Research And System Implementation Of Brain Computer Interface Combined With Deep Learning

With the development of science and technology,more and more people begin to pay attention to the field of brain-computer interface technology.Brain-computer interface(BCI)establishes the connection between brain and external equipment through computers,and it realizes the information exchange between brain and external environment.The research of brain-computer interface not only realizes the desire of the disabled to re-understand the world,but also provides the possibility for science and technology to promote the development of life.Neural networks and deep learning methods have been gradually applied to the recognition of EEG signals with the wide application of deep learning in recent years.In this thesis,the deep learning method is introduced to study the preprocessing of various EEG signals and to research the classification and recognition model,which has theoretical significance and research value.Firstly,the research status of brain-computer interface(BCI)technology based on multi-type EEG signals at home and abroad is described.The key technologies of brain-computer interface are briefly introduced.The traditional brain-computer interface system and EEG signal processing methods are introduced.The principle and advantages of deep learning is analyzed.And then designed a comprehensive solution of brain-computer interface system combined with deep learning.The acquisition of EEG signals is completed.Secondly,considering the problem of low recognition rate caused by both the large number of artifacts in EEG signals and the lose of effective EEG signals when using the traditional EEG signal processing methods,a deep contractive sparse auto encoder method based on fuzzy reasoning is proposed to remove EEG artifacts.This method improves the deep contractive sparse auto encoder by using the fuzzy reasoning algorithm,and then it was applied to remove artifacts of EEG signals.The experimental results show that the improved method can effectively improve the effect of the EEG artifact removal.Thirdly,considering the spatio and temporal characteristics of EEG signals,convolution based on bidirectional long-and short-term algorithms is proposed in order to solve the problem of the low recognition rate and the poor robustness when researching multi-class EEG signals by using some traditional classification models.CNN is used to deal with the spatial frequency characteristics of EEG signals,while BLSTM processes the temporal correlation of EEG signals,and because the deep network models are difficult to train,a residual network is added to CNN.The experimental results show that the EEG recognition model proposed in this thesis is superior to the traditional classification models such as SVM and CNN,and the EEG signal recognition performance is effectively improved.Finally,offline and online test verifications are carried out respectively by using the brain-computer interface system proposed in this thesis,and finally the system is applied to the control of intelligent service robots.The experimental results show that the recognition rate is improved,the robustness is better and the generalization ability is stronger than the traditional BCI systems.