跳到主要內容

臺灣博碩士論文加值系統

(44.221.73.157) 您好!臺灣時間:2024/06/20 12:07
字體大小: 字級放大   字級縮小   預設字形  
回查詢結果 :::

詳目顯示

我願授權國圖
: 
twitterline
研究生:陳信瑀
研究生(外文):CHEN,XIN-YU
論文名稱:基於捲積網路與長短期記憶模型之異常心率與心肺音偵測
論文名稱(外文):Abnormal Detection of Heart Rate and Lung Sounds Based on CNN and LSTM models
指導教授:張貴忠
指導教授(外文):CHANG,KUEI-CHUNG
口試委員:張哲誠游景盛
口試委員(外文):CHANG,CHE-CHENGYU,CHIN-SHENG
口試日期:2020-07-20
學位類別:碩士
校院名稱:逢甲大學
系所名稱:資訊工程學系
學門:工程學門
學類:電資工程學類
論文種類:學術論文
論文出版年:2020
畢業學年度:108
語文別:英文
論文頁數:40
中文關鍵詞:聲音辨識捲積神經網路深度學習長短期記憶網路
外文關鍵詞:Voice recognitionCNNDeep learningLSTM
相關次數:
  • 被引用被引用:0
  • 點閱點閱:133
  • 評分評分:
  • 下載下載:0
  • 收藏至我的研究室書目清單書目收藏:0
近年來醫學越來越電子化,可以透過各種儀器和設備來輔助醫生進行診斷,甚至有許多的論文透過資料分析來達到穩定診斷疾病的狀況,但是這些論文都應用在相對單純且沒有噪聲的環境。在本篇論文中,可能要面對穿戴式設備隨時會產生環境音與噪聲的狀況。為了解決上述的問題,本文提出導入深度學習來處理躁聲上面的問題,CNN是個很強大端到端分類器,可以學習如何在不同的環境和病人下也能有個相對穩定的結果。
傳統上,肺部呼吸聲音診斷主要是透過採樣後的特徵提取,進行一定的分析進而可以得到在某個閥值區間去判斷是某個疾病,但是導入了深度學習之後,可以透過資料集的學習,讓定義閥值區間這件事透過模型學習的特徵來推論。本論文提出了兩個模型。在輸入的特徵上,本文使用了MFCC的特徵來做CNN模型的輸入。方法一是使用單個時間段輸入的CNN模型讓我們能夠透過多個特徵提取來面對不同的環境和分類。方法二嘗試使用多個時間段輸入的CNN加上LSTM的模型,用來強化時間序上的概念並且用來分類異常呼吸音。
除了肺部呼吸音的分類之外,本文還特別提出了心率演算法來對心音特徵做採樣,並且提出透過計算出的心律,來自動標注病人此段心音是否異常的方法。
在心率演算法的準確度上,正常病人的心律誤差可以達到3以下,呼吸分類中,對於肺泡音和爆裂音在訓練時最高都曾經達到80%以上的精準度,而喘鳴聲精準度只能在50%左右。

In recent years, medicine has become electronic, and various instruments and equipment can be used to assist doctors in diagnosis. There are many researches that use data analysis to achieve stable environment for diseases diagnosis. In this thesis, we may have to face the situation where the wearable device generates ambient sound and noise at any time. In order to solve the above problems, this thesis proposes a deep learning approach to deal with the above problems. CNN is a very powerful end-to-end classifier that can learn how to have a relatively stable result in different environments.
Traditionally, medical sound diagnosis is mainly through feature extraction after sampling, and a certain analysis can performed to obtain a certain threshold range to determine whether it is a disease. However, after deep learning is introducing, you can learn the dataset to let the problem of defining the threshold range captured by the characteristics of the model learning. In the thesis, two deep learning models are proposed. The first approach is to use a CNN model input in a period to allow the system to face different environments and classifications through multiple feature extractions. The second approach attempts to use the Cascade CNN-LSTM model to use multiple time slots input to strengthen the concept of time sequence and use it for classification.
In addition, this thesis also specifically proposed a heart rate algorithm to sample heart sounds. It can automatically mark whether the patient's heart sounds are abnormal after calculate heart rate.
In the accuracy of the heart rate algorithm, the heart rate error of normal patients can reach less than three. In the respiratory classification, the highest accuracy of the normal and crackle sounds has reached more than 80% during training, while the accuracy of wheezing around 50%.

摘  要 i
Abstract ii
Chapter 1 1
1.1 MOTIVATION AND OBJECTIVES 1
1.2 THESIS ORGANIZATION 2
Chapter 2 Related Work 3
2.1 WELCH FOR CALCULATE HEART RATE 3
2.2 USE MFCC FOR FEATURE EXTRACTION TO ANALYZE LUNG SOUNDS 4
2.3 CNN COMBINE TIME-DOMAIN AND FREQUENCY-DOMAIN 6
2.4 CNN AUTOMATICALLY CLASSIFIES FETAL HEART RATE 7
2.5 CNN FOR LUNG SOUND AND HEART SOUND CLASSIFICATION 9
2.6 CASCADE CNN-LSTM MODEL FOR CLASSIFICATION SOUND 11
Chapter 3 Proposed methods 12
3.1 ARCHITECTURE OF LUNG ABNORMAL SOUND DETECTION SYSTEM 12
3.2 WELCH HEART ALGORITHM TO CALCULATE HEART RATE 14
3.3 BASIC CNN MODEL AND CASCADE CNN-LSTM LUNG ACOUSTIC MODEL 20
Chapter 4 Experimental Results 24
4.1 HEART DIAGNOSIS ANALYZING 26
4.2 RESPIRATORY DIAGNOSIS ANALYZING 29
4.3 HEART RATE ALGORITHM TO COLLECT DATA ANALYZING AUTOMATICALLY 31
4.3.1 HEART RATE ALGORITHM FOR AUTOMATIC LABELING 31
4.3.2 HEART RATE ALGORITHM FOR ABNORMAL DATA 32
4.4 BASIC CNN MODEL AND CASCADED CNN-LSTM MODEL ANALYZING 33
4.5 ENHANCED CNN MODEL AND CASCADE CNN-LSTM MODEL ANALYZING 36
Chapter 5 Conclusion and Future Work 38
References 39


[1]Vipul Jindal, Varun Agarwal, S. Kalaivani, “Respiratory Sound Analysis for De-tection of Pulmonary Diseases,” 2018 IEEE Applied Signal Processing Confer-ence (ASPCON), July.2019
[2]Nabila Husna Mohd Johari, Noreha Abdul Malik, Khairul Azami Sidek, “Dis-tinctive Features for Classification of Respiratory Sounds Between Normal and Crackles Using Cepstral Coefficients,”2018 7th International Conference on Computer and Communication Engineering (ICCCE), November.2018
[3]V. Tyagi,C. Wellekens,“On desensitizing the Mel-cepstrum to spurious spectral components for robust speech recognition,” Proceedings. (ICASSP '05). IEEE In-ternational Conference on Acoustics, Speech, and Signal Processing, 2005., May.2005
[4]Kimitake Ohkawa,Masaru Yamashita,Shoichi Matsunaga,“Classification Be-tween Abnormal and Normal Respiration Through Observation Rate of Heart Sounds Within Lung Sounds,” 2018 26th European Signal Processing Confer-ence (EUSIPCO), December.2018
[5]Laszló Tóth, “Combining time- and frequency-domain convolution in convolu-tional neural network-based phone recognition,” 2014 IEEE International Con-ference on Acoustics, Speech and Signal Processing (ICASSP), July.2014
[6]Jianqiang Li, Zhuang-Zhuang Chen, Luxiang Huang, Min Fang, Bing Li, Xianghua Fu,“Automatic Classification of Fetal Heart Rate Based on Convolu-tional Neural Network,” IEEE Internet of Things Journal,Vol.6,Issue: 2 , April.2019.
[7]Valentyn Vaityshyn, Hanna Porieva, Anastasiia Makarenkova,“Pre-trained Con-volutional Neural Networks for the Lung Sounds Classification,” 2019 IEEE 39th International Conference on Electronics and Nanotechnology (ELNANO), August. 2019
[8]Funda Cinyol, Huseyin Emre Mutlu, Ugur Baysal, “Classification of lung sounds with convolutional neural network,” 2017 21st National Biomedical Engineering Meeting (BIYOMUT), October.2018
[9]Renyu Liu, Shengsheng Cai,Kexin Zhang, Nan Hu, “Detection of Adventitious Respiratory Sounds based on Convolutional Neural Network,” 2019 Internation-al Conference on Intelligent Informatics and Biomedical Sciences (ICIIBMS), February.2020
[10]Qiyu Chen, Weibin Zhang,Xiang Tian , Xiaoxue Zhang, Shaoqiong Chen, Wenkang Lei,“Automatic heart and lung sounds classification using convolu-tional neural networks,” 2016 Asia-Pacific Signal and Information Processing Association Annual Summit and Conference (APSIPA), January.2017
[11]Chuya China, Dipjyoti Bisharad, Rabul Hussain Laskar, “Automatic Classifica-tion of Indian Languages into Tonal and Non-tonal Categories Using Cascade Convolutional Neural Network (CNN)-Long Short-Term Memory (LSTM) Re-current Neural Networks,” 2018 International Conference on Signal Processing and Communications (SPCOM), July.2018
[12]Koki Minami, Huimin Lu, Hyoungseop Kim, Shingo Mabu, Yasushi Hirano, Shoji Kido, “Automatic Classification of Large-Scale Respiratory Sound Dataset Based on Convolutional Neural Network,”2019 19th International Conference on Control, Automation and Systems (ICCAS), January.2020
[13]N. Jakovljević, T. Lončar-Turukalo, “Hidden Markov Model Based Respiratory Sound Classification,” ICBHI 2017: Precision Medicine Powered by pHealth and Connected Health, November.2017
[14]Ana Oliveira, Alda Marques “Respiratory sounds in healthy people: A systemat-ic review,” Respiratory Medicine Volume 108, Issue 4, April.2014
[15]Sergey Ioffe, Christian Szegedy, “Network In Network,” Neural and Evolution-ary Computing Computer Vision and Pattern Recognition Machine Learning, March.2014
[16]Min Lin, Qiang Chen, Shuicheng Yan, “Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift,” Proc. 32nd Int. Conf. Mach. Learn. (ICML), March.2015

電子全文 電子全文(本篇電子全文限研究生所屬學校校內系統及IP範圍內開放)
QRCODE
 
 
 
 
 
                                                                                                                                                                                                                                                                                                                                                                                                               
第一頁 上一頁 下一頁 最後一頁 top
無相關期刊