Research On Brain Computer Interface Based On Deep Learning Algorithm

Brain-computer interface has created a unique communication system that achieves the direct interaction between individual intentions and external equipment without depending on peripheral nerves and muscle tissues.This new type of human-computer interaction can effectively enhance the communication ability between the disabled and the outside world,and raise patient life quality.The motor imagery based brain-computer interface system to control the external equipment by processing and analyzing the collected neuronal activity signals and then translating those brain signals into user commands.Because of its practical medical significance,more and more researchers have paid attention to the study of this unique communication mode,and it has become a hotspot in the study of artificial intelligence,rehabilitation medicine and computer science,etc.The application of brain-computer interface system technology depends on the accuracy of brain signal recognition.The performance of feature extraction and classification of brain signal has in the face of the challenge because of brain signal is easy to be interfered by noises.Meanwhile,there are many problems such as channel selection relying on experience,time-consuming feature engineering,and incomplete feature extraction in brain signal processing.Therefore,how to effectively extract the information that can represent the brain activity characteristics and how to effectively train a classifier with good generalization ability are the major research content of motor imagery based brain-computer interface systemThis thesis is based on the research of the algorithms applied in feature extraction and classification stages,and the deep learning algorithm that can learn is employed to process brain signals.And we designed the experiment of brain signals acquisition to understand and analyze the whole process of the brain-computer interface system all-sided and detailedly.1.We build a motor imagery classification model based on convolutional neural network framework.According to their characteristics in the spatial domain,this model can represent brain signals.Experimental results and analysis show that the algorithm has better classification results and lower complexity.2.We build a motor imagery classification model based on long short-term memory-fully connected framework.This model can neatly combines the fully connected layer network and long short-term memory network to capture the representation of brain signals from the breakthrough point in the temporal domain.Experimental results and analysis show that the algorithm has high classification results,good robustness and strong generalization.3.We designs the electroencephalogram signals acquisition experiment and researchthe the application of BCI system.Meanwhile,more representative,interpretable and differentiated features in the brain signals are captured by the convolutional neural network-long short-term memory framework that combines convolutional neural network framework and long short-term memory-fully connected framework.The model researched in this thesis has achieved good results in a number of experimental data as well as can be used as a general brain signals processing method,and lay the algorithm foundation for a real-time brain-computer interface system.