| Summary: | 碩士 === 義守大學 === 電機工程學系 === 100 === In this paper, we write a program of heart sounds segmentation using MATLAB software, and extract features from a single heartbeat cycle that has been segmented. The heart sounds for segmentation and feature extraction are recorded by the stethoscope that sold by 3M Company. We compare five methods that are used to process heart sound signals, and confirm that which method has a higher accuracy rate in the implementation of heart sounds segmentation. The 5 methods are Envelope, Short Time Fourier Transform, Continuous Wavelet Transform, Discrete Wavelet Transform and Hilbert–Huang Transform, respectively. The heart sounds process by these five methods and pass through a final-selection procedure to identify each first heart sound of heart sound signals, and then we can segment heart sounds into a single heartbeat cycle. The final-selection procedure can automated segment heart sounds into a single heartbeat cycle by setting of the Threshold and the minimum distance between Peaks. After heart sounds process by HHT and implement the final-selection procedure, we got the highest accuracy rate of heart sounds segmentation is 83.39%. Then we compared the two methods of features extraction of a single heartbeat cycle. In order to confirm which method of feature extraction is better, we use Cross-Correlation to observe whether these features are representative. The result is that, the feature of a single heartbeat cycle that obtained by the second method of feature extraction with the sixth detail coefficients is more representative.
|
|---|
