應用深度學習於運動區腦波之手部動作預測

NCU Institutional Repository > 資訊電機學院 > 電機工程研究所 > 博碩士論文 > Item 987654321/81927

請使用永久網址來引用或連結此文件: http://ir.lib.ncu.edu.tw/handle/987654321/81927

題名:	應用深度學習於運動區腦波之手部動作預測
作者:	徐則林;Hsu, Tse-Lin
貢獻者:	電機工程學系
關鍵詞:	腦電波;腦波人機介面;深度學習網路;Electroencephalography (EEG);Brain Computer Interface (BCI);Deep Learning Neural Network
日期:	2019-10-15
上傳時間:	2020-01-07 14:38:59 (UTC+8)
出版者:	國立中央大學
摘要:	心電圖、腦電波、肌電波等是臨床上重要的診斷儀器，傳統溼式 Ag/AgCl電極雖然在訊號接收上相當穩定，但實驗中造成使用者的不適，特別是用於降低電極與皮膚間阻抗的電解凝膠，此介質可能對使用者的皮膚造成刺激，甚至是產生過敏反應，故我們使用乾式電極進行實驗。由於在醫學研究中指出 EEG中特定的頻寬具有大腦活動與動作技能表現的關聯，因此本研究希望藉由整合腦波訊號處理與慣性感測器系統，提供足夠且標記的腦波訓練資料，發展貼近日常動作的腦波分析系統。本實驗以慣性感測器作為標記工具，透過不同動作間的角度變化作為腦波的時間標記，目的在結合慣性感測器與腦波機來提升腦波人機介面的準確率。我們在受試者雙手各裝置感測器，結合腦波電極設置在 10-20 EEG System之 C3、Cz、C4、F3、F4位置的腦波，每隔 8 秒做一次手臂動作，紀錄受測者的動作姿態與腦波，透過肢體間的角度變化進行腦波的時間標記，標記方式為抓取動作瞬間作為基準點，以此基準點向前取兩秒的資料作為分析腦波的區間，透過小波轉換 (wavelet transform) 的方式取出此腦波區間的 (Event RelatedDesynchronization/Event RelatedSynchronization, ERD/ERS)，並將五個通道所取出的頻率與時間關係做疊加，作為疊加後的二維時頻圖像，輸入卷積神經網路 (Convolution Neural Networks, CNN)及長短期記憶神經網路(Long Short-Term Memory, LSTM)進行分析，達到 CNN 80% 及 LSTM 89% 準確率，並以此架構找出動作與腦波相對應的連結。;Electrocardiogram (ECG), electroencephalogram(EEG), electromyography(EMG) are important diagnostic instruments in clinic. Although the Ag/AgCl electrode is quite stable in signal reception, it causes user discomfortable in the experiments, especially for using Electrolyte gel to reduce the impedance between the electrode and the skin. The use of wet type electrolyte may cause stimulation to the user′s skin, and even produce allergic reaction, so we use dry electrodes for experiments. Since it is pointed out in medical research that the specific bandwidth in EEG has a correlation between brain activity and motor performance, in this study we propose to develop a brain wave analysis system close to daily movements by integrating brain wave signal processing and IMU system to provide sufficient and marked brain wave training data. In this experiment, IMU is used as the marking tool, and angles change between different actions are used as the time mark of brain wave. The purpose of this experiment is to improve the accuracy of brain computer interface (BCI) by combining IMU and brain wave system. In this research, we mounted two IMU on subject’s left and right arm. The EEG electrodes were attached on C3,Cz,C4,F3, and F4 positions, according to international 10-20 EEG system. Subjects were asked to do specified motion between 8sec, and timing of subject’s posture data was wirelessly transmitted for EEG labeling. EEG data were segmented into epochs from -2sec anchored to subject’s movement onsets. Labeled EEG data were extracted Event RelatedDesynchronization/Event RelatedSynchronization (ERD/ERS) by wavelet transform, and we combine the five channels (C3,Cz,C4,F3, and F4) time-frequency relationship as two-dimension image. Using this image as Convolution Neural Networks(CNN) and Long Short-Term Memory(LSTM) input attained CNN 80% and LSTM 89%, to exploring the connections between subject’s movement and brain wave.
顯示於類別:	[電機工程研究所] 博碩士論文

文件中的檔案:

檔案	描述	大小	格式	瀏覽次數
index.html		0Kb	HTML	158	檢視/開啟

在NCUIR中所有的資料項目都受到原著作權保護.

社群 sharing

資料載入中.....