情緒腦波分析及腦機介面之開發

Analysis of Emotional EEG and Development of Brain-Computer Interface

高永樺(Yung-Hwa Kao)

指導教授 ： 劉益宏

中原大學/工學院/機械工程研究所/碩士(2013年)

https://doi.org/10.6840/cycu201300950

摘要

本論文開發了情緒辨識之腦機介面(Brain-Computer Interface, BCI)系統，在情緒辨識中達到即時診斷的技術，對未來應用在憂鬱症(Depression)、躁鬱症(Bipolar Disorder)、強迫症(Obsessive-Compulsive Disorder)、精神分裂症(Schizophrenia)等精神疾病，有著相當大的貢獻。
本論文基於情緒辨識而設計了一套情緒誘發的實驗流程，流程中使用了100張的情緒圖片提供受測者進行實驗，其情緒圖片皆源自於國際情緒圖片系統(International Affective Picture System, IAPS)，而實驗中的每次試驗皆依照受測者的感受填選自我評估表(Self-Assessment Form, SAF)，其中針對受測者填選的Arousal及Valence進行主觀情緒的類別標籤，將受測者主觀的情緒當作機器學習分類的依據。其中分類的方式可將標籤分為四個類別：第一類為High Valence–High Arousal(HVHA)；第二類為Low Valence–High Arousal(LVHA)；第三類為Low Valence–Low Arousal(LVLA)；第四類為High Valence–High Arousal(HVLA)，而在過去的文獻中提及：利用Arousal與Valence可使得分類率得到較大的改善，故本論文將分類方式分為兩個階段，分別為兩個類別的分類問題，第一階段為High Arousal及Low Arousal；第二階段為High Valence及Low Valence。
本實驗利用了快速傅立葉(Fast Fourier Transform, FFT)將腦波轉換至頻率域分析，並利用頻帶功率(Band Power, BP)搭配主成分分析(Principal Component Analysis, PCA)的特性，在Gamma頻帶中使用支持向量機(Support Vector Machine, SVM)得到了很棒的結果。結果表明平均分類率Arousal可達70.2%、Valence可達77.5%，而分類率最佳的受測者在Arousal可達到84%，Valence更可達到89%，此分類結果對情緒辨識有著極大的幫助。

關鍵字

腦波訊號 ； 腦機介面 ； 情緒辨識 ； 快速傅立葉轉換 ； 主成分分析 ； 支持向量機

並列摘要

This paper developed an emotion-recognition Brain-Computer Interface (BCI) system, a technology that can instantly detect and diagnose emotions, which provides a great contribution to future applications of mental illness such as Depression、Bipolar Disorder、Obsessive-compulsive disorder、Schizophrenia, etc.
This paper designed a procedure based on emotion recognition to induce emotions, this procedure used 100 emotion pictures, all emotion pictures came from the International Affective Picture System (IAPS), each trial the subject fills a self-assessment form (SAF), subjective emotions will be labeled according to the Arousal and Valence from the SAF each subject has filled, using the subject’s subjective emotion as the criteria of machine learning classification. There are four classes in classification: first class is High Valence–High Arousal (HVHA); second class is Low Valence–High Arousal (LVHA); third class is Low Valence-Low Arousal (LVLA); fourth class is High Valence–High Arousal (HVLA), early studies show that using Arousal and Valence can improve the classification rate, thus in this paper there are two stages of classification, the first stage is classification of High Arousal and Low Arousal; the second stage is classification of High Valence and Low Valence.
The experiment used Fast Fourier Transform (FFT) to transform and analyze EEG signal in frequency-domain, using Band Power (BP) along side with Principal Component Analysis (PCA), the result of Support Vector Machine (SVM) is great in Gamma band. The result show an average accuracy of Arousal is 70.2%、Valence is 77.5%, the subject with the highest accuracy has a 84% accuracy in Arousal, and 89% accuracy in Valence, this result has significant contribution to emotion recognition.

並列關鍵字

BCI ； EEG ； Emotion Recognition ； Fast Fourier Transform ； PCA ； SVM

參考文獻

[5] 翁瑞宗, 腦機介面之P300拼字器開發及特徵分析, 中原大學機械工程研究所碩士學位論文, 2010.

[37] 林威志, 基於前額不對稱性之情緒腦波研究, 中原大學機械工程研究所碩士學位論文, 2011.

[38] 王志翰, 情緒腦波誘發範例建立及辨識, 中原大學機械工程研究所碩士學位論文, 2012.

[39] 鍾文祥, 針對運動想像腦機介面之特徵分析, 中原大學機械工程研究所碩士學位論文, 2011.

[45] 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, 中原大學機械工程研究所碩士學位論文, 2009.