配置有訊號增強與分類方法之智慧型聽診器

Abstract

本論文提供一運用在肺音之聽診器系統，其整合有應用麥克風陣列進行之訊號增強技術與智慧型分類方法。由麥克風陣列收集到之肺音訊號可利用時間對齊之方法(GCC 與Blind Beamforming)將各聲道間的延遲時間消除並增強訊號的品質。在進行分類之前，可以從被增強後的肺音訊號進行前處理，而從中萃取出一些時間域及頻率域的音訊特徵。建立在支持向量機(SVM)分類學習機器基礎下之適應性多樣特徵子空間方法(AMFES)可應用在篩選就具影響力之特徵上，並以此執行分類。支持向量機提供分類器來對不同病徵之肺部疾病進行分類。在訓練分類器的階段中，可透過支持向量機對訓練音訊建立其特徵空間模型，而在接下來的測試階段，可計算測試音訊的特徵來與訓練音訊之特徵空間模型進行比對。適應性多樣特徵子空間方法可以減少特徵空間的維度並改善訊號處理效率，而利用麥克風陣列進行訊號增強之方法可以增加分類的正確率。在分類的實驗結果中顯示了本聽診器系統對於各種肺病音訊具有相當高的分類準確率，其中這些準確率都高達八成以上。

This thesis proposed a stethoscope system that integrates signal enhancement technique based on a microphone array and intelligent classification method for lung disease symptoms. Lung sound signals acquired from the array are enhanced by using time alignment methods based on the generalized cross-correlation (GCC) and beamforming algorithm. Audio features in the time and frequency domains are extracted from the strengthened signals as preprocessing prior to classification. The adaptive multiple feature subset method (AMFES), based on the support vector machine (SVM), is applied to screening the beneficial features to execute classification. The SVM method provides a classification method to classify categories corresponding to lung disease symptoms. A training phase is required to establish a feature space model by above SVM method, followed by a testing phase in which the audio features of the test data are calculated and matched to the feature model. In particular, the AMFES method can reduce the dimension of feature set to improve the processing efficiency and the array signal enhancement can increase the detecting rate of classification. Experiments were conducted to demonstrate that our stethoscope system has high detecting rate for classification of lung disease, wherein these detecting rates are more than 80% for each lung sound case.