針對運動想像腦機介面之特徵分析

Feature Analysis for Motor Imagery-Based Brain-Computer Interface

鍾文祥(Wun-Siang Jhong)

指導教授 ： 劉益宏

中原大學/工學院/機械工程研究所/碩士(2011年)

https://doi.org/10.6840/cycu201100498

摘要

本論文以設計互動式腦控開關之大腦人機介面的前提下，觀察放鬆狀態與運動想像狀態下的腦電波變化，尋找能有效提升正確率以及偵測率，並且偽正率也會下降的特徵組合。本研究也設計出運動想像之訓練面板，可以即時的偵測給定的想像任務受測者是否有成功的執行，其中想像的時間長度為2秒，以放鬆狀態與想像狀態其腦波的差異進行偵測。選擇大腦皮質區的運動感覺區上的C3與C4頻道為訊號量測源，利用想像前後左右腦的μ節律與β波會產生不對稱性，針對Graz大學所提供的腦波資料分析自迴規模型(Autoregressive Model，AR)、功率頻譜熵值(Power Spectral Entropy，PSE)、小波包熵值(Wavelet Packet Entropy，WPE)、相位鎖定值(Phase Locking Value，PLV)與共同空間型態(Common Spatial Pattern，CSP)等特徵組合與事件相關移動平均來增加事件相關非同步(Event-Related Desynchronization，ERD)與事件相關同步化(Event-Related Synchronization，ERS)的特性，再加入主成份分析(Principal Component Analysis，PCA)與線性鑑別分析(Linear Discriminant Analysis，LDA)兩種特徵抽取方法，並且提出機器學習方法中的支持向量機器(Support Vector Machine，SVM)取代以往需要經過經驗法則來調整的閥值偵測。將Graz資料的分類結果中較好的特徵組合搭配拿來分析本研究室的腦波資料，觀察Graz與本實驗所收集到的腦波的分類結果，交叉驗正所提出的特徵搭配與訓練運動想像實驗可以被應用及使用的。本實驗共有5位受測者。結果顯示特徵AR搭配PLV、PSW與WPE後，使用3筆資料(epoch)的事件相關移動平均再經過LDA轉換後會有較好的效果，其平均的偽正率(FPR)為2.86%、偵測率為96.79%、正確率為96.13%。

關鍵字

支持向量機器 ； 特徵組合 ； 腦電波 ； 事件相關移動平均 ； 運動想像

並列摘要

This paper is for observation of the change in EEG (Electroencephalogram) under relaxed state and imaginary action state with the proposition of designing interactive brain control switch brain-computer interface (BCI) for brain to seek for combination of characteristics of effectively promoting accurate rate and detecting rate and reducing false positive rate (FPR) at the same time. In this study, we also designed the training panel for motor imagery for instantly detecting whether the tested object has successfully execute the given motor imaginary task, where the time length for imagination is 2 seconds, and it is detected with the difference in brain wave under relaxed state and imaginary state. Channels C3 and C4 of motor area of Cerebral Cortex are selected as signal measuring source. The asymmetry μ rhythm and β wave of left and right brain is used in EEG analysis of Graz University, the Autoregressive Model (AR), Power Spectral Entropy (PSE), Wavelet Packet Entropy (WPE), Phase Locking Value (PLV) and Common Spatial Pattern (CSP) combined characteristic and event-related moving average to increase Event-Related Desynchronization (ERD) and Event-Related Synchronization (ERS) features and add Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA) characteristics extraction method and proposed the Support Vector Machine (SVM) of machine learning method to replace the threshold detection adjustment with thumb rule. The better characteristic combination of the Graz data categorization result to analyze the EEG data of this study to check the results of categorization collected in this experiment and of Graz to cross verify if the characteristics combination and training motor imagery experiment can be applied and used or not. In this experiment, we have 5 subjects. The results indicated that characteristic AR combined with PLV, PSW and WPE, use related moving average of 3 epochs and converted with LDA will have better effect. The average FPR is 2.68% and Detecting Rate is 96.79% and Accurate Rate is 96.13%.

並列關鍵字

Characteristic Combination ； Motor Imagery ； SVM ； EEG ； Event-Related Moving Average

參考文獻

[1] C. Andrew and G. Pfurtscheller, “On the Existence of difference Alpha Band in The Hand Area of Man,” Neuroscience Letters, Vol.222, no.2, pp.103-106, Jan 1997.

[2] G. Edlinger, P. Wach, and G. Pfurtscheller, “On the Realization of an Analytic High-Resolution EEG,” IEEE Transactions on Biomedical Engineering, Vol.45, NO.6, pp.736-745, June 1998

[3] J. Duchen and T. Y. Hohrel, “A EMG generation,” IEEE Transactions on Biomedical Engineering, Vol.47, no.2, pp.192-201, Feb 2000.

[5] J. R. Wolpaw, N. Birdaumer, and W. J. Heetderks, ”Brain computer interface technology：Review of the first international meeting.” IEEE Transactions on Rehabilitation Engineering, Vol.8, No.2 pp.164-173, June 2000.

[7] B. Mensh, J. Werfel, and S. Seung, “BCI Competition 2003—Data set Ia: Combining gamma-band power with slow cortical potentials to improve single-trial classification of Electroencephalographic Signals,” IEEE Transactions on Biomedical Engineering, vol. 51, pp. 1052–1056, June 2004