運用鏡像運動資料之遷移學習於想像運動為基礎的即時回饋腦波人機介面;Transfer Learning of real-time Motor Imagery BCI using Action Observation EEG data with Transformer-based Spatial-Temporal Fearure

NCUIR > College of Electrical Engineering & Computer Science > Graduate Institute of Electrical Engineering > Electronic Thesis & Dissertation > Item 987654321/90128

Please use this identifier to cite or link to this item: http://ir.lib.ncu.edu.tw/handle/987654321/90128

Title:	運用鏡像運動資料之遷移學習於想像運動為基礎的即時回饋腦波人機介面;Transfer Learning of real-time Motor Imagery BCI using Action Observation EEG data with Transformer-based Spatial-Temporal Fearure
Authors:	張子謙;Chang, Tzu-Chien
Contributors:	電機工程學系
Keywords:	腦電波;腦機介面;想像運動;鏡像神經元;深度神經網路
Date:	2022-09-22
Issue Date:	2022-10-04 12:11:47 (UTC+8)
Publisher:	國立中央大學
Abstract:	腦機介面(Brain Computer Interface, BCI)提供神經或是肌肉損傷的人士一個利用腦電波與外界交流的管道。過去要求受試者進行想像運動，然而對於癱瘓病人很難對動作進行想像，因此我們以鏡像神經元系統(human Mirror Neuron System, hMNS)作為訓練想像運動的方法，結合VR(virtual reality)觀看影片建立想像依據，分別建立個人的模型(individual model)、共同的模型(global model)，接著讓受試者回想VR場景進行想像運動，最後建立一個回溯想像的即時回饋腦機介面，使想像運動的訓練過程更方便、容易。實驗分成三個部份，第一部份是建立鏡像神經觀察腦波模型，第二部份建立回溯想像遷移學習模型，第三部份利用前面訓練出的模型建立BCI即時回饋系統。實驗中將乾式腦波電極設置在 10-20 EEG System 之 F3、F4、 C3、Cz、C4、P3、Pz、P4的位置，模型需要兩秒的腦波資料，並以每0.1秒的移動視窗(sliding window)，利用動作觀察、回溯想像、即時回饋加強後的模型，進行即時BCI的判斷。實驗受試者有五人，皆為年齡介於20~22的健康男性，慣用手皆為右手。結果顯示在個人的模型上，觀察模型的平均準確率達到55.8%，回溯想像的平均準確率達到63.6%，即時回饋的平均準確率達到73.1%。在共同的模型上，觀察模型的平均準確率達到50%，回溯想像的平均準確率達到61%，即時回饋的平均準確率達到66.5%。 ;Brain Computer Interface (BCI) provides a way for people with nerve or muscle damage to communicate with the outside world using brain waves. In the past, subjects were required to perform motor imagery. However, it is difficult for paralyzed patients to do motor imagery. Therefore, we used the human Mirror Neuron System (hMNS) as a method to train motor imagery, combined with VR (virtual reality) to watch videos establish an imaginary basis, respectively establish an individual model and a global model, and then ask the subjects to recall the VR scene to perform motor imagery, and finally establish a real-time feedback brain-computer interface for retrospective imagination, so that the motor imagery training process is simpler and more convenient. The experiment is divided into three parts, the first part is to establish a mirror nerve action observation brain wave model, the second part is to establish a retrospective imaginative transfer learning model, and the third part uses the previously trained model to establish a BCI real-time feedback system. In the experiment, the dry brainwave electrodes are set at the positions of F3, F4, C3, Cz, C4, P3, Pz, and P4 of the 10-20 EEG System. The model requires two seconds of brainwave data, and the sliding window moves every 0.1 seconds, using action observation, retrospective imagination, and real-time feedback to strengthen the model to make real-time BCI judgments. There were five subjects in the experiment, all of them were healthy males between the ages of 20 and 22, and all of them were right-handed. The results show that on the idividual model, the average accuracy of the observation model is 55.8%, the average accuracy of retrospective imagination is 63.6%, and the average accuracy of instant feedback is 73.1%. On the global model, the average accuracy of the observation model is 50%, the average accuracy of retrospective imagination is 61%, and the average accuracy of instant feedback is 66.5%.
Appears in Collections:	[Graduate Institute of Electrical Engineering] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	70	View/Open

社群 sharing

Loading...