Seismic Deconvolution Via Sparse Coding

High-resolution processing of seismic data plays a crucial role in gaining a clear understanding of subsurface geological structures and accurately describing hydrocarbon reservoir environments.Deconvolution processing,as a primary technique for highresolution seismic data processing,holds significant practical significance.If seismic data can be represented as a linear time-invariant system convolved with a wide-band reflectivity using a finite bandwidth seismic wavelet,the objective of deconvolution is to recover the wide-band reflectivity through inverse problem-solving methods.However,due to the limited bandwidth of seismic data,the deconvolution problem faces issues of non-uniqueness and uncertainty.Currently,the solution to the ill-posedness of deconvolution mainly involves adding various constraint terms to the objective function to narrow down the range of solutions and reduce uncertainty.However,the mainstream constraint methods still have several issues:(1)Assuming that the underground reflectivity satisfies a specific mathematical distribution to constrain the range of solutions,which cannot fully describe complex reservoir features,especially with inadequate representation capability for weak reflectivity.(2)Existing methods struggle to simultaneously address the issues of vertical resolution and lateral continuity,resulting in a trade-off situation.To address the aforementioned issues,this thesis introduces sparse coding techniques into the deconvolution problem to extract reflectivity features from the data,avoiding inaccuracies in conventional analytic regularization terms’ description of geological features.Simultaneously,a novel sparse representation method is developed to resolve the issues of vertical resolution and lateral continuity,thereby enhancing the resolution of deconvolution results.The main innovative contributions of this thesis are as follows:(1)The development of an error-constrained joint sparse coding seismic deconvolution method.In response to the limitation of conventional methods that rely on statistical models to construct regularization terms and struggle to adapt to complex reservoirs,this method assumes a consistent sparse coding between well-side seismic data and well log data.By jointly learning dictionaries and obtaining a joint representation of wide-band reflectivity and seismic data.The sparse coding is modified by correcting the error between observed seismic data and synthetic seismic records.The modified sparse coding is then used to reconstruct a new data-driven regularization term,which is combined with the deconvolution objective function,resulting in a novel seismic deconvolution method driven by both data and model.Model analysis and practical data evaluation demonstrate that this method significantly improves data resolution,particularly in representing weak signals.(2)The development of a sparse deconvolution method based on dual-sparse coding.Addressing the issue of neglecting lateral continuity in deconvolution processing,this thesis extracts lateral structural features from various frequency components of seismic data using 2D K-SVD and convolutional sparse coding.In the high-frequency component,the lateral continuity is further enhanced by incorporating gradient regularization in the process of convolutional sparse coding.Then,the lateral features of the 2D high-and lowfrequency components are used as constraints and added to the inversion objective function,improving the lateral continuity of the deconvolution method.By leveraging the advantages of high vertical resolution in well log reflectivity and strong lateral continuity in seismic data,a new deconvolution method is developed using an iterative approach,achieving high-resolution processing goals in both the vertical and lateral axes.