الفهرس 
TitleOnly 14 pages are availabe for public view
 |  | from | 89 |  |  |
| | | from | 89 | | |
Abstract
This study deals with three comparative pre-processing methods to extract features from sounds and then classify them to one of the three categories: normal, wheeze, or stridor. Features were extracted using three different techniques in separate ways to compare the effectiveness and to decide the best way to extract a set of statistical features from lung sound signals. The feature extraction methods used were Wavelet transform, Short Time Fourier Transform (STFT) and Mel-Frequency Cepstral Coefficients (MFCCs). The feature extraction methods produce three different matrixes which enter the classification phase; the sounds are categorized using four different classification techniques which include Artificial Neural Network (ANN), Support Vector Machine (SVM), K-Nearest Neighbor (KNN) and Naïve Bayes. The main aim of this research is to choose the best signal processing technique with the most suitable classifier from twelve different combinations among three distinctive feature extraction methods and four various classification techniques to categorize the lung sound disease especially in infants and children from a particular environment and this is considered one of the advantages of this research. Moreover, extra 146 wheezes were used to validate the usefulness of the classifiers. The purpose of this research is to decide the best feature extraction method with the most suitable classification techniques in infant lung sounds diagnosis. The classification performance of the classifier in both techniques used here is evaluated in terms of its ability to identify true positives (sick people correctly diagnosed as sick) and true negatives (healthy people correctly diagnosed as healthy), as well as to reject false positives (healthy people incorrectly identified as sick) and false negatives (sick people incorrectly diagnosed as healthy). The results revealed that the DWT has 100% accuracy with Artificial Neural Network (ANN) in the three categories (normal, wheeze and stridor) with no error result. Furthermore, MFCC has the best result in the wheeze detection phase with no error result when combined with Artificial Neural Network (ANN). The MFCC and DWT signal analysis techniques outperformed the STFT in all cases. In addition one of the findings of this research in wavelet as feature extraction method, we compare five different wavelet families (Scaling function) we announce that the db8 had a excellence with better accuracy than db10, sym10 and, bio orth 1.5. from each sound point of view in wheeze and stridor the ANN with wavelet combination had a perfect accuracy result, while in normal best accuracy in both ANN and SVM combined with Wavelet and MFCC, also with the MFCC combination KNN, although, the worst accuracy in all tests was in stridor with STFT and Naïve Bayes combination. However the STFT accuracy results was the lowest in all cases which was very disappointed. Finally this thesis is organized as follow a brief introduction and the aim of the work in chapter 1, in chapter 2 a review of literature and the related work are surveyed and the need for extending the surveyed techniques is established as well. Materials and methods are summarized and discussed in chapter 3. The results and discussions are detailed in a section in chapter 4. Finally in chapter 5, the research is concluded along with suggestions for some future extensions of the work.