Features extraction of heart sounds using timefrequency distribution and mel-frequency cepstrum coefficient
Heart sounds analysis can provide lots of information about heart condition whether it is normal or abnormal. Heart sounds signals are time-varying signals where they exhibit some degree of non-stationary. Due to these characteristics, therefore, two techniques have been proposed to analyze them. Th...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Published: |
2006
|
| Subjects: | |
| Online Access: | http://eprints.utm.my/id/eprint/2724/1/MasnaniMohamedMFKE2006.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Heart sounds analysis can provide lots of information about heart condition whether it is normal or abnormal. Heart sounds signals are time-varying signals where they exhibit some degree of non-stationary. Due to these characteristics, therefore, two techniques have been proposed to analyze them. The first technique is the Time-Frequency Distribution using B-Distribution, used to resolve signal’s components in the time-frequency domain and specifies the frequency components of the signal that changing over time. Another proposed technique is the Mel- Frequency Cepstrum Coefficient, used to obtain the cepstrums coefficients by resolving signal’s components in the frequency domain. An experiment is presented to extract features of heart sounds using both mentioned techniques and compare their performances. Both techniques are discussed in details and tested against ideal simulations of 50 heart sound signals including normal and abnormal signals. All simulations are done using Matlab software except for MFCC where it has used the Microsoft Visual C++ software. A brief description of SVD is included to the technique using time-frequency distribution. Also, a brief description of Neural Network is used to verify and to compare the performances results of the two techniques with regard to the values of hidden node, learning rate and momentum coefficient. The results showed that performance of the TFD can be achieved up to 90% whereas MFCC is only 80%. Therefore, the TFD technique is chosen as the best technique to analyze and to extract features of the non-stationary signals such as the heart sounds signals |
|---|