- 學位論文
情緒腦波誘發範例建立及辨識
The Establishment and Recognition of the Emotional EEG Induced Paradigm.
摘要
由於現代科技發展快速、職場上競爭激烈加上現代人生活步調快速，使得很多人在精神上或心理上受到很多壓力，使得罹患憂鬱症或躁鬱症的人日俱增多，若我們能將不同的情緒作準確的分類及辨識，對於在一般家庭中的患者自我照護方面肯定是一大研究貢獻。 本論文研究核心在於利用腦機介面(Brain-Computer Interface, BCI)的概念來作情緒分類，利用情緒圖片對8位受測者誘發情緒，本實驗所使用之圖片皆來自國際情緒圖片系統 (International Affective Picture System, IAPS)所提供的情緒圖片來進行情緒誘發，並從量測到的腦波訊號進行情緒分析。同時找出最有鑑別力之時間區段，並利用放鬆情況以及不同情緒下的腦波訊號(Electroencephalogram, EEG)進行情緒分類。本實驗之情緒分類分為四大類：第一類：High Valence – High Arousal(HVHA)；第二類：Low Valence – High Arousal(LVHA)；第三類：Low Valence – Low Arousal(LVLA)；第四類：High Valence – Low Arousal(HVLA)，第一類HVHA主要情緒有愉悅、欣喜、快樂、幸福；第二類LVHA主要情緒有恐懼、緊張、害怕、壓力；第三類LVLA主要情緒有傷心、悲傷、愁苦、沮喪； 第四類HVLA主要情緒有平靜、放鬆、滿足、安詳。 透過腦波擷取儀器所獲得的情緒腦波訊號進行特徵抽取，特徵抽取方式為先將腦波訊號進行離散傅立葉轉換(Discrete Fourier Transform, DFT)，取出α(8~13Hz)以及β(14~30Hz)兩種頻帶的腦波訊號，計算其功率頻譜密度(Power Spectral Density, PSD)值，並利用大腦前額不對稱性、不同電極組合方式、不同情緒時間區段也使用傅立葉頻譜重組圖片(Fourier Shuffle Picture)進行分析，再將不同特徵抽取方式的資料進行K個最近鄰居分類法(K-Nearest Neighbor Classification, K-NNC)及支持向量機器(SVM)求得分類率，本篇論文所有受測者K-NNC平均分類率為65.71%，最高受測者達80.22%；所有受測者SVM平均分類率為70.80%，最高受測者達87.93%，此結果將有助於開發線上情緒辨識系統，對於憂鬱症或躁鬱症病患更是一大幫助。
並列摘要
As science and new technologies develop rapidly, working environment becomes more competitive, and the pace of modern life turns to be faster, people nowadays suffer from spiritual or psychological pressure. This phenomenon has made the number of Depression and Bipolar Affective Disorder patients increase day by day. If we are able to get an accurate identification and classification of different emotional states, it will be a research to improve patients’ self-healthcare program within the family. The topic of this paper is about how to classify different emotional states by using the brain – computer interface (BCI) concept. We use emotional pictures to induce different emotions in 8 subjects. The pictures are from the International Affective Picture System (IAPS). The research is based on the analysis of the Electroencephalogram (EEG) . At the same time, the most discerning time period is identified and the EEG of different emotional states, such as relaxed and others, is used to classify emotional states. In this paper, emotional states are classified into four main categories: first, High Valence – High Arousal (HVHA); second, Low Valence – High Arousal (LVHA); third, Low Valence – Low Arousal (LVLA); and fourth High Valence – Low Arousal (HVLA). The first category consists of pleasure, gladness, joy, and happiness; the second one contains horror, nervous, fear, and pressure; the third includes sadness, sorrow, distress, and depression; and the fourth category comprises calm, relaxation, satisfaction, and serene. The feature extraction of the emotional EEG is obtained through the NeuroScan. The first procedure of feature extraction is to transform the EEG with Discrete Fourier Transform (DFT), by taking the alpha band (8Hz ~ 13Hz) and beta band (14Hz ~ 30Hz) to calculate the Power Spectral Density (PSD). And then, the theorem of Frontal EEG Asymmetry, the different electrodes combinations, the different emotion time period, and the Fourier spectrum reorganization pictures (Fourier shuffle picture) are analyzed. Finally, by using different data of feature extraction to obtain a classification rate by K-Nearest Neighbor Classification (K-NNC), and Support Vector Machine (SVM). In this paper, the emotional average classification rate of each subject has reached 65.71%, the highest classification rate reached 80.22% by K-NNC, and the emotional average classification rate of each subject has reached 70.80%, the highest classification rate reached 87.93% by SVM. In the future, it will contribute to the development of online emotion recognition system, and be a huge help for Depression or Bipolar Affective Disorder patients.