected by performing the boundary avoidance task with energy constraints and visual guidance.(2)For noise and artifacts in EEG signals,the method of polynomial fitting and spatial filtering are used to remove baseline drift and artifacts respectively.The common spatial patterns was used to extract the EEG features and support vector machine was used to recognize the features.By comparing the EEG of the boundary avoidance task with the characteristic recognition results of the fourth BCI competition dataset 1,the superiority of the boundary avoidance task induced EEG was demonstrated.In addition,in order to shorten the training time,the classification of the boundary avoidance task EEG under small samples was explored,and regularization parameters and contraction parameters were introduced to solve the overfitting problem.The results show that the regularization method can improve the control precision of small sample intention recognition.(3)In order to improve the recognition results of EEG,the proposed feature extraction method based on nonnegative CP decomposition model.Through the continuous wavelet transform to obtain the frequency of the EEG components,EEG tensor can be generated.Nonnegative CP decomposition model is adopted to extract EEG tensor component characteristics.With a two-dimensional principal component analysis optimization feature,support vector machine is used to realize feature recognition.Furthermore,compared the recognition results based on common spatial patterns and support vector machine,the effectiveness of the proposed method in the boundary avoidance task and the fourth BCI competition 1 data sets.The dynamic complex task model and the recognition of EEG signals established in this paper can provide theoretical support for BCI research and improve the overall performance.