Estimation And Applications Of Local Dips In Seismic Data Processing

Raw seismic data are often contaminated with random noise and coherent noise such as ground roll.The noise seriously affects the seismic data quality and subsequently the processing and interpretation of seismic data.So,random noise attenuation and coherent noise separation always play important roles in seismic processing.Dip angles of seismic events are widely used for different tasks in seismic data processing,including random noise attenuation,coherent noise separation,imaging,structure-oriented filtering and seismic data reconstruction.Besides,estimation of seismic local dips provides important information for seismic interpretation,such as seismic fault detection and.horizon picking.The study of robust local-dips estimation method,and then attenuating random and coherent noise based on the dip information,have important significance in seismic data processing.Based on the idea of single-dip signal decomposition,a robust multi-dips estimation method is proposed,and applications of dip information in seismic random noise attenuation and ground rolls separation are studied.Based on the idea of multidirectional component analysis(MDCA),a local-dips estimation method is proposed,which can provide robust local-dips estimation results for noisy seismic data,and at the same time can attenuate random noise effectively.There are two main problems for traditional dip estimation methods,the first one is about crossing signals.The second problem is that most current algorithms typically lack in robustness to noise.The proposed local-dips estimation method integrates the denoising process and the dip estimation process together,and builds an optimization model with both the denoised single-dip signals and the estimated dips as optimization variables,so as to attenuate random noise and estimate dips simultaneously.Applying the MDCA method to synthetical and real field data sets indicate that,compared with traditional dip estimation methods,the proposed method is more robust to noise,and is applicable to multi-dips estimation problem.The denoising results of the proposed method is better then the denoising results of the traditional frequency-space and time-space predictive filtering method.A gradient vector rank-one regularization model(GVRO)is proposed,and applications to seismic random noise attenuation and coherent noise separation show its effectiveness.Low rank matrix approximation can only be used when a suitable transformation exists,which should guarantee the transformed data matrix has low rank.By decomposing a local seismic data into several single-dip components,the gradient vector matrices for the component signals are built,which should be approximately rank one matrices.Then,an optimization model is built and a fast algorithm is developed.Applying the GVRO method to synthetical and real field data sets indicate that random noise in seismic data can be attenuated effectively.When coherent signals exist and have dip difference,the GVRO method can be used to separate coherent signals with different dips.By introducing a sampling matrix to the optimization model,the GVRO method can be used for seismic data restoration.Based on dips difference,methods are proposed for ground rolls separation,which attenuate ground rolls effectively,and at the same time protect reflections.Two methods are developed for ground rolls separation.The first method is based on gradient flow regularization(GFR).Firstly,gradient flows for part of the reflections and part of the ground rolls are calculated via the traditional structure tensor method,which indicate the dip difference between the reflections and the ground rolls.Then,taking the gradient flows as regularization terms,an optimization model is built.Based on the specific structure of difference matrix,a fast algorithm is developed via the discrete Fourier transform.Applying the GFR method to synthetical and real field data sets indicate that ground rolls in seismic data can be attenuated effectively,and the reflections can be protected well.The lateral coherence of restored reflective signals is also enhanced.Taking both the dip difference and spectrum difference into consideration,a second method for ground rolls separation is proposed,which is based on incoherent dictionary learning(IDL).Sparse coding dictionaries for the ground rolls and the reflections are learned from the data.Atoms of the two dictionaries should have very different wave form,so,the two dictionaries should have low coherence.The proposed IDL model is solved via incoherent dictionary learning algorithm and the sparse optimization algorithm.Applying the IDL method to synthetical and real field data sets indicate that ground rolls can be attenuated effectively,and the low frequency energy of the reflections can be protected well.The random noise can be attenuated simultaneously.