The Research And Implementation Of Online Brain-Computer Interface System Based On Independent Component Analysis

Brain-Computer interface (BCI) is a new type of Human-Computer interaction technique. By establishing channels between the human brain and external electronic devices, BCI can convert the neural activity of human brain into control commands to complete a predetermined operation. In the fields such as medical rehabilitation and amusement games, etc., BCI has a very broad application prospects. However, there are still some crucial problems related to BCI system implementation need to be solved, such as the slower response speed of system, the lower recognition accuracy and so on. Thus, the researches of efficient EEG processing algorithms have significant meaning in building online BCI system.In the research of non-invasive BCI, independent component analysis (ICA) has been considered as a promising method for electroencephalogram (EEG) preprocessing and feature enhancement. However, so far, the most researches about ICA-BCI system are offline analysis based on Matlab platform.This thesis investigated the ICA-based motor imagery BCI (MIBCI) system, combining the unsupervised learning characteristics of ICA and Event-Related Desynchronization (ERD) effect induced by motor imageries, a simple and practical calculation method of ICA-based= spatial--filter-and-the-discriminate-criterion of three-class motor imageries were constructed. On this basis, the online ICA-MIBCI experimental system were implemented based on NeuroScan EEG amplifier and VC++ platform. The main contents of the dissertation are as following:1. The Event-Related Desynchronization/Synchronization (ERD/ERS) phenomena related to motor imagery were analyzed in time/frequency/spatial domain. The typical MI-EEG feature extraction and pattern classification algorithms were described.2. The basic model of ICA and its feasibility for EEG processing were introduced, some adjustments on extended infomax ICA algorithm were achieved by C++ programming language.3. A simple and practical method of ICA spatial filter design and discriminant criterion of three-class motor imageries were implemented. On this basis, an online synchronize ICA-IMBCI system were achieved completely.4. The EEG acquisition paradigm were designed independently, Some parameter optimization problems, such as, the frequency band selection of feature extraction and the ICA electrodes selection were discussed by offline analysis. The influence of the quality of training samples on the ICA filter performance and the influence factors of cross-subject BCI system performance were analyzed. Experimental results showed that the average online classification accuracies of three-class motor imageries for the well-trained subjects could achieve more than 80%. The online ICA-MIBCI experimental system was implemented integrally based on NeuroScan EEG amplifier and VC++ platform, the complexity in time and space of the proposed core algorithms is low, which has the potential to be ported to wearable and mobile platform.