| DC 欄位 |
值 |
語言 |
| DC.contributor | 電機工程學系 | zh_TW |
| DC.creator | 陳俞豪 | zh_TW |
| DC.creator | Yu-Hao Chen | en_US |
| dc.date.accessioned | 2023-7-20T07:39:07Z | |
| dc.date.available | 2023-7-20T07:39:07Z | |
| dc.date.issued | 2023 | |
| dc.identifier.uri | http://ir.lib.ncu.edu.tw:444/thesis/view_etd.asp?URN=110521075 | |
| dc.contributor.department | 電機工程學系 | zh_TW |
| DC.description | 國立中央大學 | zh_TW |
| DC.description | National Central University | en_US |
| dc.description.abstract | 腦電訊號（Electroencephalography, EEG）存在著不平穩(Non-stationary)的特性，由於受試者不同或相同受試者但錄製腦波的時間、環境不同，使腦電訊號之間存在巨大差異，造成想像運動分類準確率不佳，最終難以訓練出通用性高的分類模型。若是能夠運用遷移學習法，利用事先錄製的 EEG 訊號遷移至當前的分類任務，不僅可以減少大量錄製腦波資料所需要的時間，還能使分類模型有足夠的訓練集資料提升分類準確度。
為此本論文提出基於領域自適應方法將大量帶有標籤(Label)的腦波從
源域遷移至少量標籤的目標域中，透過歐式空間與黎曼流行兩邊各自資料
對齊的方式，完成多分類遷移學習之想像運動，提高跨時段(Cross-sessions)及跨受試者(Cross-subjects)在想像運動上的分類性能。首先計算電極通道與各個類別黎曼均值之間的黎曼距離進行腦波電極通道選擇，通過濾波器組(Filter Bank)並基於互訊息之頻帶選擇法選擇最佳頻帶，再結合歐式空間的資料中心對齊法(Euclidean Space Data Alignment, EA)使源域與目標域資料圍繞同一中心，降低源域和目標域之間的差異性，接著由共空間模式(Common Spatial Patterns, CSP)提取一部分特徵，同時經由共變異數矩陣中心對齊(Covariance Matrix Centroid Alignment, CA)後從黎曼流形切線空間映射(Tangent Space Mapping, TSM)方法提取另一部分特徵，結合兩種特徵後透過領域可遷移性評估(Domain Transferability Estimation, DTE)選擇腦波資料，最終透過基於支持向量機(Support Vector Machine, SVM)的分類模型訓練以及分類。由 BCI 競賽資料集驗證，實驗結果顯示基於領域自適應的想像運動分類準確率與近幾年提出的架構相比準確度最高。 | zh_TW |
| dc.description.abstract | Electroencephalography (EEG) has non-stationary characteristics, leading to a large difference in EEG classification accuracy due to different subjects or the same subject′s brainwaves at different times and environments. Finally, it is difficult to train a robust classification model. If the transfer learning method can be used to transfer pre-recorded EEG signals to the current classification task, it can make the classification model have enough training set data to improve the classification
accuracy. Therefore, this paper proposes a domain adaptive method to transfer a large number of brainwaves with labels from the source domain to the target domain with a small number of labels. To improve the classification performance of Crosssessions and Cross-subjects in motor imagery, the Riemannian distance between the electrode channel and the Riemannian mean of each category was calculated to
screen the brainwave electrode channel, and the optimal frequency band was screened by Filter Bank and based on the mutual information band screening method.
Combined with Euclidean Space Data Alignment (EA), the data of the source domain and target domain revolve around the same center, reducing the difference between source domain and target domain. Some features were extracted from Common
Spatial Patterns (CSP). The other features are extracted from Tangent Space Mapping (TSM) method of Riemannian manifold after being aligned through Covariance Matrix Centroid Alignment (CA) at the same time. Combined with two features, brainwave data are screened through Domain Transferability Estimation (DTE). Finally, the classification model is trained and classified based on Support Vector Machine (SVM). The results showed that the accuracy of classification was obviously better than other algorithms through BCI competition EEG dataset. | en_US |
| DC.subject | 腦電圖 | zh_TW |
| DC.subject | 腦機介面 | zh_TW |
| DC.subject | 想像運動 | zh_TW |
| DC.subject | 遷移學習 | zh_TW |
| DC.subject | 領域自適應 | zh_TW |
| DC.subject | 資料對齊 | zh_TW |
| DC.subject | EEG | en_US |
| DC.subject | brain-computer interface | en_US |
| DC.subject | motor imagery | en_US |
| DC.subject | transfer learning | en_US |
| DC.subject | domain adaptation | en_US |
| DC.subject | data alignment, | en_US |
| DC.title | 基於領域自適應法結合多階層歐式空間與黎曼流形對齊之想像運動腦波分類 | zh_TW |
| dc.language.iso | zh-TW | zh-TW |
| DC.title | Classification of Motor Imagery EEG using Multi-Stage Euclidean Space and Riemannian Manifold Alignment based on Domain Adaptation | en_US |
| DC.type | 博碩士論文 | zh_TW |
| DC.type | thesis | en_US |
| DC.publisher | National Central University | en_US |