Study On Algorithms Of Feature Extraction And Classification For Motor-Imagery-Based Brain-Computer Interface

Brain-computer interface (BCI) is a new kind of human-computer interaction system. BCI translates brain signals reflecting user intentions into commad and drives output devices. BCI has a wide range of applications in medical, military, transportation and other fields. Among them, the motor-imagery-based BCI become a research hotspot recently. The key to improve the reliability and performance of this type of BCI is the pattern recognition of EEG, including preprocessing, feature extraction and classification. The emphasis of this paper is study on algorithms of feature extraction and classification for the event-related desynchronization/synchronization (ERD/ERS) and the movement-related potentials (MRP), our research results and main contribution include:(1) In this dissertation, we prove the inconsistency of DSP filtering with feature extraction and classification. The goal of discriminative spatial pattern (DSP) filtering is to maximize the separability of different class samples by adjusting the distance between each pair of time series. However, after spatial filtering, DSP uses the average value of each pair of time series. Therefore, feature extraction of DSP is not exactly consistent with DSP filtering. In addition, DSP filtering and subsequent classifiers, such as support vector machines, are optimized toward different goals. These two drawbacks may degrade overall classification performance.(2) We propose a unified framework based on a logistic regression model for the feature extraction and classification of MRP. This framework directly extracts the distance between each pair of time series as a feature, and unifies spatial filtering and classification under a regularized empirical risk minimization problem. This framework is the extension of DSP algorithm. In this framework, the feature extraction, spatial filter and classifier are optimized toward the same criteria. Experiment results show that spatial filtering in our framework can extract spatial and temporal information from MRP more exactly, and improve classification accuracy and robustness of BCI system.(3) We propose a unified framework based on a linear ridge regression model for the feature extraction and classification of ERD. We investigate the nonlinear correlation between channels with kernel technique, and propose a unified prediction framework based on linear ridge regression model. And this framework integrates preprocessing, feature extraction and classification, automatically selects the time windows, frequency bands and regularization parameter by minimizing leave-one-out crossvalidation error through gradient descent. Experiment results show that the unified framework can extract more physiological meaningful information with nonlinear spatial correlation and improve classification accuracy.