Research And Application Of EEG Recognition Base On Deep Learning

In recent years, the global aging of population, has becoming more and more serious, and the elderly are the high incidence of stroke. Hemiplegia or limb motor dysfunction has not only brought great inconvenience and mental pain to patients, but also heavy burden to their family and society. How to help the patients to carry out effective rehabilitation treatment and to regain some independent living ability is an urgent problem to be solved in today's society, and it is also a hot topic in the research of rehabilitation engineering and artificial intelligence.Brain computer interface(BCI) technology provides a good solution to the problem. Medical theory and practice have proved that the brain has plasticity, that is to say, if patients actively participate in rehabilitation training it is conducive to the rehabilitation. BCI is a new active rehabilitation method based on motor imagery, which has obvious advantages for the recovery of limb motor function in patients. In this paper, studies were conducted from the aspects of EEG feature extraction, classifier design, personalized EEG acquisition scheme and the online arm rehabilitation system design. The main achievements are listed as following:(1) EEG feature extraction based on WPT and DBNEEG is very weak, low recognition rate, poor adaptive capacity and other issues. In this paper, DBN is integrated with WPT to yield a novel recognition method, denoted as WPTDBN. Firstly, the MI-EEG is transformed into power signal and analyzed the effective time domain. Then, WPT is applied to each channel of MI-EEG to obtain the effective time-frequency information. Finally, DBN is used for the identification and classification simultaneously. Experiments are conducted on a publicly available dataset, and the 5-fold cross validation experimental results show that WPTDBN yields relatively higher classification accuracies compared to the existing approaches.(2) EEG recognition method based WPT and LSTM neural networkEEG is time-varying and subject-specific, its recognition needs the perfect adaptability and combination of feature extraction method and classifier. EEG is a typical of time sequence signal, and extract the time-frequency feature of EEG remains the timing information. In this paper LSTM is integrated with WPT to yield a novel recognition method, denoted as WLR, WPT was applied to analyze the each EEG channel and extract the effective MI-EEG time-frequency feature. LSTM networks was used as classifier to analyze the observed MI-EEG data. Experiments are conducted on a publicly available dataset, and the 5-fold cross validation experimental results show that WLR yields relatively higher classification accuracies compared to the existing approaches.(3) Experimental design of EEG acquisition based on face recognition technologyIn order to promote the application of MI-BCI rehabilitation research,and to solve the problem that the existed boring EEG signal acquisition scheme which can easy make patients lose initiative of rehabilitation, this paper designs a personalized EEG signal acquisition system based on face recognition. The system used MFC programming interface, Open CV library of camera and image processing, mixed programming of Matlab and C++ for EEG offline analysis processing, face recognition based on Face++ face recognition database, curl network library of network related operations, Json Library of face recognition data for online analysis and g.MOBIlab+ electroencephalograph EEG acquisition device to control. First of all, the system can acquire real-time face images and personalized features through online recognition test of age, gender and the degree of smile, then personalized experimental acquisition mode is set up according to the personalized features. Then, the EEG data of the arm extension / flexion movement were collected for several times. And the WLR method was used to conduct off-line analysis and processing of the collected data.(4) Design of online arm rehabilitation system based on MI-BCI TechnologyOn the basis of the above research, an online arm rehabilitation system was designed. The system is built based on the recognition of personalized EEG acquisition system; then through the MFC and multi thread technology and using g.tec company g.MOBIlab+ acquisition instrument to acquire real-time of EEG data; then, C++ and Matlab mixed programming were used to achieve automatic recognition of online EEG. Finally, through Win32 API technology the classification results were transmitted to the robot arm control system which used 51 single chip microcomputer as the control chip, and then the extension / flexion movement of the robot arm was realized. Experimental results proved the feasibility of the arm rehabilitation system based on MI-BCI. In this way, the initiative of patients can be promoted, which associated with the increasing efficacy. The results show its potential application prospect and value in arm rehabilitation area.