The Research Of Electroencephalogram Recognition Based On Deep Belief Net And Its Application In Intelligent Wheelchair

The Brain Computer Interface(BCI)achieves direct information exchange between the human brain and external devices.It decodes the user's intent into control commands to manipulate the device,which is a new type of data interaction.With the rapid development of brain science,BCI technology has become a research focus in the field of assisted medicine.The key to BCI system research is how to efficiently and accurately extract the characteristics of the brain's thinking activity signal and translate it into output instructions that can control external devices.In this paper,the problem that the feature extraction of EEG signal is difficult and the recognition rate of it is not high enough,we make a deeply research which has very important theoretical significance and practical value.First of all,in the process of EEG feature extraction.To solve the problem of long training time and over-fitting of small sample EEG signal processing,this paper proposed a DBN based on random retreat algorithm,which can classify and identify the left and right hand motion imaginary EEG signals.Firstly,the original EEG data are processed by dimension reduction,and use the random DBN model to train the reduced EEG data,then get the optimal parameter values for classification and recognition.The experimental results show that the DBN algorithm based on random retreat can maintain the high recognition rate and reduces the training time,compared with CSP,PCA and single DBN network.It proves the effectiveness of the method.Secondly,inspired by the individual differences in migration learning,in order to improve the recognition rate of EEG signals,the Multi-bands FDBN algorithm is proposed on the basis of DBN and its related improved algorithms.Due to the individual differences in different bands,the contribution to the classification results is not exactly the same.In this paper,the band-pass filter is used to divide the original EEG signal into multiple frequency bands,and then the FFT is used to convert the time domain signal into frequency domain signal and normalized.Finally,the frequency domain data of each frequency band is input into DBN for training and identification.Compared to FDBN,the accuracy of Multi-bands FDBN increases by 3.3% in average and the variance is smaller and has better robustness.Finally,to verify the validity of the improved algorithms,the system of intelligent wheelchair human-computer interaction based on Motor Imagery Brain Computer Interface(MI-BCI)is designed and realized.The experimental results show that the a DBN based on random retreat algorithm and the Multi-bands FDBN algorithm are used to extract and classify the motor imagery Electroencephalogram respectively,the experiments of EEG-controlled wheelchair movement can be well completed.This paper compares the difference between a DBN based on random retreat and Multi-bands FDBN by the experiment controlling intelligent wheelchair of a fixed trajectory with“8”glyph,and the results demonstrate that the two improved algorithms are effective in the BCI system.