A Study On The Classification Of Electroencephalographic Signals In Motor Imagery Based On Transfer Learning

Brain-computer interface(BCI)is a communication system that can use computer and other external devices to interact with human machine.BCI system has been widely used in many fields,such as medical rehabilitation,public consumption and entertainment,education and consultation.However,due to the non-stationarity of EEG signals and the differences among different subjects,the common subject classification method cannot produce high accuracy in the classification of different subjects.It is challenging for adapting classifiers from are user to another in brain-computer,which is necessary to reduce training time for new users.Inspired by transfer learning,two improved algorithms are put forward to solve this problem in this paper.The main research contents and innovations of the paper are as follows:In this paper,subspace alignment and adaptive common space pattern(CSP)is proposed.The algorithm updates the covariance matrix by calculating and selecting suitable candidates,constructs a spatial filter to extract classification features,and improves the subspace alignment algorithm for the extracted feature vectors to reduce the distribution difference between the training set and the test set,so as to improve the performance of the classifier.Finally,the performance of the proposed algorithm is verified on the public EEG data set.In this paper.an algorithm based on subspace weighting and improved common space pattern is proposed by improving subspace algorithm.The algorithm can improve the performance of the classifier by weighting the samples according to the similarity between the samples.Then,the performance of the algorithm is verified by experimental analysis on public data.