基於鏡像神經元訓練之深度學習想像運動模型;A Deep Learning Model for Motor Imagery based on Mirror Neuron Training

NCUIR > College of Electrical Engineering & Computer Science > Graduate Institute of Electrical Engineering > Electronic Thesis & Dissertation > Item 987654321/88364

Please use this identifier to cite or link to this item: http://ir.lib.ncu.edu.tw/handle/987654321/88364

Title:	基於鏡像神經元訓練之深度學習想像運動模型;A Deep Learning Model for Motor Imagery based on Mirror Neuron Training
Authors:	陳泓夬;Chen, Hong-Guai
Contributors:	電機工程學系
Keywords:	鏡像神經元;深度學習;腦波;想像運動
Date:	2022-04-06
Issue Date:	2022-07-14 00:24:12 (UTC+8)
Publisher:	國立中央大學
Abstract:	隨著腦科學的研究發展，腦機介面以成為腦科學的發展重點，而具有非侵入式優點的 EEG 腦電訊號，正是腦機介面的主要方案之一。然而腦電訊號通常被掩沒在噪聲中，其訊號的信噪比(SNR)非常低，無法以肉眼替不同動作的腦波進行分類，這對運動圖像分類立下了巨大的挑戰。而近年來分析的方式也相當多元，從自動編碼器(Auto Encoder)、受限玻爾茲曼機，到人工智慧的 CNN 和 RNN 深度學習網路，這些用於 EEG 運動圖像分類的方式皆極大地提高了運動圖像分類的準確性。然而多分類依然為一大挑戰，不僅想像運動的方式無法訓練出多種不同的分類，且訓練階段也將耗費相當多的時間，因此本論文針對訓練方式提出了不同的方法，利用鏡像神經元的方式訓練想像運動，希望藉由觀看學習的方式達到更好的效果，同時縮短訓練的時間並提高不同動作的判斷率，與此同時在神經網路上也將採用新穎的時序性架構進行訓練。實驗過程採用 13 通道的腦波機收集資料，於最後的辨識成果中，透過五位受試者對五種動作各 300 次的動作觀看，三分類最高可達 56.2%，而五分類也有 40.3% 的辨識率，除此之外，以純想像運動資料放入鏡像神經元所訓練出的模型中，三分類辨識率可達 52%，而五分類則為 33.1%。證實鏡像神經元與像想運動之間並非毫無關聯，前者所訓練出的網路對後者依然有一定的辨識度。 ;With the development of EEG science research, the advanced techniques for brain wave analysis enable people to probe the profound neurocircuitry inside human brain. One promising technique is the use of brain waves to control peripheral machines through user’s intentions, which is called brain copmuter interface (BCI). But how to extract useful information from brain wave is still a big challenge. In recent years, some deep learning solutions based on AutoEncoder, Restricted Boltzmann Machine, CNN and RNN have been proposed for EEG imagery classification which have will improved the accuracy. However, multi-task problem remains a challenge. Because of imagery movement can’t use for multi-task classification, so we try to use mirror neuron training for the better result. In this paper, we also use a new solution based on Temporal Convolutional Network(TCN) with much higher computational efficiency and accuracy. Our experiment result have shown that the propose TCN solution has obtained state of the great performance on multi-task of mirror neuron training. A high classification accuracy as 56.2% on 3 tasks and 40.3% on 5 tasks. Besides, imagery movement data is also recognizable in the model. The three task recognition rate can reach 52% and five task recognition rate can reach 33.1%. It shows that there are some similar associations between mirror neuron training and motor imagery.
Appears in Collections:	[Graduate Institute of Electrical Engineering] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	55	View/Open

社群 sharing

Loading...