利用情感性視覺誘發腦磁波之情緒辨識

Emotion Recognition using Magnetoencephalographic Recordings Evoked by Affective Visual Stimuli

李欣芳(Hsin-Fang Lee)

指導教授 ： 陳永昇

國立交通大學/資訊學院/生醫工程研究所/碩士(2011年)

https://doi.org/10.6842/NCTU.2011.00903

摘要

近年來因為情緒和壓力造成的情感性疾病逐年增加，如能讓智慧型環境系統偵測
我們的情緒，當偵測到負面情緒時，系統能夠適時的改變週遭的環境，如光線、
溫度、音樂等，讓我們的情緒能因環境的舒適而改善，將有助於降低情感性疾病
的發生。因此情緒的辨識系統變得越來越重要，關於情緒辨識的各方面研究也在
近幾年內蓬勃發展。在情緒辨識的領域上，過去大都採用臉部表情、語音及肢體
語言等外在表徵來辨識情緒，本研究提出一個基於腦磁波訊號的情緒辨識系統，
採用情感性視覺刺激的誘發，觀察腦磁波訊號中所呈現出的腦波變化來進行情緒
辨識。本情緒辨識系統分為生理訊號的擷取、特徵擷取、分類三個主要部份，本
研究共收集14 位正常人與14 位憂鬱症患者，並使用五種特徵擷取方法，分別為
事件相關磁場(event-related magnetic fields, ERF)、頻譜能量密度(power spectral density, PSD)、自迴歸模型(autoregressive (AR) model)、主成分分析(principal component analysis, PCA)、以及局部線性嵌入(locally linear embedding, LLE)，經比較這些特徵擷取方法後，我們採用結合頻譜能量密度的頻域特徵與局部線性嵌入的空間降維，並利用支持向量機 (support vector machine, SVM)來進行情緒分類。實驗結果顯示，我們的情緒辨識系統在分辨快樂和害怕這兩種情緒時，平均準確率能達到91.8%。因此，在腦波特徵擷取上，頻譜能量密度與局部線性嵌入的結合，可以達到不錯的辨識率。

關鍵字

情緒辨識 ； 腦磁波 ； 事件相關磁場 ； 自迴歸模型 ； 頻譜能量密度 ； 主成分分析 ； 局部線性嵌入

並列摘要

The prevalence of affective disorders has been increasing recently. One possible remedy is to construct an intelligent environment system which can recognize emotions and adjust the conditions of environment like temperature, light, and music to comfort the subjects when their negative emotions are detected. Therefore, automated emotion recognition has attracted more and more attentions in the field of affective computation. In the literature, there have been many research works which recognize emotions through external manifestation, such as facial expression, voice, and gestures. In this works, we propose an emotion recognition system using magnetoencephalographic (MEG) signals evoked by affective visual stimuli. Our system can be divided into three parts including
physiological signal acquisition, feature extraction, and classification. We recruited 14 normal subjects and 14 bipolar disorder patients. Five kinds of feature extraction methods were evaluated in this work, including: event-related magnetic fields (ERF), power spectral density (PSD), autoregressive (AR) model, principal component analysis (PCA), and locally linear embedding (LLE). We adopted the features and the combination of PSD and LLE as use support vector machine (SVM) to categorize happy and fear emotions. According to our experiments, the proposed system can achieve 91.8% accuracy of emotion recognition.

並列關鍵字

Emotion Recognition using ； Magnetoencephalographic ； event-related magnetic fields ； autoregressive (AR) model ； power spectral density ； principal component analysis ； linear embedding

參考文獻

[1] K. Alho, C. Escera, R. Diaz, E. Yago, and J. M. Serra. Effects of involuntary auditory attention on visual task performance and brain activity. Neuroreport, 8(15):3233–3237, 1997.

[3] Walter B. Cannon. The james-lange theory of emotions: A critical examination and an alternative theory. The American Journal of Psychology, 39(1):106–124, 1927.

[4] C. Cherniss, M. Extein, D. Goleman, and R. P. Weissberg. Emotional intelligence: What does the research really indicate? Educational Psychologist, 41(4):239–245, 2006.

[6] P. Ekman, R. W. Levenson, and W. V. Friesen. Autonomic nervous-system activity distinguishes among emotions. Science, 221(4616):1208–1210, 1983.

[7] C. A. Frantzidis, C. Bratsas, C. L. Papadelis, E. Konstantinidis, C. Pappas, and P. D. Bamidis. Toward emotion aware computing: An integrated approach using multichannel neurophysiological recordings and affective visual stimuli. IEEE Transactions on Information Technology in Biomedicine, 14(3):589–597, 2010.