Seismic data preprocessing significantly benefits from advanced sparse representation and domain transformation techniques to enhance denoising, wavefield separation, and data reconstruction. This study introduces a novel approach utilizing a deep learning framework for discrete sparse representation of seismic data. Our method utilizes a Swin Transformer-based encoding-decoding framework, which combines the hierarchical structures of CNNs with the self-attention mechanism of Transformers, to model both local and global information efficiently. This integration enables the precise characterization of seismic reflection events and the reconstruction of seismic records from a constructed sparse feature space. The proposed model has been rigorously tested on both simulated and field datasets, demonstrating its robustness, and potential provides superior decomposition of seismic data.

In seismic exploration, the multiple suppression is crucial for accurate subsurface imaging and resource identification. Internal multiples, generated by multiple reflections at impedance interfaces, act as interference signals that can mislead resource exploration. Compared to traditional methods, the conventional Marchenko multiple elimination (C-MME) method allows for the direct extraction of primary waves from seismic records without requiring a macro velocity model or predictive-subtraction, thereby preserving effective signals. However, challenges such as low signal-to-noise ratios (SNR) and high-density sampling requirements have hindered its application to field land seismic data. To address these challenges of C-MME in field seismic data processing, we propose a compressive sensing-based Marchenko multiple elimination (CS-MME) method, which incorporates efficient denoising, reconstruction, and deconvolution capabilities. In this study, the CS-MME method has demonstrated exceptional performance in processing field land seismic data, successfully overcoming the aforementioned challenges. marking the first successful implementation of Marchenko multiple elimination on field land data.

This study addresses the challenges in seismic interferometry (SI), particularly the deghosting of virtual-shot gathers affected by nonphysical reflections in marine seismic exploration. SI’s effectiveness largely depends on the distribution of seismic sources, which in field situations often leads to complex wavefield interactions and nonphysical reflections, adversely impacting data processing. Traditional methods are limited with these nonphysical reflections, and similarity-based approaches have shown potential in overcoming this challenge. In this paper, our study introduces an innovative deghosting framework based on Regularization by Denoising (RED). This approach effectively mitigates the influence of nonphysical reflections and enhances the accuracy of deghosting. We propose a simultaneous multi-shot deghosting framework that integrates RED, focusing on optimizing data similarity and computational efficiency. This is complemented by a novel denoising engine based on an enhanced BM3D method, specifically tailored for virtual-shot gathers. The efficacy of our framework is demonstrated through synthetic and field data, showcasing significant improvements in seismic data processing and offering a new paradigm for the application of seismic interferometry. This research contributes not only to the advancement of deghosting techniques but also to broadening the scope of seismic exploration.

Wave equation-based inversion, such as Least-Squares Reverse Time Migration (LSRTM), is a powerful seismic imaging technique. However, the performance of LSRTM in constructing high-resolution subsurface models is closely tied to the design of the gradient. The commonly usedL2gradient, derived within theL2space framework, often leads to insufficient illumination and slow convergence in inversion. To confront these issues, we introduce a novel Sobolev gradient based on Sobolev space and propose a novel inexact scaled-Sobolev gradient projection LSRTM (ISGP-LSRTM), which can correct the conventionalL2gradient, accelerates convergence and mitigates illumination deficiencies. Specifically, we construct a scaled-Sobolev gradient operator to optimize the direction of the conventional adjoint gradient, addressing the low-resolution caused by insufficient gradient information during iterations. Futhermore, by incorporating an inexact backtracking strategy, a proximal backtracking projection operator based on scaled- Sobolev gradient is constructed to refine the traditional update process in LSRTM. Numerical tests on multiple models, including the wide-angle Valley model, the truncated Marmousi model, and the fault-rich truncated Sigsbee2B model, demonstrate the superior performance of the ISGP-LSRTM inversion. The proposed approach significantly improves imaging illumination and event continuity, achieves faster convergence in iterative curves, and effectively enhances migration imaging quality.

The velocity analysis requires a significant degree of automation due to its time consuming and labor-intensive nature. Recently, clustering algorithms, an unsupervised learning method, have been used in velocity analysis to achieve automated velocity picking. Unlike supervised learning methods, this approach does not require a large amount of high-quality labeled data and high training costs. Meanwhile, existing clustering-based velocity analysis requires high-signal-to-noise ratio inputs and is particularly sensitive to multiple reflection noise, which severely limits the effectiveness of automatic velocity analysis (AVA) methods. To provide an adaptive AVA method for seismic data that contains multiple reflections, we introduce physical prior knowledge constraints within the framework of clustering algorithms. In particular, we transformed the peak picking problem in the velocity spectrum into a classification problem between primary and multiple reflections. Meanwhile, three physical attributes, velocity, amplitude, and primary similarity, are used as physical prior information for the clustering algorithm to help solving the classification problem. The synthesized and field data examples have proven the effectiveness of the proposed method.

This article proposes a new surface wave suppression method based on nonsubsampled Shearlet transform (NSST). First, we use a frequency wavenumber (FK) Filter to extract surface wave noise from seismic data. We then apply a Shearlet transform to both the original data and the isolated surface wave, facilitating the creation of distinct Shearlet domain subbands characterized by diverse scales and directions. Next, we calculate the similarity coefficients between each corresponding subband of the original data and separated surface wave and use the standard deviation as the criterion for subband selection to enable adaptive selection. Then, mean thresholding and Bayesian separation algorithms are applied to suppress the surface wave components in the subbands. Finally, we integrate the selected subbands and perform an inverse NSST to reconstruct the surface wave-free seismic data. Replacing the traditional subband selection method with the adaptive method can effectively improve the efficiency of surface wave suppression, and the combination of mean value thresholding and Bayesian separation algorithm suppresses the surface wave noise in the subbands, resulting in high-quality reconstructed data.

Model-based reconstruction offers an effective framework for solving diverse inverse problems in geophysical imaging, and least-squares reverse time migration (LSRTM) can be regarded as a model-based reconstruction method. While LSRTM can produce higher resolution compared to reverse time migration (RTM), it remains prone to challenges such as noise interference, computational complexity, and problems, e.g., artifacts and incomplete illumination. This study adopts the regularization by denoising (RED) strategy to alleviate these issues and further improve imaging resolution and inversion efficiency. The RED technique is notably flexible and requires just a single denoising engine. In our case, it is a filter constructed based on the curvelet transform (CT), which decomposes the imaging into wavelet coefficients of different scales and directions and applies threshold processing to achieve noise suppression. Subsequently, the CT denoising engine is integrated into the explicit RED term, which is formed by the inner product of the imaging result and its denoising residuals. The inversion result is obtained by solving the objective function using the alternating directions method of multipliers (ADMM). The effectiveness of RED-LSRTM is initially tested with the simple flat model, followed by an assessment of its imaging performance on large-angle reflection layers with a layered model. Finally, the RED-LSRTM is assessed in complex geological structures using a salt model.