心音圖壓縮與辨認系統設計

Abstract

一個理想的居家照護系統應該建立在一個可攜式的平台上，能夠在此平台上處理各種的生醫訊號，並且可以提供即時的影音處理。在本論文中，我們使用了一個消費型的數位相機，並在此相機上作生醫訊號的即時處理。而且，我們利用相機執行正常功能（例如作JPEG檔案的壓縮，或是MPEG-1/MPEG-4錄影功能）的時候，可以同時進行心音圖的壓縮與辨認計算。 我們將心音圖之壓縮辨認演算法實現在嵌入式軟體平台中，這個平台建構在一個數位相機上。以小波轉換為基礎的心音圖壓縮演算法不但可以做到壓縮檔案的效果並且可以降低不必要的雜訊。以支持向量機為基礎之心音圖辨認演算法可以成功的將九種心音辨認出來。 我們所使用的心音圖資料庫包含42個心音圖測試資料，測試資料可以分為訓練組及測試組，而可以達到每次處理1.85秒的心音訊號僅需花0.16秒的時間，在這個速度之下足夠提供即時的系統運算，而且可以達到100%的辨識率。

An ideal homecare system needs to process multiple diagnostic signals using a portable battery-driven device and provide a real time audio/video interface for telemedicine applications. In this thesis, we present a low cost real-time homecare system based on a commercial digital camera platform. In addition to the typical multimedia recording functions of a digital camera, such as JPEG and MPEG-1/MPEG-4 recording, the proposed system can further process and compress phonocardiogram (PCG) signal concurrently. We include the PCG signal compression and analysis function into a robust embedded software platform for digital camera systems. The proposed wavelet transform-based PCG compression algorithm efficiently reduces the undesired noises as well as signal data size. The PCG analysis algorithm based on support vector machines successfully identify nine types of heart disease. To demonstrate the performance of the proposed system, we have tested with the PCG database which includes 42 samples of PCG signals. The samples are partitioned into two sets for training and test purposes. The average processing for 1.85 seconds of PCG signal time on the camera is 0.16 seconds, which is faster enough for the real time applications. The MPEG-4 video can be recorded simultaneously in the DSP subsystem and the recognition rate is 100%.