是非題及選擇題答題之腦電波分析

Abstract

摘要

大腦認知活動的分析，目前在教育心理學和認知神經科學等領域已被廣泛的研究。本研究目的除了將腦電波訊號做資料分類的分析，以便於應用在大腦人機介面(Brain Computer Interface, BCI)之外，也討論了實驗設計所給予的不同類型問題對大腦認知活動的影響。 本研究利用所設計的問題當刺激，來探討受測者在思考不同類型問題時腦電波的差異，其中以智力測驗為主的選擇題實驗研究中發現Theta在做思考數學問題時的能量，均高於思考圖形幾何問題的能量，但在Alpha頻段則恰好相反。另外藉由是非題實驗我們卻也發現Gamma頻段對於不同類型的問題在認知活動時並無差異。 在辨識的部份，本實驗目的為找出特徵擷取的方法，對受測者在想像回答「是」與「否」的腦電波做分類辨識，以及找出最適當且少量的電極組合來降低運算量。腦電波辨識是否成功的關鍵，在於特徵擷取與分類兩個議題。過去文獻將重點放在分類演算法的改良上，然而找出重要的特徵，可以獲得更高的辨識率。研究中發現在時域部份所擷取的腦電波具有相當好的鑑別性，藉由線性鑑別分析法（LDA）找出最佳的轉換向量，讓資料更具鑑別性，再計算特徵像量間的歐氏距離就可分類腦電波。結果顯示本實驗使用了C3、C4及F3三個電極，腦電波辨識的準確率大幅提升至99%。Abstract

In the field of educational psychology and cognitive neuroscience, the analysis on the activities of cerebrum cognition is generally researched at present. This study is not only to analyze and classify the electroencephalography signals in favor of being applied on the Brain Computer Interface, or BCI, but also to discuss the effect on the cognitive thinking which is due to being inducted by different kinds of questions. The analysis of this study is based on the frequently used EEG bands in recent years. By means of the multiple-choice questions, we use intelligence test, this study discusses the differences on the energy of these bands for being tested by different kinds of questions. In contrast with the energy of Alpha band, the study results show that the energy of Theta band as the testee doing the math questions is much higher than that of Theta band as the testee doing the geometry questions. The energy of the Gamma band shows no differences on cognitive activities for being tested by different kinds of questions. At the identification portion, this experiment is to find out the method of characteristic acquisition, to identify the electroencephalography of the testee as he imaging answering for ‘yes’ or ‘no’, and to find out the least and the most suitable channel in order to minimize the quantity of operation. Algorithms of feature detection and classification are the two keys to EEG classifying. In the past, most articles focused on the improvement of classifiers, but selecting simper and more important feature is an alternative way to get a high accuracy. The feature extraction can be obtained by the Linear Discriminant Analysis (LDA). The method also uses Nearest Neighbor Rule (NNR) to classify the processed data.The experimental results show that the average accuracy rate is improved to 99% by C3、C4 and F3 channels.