Seismic denoising using curvelet analysis

Abstract

A curvelet is a new and effective spectral transform, that allows sparse representations of complex data. It has many applications in several fields, including denoising, wave propagation in disordered media and pattern recognition. This spectral technique is based on directional basis functions that represent objects having discontinuities along smooth curves. In this work we apply this technique to the removal of Ground Roll, which is an undesired feature signal present in seismic data obtained by sounding the geological structures of the Earth. In this methodology the original seismic data is decomposed by curvelet transform in scales and angular domains. For each scale the curvelet denoising technique allows a very efficient separation of the Ground Roll in angle sections. The precise identification of the Ground Roll pattern allows an effective erasing of its coefficients. In contrast to conventional denoising techniques we do not use any artificial attenuation factor to decrease the amplitude of the Ground Roll coefficients. We have estimated that, depending on the scale, around 75% of the energy of the strong undesired signal is removed.