Research On Classification Of Motor Imagery Brain-Computer Interface Tasks Based On Transfer Learning

Brain-computer interface(BCI)technology can provide a new way of communication for humans.By directly connecting the brain to external devices,information interaction between the brain and machines can be achieved without relying on the peripheral nervous system.Among them,the motor imagery brain-computer interface is a spontaneous BCI mode that is more in line with the "brain-computer interface" form in the cognitive process.Users generate EEG signals similar to real movements by imagining their own or others’ limb movements,analyze these EEG signals,and convert them into control commands to control external devices.The analysis of motor imagery EEG signals has important research value in BCI applications.However,due to the high cost of collecting motor imagery EEG signals and significant individual differences in EEG signals among subjects,there are problems of small sample size and difficulty in cross-subject classification when recognizing motor imagery tasks,resulting in a long calibration time and poor recognition accuracy for motor imagery BCI systems.In response to these issues,this paper designs two motor imagery EEG signal classification models based on transfer learning,achieving effective recognition of small sample and cross-subject right-hand and right-foot motor imagery tasks.The specific research work is as follows:(1)In response to the problem of a small sample size of motor imagery EEG data,an method for motor imagery task classification based on data alignment and feature transfer,Euclidean Alignment-Regularized Common Spatial Patterns(EA-RCSP)is proposed.Data alignment enables different subjects’ EEG data to be centered on the identity matrix,reducing the distribution differences between subjects.Regularized transfer learning is used to improve the common spatial pattern,extracting more discriminative spatial features.Multiple comparative experiments are set up on the BCI competition motor imagery EEG dataset to verify the effectiveness of the model in small sample and cross-subject classification tasks.The average classification accuracy of the right-hand and right-foot motor imagery binary classification tasks reaches 87.10%.The EA-RCSP method improves the average classification accuracy by 14.43% compared to the common spatial pattern and by 23.68% for cross-subject motor imagery tasks.The experimental results show that the method can effectively improve the small sample motor imagery task classification effect.(2)In response to the difficulty of cross-subject classification of motor imagery EEG signals,a model for motor imagery EEG signal classification based on pre-alignment strategy and adversarial transfer learning,Pre-alignment Strategy Weighted Multi-Source Adversarial Networks(PS-WMSAN).Combining transfer learning and adversarial training concepts,the domain adversarial neural network is extended to multiple source domains.Different source domains are weighted using the Pearson correlation coefficient to achieve weighted alignment of the source and target domains in features.The pre-alignment strategy is used to improve the consistency of data distribution between domains,and a sliding time window is used to expand the EEG samples.On the BCI competition motor imagery EEG dataset,the PS-WMSAN method achieves a recognition accuracy of 84.43% for cross-subject motor imagery tasks,an improvement of 5.00%compared to the domain adversarial neural network,and 5.86% higher than the EA-RCSP method.The experimental results show that this method can effectively reduce the differences in EEG data distribution and feature distribution among different subjects,achieve double alignment of data and features,and thus improve the cross-subject classification performance of motor imagery EEG signals.This paper investigates the construction of small sample and cross-subject motor imagery EEG classification models,providing insights for the design and application of brain-computer interface systems.The research results have important theoretical and practical value in the fields of medical rehabilitation and treatment,entertainment,military,and more.