Separation of seismic multiple reflection-refraction based on morphological component analysis with high-resolution linear Radon transform

Investigating coherent noise attenuation is a continuing concern within seismic signal processing. As a common type of linear coherent noise, the multiple reflected refraction (MRR) occurs in seismic records where low-velocity strata overlie high-velocity strata, such as deserts, Loess Plateaus, etc. Due to its high velocity and strong energy, MRR seriously distorts the reflections and affects the interpretation. MRR has linear morphological characteristics on the shot gathers. Thus, a linear Radon transform with a surgical mute is usually applied to suppress MRR. However, significant numbers of shallow reflections are removed unintentionally in the Radon domain when reflections overlap MRR. Therefore, a robust method that reduces the leakage of reflection energy is required. We propose a novel method to attenuate MRR by examining the morphological difference between MRR and useful signals in the t-p domain. MRRs are oblique linear events on the shot records, while useful signals are quasi-hyperbolic events under the assumption of horizontal layers. After the high-resolution linear Radon transform, MRRs are ideally mapped into point features while the useful signals are aligned with narrow curve bands in the ellipse. To better separate them in the t-p domain, we employ the morphological component analysis (MCA) theory and select the 2D stationary wavelet transform and the Shearlet Transform as sparse representation sub-dictionaries of point features and curved features, respectively. After MCA separation, we apply the inverse Radon transform to the separated MRRs and subtract them from the original seismic data. Subtraction can better preserve the amplitude of reflections. We use both synthetic data and field data to illustrate the effectiveness of our method and demonstrate that making full use of the above morphological differences can improve the results of MRR attenuation.