高斯混合模型與K-最近鄰居分類法在腦波辨識研究上的應用

Abstract

在本論文中，我們利用腦電圖技術（Electroencephalography, EEG）來辨識不同狀況之下的腦部反應。在訊號辨識的研究中，高斯混合分配（Gaussian mixture distribution）是一個常用的模型。在本論文中，我們使用高斯混合模型來配適腦波資料。並且提出利用K-最近鄰居分類法（K-nearest neighbor classifier）來協助分類。我們收集了17位健康受測者想像運動和想像放鬆的腦波資料，並將腦波資料配適高斯混合模型，接著以K-最近鄰居分類法與最大概似分類法來辨識腦波。最後，K-最近鄰居分類法達到辨識正確率82.35%，最大概似分類法達到辨識正確率76.47%。

In this thesis, we investigate the feasibility of using Electroencephalography（EEG） to recognize the brain response under different conditions. In signal recognition, the Gaussian mixture distribution is a widely-used model. In this study, we adopt the Gaussian mixture model（GMM） to fit the EEG data, and propose using the K-nearest neighbor method to classify the data. We collect 17 healthy individuals’ EEG under the condition "Imagine sporting" and "Imagine relaxing", respectively. Then we use the GMM to fit the data. Then, we adopt the K-nearest neighbor classifier and maximum likelihood method to recognize the EEG data. Finally, the accuracy of K-nearest neighbor classifier is 82.35%, and the accuracy of maximum likelihood method is 76.47%.