Automatic noise attenuation based on clustering and empirical wavelet transform

Abstract

Strong noise in seismic data seriously affects many steps in seismic data processing and imaging. While most traditional methods depend on carefully tuned input parameters by human, the method proposed in this paper is an automatic noise attenuation algorithm to facilitate a fast preprocessing of massive prestack seismic data. In the proposed algorithm, the non-stationary seismic data is first adaptively decomposed into several empirical components adaptively via empirical wavelet transform (EWT) according the frequency contents in the data. Then, the first component is selected to stand for the useful signals. To deal with the residual noise in the roughly estimated signal from the previous step, we propose a clustering based thresholding method. The most dominant signals are detected via a simple clustering step and other components are damped with an adaptive percentile threshold. The two steps refer to a new automatic algorithm to denoise the seismic data with high fidelity. We demonstrate the performance of the proposed method via both synthetic and field data examples.