A new type of wavelet de-noising algorithm for lung sound signals

Abstract

With the development of digital auscultation, the computer-based intelligent auscultation of respiratory sounds has drawn attention of researchers. However, the noises of acquired signals influence the further analysis of lung sound, so there is necessity to develop the noise reduction algorithm of lung sound signals. In this paper, a new type of noise reduction is proposed. The original signals are decomposed into 7 layers by wavelet transform. The locations of lung sound part are obtained in the sub-signals by the mean values of autocorrelation coefficients. The noises between lung sound parts are reduced by setting zero directly. The noises in the lung sound parts are filtered by a Chebyshev type I band-pass filter. The de-noising results are judged by two means. One is the subjective judgement of internal physicians and the de-noising effect is accepted by doctors without distortion. The other is the classification result of sound types in the further research by BP neural network and the classification accuracy can reach 85%.