LAUSR

AbstractSingular value decomposition (SVD) is an efficient method to suppress random noise in seismic data. The performance of noise attenuation is typically affected by choosing the rank of the estimated signal using SVD. That the rank is fixed limits noise attenuation especially for a low signal-to-noise ratio data. Therefore, we propose a modified approach to attenuate random noise based on structure-oriented adaptively choosing singular values. In this approach, we first estimate dominant local slopes, predict other traces from a reference trace using the plane-wave prediction and construct a 3D seismic volume which is composed of all predicted traces. Then, we remove noise from a 2D profile whose traces are predicted from different reference traces via adaptive SVD filter (ASVD), which adaptively chooses the rank of estimated signal by the singular value increments. Finally, we stack every 2D denoised profile to a stacking denoised trace and reconstruct the 2D denoised seismic data which are composed of all stacking denoised traces. Synthetic data and field data examples demonstrate that the proposed structure-oriented ASVD approach performs well in random noise suppression for the low SNR seismic data with dipping and hyperbolic events.