| 摘要: | 腦機介面(Brain Computer Interface, BCI)作為殘疾人士與現實世界進行溝通的橋樑。想像運動(Motor Imagery)是BCI架構中一個非常重要的分支，提供大腦與機器之間一種控制方法。其中想像運動與動作觀察的鏡像運動皆被認為是學習運動的有效工具。在複雜的協調任務學習中，動作觀察(Action Observation, AO)比想像運動能夠更有效的達到大腦訓練的效果。本論文以鏡像神經元系統(human mirror neuron system, hMNS)作為訓練想像運動的方法，讓受試者有一個視覺化的動作參考對象，結合虛擬實境探討左右手不同角度的動作。實驗分成兩部份，第一部份以動作觀察激發鏡像神經元系統，第二部分將人物換成箭頭，以箭頭指示想像運動，最後使用鏡像運動的腦波資料模型預測想像運動的腦波資料，此二階段資料量測的目的在於嘗試建立一個利用鏡像神經元訓練想像運動的新方法。實驗中將乾式腦波電極設置在10-20 EEG System之Fp1、Fp2、F3、Fz、F4、C3、Cz、C4、P3、Pz、P4、O1、O2的位置，以動作後兩秒的腦波資料，經過重疊頻帶濾波器組分成10個子頻帶，並計算協方差矩陣投射到黎曼空間中，使用黎曼均值搭配切線空間投影法將資料轉到歐氏空間中進行預測及分類。結果顯示經過鏡像學習的受試者在想像運動上的表現有明顯的提升，能夠分類細微動作的差異。未來有望提升BCI在各領域的應用。;Brain Computer Interface(BCI) is provided as a bridge for disabled people to communicate with the world. Motor imagery(MI) is an important part of the BCI. MI and action observation (AO) have been considered as effective ways for motor learning. In complex learning tasks, AO is considered a more effective method to train brain motor cortex, compared to MI. In our study, we constructed a human mirror neuron system (hMNS) as pre-trained task for suject’s MI training. The hMNS provided subjects reference images for MI, and tried to guide the movements of the left/right hand in different angles under virtual reality(VR) environment. Our EEG experiment contained two parts. In the first part, subjects were requested to performed a hMNS task by viewing AO videos, and EEG data were collected to train a pre-trained model for the subsequent MI task. In the second part, MI task was given to subjects and the MI classification was performed using the pre-trained hMNS deep learning model obtained from the experiment of the first part. In the MI task, instead of view hand motions, an arrow indicator was used to indicate the direction for MI. The purpose of the aforementioned two-step EEG experiment is trying to build a new MI training process based on a hMNS pre-training approach. A thirteen-channel dry-electrode wireless EEG system was used to measure EEG signals from electrode positions at Fp1, Fp2, F3, Fz, F4, C3, Cz, C4, P3, Pz, P4, O1, and O2 according to the international 10-20 montage system. The EEG data were real-time filtered into ten frequency bands as features. The covariance matrixes obtained from the features of ten frequency bands were calculated and projected into the Riemann space. The mean of the projected values in Riemann were calculated and the tangent space mapping(TSM) method was used to transfer EEG data to the Euclidean space for prediction and classification. The results showed that the subjects, who participated in pre-trained by hMNS task, had significant improvements in the following MI tasks. In the future, it is expected to enhance the application of BCI in various fields. |
