Font Size: a A A

Pattern Recognition Techniques And Experiment Research For Spontaneous EEG-based Brain Computer Interface

Posted on:2009-11-09Degree:DoctorType:Dissertation
Country:ChinaCandidate:T WuFull Text:PDF
GTID:1118360305956642Subject:Precision instruments and machinery
Abstract/Summary:PDF Full Text Request
The brain is the most complex part of human body and one of the most complicated structures of the universe, the research on the brain has become a hotspot problem in scientific fields, and it is also a rather challenging problem due to the diversity of nerve relations. There are many kinds of methods that applied to the research on human brain. Electroencephalography (EEG) is a common used technique for externally recording the diversified state of human brain, and the brain computer interface based on EEG is also become a creative application in brain science.Brain-computer interface (BCI) is an alternative and a novel interface between human and computers. BCIs can provide their users communication and control channels that do not depend on the brain's normal output channels of peripheral nerves and muscles. It is an important approach for understanding the human brain and improving its function. Current interests in BCI development mainly derive from rehabilitation engineering, assistant control for normal people, entertainment, brain cognition, and etc. The essence of a BCI is to deduce human thoughts or intentions via EEG signal and so to realize the communication between human and computers. The system of BCIs consists of three main modules, collecting and recording EEG, data processing, peripheral equipment and interface. The pattern recognition of spontaneous EEG mainly includes three steps: the feature extraction, the feature selection and the classification. To improve the recognition accuracy of BCIs system, this paper focuses on the study of spontaneous EEG signal processing and recoginition methods, including a system of movement control based on spontaneous EEG. The main topics studied in this thesis are as follows:Research on the algorithms of EEG feature extraction. Considering EEG signal is a typical non-static signal, the thesis presents the following three feature extraction methods. 1) the feature extraction based on wavelet packet decomposition coefficients combining with sub-space energies, 2) the feature extraction based on EMD and Hilbert transform, 3) the feather extraction for two kinds of thoughts based on threshold value. The proposed methods are compared with other existing feature extraction methods such as the feature extraction based on the autoregressive model, the feature extraction based on wavelet transform coefficients, the computational results demonstrate that the proposed methods can obtain the optimal features, and are effective for further feature selection and classification.Research on the algorithms of EEG feature selection. The feature selection is to select a feature subset which can provide the best classification performance. In this thesis, a method based on gene optimization is applied to BCIs. The method is compared with other existiong methods such as the feature selection based on the genetic algorithm, the feature selection based on the filter algorithm of Fisher distance. The effectivity can be proved by the results of recognition.Research on the algorithms of EEG feature classification. The classification is to yield the corresponding class label according to the input feature subset. The thesis presents two feature classification methods: 1) the classification based on the probabilistic neural network (PNN) with supervised learning, 2) the classification based on optimal support vector machine (SVM) parameters with gene optimization. The two methods are compared with existing classification methods such as the classification based on experiential SVM model parameters and the classification based on optimal SVM parameters with genetic algorithm. The effectivity of the proposed methods can be verified by the classification results.The system design of brain computer interface which control the movement based on spontaneous EEG signal. The research is concerntrated on building a spontaneous EEG signal collector and a robot controlling organization. We have finished the overall system architecture and the specific module designs which include amplifier network, filter network, electrical isolation system based on wireless communicaton, interface circuit of input or output based on USB and movement control system based on wireless communication. The work can provide a guideline for further research on the expansion of brain computer interface.The researched methods are analyzed theoretically and the effectiveness of these methods is verified through experiments. Firstly, the data of the BCI competition 2005 are analyzed with these methods. Then the experiment data obtained from the designed experiment paradigm are analyzed. The thesis designes an experiment paradigm based on three motor imagery tasks (playing basket using left hand, playing basket using right hand, braking using right foot). The researched pattern recognition methods are compared with other existing methods and the results demonstrate that the researched methods can effectively improve the classification accuracy.The robustness of the researched methods needs to be further verified due to lack of more experimental data. Finally, conclusions of the dissertation are deduced and the future work is discussed.
Keywords/Search Tags:Spontaneous electroencephalography (EEG), Brain computer interface(BCI), wavelet packet transform, Gene optimization algorithm, Support vector machine(SVM), Supervised learning, probabilistic neural network (PNN), Feature extraction, Feature selection
PDF Full Text Request
Related items