Year: 2026
Type: dissertation
Abstract: In this thesis, a brain-computer interface (BCI) wheelchair control system to support the autonomy of individuals with motor disabilities, and a BCI system to understand the emotions of patients such as those with dementia, are developed.First, regarding the BCI wheelchair control system for users with motor disabilities, the world's first "brain-controlled wheelchair" was developed in 2008; however, it still had the problem of imposing strong physical constraints on users.This study aims to realize a system with higher freedom, improved safety, and better operability.Specifically, (1) a wireless command converter was designed and prototyped to improve data transmission performance through real-environment experiments and to overcome the limitations of wired connections.(2) While maintaining the accuracy of signal acquisition, feature extraction, and classification, the gel-based electrode cap was replaced with a lightweight EEG device.(3) A new stimulator and graphical user interface (GUI) were designed and developed, with an additional safety mechanism enabling immediate stop in case of misclassification.As a result, system operation experiments involving five participants achieved a classification accuracy of 92%, confirming the effectiveness of the proposed system.Next, in emotion recognition for elderly people such as dementia patients, traditional facial expression recognition technology is difficult to apply.Therefore, this study adopts an EEG-based emotion recognition approach.However, due to the insufficient availability of labeled emotional data and the diversity of emotional states, identifying all emotions is not practical.Accordingly, a practical method focusing on specific emotions is proposed.Specifically, (1) emotion is formulated as a binary classification problem of "positive / non-positive".( 2) Positive and Unlabeled (PU) learning is introduced to construct a classifier from a small number of positive labels and a large amount of unlabeled data.(3) Public datasets and in-house collected data are integrated, and differential entropy (DE) is adopted as the feature.The experimental results show that the classification accuracy reaches 93.95%, confirming the effectiveness of the proposed method.Furthermore, considering the practical needs related to emotion recognition for elderly individuals in nursing care facilities, this study aims to implement a real-time system.Specifically, (1) a classifier based on PU learning is constructed.(2) A lightweight EEG device suitable for daily use is employed.(3) A personalized adaptive model is retrained to optimize performance for each participant.As a result of system operation, the real-time classification accuracy achieved 86.20%, indicating the potential of the proposed system toward practical application.
Language: English
Subfield: Cognitive Neuroscience
Field: Neuroscience
Domain: Life Sciences
Open Access status: green
OpenAlex ID: W7208080410
