Motor imagery classification based on deep learning

• Main Author: Geng, Zhiheng
• Other Authors: Mao Kezhi
• Format: Thesis-Master by Coursework
• Language: English
• Published: Nanyang Technological University 2023
• Subjects: Engineering::Computer science and engineering::Computing methodologies::Image processing and computer vision; Engineering::Computer science and engineering::Computing methodologies::Pattern recognition
• Online Access: https://hdl.handle.net/10356/163993
• Institution: Nanyang Technological University

The Brain Computer Interface (BCI) is a device that captures Electroencephalograms (EEG) from human beings. By using brain wave signals acquired from brain-computer interfaces to control devices, direct communication between the human brain and physical platforms (such as wheelchairs) can be built. This is an emerging, promising, and valuable technology. One of the most promising areas of BCI research is the control of physical devices (vehicles, wheelchairs, prosthetics, etc.) based on recorded motor imagery (MI) signals. In recent years, the end-to-end deep learning model is more suitable to replace manual feature extraction to complete the task of feature extraction and classification. The research direction of this dissertation is to complete the pattern recognition and classification of signals employing convolutional neural networks (CNN). This dissertation aims to evaluate three deep learning models for the multi-classification of MI signals based on CNN, namely 2D CNN, 1D CNN, and TCN. These three models show advanced performance on two datasets. In addition, tuning the hyperparameters of the model and the overall model architecture to evaluate the effectiveness of the hyperparameters and model architecture. Make personal summaries and opinions on the CNN model based on MI classification.