Recognizing EEG signals for brain-computer interface based on machine learning

• Main Author: Liu, Chang
• Other Authors: Jiang Xudong
• Format: Theses and Dissertations
• Language: English
• Published: 2019
• Subjects: DRNTU::Engineering::Computer science and engineering::Computing methodologies::Pattern recognition
• Online Access: http://hdl.handle.net/10356/78407
• Institution: Nanyang Technological University

In recent years, the Brain-Computer Interface (BCI), has been a very popular topic globally. BCI is a method that can establish a direct communication pathway between the human’s brain and external devices by analyzing the EEG (Electroencephalograph) signals, without any help from peripheral nerves and muscles. Thus, it can be used to help patients whose motor ability is highly damaged. Meanwhile, Machine Learning is also a hot research area that has been adopted in many other fields, including BCI. In this dissertation, the EEG signal processing in the Brain-Computer Interface-assisted Motor Imagery (MI-BCI) system based on machine learning is mainly studied. The main contributions include the following aspects: 1. The structure of typical BCI systems is reviewed. In this dissertation, a BCI system is divided into 5 parts: capturing, feature extracting, classifying and outputting. A survey of algorithms in two key parts, feature extracting and classifying, is done, and several representing algorithms, like CSP, LDA and SVM are introduced in detail. 2. In the dissertation, we proposed a novel feature extracting approach: CSP in Cells. Using this approach, the accuracy of classifying is significantly increased comparing to the original CSP algorithm. 3. Further, an experimental and comparative test is done on three datasets. Based on the discussion of the result we concluded some difficulties and bottlenecks for the current BCI study and give some suggestions.