本論文以開發非同步(Asynchronous)腦控開關(Brain-Controlled Switch)之腦機介面(Brain Computer Interface, BCI)為主軸,而非同步腦機介面系統或稱作Self-paced BCI系統是一種能用於現實生活的BCI系統。在之前的研究中,已經提出許多方法去改善非同步BCI系統的效能,而本研究也專注於提升非同步BCI的效能,藉由使用不同的特徵抽取方法與分類器方法於兩類別的運動想像分類,希望能找到具有低偽正率(False Positive Rate, FPR)與高偵測率(True Positive Rate, TPR)的方法用於非同步BCI系統中。特徵抽取方法包括頻帶功率(Band Power, BP)、自迴歸模型(Autoregressive Model, ARM)、共同空間模式(Common Spatial Pattern, CSP)和CSP與線性鑑別分析(Linear Discriminant Analysis, LDA)和主成份分析(Principle Component Analysis, PCA)特徵抽取串聯方法,而分類器有K個最近鄰居法(K-nearest neighbor, K-NN)、支持向量機器(Support Vector Machine, SVM)、非平衡式支持向量機器(Imbalanced Support Vector Machine, ISVM)、支持向量資料描述(Support Vector Data Description, SVDD)、線性鑑別分析(Linear Discriminant Analysis, LDA)、二次鑑別分析(Quadratic Discriminant Analysis, QDA)與馬氏鑑別分析(Mahalanobis Discriminant Analysis, MDA)。本實驗共有4位受測者,實驗結果顯示CSP特徵抽取方法配合ISVM分類器方法具有低偽正率且偵測率有一定的程度,可用於非同步BCI系統,其最好的離線分析分類結果偽正率為3.33%、偵測率為67.78%、正確率為82.22%。本論文也發展了一套非同步的即時腦控開關之腦機介面,藉由離線分析得到的方法套用於即時腦控開關系統中使用,此套即時腦控開關系統能即時判斷出 受測者當下的腦波類別,未來能用於現實生活中使用。
This study is based on the development of an Asynchronous Brain-Controlled Switch Brain Computer Interface, as to Asynchronous Brain Computer Interface, or so called Self-paced BCI system, is a BCI system that can be used in real live. In early studies, many methods had been brought out to increase the performance of asynchronous BCI system, and this study also focused on improving asynchronous BCI system’s performance, by using various means of feature extraction and classification methods on two classes of motor imagery classification, hoping to explore a method that can be used in asynchronous BCI system with Low Positive Rate(FPR) and high True Positive Rate(TPR). Feature extraction methods include Band Power(BP)、Autoregressive Model(ARM)、Common Spatial Pattern (CSP) and the integration us of Common Spatial Pattern、Linear Discriminant Analysis(LDA) and Principle Component Analysis(PCA), classifiers we used K-nearest neighbor(K-NN)、Support Vector Machine(SVM)、Imbalanced Support Vector Machine(ISVM)、Support Vector Data Description(SVDD)、Linear Discriminant Analysis(LDA)、Quadratic Discriminant Analysis(QDA) and Mahalanobis Discriminant Analysis(MDA). There are 4 subjects participating in this study, experiment results showed that CSP feature extraction method coordinating with ISVM classifier has low FPR with TPR at a level, which can be used in asynchronous BCI system, its best result is FPR 3.33%、TPR 67.78% and an accuracy of 82.22%.This paper also developed an asynchronous real-time brain-controlled switch BCI, by offline analysis we obtained a method, which was applied to the real-time brain-controlled switch system, this real-time brain-controlled switch system can judge the subject's EEG class at a certain moment, and can be used in real life in the future.