English  |  正體中文  |  简体中文  |  Items with full text/Total items : 889/889 (100%)
Visitors : 14578726      Online Users : 15
RC Version 7.0 © Powered By DSPACE, MIT. Enhanced by NTU Library IR team.
Scope Tips:
  • please add "double quotation mark" for query phrases to get precise results
  • please goto advance search for comprehansive author search
  • Adv. Search
    HomeLoginUploadHelpAboutAdminister Goto mobile version
    Please use this identifier to cite or link to this item: http://ccur.lib.ccu.edu.tw/handle/A095B0000Q/382

    Title: 一個基於堆疊式降噪自動編碼器的光體積變化描記圖專注辨識系統;A Photoplethysmography Attention Recognition System Based on Stacked Denoising Autoencoders
    Authors: 董泱佃;DONG, YANG-DIAN
    Contributors: 電機工程研究所
    Keywords: 光體積變化描記圖;堆疊式降噪自動編碼器;波形特徵;改良式基因演算法;Photoplethysmography;Stacked Denoising Autoencoders;Waveform Feature;Modified Genetic Algorithm
    Date: 2017
    Issue Date: 2019-07-17
    Publisher: 電機工程研究所
    Abstract: 本研究提出一個利用光體積變化描記圖(Photoplethysmography, PPG)作為數據並使用堆疊式降噪自動編碼器(Stacked Denoising Autoencoders, SDAE)分類的專注辨識系統。本研究中我們自行設計專注誘發實驗並同時量測數據,在專注辨識系統的研究內容方面主要有兩大方向,分別為傳統機器學習(Traditional Machine Learning, TML)及深度學習(Deep Learning, DL)。在傳統機器學習實驗架構過程可分為訊號前處理、特徵擷取、特徵選取和分類,訊號前處理為濾除雜訊和偵測重要點;特徵擷取的特徵主要有四大類:心率變異(Heart Rate Variability, HRV)、時域、波形和個體差異,總共224個;特徵選取的方式有三種:P值法(P-value)、基因演算法(Genetic Algorithm, GA)和改良式基因演算法(Modified Genetic Algorithm, MGA),用以降低架構複雜度及提升辨識正確率;分類使用倒傳遞類神經網路(Back Propagation Neural Network, BPNN)。在深度學習實驗架構過程可分為訊號前處理和分類,訊號前處理為濾除雜訊、波形擷取、頻譜處理和個體差異處理,總共12種波形;分類使用堆疊式降噪編碼器,堆疊方式分為類別堆疊(Stacked Categories)和編碼堆疊(Stacked Encoding)兩層次。結果顯示,在傳統機器學習方面,以倒傳遞類神經網路搭配基因演算法於留一人測試交叉驗證分類辨識正確率為92.23 %,特徵數為96;搭配改良式基因演算法於留一人測試交叉驗證分類辨識正確率為86.25 %,特徵數為20。在深度學習方面,以調整過的全階層堆疊式降噪自動編碼器於留一人測試交叉驗證辨識正確率為84.25%。
    This study proposes an attention recognition system using photoplethysmography(PPG) as data and using stacked denoising autoencoders(SDAE) for classification.In the study, we design experiment, to induce attention and measure data. The main development of attention recognition system has two major directions, namely, traditional machine learning(TML) and deep learning(DL). The TML experiment architecture processes can be divided into signal pre-processing, feature extraction, feature selection, and classification. Signal pre-processing is to filter out noise and detect important points. There are four main types of features to be extracted, including heart rate variability(HRV), time domain, waveform, and individual differences, resulting in a total of 224 features. There are three methods of feature selection, namely, P-value method, genetic algorithm(GA), and modified genetic algorithm(MGA) to reduce the complexity of the architecture and improve the accuracy of recognition. The classification uses back propagation neural network(BPNN) as classifier. The DL experimental architecture processes can be divided into signal pre-processing and classification. Signal pre-processing is to filter noise, extract waveform, have spectrum processing and have individual differences processing. A total of 12 kinds of waveforms can be extracted. The SDAE is used as classifier. The stacked method can be further divided into stacked categories and stacked encoding.The results show that with TML, the recognition accuracy using BPNN and GA is 92.23% and the number of features is 96 by using leave one person out cross-validation. The recognition accuracy using BPNN and MGA is 86.25% and the number of features is 20 using in leave one person out cross-validation. With DL, the recognition accuracy is 84.25% using adjusted all hierarchy SDAE with leave one person out cross-validation.
    Appears in Collections:[電機工程研究所] 學位論文

    Files in This Item:

    File Description SizeFormat

    All items in CCUR are protected by copyright, with all rights reserved.

    版權聲明 © 國立中正大學圖書館網頁內容著作權屬國立中正大學圖書館


    DSpace Software Copyright © 2002-2004  MIT &  Hewlett-Packard  /   Enhanced by   NTU Library IR team Copyright ©   - Feedback