透過您的圖書館登入
IP:18.223.172.184
  • 學位論文

同步腦機介面特徵分析與研究

Analysis of Synchronous BCI Features and Studies

指導教授 : 劉益宏

摘要


近幾年來,隨著電腦運算技術提升和分析處理方法改良,生理訊號相關研究領域逐漸被重視,腦機介面(Brain Computer Interface, BCI)就是當中的新穎技術之一。腦機介面是一種利用腦波(Electroencephalogram, EEG)訊號的變化,讓使用者可以直接與外界溝通。 本論文主要在建立一套同步化系統,分析受測者在經過不同的運動想像手動(想像左手或是右手移動)情境(或是有真實手動)時,大腦當中的運動想像區域(Somatosensory Cortex)之腦波律動訊號的變化,選取當中最有鑑別力的腦波變化頻率區段作為辨識資料,並將選取後的腦波資料做為分類。訊號前處理中,本研究利用了有限頻寬(Finite Impulse Response, FIR)濾波器將環境雜訊和一些生理訊號所去除。本文將經過訊號前處理的腦波資料利用共同空間型態(Common Spatial Pattern, CSP)的方法作線性轉換,並且將這些線性轉換後的腦波資料分別作時間域(Time Domain)和頻率域(Frequency Domain)的特徵抽取動作。而在特徵抽取方法上,本研究分別使用了自迴歸模型(Autoregressive Model, ARM)和功率頻譜熵值(Power Spectral Entropy, PSE)方法去做腦波資料特徵抽取並找出具有鑑別力的特徵值。最後,本文將較有鑑別的特徵值利用K個最近鄰居分類器(K Nearest Neighbor Classifier, K-NN)和支持向量機器(Support Vector Machine, SVM)做分類的動作並且做統合性的比較。 經由實驗結果顯示,在不同的運動想像手動情境實驗中,利用了自迴歸模型方法為特徵抽取方法而分類器為SVM分類器時,分類率皆可達到70%以上。而在食指上下移動想像實驗中,分類率甚至可達到80%。

並列摘要


The researches on analysis of biosignal related areas have received increasingly attention in recently years because the skills of technical computing promotion and the analysis method improvement. Brain-computer interface (BCI) is one of the novelty skills. The main purpose of BCI is to use the variation of electroencephalogram (EEG) signals and the user can use them to communicate with the external environment directly. The main purpose of this thesis is to build the synchronous system to analyze the variation of EEG signals in somatosensory cortex area when the subject made different kinds of motor imagery (imagined right-hand or left-hand movements) experiments (or real hand movements). Then, we chose the most discriminated EEG signals from the specific frequency band as the recognition datasets and used them to classify. We used the finite impulse response (FIR) filter to reduce the ambient noises and some of physiological signals in signal preprocessing. We used Common Spatial Pattern (CSP) method to do a linear transform and extracted the discriminative features on time and frequency domain, respectively. In the feature extraction methods, we used Autoregressive Model (ARM) and Power Spectral Entropy (PSE) to extract the features and found the most discriminative features to classify. Finally, we used K Nearest Neighbor classifier (K-NN) and Support Vector Machine (SVM) to classify these features and made the comprehensive comparison. The experimental results show the classification rates in different motor imagery tasks all can reach over 70 % when using the ARM as the feature extraction method and SVM as the classifier. The classification rate in the imagined finger movement task can even achieve 80%.

參考文獻


[29] 薛昭能,“應用機器視覺技術於雙穩態膽固醇液晶顯示面板之瑕疵檢測,” 中原大學機械工
[33] 王茲元,“應用區域化變數理論於短期風速估測,” 中原大學電機工程學系碩士學位論文,
[4] 堺章,“透視人體醫學地圖,” 瑞昇文化事業股份有限公司,2008年。
[15] 劉勇均,“應用於指動偵測之腦波訊號分析系統,” 國立成功大學資訊工程學系碩士班碩士
[2] 方振隆,“以腦波控制之主動式義手,” 國立成功大學機械工程學系碩士論文,2004。

延伸閱讀