肺音信号のスパース表現と断続音分離への応用

概要

論文の詳細を見る
This paper reports on preliminary analysis of pulmonary auscultation sounds via sparse representation of the sound signals for diagnostic classification. The pulmonary adventitious sounds associated with respiratory disorder are typically classified as continuous or discontinuous. The sound signals can therefore be represented as mixture of a small number of continuous and discontinuous signal components, which is called the sparse representation. A simple overcomplete basis set of the Fourier sinusoids and the Daubechies wavelets is suitable for the sparse representation of breath sounds and crackles. It is experimentally shown that the crackles can be separated from breath sounds using the sparse representation even if they overlap each other in time-frequency domain. This separation is demonstrated to be highly robust against random noise.
長崎大学の論文