2026· IEEE Transactions on Geoscience and Remote Sensing· Vol 64, pp. 5919821-5919821· 0 citations· 39 references
Abstract
The limited seismic resolution significantly degrades thin beds identification and reservoir characterization. Conventional deconvolution methods often rely on manual parameter tuning, while supervised deep learning (DL) methods require extensive high-fidelity labels. Although some unsupervised methods reduce the dependence on labels, they often lack geophysical interpretability. In addition, most DL-based deconvolution methods assume a known wavelet, which limits their applicability in practical scenarios. To address these challenges, we propose a physics-informed blind deconvolution workflow for seismic resolution enhancement based on zero-shot learning. Specifically, a three-stage progressive optimization strategy is proposed. An initial wavelet is first estimated by amplitude spectrum constraints and phase scanning. A dual-stream time–frequency network (DSTFNet) is then warmed up with the wavelet fixed to obtain a reliable initial reflectivity. Finally, the wavelet and the network parameters are alternately updated to achieve joint refinement of the wavelet and the reflectivity. In addition, we construct a joint loss function consisting of a physics-constrained Huber loss, and an L1 sparsity loss, which enhances physical interpretability and robustness to noise. Experiments on synthetic and field data demonstrate that the proposed method exhibits strong generalization and robustness.
Enhancing seismic resolution is crucial for detecting thin geological layers. Traditional physics-driven methods are generalizable but rely heavily on physical approximations, accurate wavelet estimation, and stationary assumptions. Data-driven approaches, such as deep learning (DL), offer model-independent mappings fr...
Seismic data are widely used in subsurface energy exploration. However, several factors affect seismic data quality. Seismic data reconstruction is essential when acquisition is incomplete because missing sources or receivers produce subsampled shot-gathers with incomplete trace coverage that degrades data quality an...
L. Rodríguez-López, Paul Goyes-Peñafiel, L. Galvis et al.· The Leading Edge· 0 citations
Seismic data are inevitably contaminated by random noise arising from various sources, which seriously degrades the quality of subsequent data processing, inversion, and interpretation. Deep learning has emerged as an effective approach for seismic random noise attenuation. However, these methods are often susceptible...
Yun-Hong Wang, Rui-Hua Chen, Jun Lin et al.· IEEE Geoscience and Remote S...· 0 citations
Seismic data augmentation is critical for training robust deep learning (DL) models for fault detection and characterization, yet conventional approaches remain inadequate. Geometric transforms, e.g., flips or rotations, and noise injection fail to capture realistic seismic textures, while physics-based forward modelin...
Tolulope Q. Agbaje, N. Pham, A. Sansal et al.· 0 citations
Full waveform inversion (FWI) has significant potential to reconstruct the subsurface structure at high resolution, but it is a nonlinear and ill-posed problem. Traditional physics-based FWI methods that employ gradient-based optimization algorithms to minimize the discrepancy between observed and synthetic data face...
Cong Wang, Xing-Guo Huang, Stewart A. Greenhalgh et al.· Geophysical Journal Internat...· 0 citations
High-resolution seismic data is crucial for accurately characterizing thin reservoirs and small-scale faults, providing the foundation for lithology and fluid identification. However, image super-resolution techniques based on deep-learning methods, particularly convolutional neural networks, are fundamentally constr...
Kun Li, Wei Ge, Xing-Yao Yin et al.· Geophysics· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.