| Brain-Computer Interface(BCI)is a new type to decode the electroencephalogram(EEG)signals generated by the brain’s thinking activity through the use of computers or other devices to achieve the purpose of communication with the external environment.Therefore,as one of thethe most promising research paradigm in brain-computer interface systems,how to achieve reliable decoding of intent recognition brain-computer interaction control with limited number of training samples is an urgent research problem.In this paper,the following main work is carried out to address the issues of intent recognition and data augmentation of EEG signals in brain-computer interface systems.(1)This paper proposes a new experimental task based on dynamic complex object control—— "bowl-ball" system,and conducts a study on the intention recognition of the experimental paradigm of motion perception by acquiring the obtained EEG signals,aiming to achieve precise control of complex boundary avoidance tasks through EEG signals.(2)Addressing the problem of EEG sample size,how to achieve brain-computer interaction control and improve decoding performance is a key issue to be addressed.One of the effective methods to solve the issues of data calibration and sample size is applying different subjects’ EEG data to new subjects by using data augmentation methods during the execution of the relevant control task.In the data augmentation study,a data augmentation method for cross-subject EEG features is constructed by combining the two methods based on the fast fitting of the broad learning system to the EEG feature model and the similarity measure of the feature space by the twin neural network.The broad learning system(BLS)and Siamese neural network are combined to solve the problem of feature space similarity of different subjects’ EEG data,and the set of features of the two subjects with the highest similarity is used to generate a new EEG data feature set to achieve the purpose of enhancing the original EEG data.The experimental results in various machine learning classification methods show that this method can effectively improve the classification accuracy of intent recognition EEG data.(3)How to achieve fast and reliable decoding of EEG signals has become a key issue in the study of intent recognition research in brain-computer interface system.Therefore,a new Transformer-BLS network combining BLS and neural network is proposed to be applied to the study of intent recognition of EEG signals.By obtaining the weight models of EEG signals in the Transformer-BLS network,the intent of the original EEG signals can be successfully identified.Then the recognition effects were compared with our proposed model using various research methods.The experimental results show that the classification results using the Transformer-BLS algorithm outperform many existing classification methods,both in terms of accuracy and in terms of training time and computational resources.Overall,the results of this study can help to better analyze the potential intent of patients by EEG signals and provide research support for practical applications such as intent recognition and data enhancement. |
