| DC 欄位 |
值 |
語言 |
| DC.contributor | 電機工程學系 | zh_TW |
| DC.creator | 蘇昱綸 | zh_TW |
| DC.creator | Yu-Lun Su | en_US |
| dc.date.accessioned | 2021-8-2T07:39:07Z | |
| dc.date.available | 2021-8-2T07:39:07Z | |
| dc.date.issued | 2021 | |
| dc.identifier.uri | http://ir.lib.ncu.edu.tw:88/thesis/view_etd.asp?URN=108521112 | |
| 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(EEG)會隨時間變化造成的不平穩特性與受試者之間不同的身心狀況帶來的差異性，導致難以訓練出高通用性的分類模型，並且會因錄製訊號時間長造成使用者疲勞，使腦機介面(BCI)在實際應用上受到限制，若能有效利用先前錄製的EEG遷移至當前的分類任務，使分類器能有足夠的訓練資料藉以提升分類準確度即可解決上述問題。為此本論文提出ELA-TSM領域自適應方法用於二分類想像運動訊號，藉由將大量帶有標籤樣本的源域遷移至少量標籤樣本的目標域中，首先會以歐基里德中心歸一(Euclidean Recenter)讓源域與目標域資料圍繞在同一中心，降低領域之間的差異性以最佳化共同空間型樣法(CSP)在不同領域的性能並使兩類訊號盡可能地分開，再結合標籤對齊法(Label Alignment)以兩類訊號黎曼均值為基準點對齊，最後用黎曼切線空間投影法(Tangent Space Mapping)提取黎曼幾何特徵並投影回歐式空間做分類。實驗使用BCI競賽IV資料集IIa驗證本論文演算法用於跨受試者、跨時段以及跨類別任務，結果顯示在目標域具有30筆帶標籤訓練資料的條件下平均分類準確率分別可達78.38%、80.37%及76.72%，且在資料集IVa於跨受試者任務可達85.22%。此外也會使用t-SNE視覺化做佐證，提供直觀的效果展示。 | zh_TW |
| dc.description.abstract | The characteristics of EEG signals are non-stationary caused by highly subject various individual differences, such as mentality and different individuals. Which makes it difficult to train a highly robust classification model and spend a lot of time recording EEG. It cause brain-computer interface (BCI) limited in practical applications. If the previously collected EEG can be transfered to the current classification task, many shortcomings can be solved. How to transfer the signal so that the classifier can have enough training data to improve the accuracy rate is a big challenge. This paper proposes a domain adaptation method called ELA-TSM, which can transfer source domain to the target domain. First, we use Euclidean Recenter on the source domain and the target domain. All data will be centered on the same point to reduce the difference between the domain to optimize the common space pattern method (CSP), and then separate two classes data. Next step is to use the Label Alignment method align the source label space with the target label space. Finally use the Riemann Tangent Space Mapping(TSM) to project the data from the Riemann space back to the Euclidean space for classification. In the part of experiment we use two BCI competition datasets to verify the feasibility of our method in Cross-subjects, Cross sessions and Crossclasses tasks. We also use t-SNE visualization as a proof to provide an intuitive effect display. | en_US |
| DC.subject | 腦機介面 | zh_TW |
| DC.subject | 領域自適應 | zh_TW |
| DC.subject | 黎曼幾何 | zh_TW |
| DC.subject | 腦電圖 | zh_TW |
| DC.subject | 想像運動 | zh_TW |
| DC.subject | Brain-computer interface | en_US |
| DC.subject | Electroencephalographic | en_US |
| DC.subject | Motor Imagery | en_US |
| DC.subject | Transfer learning | en_US |
| DC.subject | Domain adaptation | en_US |
| DC.title | 結合黎曼幾何與領域自適應方法於想像運動腦波分類 | zh_TW |
| dc.language.iso | zh-TW | zh-TW |
| DC.title | The Classification Of Motor Imagery EEG Based On Riemannian Geometry And Domain Alignment | en_US |
| DC.type | 博碩士論文 | zh_TW |
| DC.type | thesis | en_US |
| DC.publisher | National Central University | en_US |