An automatic system for crackles detection and classification

In this paper, an automatic system for crackles detection and classification is developed. The proposed system is preceded by a stationary-nonstationary filter based on the wavelet packet transform (WPST-NST) which isolates the crackles from the vesicular sounds. The crackle analysis consists of three major steps: Firstly, a denoising filter is applied to suppress the stationary residual noise in non-stationary signal. Secondly, a new version of crackles detection based on the fractal dimension is presented. The avantage of this method is to detect crackles even they are week or overlapped. Finally, the extracted crackles am classified into fine or coarse crackles. The time-frequency analysis, the Prony model and matched wavelet analysis techniques are tested and compared in this paper.