Research And Application Of Manifold Learning For Feature Extraction In EEG

In response to an external stimulus or active thinking activity, nerve cells in the brain cortex can produce Motor Imagery Electroencephalography(MI-EEG) signals with the characteristics of specificity and rhythmicity, and they not only contain a large number of physiological or disease information but also have a close correlation with the state of consciousness. As a result, much attention has been paid to the application of MI-EEG in brain cognition; meanwhile, the correct interpretation and accurate extraction of MI-EEG features is the key to its successful applications.With the tremendous development and wide application of information technology, the quantity and species of data in numerous fields have increased at a speed exceeding the imagination. The curse of dimensionality restricts many machine learning algorithms from achieving more applications in many research areas including the analysis and feature extraction of Motor Imagery Electroencephalography(MI-EEG). Many approaches have been proposed for dimensionality reduction. Unfortunately, most of their nature are linear, definitely resulting in information losing in non-linear real-world- data processing. Manifold learning emerges at the right moment. As a nonlinear data dimension reduction method, manifold learning has been widely used in diverse fields such as data visualization, information retrieval, image processing, pattern recognition and data compression, and so on. In order to perform data mining for MI-EEG from the perspective of intrinsic geometric distribution in data sets with remarkable time-frequency characteristics and neural physiological characteristics, manifold learning is applied in feature extraction of MI-EEG and subsequently in BCI system with application significance. Specifically speaking, the following works are aehieved in this thesis:1、A novel feature extraction method is proposed based on Locally Linear Embedding algorithm(LLE) and Discrete Wavelet Transform(DWT), which denotes as DWT-LLE. The neurophysiological characteristics of MI-EEG are analyzed, and the valid time and frequency segments are determined by Wigner-Ville distribution and power spectrum; Then MI-EEG are processed by using DWT multi-scale analysis, and the statistics of detail components are calculated to obtain the time-frequency features. Furthermore, LLE is applied to the approximation components to get nonlinear features, and they will be fused with time-frequency features in serial. Finally, a support vector machine is used as classifiers, and many experiments are conducted on a publicly available dataset. The classification accuracy and standard deviation of 10-fold cross-validations show that the proposed method has better performance and more stability against conventional DWT-based feature extraction methods. The method keeps the balanced consideration of nonlinearity, time and frequency characteristics, rhythmicity, neural physiological characteristics and simultaneously achieves data dimension reduction.2、In the reseach on manifold learning, we found that the main limitation of traditional manifold learners is that they are incapable of providing explicit mappings between the high-dimensional data space and the low-dimensional latent space, which is non-trivial for the out-of-sample extension of these techniques. Encompass the problem, a new unsupervised nonlinear dimensionality reduction technique termed parametric t-Distributed Stochastic Neighbor Embedding(P. t-SNE) is employed to extract the nonlinear features from MI-EEG. Experiments are conducted on a publicly available dataset, and the experimental results show that the nonlinear features have great visualization performance with obvious clustering distribution, and the feature extraction method indicates excellent classification performance.3、In order to verify the availability and effectiveness in the field of BCI system, a appliance control system based on MI-EEG by using DWT-LLE and a A MI-BCI based online prototype system for rehabilitation of upper limb motor function are disined. In the process of system development, C# was used as a main language for PC-End software. The real-time acquisition and processing of MI-EEG is applicable by hybrid programming with Matlab and C language, and as well as multithreading technology. The control module is designed for mechanical arm based on a microprocessor ‘atmega168’. The two systems introduce manifold learing algorithm to BCI application, especially in feids of daily life application and medical rehabilitation.The research in this work enhance the performance of reliability and real-time, which means a lot to access the nature of MI-EEG from a perspective of nonlinear nature and time-frequency information in the data, helping us understand the relationship beteen EEG and motor imagine consciousness. It provide a novel path to further practical application in classification and visualization.