Abstract
An intelligent stethoscope system that makes use of feature extraction and classification of sound signals for lung conditions is presented in this thesis. First, the method of harmonic/percussive sound separation (HPSS) separate a monaural lung sound signal into harmonic and percussive components and then calculate the features on separated signals which is helpful to classification. Each separated component, the acoustic features are extracted by using Mel-scale frequency cepstral coefficient (MFCC). Then, audio signals are converted into spectrograms to extract texture features from time and frequency image features which are then used for lung conditions prior to classification. The texture features are based on gray level co-occurrence matrix (GLCM) and local binary patterns (LBP) which have been proved to be a promising method for image classification research. In order to find the most discriminative features the method are used with sequential forward selection (SFS) and k-nearest-neighbor classifier (KNNC) as the classifier. Finally, based on these features, three classification approaches are employed, including k-nearest-neighbor classifier (KNNC), support vector machine (SVM), and hidden markov models (HMM). The experimental results reveal that the performance of features with above technique is satisfactory, for lung conditions diagnosis.