基於前額不對稱性之情緒腦波研究

A Study of Emotional Brain-Wave based on Frontal Asymmetry

林威志(Wei-Zhi Lin)

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/cycu201100766

摘要

近年來隨著腦神經科技發展以及訊號擷取技術的成熟，使得腦機介面(Brain Compute Interface, BCI)成為熱門的研究主題。研究主要分為閃光刺激、運動想像以及情緒辨識三大部分的應用。
本研究著重於情緒應用，藉著影片方式對8位受試者激發情緒，利用影片進行情緒激發流程，並從前額上5組電極進行分析選取最有鑑別力之電極偶。同時選取最有鑑別力之時間區段。量取放鬆情況以及不同情緒下的腦波訊號(Electroencephalogram ,EEG)進行正向、負向情緒分類。特徵抽取方式先將腦波進行離散傅立葉轉換(Discrete Fourier Transform, DFT)，取出α(8~13Hz)以及β(14~30Hz)兩種頻帶的資料，計算腦波不對稱關係比(Asymmetry Relation Ratio, ARR)，並用放鬆狀態腦波對ARR特徵進行三種修改，將兩種特徵進行比較；另一方面將各頻帶資料進行反離散傅立葉轉換(Inverse Discrete Fourier Transform, IDFT)重建，將右側電極減去左側電極並取α、β、α+β(8~30Hz)三種頻帶組合的重建訊號進行碎形維度的計算，再將ARR、碎行維度以及自回歸模型作為特徵使用Leave One out方式進行支持向量機器(Support Vector Machine, SVM)測試。
從中得知FP1-FP2、F3-F4此兩對電極偶個別取α頻帶以及β頻帶會有最佳鑑別力，使用ARR以及碎形維度為特徵的初步分類率為65%上下，將ARR進行修正後分類率上升接近70%，得知ARR數值會受到各受試者腦部成長影響，將放鬆狀態腦波考慮進來，能有效減少受測者間的變異性。

關鍵字

離散傅立葉轉換 ； 情緒辨識 ； 碎形維度 ； 腦波不對稱關係比 ； 腦機介面 ； 自回歸模型

並列摘要

This study focuses on the emotional feature analysis. We use video clips stimulate 8 subjects on the protocol that has been designed with more dynamic emotional content for inducing discrete emotions (joy, disgust, sad and angry). EEG signals will be collected from 8 subjects by using International 10-20 system. The electrodes are the best things to discriminate emotional features that have been selected by frontal EEG Asymmetry. We also select the most discriminating period at the same time. By using Discrete Fourier Transform (DFT), first EEG signals decompose into two different frequency bands (α and β). And second EEG signals which form relaxation and emotional state that were calculated as the Asymmetry Relation Ratio (ARR). We will use ARR of relaxation state to modified ARR of emotional state in three different ways. To calculate fractal dimension, we make the right electrode signal subtracts the left electrode signal and take three different frequency bands (α, β and α + β) EEG signals. And then, take all features ARR, fractal dimensions and auto-regression model to the Support Vector Machine (SVM) by using “Leave-One-Out” way.
The results that α bands from FP1-FP2 electrodes have the best discrimination and F3-F4 electrodes also have the best result to β bands. By using SVM, which bases on ARR and the fractal dimension, the accuracy is 65%. After modifying ARR, the SVM accuracy will be up to nearly 70%. It proof that the ARR values will be affected by the subjects' brain growth. Considering the relaxation of EEG signal could increase the accuracy of ARR, and subtract the variability of emotional quantitative indicators.

並列關鍵字

fractal dimension ； asymmetry relation ratio ； brain-machine interface ； discrete Fourier transform ； autoregressive model ； emotion recognition

參考文獻

[16] 堺章，透視人體醫學地圖，瑞昇文化事業股份有限公司，2008年。

[2] P. Ekman, “An Argument for basic Emotion”, Cognition and Emotion, 6, pp.169-200, 1992

[3] M.Murugappan, M.Rizon, RNagarajan, S.Yaacob, I.Zunaidi, and D.Hazry, “EEG Feature Extraction for Classifying Emotions using FCM and FKM”, International Journal of Computers and Communications, Issue 2, Volume 1,pp21-25, 2007

[5] M Murugappan, R Nagarajan, and Sazali Yaacob, “Appraising Human Emotions using Time Frequency Analysis based EEG Alpha Band Features”, Conference on Innovative Technologies in Intelligent Systems and Industrial Applications, 2009

[6] M. Murugappan, N. Ramachandran, and Y. Sazali, “Classification of human emotion from EEG using discrete wavelet transform”, J. Biomedical Science and Engineering, 3, pp.390-396, Apr. 2010.