Seismic random noise attenuation using FDOC-seislet transform and threshold for seismic data with low SNR
Zhang YaChen,Liu Yang,Liu Cai,Wu Shang
DOI: https://doi.org/10.6038/cjg2019M0062
2019-01-01
Abstract:In seismic exploration, seismic data are essentially time-varying, not only effective events display time-varying characteristics, but also complex surface and subsurface conditions and deep exploration environment always create time-varying and nonstationary random noise. Industrial standard prediction filter in the frequency-space (f-x) domain can only suppress stationary random noise under the assumption of stationary plane waves, for which cutting data into overlapping windows is a common method to handle nonstationary problem. However, such an approach cannot produce good results. On the other hand, random noise attenuation methods without prediction cannot provide good adaptability for different data. It is important to develop nonstationary random noise suppression methods that are suitable for time-varying characteristics of seismic data. Compressive sensing is a new sampling theory, which has found different applications in seismic data processing, for example, missing data interpolation and noise attenuation. In this paper, we analyze the problem of seismic random noise attenuation under compressive sensing framework and establish the objective function for threshold denoising. Aiming at the compressible characteristics of time-varying seismic signal, we solve the offset continuation equation by using finite difference algorithm to construct a finite-difference offset-continuation operato. Then a new fast sparse transform, FDOC-seislet transform, is proposed under seislet transform framework, which provides a highly sparse representation of seismic signals. According to the feature that nonstationary random noise is less compressible, we propose a new iterative shaping denoising method, a kind of generalized iterative shrinkage threshold methods, which is able to recover the time-varying signal under strong random noise environment, even when the adjoint operator of sparse transform cannot be calculated. Compared with conventional denoising methods, the examples of synthetic and field data demonstrate that the random noise attenuation method based on FDOC-seislet transform and iterative shaping can protect complex seismic signal and attenuate random noise even with strong amplitudes.