請用此 Handle URI 來引用此文件:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/67488
標題: | 時頻應用於醫學信號分類器設計與聲音信號壓縮 Time-Frequency Applications for Medical Signal Classifier Design and Compressive-Based Vocal Signal Compression |
作者: | Chi-Lin Kuo 郭起霖 |
指導教授: | 丁建均 |
關鍵字: | 時頻分析,分類器,聲音信號壓縮,壓縮感知, Time-Frequency Analysis,Classifier,Vocal Signal Compression,Compressive Sensing, |
出版年 : | 2017 |
學位: | 碩士 |
摘要: | 對於信號處理領域來說,時頻分析一直是一項重要的分析工具。在這篇碩士論文中,我們將會運用時頻分析方法去實作出兩項有關信號處理的應用。第一項應用是針對於聲音訊號去做壓縮;第二項應用是設計一個分類器可以有效將兩組不同性質的醫學平衡信號種類分開。
對於聲音訊號壓縮的研究中,我們藉由Matching pursuit和Compressive sensing 的概念而達到壓縮的目的。對於Matching pursuit來說,運算時間與效能會相互影響使得我們必須做出衡量,但由於運算時間過長的問題導致這領域的許多研究學者不喜歡使用Matching pursuit,因此我們提出一個演算法藉由修改Matching pursuit與結合Compressive sensing的概念來達到在不影響效能的情形下還能有效降低運算時間的成效。模擬結果將顯示相較於以往的Matching pursuit來說,對於運算時間上我們確實有顯著的貢獻,且對於壓縮結果來說,比現今廣為人知的MP3壓縮格式還能達到更低的資料量。 在有關肢體平衡的醫學領域上,Myelopathy和Radiculopathy是兩種影響肢體平衡的疾病,我們藉由Signal processing與Generalized spectrogram兩項與時頻分析有關的手法去計算出肢體平衡信號在時間與頻率上的參數特徵,結合統計學上的Z-score與K-value對這些統計參數設計出能用於分類兩類疾病的分類器,更進一步引入KNN演算法對這個分類器做出準確率上的改良,模擬結果將顯示出我們設計出來的分類器與現今廣為人知的SVM分類器比較下,是具有較佳的準確率的。 Time-frequency analysis is an important tool for signal processing. In this thesis, we use the concepts of time-frequency analysis to implement two topics, one is for vocal signal compression, another is for medical signal classifier design. In our proposed method for vocal signal compression, we use the concepts of the matching pursuit algorithm and the compressive sensing to implement it. In the matching pursuit algorithm, there is a trade-off between the computation time and the performance. The problem of the computation time often makes people do not want to use it. Hence, we propose the method which modify the matching pursuit algorithm and the compressive sensing to reduce the computation time and do not affect the performance. The simulation results will show that we have a significant improvement in operational efficiency and the better performance than some audio compression format. In medical field of postural steadiness, cervical myelopathy and cervical radiculopathy are common disease that affect the balance of limb. We use the signal processing methods and the generalized spectrogram to calculate a variety of time and frequency domain measures of postural steadiness between myelopathy and radiculopathy under both eyes-open and eyes-closed. We utilize that information to observe the difference between myelopathy and radiculopathy. We design the statistical method which combine the concepts of the Z-score and the K-value to determine that the subject belongs to which type and use KNN (K Nearest Neighbor) algorithm to judge fuzzy case which statistical method can not judge. The simulation results will show that we have a better performance than SVM classifier. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/67488 |
DOI: | 10.6342/NTU201702323 |
全文授權: | 有償授權 |
顯示於系所單位: | 電信工程學研究所 |
文件中的檔案:
檔案 | 大小 | 格式 | |
---|---|---|---|
ntu-106-1.pdf 目前未授權公開取用 | 3.08 MB | Adobe PDF |
系統中的文件,除了特別指名其著作權條款之外,均受到著作權保護,並且保留所有的權利。