Classification And Parameter Estimation Of Eeg Signals Based On Deep Learning

Electroencephalogram(EEG)signals contain abundant physiological and pathological information.EEG signals can not only reflect the cognitive function of brain nervous system,but also the physiological state of the human body.Therefore,combined with appropriate signal processing methods,EEG signals play an important role in the brain science research and human disease diagnosis.On one hand,EEG signals can be used in Brain-Computer Interface systems which can be applied in medical rehabilitation,education,entertainment and other fields;on the other hand,by analyzing some important components of EEG signals,doctors can realize the diagnosis of diseases.At present,EEG signal processing technology has made great progress in these two applications,but there are still some problems worth studying.For example,the accuracy and real-time performance of EEG classification in Brain-Computer Interface systems need to be further improved;the capabilities of EEG denoising and analysis need to be further enhanced and so on.At the same time,with the rise of artificial intelligence,deep learning has become one of the important tools for biomedical signal processing.Aiming at the existing problems in EEG analysis and processing,this thesis systematically studied the classification of EEG signals based on deep learning in P300 spelling Brain-Computer Interface and motor imagery Brain-Computer Interface system,and the new method for dynamic estimation of P100 latency of single-channel visual evoked potential based on deep learning.(1)To improve the recognition rate of characters in the P300 spelling Brain-Computer Interface system,a method which is based on the improved convolutional neural network is proposed.By transforming the second serially connected convolutional layer in a traditional convolutional neural network to three parallel connected convolutional layers,the method widens the network to enhance the ability of feature extraction in the proposed network.Combining the extracted features with the fully connected layers and sigmoid function,a P300 event-related potential classifier is constructed.Aiming at the problem of unbalanced data volume between target and non-target stimulus data in Brain-Computer Interface Competition data set,this thesis partially averages the EEG data which contains P300 event-related potential to increase the amount of EEG data.The experimental results show that this method can effectively recognize P300 event-related potential and improve the accuracy of character recognition.The accuracy of character recognition is not less than 90% when the number of experiments is more than 9 times.(2)Aiming at the classification problem of motor imagery EEG signals,this thesis uses Wavelet Packet Transform(WPT)and Short Time Fourier Transform(STFT)in the time-frequency analysis to extract the features of the preprocessed motor imagery EEG signals according to the difference in EEG signal energy among different channels.Both kinds of features are combined and applied to achieve the classification for left and right hand motor imagery EEG signals by using a convolutional neural network.The experimental results show that the convolutional neural network based on multi-feature combination is beneficial to the classification of EEG signals and improves the classification accuracy to some extent.(3)Aiming at the difficulty in estimating latency of single-channel evoked potential,a new method of dynamic estimation and tracking P100 latency of single-channel visual evoked potential based on deep learning is proposed.In the research experiment,taking individual visual evoked potential as an example,this thesis constructs visual evoked potential signal sequence under multiple stimulus and use convolutional neural network in deep learning to dynamically estimate and track P100 latency of single-channel visual evoked potential.This method can reflect the dynamic changes in the latency of visual evoked potential under each stimulus.This method can provide an idea and method for the study of clinical medicine and brain science.