應用單類別分類器於非同 步腦機介面之想像運動研究

Application of One-Class Classifier for Asynchronous BCI – Motor Imagery Study

施孟雄(Meng-Xiong Shi)

指導教授 ： 劉益宏

中原大學/工學院/機械工程研究所/碩士(2010年)

https://doi.org/10.6840/cycu201000939

摘要

本研究旨在設計出與真實世界有著相同功能的大腦人機介面－互動式腦控開關，其中透過時間與頻率圖觀察出在放鬆狀態與想像狀態腦電波（Electroencephalogram，EEG）變化，藉由觀察時間與頻率圖找到顯著的特性，也就是在想像動作前後所μ節律與β波所產生的事件相關非同步化（Event-Related Desynchronization，ERD）與事件相關同步化（Event-Related Synchronization，ERS）特性。本研究嘗試利用2秒放鬆狀態搭配2秒想像狀態，兩種不同狀態來控制開關，其中選擇量測腦電波訊號位於大腦皮質層之運動感覺區（C3與C4 頻道），並使用離散傅立葉轉換至頻率域上抽取出兩種代表性特徵。本論文提出了ㄧ種功能強大的機器學習方法，稱為支持向量資料描述(Support Vector Data Description，SVDD)，取代以往透過簡單閥值（Simple Threshold）偵測以及需依靠經驗法則來調整受測者閥值。支持向量資料描述能夠判斷所量測之腦波屬於放鬆狀態或想像狀態。為了使支持向量資料描述具有自我學習的能力，我們也設計了在實驗之前的校準階段，其目的在於收集一組放鬆狀態的腦波去訓練此分類器。實驗結果顯示，本論文所提出的實驗設計對於未經過任何相關腦波訓練之受測者。Subject 1錯誤率低於4.75%、主動慫恿次數低於4.12次、響應時間平均16.75秒。Subject 2錯誤率低於8.25%、主動慫恿次數低於7.9次、響應時間平均31.6秒。Subject 3錯誤率低於4.25%、主動慫恿次數低於8.97次、響應時間平均35.9秒。已具初步實用價值。

關鍵字

ERD ； ERS ； 腦電波 ； 離散傅立葉轉換 ； 支持向量資料描述

並列摘要

The purpose of this study is to design a Brain-Computer Interface, Interactive Touch Switch, which possesses the same functions as the one in the actual world. The method is observing the variation of EEG (Electroencephalogram) under the relaxing and thinking statement via the time and frequent chart, then finding the prominent features which creates ERD (Event-Related Desynchronization), and ERS (Event-Related Synchronization) of μ rhythm and β wave. This study tries to use two different statements to control the switch, one is relaxing statement, the other is thinking statement, and both are for two seconds. It detects the Electroencephalogram signals which locate on the motor area of cerebral cortex (Channel C3 and C4). Then, it uses Discrete-time Fourier Transform (DTFT) to extract two representative features on frequency domain. This study brings up a functional mechanical leaning method which is SVDD (Support Vector Data Description). It replaces the Simple Threshold detection and Tester Threshold which needs to follow the experience.The SVDD can judge whether the measured EEG is under relaxing or thinking state. In order to make SVDD possess the ability of self-learning, we design a adjustment stage before experiment. The purpose is to collect one EEG under a relaxing statement and do the training on the Sorting device. The result of the experiment: Subject 1- False positive rate is lower than 4.75%, the active urging time is lower than 4.12 times, and the response time is averagely 16.75 seconds. Subject 2- False positive rate is lower than 8.25%, the active urging time is lower than 7.9 times, and the response time is averagely 31.6 seconds.Subject3- False positive rate is lower than 4.25%, the active urging time is lower than 8.97 times, and the response time is averagely 35.9 seconds.It shows that the experimental design of this study has already had the practical values initially.

並列關鍵字

DTFT ； SVDD ； ERS ； ERD ； EEG

參考文獻

BCI controls functional electrical stimulation in a tetraplegic patient,” EURASIP J. App.

“Brain–computer interfaces for communication and control,” Clin Neurophysiol, vol. 113,

pp. 767–791. 2002.

graz brain–computer interface: methods and applications, ” Comput Intell Neurosci, 2007.

self-paced brain–computer communication: navigation through virtual worlds, ” IEEE