Journal of Geophysics and Engineering
人工智能（AI）：地震处理、速度建模、偏移成像 专题合集
本合集收录 Journal of Geophysics and Engineering 在 人工智能（AI）：地震处理、速度建模、偏移成像 方向发表的 23 篇文章，附完整引用格式、英文摘要与全文网址，复制网址到浏览器打开即可阅读全文。
目录 · 共 23 篇
01A Squeeze-and-Excitation Dilated UNet for ground roll suppression02DAS–VSP coupled noise suppression method using weakly supervised structured feature extraction03Seismic random noise attenuation with diffusion transformer04Attribute-constrained unsupervised ground-roll suppression via mid-band amplitude–phase fidelity and wavefield orthogonalization05Self-supervised CNN with coordinate attention and time-frequency hybrid loss for seismic data reconstruction06Robust and efficient 3D first-arrival traveltime reconstruction using swin transformer and masked autoencoder07Deep learning-based high-resolution Radon transform and its application to multiple elimination08Multi-condition seismic data denoising using feature-expanded gradient penalty generative adversarial network09Iterative intelligent high-precision high-SNR seismic weak reflection separation method and its application10Deep learning-based closed-loop surface-related multiple elimination11Simultaneous suppression of seismic random and erratic noise using PINN with high-frequency preservation12DPN2N: A self-supervised method for seismic data denoising and key information preservation13Seismic deblending with self-supervised Unblind-trace denoising14A wavefield decomposition-guided deep-learning framework for removing smearing artifacts from VSP RTM15Seismic data reconstruction via non-uniform Fourier transform and implicit neural network denoising prior16Seismic data reconstruction via non-uniform Fourier transform and implicit neural network denoising prior17Recent advances on deep residual learning based seismic data reconstruction: an overview18Low-frequency reconstruction of seismic data using stacked-BiLSTM networks19Seismic trace interpolation based on the principle of reciprocity using a conditional generative adversarial network (cGAN)20Upgoing and downgoing wavefield separation in VSP data using CGAN based on asymmetric convolution blocks21DAS-VSP coupled noise suppression based on U-Net network22Excitation-time imaging condition reverse-time migration based on a physics-informed neural network traveltime calculation with wavefield decomposition using an optical flow vector23Regularized deep learning for unsupervised random noise attenuation in poststack seismic data
01A Squeeze-and-Excitation Dilated UNet for ground roll suppression
Authors：Fei Li, Zhenbo Guo, Yan Huang, Juan Chen, Jiayuan Zhang
Citation：Fei Li, Zhenbo Guo, Yan Huang, Juan Chen, Jiayuan Zhang, A Squeeze-and-Excitation Dilated UNet for ground roll suppression, Journal of Geophysics and Engineering, Volume 23, Issue 4, August 2026, Pages 1479–1490, https://doi.org/10.1093/jge/gxag064
Abstract（英文原文，上下滑动查看）
Ground roll in land seismic data is characterized by strong spatial coherence and long-range propagation, which pose persistent difficulties for effective signal preservation using conventional filtering techniques. Such noise components commonly exhibit large-scale structured patterns, requiring denoising methods capable of capturing extended spatial context. In this study, a Squeeze-and-Excitation Dilated UNet is employed for seismic ground-roll attenuation. By incorporating dilated convolutions into the UNet architecture, the receptive field is effectively expanded without increasing the number of network parameters, enabling improved representation of long-range ground roll features. In addition, channel-wise attention is introduced to adaptively reweight feature responses, thereby enhancing the discrimination between effective reflections and different types of coherent noise. The performance of the adopted approach is evaluated on both synthetic and field seismic datasets and compared with the conventional UNet. The results demonstrate that the SE-Dilated UNet achieves superior ground-roll attenuation while preserving effective seismic signals with reduced signal leakage.https://doi.org/10.1093/jge/gxag064
02DAS–VSP coupled noise suppression method using weakly supervised structured feature extraction
Authors：Yunhao Pan, Zichun Liu, Yang Liu, Suoliang Chang
Citation：Yunhao Pan, Zichun Liu, Yang Liu, Suoliang Chang, DAS–VSP coupled noise suppression method using weakly supervised structured feature extraction, Journal of Geophysics and Engineering, Volume 23, Issue 4, August 2026, Pages 1346–1360, https://doi.org/10.1093/jge/gxag046
Abstract（英文原文，上下滑动查看）
Coupled noise suppression is one of the major denoising challenges in vertical seismic profiling (VSP) data acquired using distributed fibre-optic acoustic sensing (DAS) technology. Traditional methods to suppress coupled noise often require complicated parameter tuning, resulting in relatively low efficiency. Purely supervised deep learning methods demand a large amount of clean data as training labels, the acquisition of which is costly. Autoencoder (AE), as a deep learning framework that does not require manually annotated labels, has been widely used for random noise attenuation. In this study, we extend the AE to the suppression of coherent coupled noise in DAS–VSP data and propose a weakly supervised two-step coupled noise suppression workflow. First, time windows dominated by coupled noise are manually selected from the target DAS–VSP data, and the AE is trained separately on these windows to reconstruct and remove the stable, coherent coupled noise component. Second, the AE with the same architecture is trained to further reconstruct the entire dataset, aiming to enhance event continuity and attenuate residual noise. We provide a mathematical argument to explain why the AE tends to reconstruct coupled noise preferentially. Both synthetic and field examples demonstrate that the proposed method can effectively suppress coupled noise in DAS–VSP data, outperforming traditional methods. Furthermore, by introducing Gaussian white noise to simulate complex noise contamination, the method is shown to be robust and generalizable to more complex noise environments. Under our experimental settings, the result further shows that for the same network depth, incorporating skip connections weakens the latent space’s preferential representation of coupled noise, leading to overall inferior performance compared with AE without skip connections.https://doi.org/10.1093/jge/gxag046
03Seismic random noise attenuation with diffusion transformer
Authors：Cheng Zhao, Zhenbo Guo, Fei Li, Tianxiang Gao
Citation：Cheng Zhao, Zhenbo Guo, Fei Li, Tianxiang Gao, Seismic random noise attenuation with diffusion transformer, Journal of Geophysics and Engineering, Volume 23, Issue 4, August 2026, Pages 1312–1321, https://doi.org/10.1093/jge/gxag042
Abstract（英文原文，上下滑动查看）
Seismic data are often contaminated by random noise, which obscures weak reflections and complicates seismic interpretation. In this work, we propose a Vision Transformer-based diffusion model for seismic random noise attenuation. The denoising task is formulated within a diffusion framework, transforming the ill-posed problem into a sequence of stable and progressive denoising steps. This generative approach effectively alleviates the mean-collapse and over-smoothing issues inherent in traditional CNN and Transformer-based regression methods, thereby facilitating the recovery of high-frequency seismic details. We adopt a Transformer-based architecture with strong global modeling capability as the backbone of the diffusion model, enabling a global receptive field through self-attention mechanisms. Comparative experiments conducted on synthetic and field 2D seismic datasets confirm the clear performance advantage of the proposed method over f–k filtering, UNet, and standard diffusion models, achieving superior signal preservation with substantially reduced noise residue and signal leakage.https://doi.org/10.1093/jge/gxag042
04Attribute-constrained unsupervised ground-roll suppression via mid-band amplitude–phase fidelity and wavefield orthogonalization
Authors：Kunxi Wang, Ying Rao, Tianyue Hu, Zhencong Zhao
Citation：Kunxi Wang, Ying Rao, Tianyue Hu, Zhencong Zhao, Attribute-constrained unsupervised ground-roll suppression via mid-band amplitude–phase fidelity and wavefield orthogonalization, Journal of Geophysics and Engineering, Volume 23, Issue 3, June 2026, Pages 1029–1049, https://doi.org/10.1093/jge/gxag054
Abstract（英文原文，上下滑动查看）
Ground roll (GR) is a high-amplitude coherent noise to be attenuated during data processing. Supervised deep neural networks (DNNs) typically rely on labeled training data and often exhibit limited interpretability. To improve network interpretability for GR suppression and avoid any dependence on labeled data, we propose an unsupervised deep neural network (UDNN) based on attribute constraints (ACs) with the mid-band amplitude–phase fidelity and wavefield orthogonalization. GR predominantly contaminates the low-frequency effective signal, and the low- and mid-frequency effective components in the original data exhibit a frequency band similarity. Accordingly, we employ the mid-frequency component to reconstruct the low–mid-frequency effective signal. However, the conventional pixelwise amplitude consistency loss term for DNNs is highly susceptible to overfitting, often leaving substantial residual GR in the suppression result. To address this issue, we use the AC loss function by adding a mid-band amplitude–phase fidelity loss term and a wavefield orthogonalization loss term. The AC loss function enforces the physically constraints on output data in terms of amplitude, frequency, and phase. Under this loss function, an encoder–decoder convolutional network adaptively calibrates the amplitude, frequency, and phase of the mid-band effective component into the low–mid-band signal, thereby enabling thorough separation of GR noise. Superior noise-suppression and signal-fidelity performance are consistently observed in both time-domain sections and frequency–wavenumber spectra. The proposed method outperforms the synchrosqueezed generalized phase-shifted S-transform method, the supervised deep learning method, and UDNN variants trained using either single loss terms or two loss terms.https://doi.org/10.1093/jge/gxag054
05Self-supervised CNN with coordinate attention and time-frequency hybrid loss for seismic data reconstruction
Authors：Fei Zhou, Honghua Wang, Minling Wang, Xin Chen
Citation：Fei Zhou, Honghua Wang, Minling Wang, Xin Chen, Self-supervised CNN with coordinate attention and time-frequency hybrid loss for seismic data reconstruction, Journal of Geophysics and Engineering, Volume 23, Issue 3, June 2026, Pages 975–989, https://doi.org/10.1093/jge/gxag034
Abstract（英文原文，上下滑动查看）
Deep learning methods for high-precision seismic data reconstruction are typically constrained by the reliance on high-quality labeled datasets and the inherent limitations of time-domain loss functions in capturing frequency features. This study introduces a self-supervised multi-layer wavelet convolutional neural network (MWCNN) enhanced with the coordinate attention (CA) module, termed as CMWCNN, for high-precision seismic data reconstruction. The self-supervised mechanism is incorporated through dynamic random masking, which generates pseudo input-label pairs from incomplete seismic data to diversify the training datasets. The CA module, which extracts spatial location details via global pooling in horizontal and vertical directions, is integrated to guide CMWCNN to accurately capture spatial features within regions of missing traces. Multi-level discrete wavelet transform is employed to preserve multi-scale frequency features, enhancing the network’s ability to reconstruct fine details. Furthermore, a novel time-frequency hybrid loss function is proposed for CMWCNN training, referred to as TF‑CMWCNN. This loss function simultaneously constrains multi-scale frequency features and time-domain characteristics to effectively enhance the network’s robustness. Experimental results demonstrate that the proposed CMWCNN achieves superior reconstruction accuracy and computational efficiency compared to ResNet, U-Net++, and MWCNN. Additionally, compared to conventional time loss functions, the proposed hybrid loss improves CMWCNN performance by ∼12 dB in signal-to-noise ratio and peak signal-to-noise ratio.https://doi.org/10.1093/jge/gxag034
06Robust and efficient 3D first-arrival traveltime reconstruction using swin transformer and masked autoencoder
Authors：Ganghoon Lee, Sukjoon Pyun
Citation：Ganghoon Lee, Sukjoon Pyun, Robust and efficient 3D first-arrival traveltime reconstruction using swin transformer and masked autoencoder, Journal of Geophysics and Engineering, Volume 23, Issue 3, June 2026, Pages 951–974, https://doi.org/10.1093/jge/gxag032
Abstract（英文原文，上下滑动查看）
First-arrival traveltimes provide crucial information about subsurface velocities but can be missing due to environmental noise, equipment malfunction, or acquisition limitations. While spatial interpolation methods like kriging and inverse distance weighting have been used to address these issues, they struggle with the nonlinearity and complexity inherent in 3D traveltime data. Recent deep learning approaches, such as fully convolutional network (FCN), have shown promise in addressing these limitations but still face challenges in capturing global data variations, especially in 3-D cases. In this study, we adopt a deep learning framework combining a Swin Transformer-based masked autoencoder (MAE) with a Swin-Unet architecture for 3D traveltime reconstruction, validating its capability to capture both local and global data relationships. The Swin Transformer’s hierarchical self-attention mechanism is leveraged to learn diverse scales of representation, while MAE-based pretraining enhances the encoder’s ability to capture the intrinsic patterns in traveltime data by reconstructing masked portions. Blind tests show up to 80% reduction in root mean square error and 76% reduction in mean absolute error compared to existing FCN-based U-Net models. The results highlight the proposed method’s performance and potential for improving 3D traveltime reconstruction in complex geological settings.https://doi.org/10.1093/jge/gxag032
07Deep learning-based high-resolution Radon transform and its application to multiple elimination
Authors：Zhina Li, Hexiang Zhang, Gang Tan, Zhenchun Li, Zilin He, Qi Zhang, Peng Wang, Lieqian Dong
Citation：Zhina Li, Hexiang Zhang, Gang Tan, Zhenchun Li, Zilin He, Qi Zhang, Peng Wang, Lieqian Dong, Deep learning-based high-resolution Radon transform and its application to multiple elimination, Journal of Geophysics and Engineering, Volume 23, Issue 2, April 2026, Pages 677–687, https://doi.org/10.1093/jge/gxag020
Abstract（英文原文，上下滑动查看）
Multiple elimination based on Radon transform has been widely applied in industrial production. Conventional Radon transform methods tend to cause primary and multiple energy overlap in the Radon domain, while existing high‑resolution Radon transform methods suffer from low computational efficiency. In recent years, deep learning methods have been introduced into the field of multiple elimination. However, the highly nonlinear nature of data in the time–space domain increases the difficulty of model training. To address these issues, this paper proposes a deep‑learning‑based high‑resolution Radon transform method. By employing a U‑net architecture, the method establishes a nonlinear mapping from the adjoint solution to the high‑resolution solution in the Radon domain, replacing the iterative inversion process in conventional high‑resolution Radon transforms. This enables the rapid acquisition of high‑resolution Radon‑domain data for multiple elimination. Compared with time–space domain data, Radon‑domain data exhibit lower complexity and higher sparsity, which enhances both the convergence speed and stability of network model training. Furthermore, adjusting the curvature parameter in the Radon transform allows for data compression during training, reducing the size of the data to be processed and further improving training efficiency. Examples using both synthetic data and field data verify the effectiveness of the proposed method and demonstrate its advantages over conventional time–space domain algorithms.https://doi.org/10.1093/jge/gxag020
08Multi-condition seismic data denoising using feature-expanded gradient penalty generative adversarial network
Authors：Xiaotian Xue, Xi Yi, Yujing Liao, Kangning Wang, Xu Deng
Citation：Xiaotian Xue, Xi Yi, Yujing Liao, Kangning Wang, Xu Deng, Multi-condition seismic data denoising using feature-expanded gradient penalty generative adversarial network, Journal of Geophysics and Engineering, Volume 23, Issue 2, April 2026, Pages 574–586, https://doi.org/10.1093/jge/gxaf158
Abstract（英文原文，上下滑动查看）
High-quality seismic data serve as a critical foundation for imaging and interpretation. However, field seismic data are contaminated by noise, leading to numerous spurious seismic signals that adversely affect subsequent processing and interpretation. Currently, supervised deep learning methods face challenges due to the scarcity of authentic labels for 3D seismic images, while unsupervised approaches suffer from a lack of guidance during training, resulting in blurred denoising outcomes. To deal with these matters, it is essential to propose an adaptive semi-supervised random noise attenuation method that leverages pseudo-labels, which are easier to generate, to guide the model effectively. In this work, we put forward a semi-supervised gradient-penalized generative adversarial network with feature expansion (FEWGAN-SGP). The network first employs a global residual structure to ensure overall convergence during training. Subsequently, a channel attention mechanism with feature expansion is adopted to extract finer details. Finally, a smooth gradient penalty term is applied to enhance the distinction between field and generated images, thereby promoting deeper model learning. Experiments were conducted on both 2D and 3D datasets, and the results indicate that, compared with mainstream traditional methods and unsupervised learning approaches, the proposed method demonstrates excellent signal preservation and noise attenuation capabilities in practical applications.https://doi.org/10.1093/jge/gxaf158
09Iterative intelligent high-precision high-SNR seismic weak reflection separation method and its application
Authors：Shiyun Ran, Gulan Zhang, Xiangwen Li, Yiliang Luo, Caijun Cao, Chenxi Liang, Liming Ma
Citation：Shiyun Ran, Gulan Zhang, Xiangwen Li, Yiliang Luo, Caijun Cao, Chenxi Liang, Liming Ma, Iterative intelligent high-precision high-SNR seismic weak reflection separation method and its application, Journal of Geophysics and Engineering, Volume 23, Issue 1, February 2026, Pages 226–239, https://doi.org/10.1093/jge/gxaf135
Abstract（英文原文，上下滑动查看）
The seismic facies-guided trace-by-trace high-precision seismic weak reflection separation method (SRSM) and the corresponding deep-learning seismic weak reflection separation method usually can obtain high-precision high-SNR (signal-to-noise ratio) seismic weak reflection separation results; however, they face great challenges when accounting for low-SNR seismic data, resulting in undesired low-precision low-SNR seismic weak reflection separation results. In this paper, in order to obtain high-precision high-SNR seismic weak reflection separation results, we propose an iterative intelligent high-precision high-SNR seismic weak reflection separation method (IIWSM) based on the SRSM and the multi-task deep-learning network. IIWSM consists of the iterative SRSM-based seismic weak reflection label automatic generation (IWLG), and the iterative multi-task high-precision high-SNR seismic weak reflection separation method (IMWSM). IWLG aims to generate the high-precision high-SNR seismic weak reflection label for IMWSM, and IMWSM aims to obtain high-precision high-SNR seismic weak reflection separation results. A synthetic 2D seismic data set and an actual 3D seismic data set example demonstrate that IIWSM can be used for high-precision high-SNR seismic weak reflection separation.https://doi.org/10.1093/jge/gxaf135
10Deep learning-based closed-loop surface-related multiple elimination
Authors：Zilin He, Zhina Li, Zhenchun Li, Yipeng Xu, Gang Tan, Yubo Yue, Peng Wang
Citation：Zilin He, Zhina Li, Zhenchun Li, Yipeng Xu, Gang Tan, Yubo Yue, Peng Wang, Deep learning-based closed-loop surface-related multiple elimination, Journal of Geophysics and Engineering, Volume 22, Issue 6, December 2025, Pages 1896–1906, https://doi.org/10.1093/jge/gxaf115
Abstract（英文原文，上下滑动查看）
Surface-related multiple elimination (SRME) is a widely adopted method in suppressing surface-related multiples. However, when primaries and multiples interfere, the adaptive subtraction process in SRME often results in the loss of primary energy. To address this limitation, the closed-loop SRME (CLSRME) method is developed by using the advantages of waveform inversion, enabling more accurate multiple suppression without relying on the adaptive subtraction. Nevertheless, the inversion process incurs significant computational costs, hindering its practical application. To resolve both the primary energy loss in SRME and the high computational cost of CLSRME, a deep learning-based CLSRME (DL -CLSRME) method is proposed in this paper. This method takes advantage of the high computational efficiency and robust nonlinear mapping capacity of deep learning, enabling accurate and cost-effective internal multiple suppression. Based on the theories of SRME and CLSRME, we first derived the relationship between the predicted multiples using SRME and the accurate multiples estimated by CLSRME, thereby establishing the mapping between the two theoretically. Then, a U-net model is trained to learn the mapping from predicted multiples to actual multiples. Once trained, the model enables efficient and accurate multiple estimation, which is then subtracted from the seismic records to achieve effective multiple suppression. Numerical experiments and field data applications demonstrate that the proposed DL -CLSRME achieves a favourable balance between accuracy and efficiency compared with the conventional methods, highlighting its potential for practical seismic data processing.https://doi.org/10.1093/jge/gxaf115
11Simultaneous suppression of seismic random and erratic noise using PINN with high-frequency preservation
Authors：Peihong Xie, Yang Liu, Cai Liu, Chao Song, Xiangjia Zhang
Citation：Peihong Xie, Yang Liu, Cai Liu, Chao Song, Xiangjia Zhang, Simultaneous suppression of seismic random and erratic noise using PINN with high-frequency preservation, Journal of Geophysics and Engineering, Volume 22, Issue 6, December 2025, Pages 1796–1808, https://doi.org/10.1093/jge/gxaf109
Abstract（英文原文，上下滑动查看）
Random and erratic noise are common in seismic data. Traditional denoising methods that combine noise suppression for both types will lead to error accumulation. Deep learning methods often require large amounts of labeled data and complex network architectures to tackle this problem. To overcome these limitations, we utilize a physics-informed neural network (PINN) to remove both types of noise in this research. This method does not require a large amount of labeled data. Because it uses slope attribute, which is based on a local plane-wave partial differential equation. We apply velocity-dependent (VD) slope estimation to get more accurate slope values. This helps the PINN make better predictions. To enhance its ability of preservation in high-frequency signal, we use Fourier feature embedding and a periodic activation function from sinusoidal representation networks. These techniques effectively preserve high-frequency signals in seismic data and accelerates convergence. We refer to our proposed method as VD-PINN with high-frequency preservation. Its application to synthetic and field data shows effectiveness in simultaneously suppressing both noise compared to other methods.https://doi.org/10.1093/jge/gxaf109
12DPN2N: A self-supervised method for seismic data denoising and key information preservation
Authors：Kewen Li, Chunlong Li, Yuan Xiao, Yimin Dou, Xinyuan Zhu, Guangyue Zhou
Citation：Kewen Li, Chunlong Li, Yuan Xiao, Yimin Dou, Xinyuan Zhu, Guangyue Zhou, DPN2N: A self-supervised method for seismic data denoising and key information preservation, Journal of Geophysics and Engineering, Volume 22, Issue 5, October 2025, Pages 1465–1474, https://doi.org/10.1093/jge/gxaf091
Abstract（英文原文，上下滑动查看）
The focus of this study is on the use of a self-supervised deep learning method called Neighbor2Neighbor for S seismic data denoising, aiming to address two key challenges: (i) the high cost of obtaining clean noise-free data, and (ii) the difficulty in balancing denoising effectiveness with the retention of critical information. We introduce a novel regularization technique that reduces noise while preserving essential high-frequency information, such as faults. Built on the ResUNet++ architecture, the model is tailored to the frequency characteristics of seismic data, enhancing its ability to extract relevant features and generalize across different datasets. We quantitatively evaluate the denoising performance by adding noise to synthetic data and simulating noisy field data for qualitative analysis. This approach eliminates the need for expensive clean samples, effectively denoises the data without blurring critical features, and is therefore well-suited for high-quality seismic interpretation tasks.https://doi.org/10.1093/jge/gxaf091
13Seismic deblending with self-supervised Unblind-trace denoising
Authors：Shaoqi Tu, Yaxing Li, Xinming Wu, Sanfu Li
Citation：Shaoqi Tu, Yaxing Li, Xinming Wu, Sanfu Li, Seismic deblending with self-supervised Unblind-trace denoising, Journal of Geophysics and Engineering, Volume 22, Issue 4, August 2025, Pages 1149–1164, https://doi.org/10.1093/jge/gxaf074
Abstract（英文原文，上下滑动查看）
Simultaneous source seismic acquisition has the potential to significantly enhance exploration efficiency and reduce acquisition costs. However, it also introduces interference among multiple source signals, thereby necessitating effective deblending techniques to achieve signal separation. At present, artificial intelligence-based deblending methods primarily fall into two categories: supervised learning and self-supervised learning. Supervised learning methods rely heavily on large volumes of labeled data, which limits their applicability to field data. In contrast, self-supervised learning methods typically achieve deblending by means of self-supervised denoising. Although they eliminate the need for labeled data, their deblending performance remains limited. To address this issue, this study investigates a deblending method that integrates a self-supervised denoising network into the inversion process. The proposed method is built upon a plug-and-play inversion framework, where the self-supervised denoising operation serves as a regularization constraint. By iteratively applying this framework, the method achieves high-quality seismic deblending. Unlike traditional blind-trace training strategies used in the denoising stage, we introduce an “Unblind-trace” training strategy to better preserve useful information in seismic data. To implement this strategy, we design a global-aware masking mechanism, which adaptively generates training sample pairs, thereby reducing reliance on labeled data. The proposed method is tested on both single-source and two-source blended seismic datasets. Experimental results demonstrate that it outperforms traditional methods such as block-wise 2D Fourier transform and damped rank reduction in deblending performance, and is comparable to state-of-the-art self-supervised inversion methods. This research provides a novel method and technical pathway for self-supervised iterative deblending of seismic data.https://doi.org/10.1093/jge/gxaf074
14A wavefield decomposition-guided deep-learning framework for removing smearing artifacts from VSP RTM
Authors：Dingding Deng, Yang Liu, Gui Chen, Jiangtao Ma, Yujiao Wang
Citation：Dingding Deng, Yang Liu, Gui Chen, Jiangtao Ma, Yujiao Wang, A wavefield decomposition-guided deep-learning framework for removing smearing artifacts from VSP RTM, Journal of Geophysics and Engineering, Volume 22, Issue 4, August 2025, Pages 1085–1095, https://doi.org/10.1093/jge/gxaf079
Abstract（英文原文，上下滑动查看）
Reverse-time migration (RTM) is advantageous for imaging steeply dipping structures, but it may introduce migration smearing artifacts when applied to vertical seismic profiles (VSP). These artifacts mainly result from the limited coverage of VSP acquisition geometry, causing uneven stacking and insufficient noise suppression across multiple shots. Most studies on denoising VSP imaging data focus on limiting the migration aperture and utilizing dip-angle information, while wave equation-based methods often face challenges in accurately determining wavefront propagation directions or rely on prior subsurface structure. Despite recent advances in deep learning, effectively suppressing smearing artifacts remains a significant challenge in VSP RTM. To address this, we propose a novel deep-learning method for suppressing smearing artifacts in VSP RTM using the removal of smearing artifacts residual neural network (RSA-ResNet). Initially, synthetic data are generated based on convolution theory, using data with smearing artifacts as inputs and data without smearing artifacts as labels for the deep-learning network. Subsequently, the RSA-ResNet model is trained to suppress the smearing artifacts associated with VSP RTM. The trained model is then fine-tuned using wavefield decomposition results from target data containing smearing artifacts. Finally, the refined model is applied to VSP RTM data. Experimental results indicate that the waveforms of the reference traces in the two RSA-ResNet-denoised examples closely match the ground-truth solution, effectively demonstrating the superiority of the proposed method over conventional wavefield decomposition methods.https://doi.org/10.1093/jge/gxaf079
15Seismic data reconstruction via non-uniform Fourier transform and implicit neural network denoising prior
Authors：Feiying Wang, Yuhan Sui, Xiaojing Wang, Tong Zhou, Jianwei Ma
Citation：Feiying Wang, Yuhan Sui, Xiaojing Wang, Tong Zhou, Jianwei Ma, Seismic data reconstruction via non-uniform Fourier transform and implicit neural network denoising prior, Journal of Geophysics and Engineering, Volume 22, Issue 3, June 2025, Pages 940–951, https://doi.org/10.1093/jge/gxaf045
Abstract（英文原文，上下滑动查看）
Irregular sampling of seismic data on non-equispaced spatial grids presents significant challenges in seismic data acquisition and imaging, resulting in reduced quality and continuity of subsurface structures. Seismic data reconstruction plays a pivotal role in transforming irregularly sampled data into equispaced grids for improved imaging quality. Conventional methods combined the curvelet transform with the non-uniform discrete Fourier transform (NDFT) to reconstruct seismic data but often suffered from spectral leakage and incomplete continuity of seismic events, limiting their practical effectiveness. To address this, we propose an advanced seismic data reconstruction approach that integrates the NDFT with an implicit denoising neural network prior. The proposed method leverages the NDFT to map irregular sampling points accurately to equispaced grids while incorporating the denoising neural network as a prior to suppress noise and enhance the continuity of seismic events. Numerical experiments on synthetic and field datasets demonstrate the effectiveness of the proposed method. The results illustrate that the proposed method achieves better reconstruction accuracy, and preserves continuity more effectively compared to traditional curvelet-based method.https://doi.org/10.1093/jge/gxaf045
16Seismic data reconstruction via non-uniform Fourier transform and implicit neural network denoising prior
Authors：Feiying Wang, Yuhan Sui, Xiaojing Wang, Tong Zhou, Jianwei Ma
Citation：Feiying Wang, Yuhan Sui, Xiaojing Wang, Tong Zhou, Jianwei Ma, Seismic data reconstruction via non-uniform Fourier transform and implicit neural network denoising prior, Journal of Geophysics and Engineering, Volume 22, Issue 3, June 2025, Pages 940–951, https://doi.org/10.1093/jge/gxaf045
Abstract（英文原文，上下滑动查看）
Irregular sampling of seismic data on non-equispaced spatial grids presents significant challenges in seismic data acquisition and imaging, resulting in reduced quality and continuity of subsurface structures. Seismic data reconstruction plays a pivotal role in transforming irregularly sampled data into equispaced grids for improved imaging quality. Conventional methods combined the curvelet transform with the non-uniform discrete Fourier transform (NDFT) to reconstruct seismic data but often suffered from spectral leakage and incomplete continuity of seismic events, limiting their practical effectiveness. To address this, we propose an advanced seismic data reconstruction approach that integrates the NDFT with an implicit denoising neural network prior. The proposed method leverages the NDFT to map irregular sampling points accurately to equispaced grids while incorporating the denoising neural network as a prior to suppress noise and enhance the continuity of seismic events. Numerical experiments on synthetic and field datasets demonstrate the effectiveness of the proposed method. The results illustrate that the proposed method achieves better reconstruction accuracy, and preserves continuity more effectively compared to traditional curvelet-based method.https://doi.org/10.1093/jge/gxaf045
17Recent advances on deep residual learning based seismic data reconstruction: an overview
Authors：Naihao Liu, Wen Deng, Fangyu Li, Zhiguo Wang, Hao Wu, Jinghuai Gao
Citation：Naihao Liu, Wen Deng, Fangyu Li, Zhiguo Wang, Hao Wu, Jinghuai Gao, Recent advances on deep residual learning based seismic data reconstruction: an overview, Journal of Geophysics and Engineering, Volume 22, Issue 3, June 2025, Pages 854–876, https://doi.org/10.1093/jge/gxaf011
Abstract（英文原文，上下滑动查看）
Seismic data reconstruction is an essential process to reduce the effects of missing traces in field acquisition. Imputation and interpolation techniques have been proposed to reconstruct the subtle missing features, which is a challenging task that remains to be solved, such as the recovery capability for weak reflections and consecutively missing parts and the computational efficiency. In recent years, deep learning (DL), especially the residual network (ResNet), has gained remarkable success in seismic data reconstruction because it can precisely extract seismic features. In this overview, we probe into the residual module architecture and discuss in-depth the characteristics of the state-of-the-art residual modules, mainly including the split-transform-merge and the squeeze-and-attention. We use the detailed ablation experiments to investigate the roles of several key hyperparameters for each residual-based model. We explore the mechanism of the residual module’s effectiveness based on extensive qualitative and quantitative comparisons involving various residual networks and state-of-the-art models for seismic data reconstruction. The irregularly sampled and consecutively sampled scenarios demonstrate that the reconstruction performance of the residual modules is promising and superior to that of state-of-the-art deep networks.https://doi.org/10.1093/jge/gxaf011
18Low-frequency reconstruction of seismic data using stacked-BiLSTM networks
Authors：Daeun Na, Dawoon Lee, Young Seo Kim, Wookeen Chung
Citation：Daeun Na, Dawoon Lee, Young Seo Kim, Wookeen Chung, Low-frequency reconstruction of seismic data using stacked-BiLSTM networks, Journal of Geophysics and Engineering, Volume 22, Issue 1, February 2025, Pages 238–249, https://doi.org/10.1093/jge/gxae127
Abstract（英文原文，上下滑动查看）
Seismic data is often contaminated with various low-frequency noises such as swell noise, instrumental noise, and tow noise. Typically, in the processing of seismic data, a high-pass filter is used to eliminate these unwanted low-frequency components. However, this filtering not only removes noise but also eliminates meaningful low-frequency signals essential for broadband seismic imaging. Numerous studies have been conducted to reconstruct low-frequency components by using machine learning techniques. However, most studies use convolutional neural networks and transform the data into image form rather than raw time-series data directly. In this study, we propose a low-frequency reconstruction method based on a recurrent neural network tailored to the sequence nature of seismic data. To improve the accuracy of the low-frequency component reconstruction network, we augment the feature information by incorporating envelope data, which provides additional aspects to seismic traces. Synthetic data tests confirm that our proposed method can reconstruct low-frequency components accurately. Furthermore, the validated network also successfully reconstructs low-frequency components in real dataset from the Arctic Sea.https://doi.org/10.1093/jge/gxae127
19Seismic trace interpolation based on the principle of reciprocity using a conditional generative adversarial network (cGAN)
Authors：Jaime A Collazos, Katerine D J Rincon, Daniel N Pinheiro, Mesay Geletu Gebre, Carlos A N da Costa, Gilberto Corso, Tiago Barros, Joao Medeiros de Araújo, Yanghua Wang
Citation：Jaime A Collazos, Katerine D J Rincon, Daniel N Pinheiro, Mesay Geletu Gebre, Carlos A N da Costa, Gilberto Corso, Tiago Barros, Joao Medeiros de Araújo, Yanghua Wang, Seismic trace interpolation based on the principle of reciprocity using a conditional generative adversarial network (cGAN), Journal of Geophysics and Engineering, Volume 21, Issue 6, December 2024, Pages 1775–1790, https://doi.org/10.1093/jge/gxae105
Abstract（英文原文，上下滑动查看）
Seismic trace interpolation is a common step in data processing used to fill in missing traces and regularize acquisition grids. In the ocean bottom node (OBN) acquisition, it is common practice to design seismic surveys with many sources and fewer receivers. However, it results in a loss of lateral resolution in the shot-gather domain, although it increases the lateral resolution in the receiver-gather domain. In this paper, we propose to utilize the information from the receiver-gather domain to enhance the lateral resolution in the shot-gather domain, based on the principle of reciprocity. Explicitly, we use receiver gathers to train a conditional generative adversarial network (cGAN), called Pix2Pix, and obtain a generative interpolator, which can be used in the shot-gather domain. We present an iterative workflow to perform seismic data interpolation in the shot-gather domain and illustrate the workflow using a synthetic 2D data model consisting of densely populated sources and sparse receivers. We also present a field OBN data example from a Brazilian pre-salt area. The workflow can be used to increase lateral resolution or even complete the geometry of a monitor acquisition to match the baseline. In contrast to conventional methods, this iterative GAN interpolation does not lead to artefacts in the case of larger filling gaps.https://doi.org/10.1093/jge/gxae105
20Upgoing and downgoing wavefield separation in VSP data using CGAN based on asymmetric convolution blocks
Authors：Danping Cao, Xin Chen, Yan Jia, Chao Jin, Xin Fu
Citation：Danping Cao, Xin Chen, Yan Jia, Chao Jin, Xin Fu, Upgoing and downgoing wavefield separation in VSP data using CGAN based on asymmetric convolution blocks, Journal of Geophysics and Engineering, Volume 21, Issue 5, October 2024, Pages 1511–1525, https://doi.org/10.1093/jge/gxae089
Abstract（英文原文，上下滑动查看）
Accurately upgoing and downgoing wavefield separation is a critical step in vertical seismic profile (VSP) data processing, as its accuracy is directly related to the imaging quality of VSP data. Traditional methods are based mainly on transforms, and their windows are manually set in the transformed domain to obtain the target wavefield. The manual operations often cause errors and affect the accuracy of wavefield separation. In contrast, deep learning algorithms are more automatic, and have achieved a lot in seismic data processing. We propose to employ a conditional generative adversarial network (CGAN) for wavefield separation in VSP data. A CGAN consists of two main components: a generator, which generates new data samples, and a discriminator, which evaluates the generated samples against real data, with both components trained simultaneously in an adversarial manner to improve the quality of generated samples. The full wavefield serves as a constraint to link the generator and discriminator, ensuring that the separated up- and downgoing wavefields align better with the full wavefield. An asymmetric convolution block is introduced to more effectively capture the directional features of the VSP wavefield. To mitigate the influence of amplitude differences between the waves on the network update, the relative downgoing wavefield (obtained by subtracting the predicted upgoing wavefield from the full wavefield) is included in the loss function. Numerical experiments demonstrate that the trained network can effectively learn the characteristics of the up- and downgoing waves, especially their propagation directions, achieving high-precision wavefield separation.https://doi.org/10.1093/jge/gxae089
21DAS-VSP coupled noise suppression based on U-Net network
Authors：Jing-Xia Xu, Hao-Ran Ren, Zhao-Lin Zhu, Tong Wang, Zhi-Hao Chen
Citation：Jing-Xia Xu, Hao-Ran Ren, Zhao-Lin Zhu, Tong Wang, Zhi-Hao Chen, DAS-VSP coupled noise suppression based on U-Net network, Journal of Geophysics and Engineering, Volume 21, Issue 3, June 2024, Pages 938–950, https://doi.org/10.1093/jge/gxae047
Abstract（英文原文，上下滑动查看）
The emerging distributed fiber-optic acoustic sensing (DAS) technology has broad prospects for application in vertical seismic profiles (VSP). However, the acquired DAS-VSP data often suffers from coupled noise that seriously affects data quality. Traditional methods for suppressing coupled noise are usually time-consuming and not suitable for the large-scale denoising of DAS-VSP data. To address this, a coupled noise suppression method based on the U-Net network is proposed, and a self-attention (SA) block is introduced to enhance the denoising ability of the network. Transfer learning is employed to achieve coupled noise suppression from synthetic data to field data. Denoising results demonstrate that the network can effectively suppress coupled noise in DAS-VSP data while preserving signal energy to a certain extent, exhibiting strong generalization capability. Upon completion of network training, denoising results can be obtained within seconds, making it more convenient and efficient compared to traditional methods.https://doi.org/10.1093/jge/gxae047
22Excitation-time imaging condition reverse-time migration based on a physics-informed neural network traveltime calculation with wavefield decomposition using an optical flow vector
Authors：Jian Li, Guoning Du, Dewen Qin, Wensun Yin, Jun Tan, Zhaolun Liu, Peng Song
Citation：Jian Li, Guoning Du, Dewen Qin, Wensun Yin, Jun Tan, Zhaolun Liu, Peng Song, Excitation-time imaging condition reverse-time migration based on a physics-informed neural network traveltime calculation with wavefield decomposition using an optical flow vector, Journal of Geophysics and Engineering, Volume 21, Issue 1, February 2024, Pages 200–220, https://doi.org/10.1093/jge/gxad106
Abstract（英文原文，上下滑动查看）
Although the excitation-time imaging condition offers a lower memory consumption and higher computational efficiency compared to cross-correlation imaging condition, it has not been widely used in industrial applications because of the accuracy problem of traveltime calculation and the influence of low-wave-number noise. In this paper, we introduce the physics-informed neural network (PINN) algorithm to achieve a high-precision traveltime calculation of the source forward wavefield. Subsequently, we introduce a technique for high-precision wavefield decomposition of the reverse-time wavefield via the optical flow vector, enabling us to realize a correlation-weighted stacking imaging of each wavefield. Model experiments and real data processing show that the proposed traveltime calculation algorithm based on PINN offers high accuracy and good applicability in the excitation-time reverse-time migration imaging of complex models, and correlation-weighted stacking imaging based on optical flow vector-based wavefield separation can significantly suppress the noise with low wave-number and achieve high-precision imaging of complex models.https://doi.org/10.1093/jge/gxad106
23Regularized deep learning for unsupervised random noise attenuation in poststack seismic data
Authors：Chengyun Song, Shutao Guo, Chuanchao Xiong, Jiying Tuo
Citation：Chengyun Song, Shutao Guo, Chuanchao Xiong, Jiying Tuo, Regularized deep learning for unsupervised random noise attenuation in poststack seismic data, Journal of Geophysics and Engineering, Volume 21, Issue 1, February 2024, Pages 60–67, https://doi.org/10.1093/jge/gxad094
Abstract（英文原文，上下滑动查看）
Deep learning methods achieve excellent noise reduction performances in seismic data processing compared with traditional methods. However, deep learning usually requires a large number of pairwise noisy-clean training data, which is an extremely challenging task. In this paper, an unsupervised approach without clean seismic data is proposed to suppress random noise. Seismic data is divided into odd and even traces, which serve as the input and output of the depth network, so that the proposed algorithm can be trained directly on the original data. What is more, the proposed method introduces two regularization terms to solve the over-smoothing problem caused by reconstruction of adjacent traces. The first term considers an ideal denoising network that does not cause oversmooth as a constraint, while the second term considers the structural information existing in seismic data. Experiments on synthetic post-stack data illustrate that the proposed method obtain a higher signal-to-noise ratio than the comparison methods. In the application of field post-stack seismic data, the proposed method can effectively maintain the seismic amplitude and generate good spectral characteristics.https://doi.org/10.1093/jge/gxad094
Journal of Geophysics and Engineering (JGE)是由中石化石油物探技术研究院有限公司主办，英国帝国理工大学（Imperial College London）协办，牛津大学出版社(Oxford University Press)合作出版的国际英文SCI收录期刊。
JGE主要登载地球物理与油气藏工程方面的高质量文章，内容涉及固体地球物理、油气地球物理、矿产勘探地球物理、煤矿矿山地球物理、水文工程环境地球物理、岩石物理、测井技术、钻井技术、考古、遥感、CCUS、地热、氢能源、GPR、仪器和传感器设计及相关的地球物理等。
JCR 2025影响因子：2.0
JCR 分区：Q3
收录：SCIE, Scopus, DOAJ
Editors-in-Chief：
Yanghua Wang Principal of Resource Geophysics Academy, Imperial College London, UK
Xuhui Xu Director of SINOPEC Geophysical Research Institute Co., Ltd., Nanjing, China
官方网址：https://academic.oup.com/jge
官方邮箱：njjge@163.net
联系电话：025-68109538
长按识别二维码，关注 JGE，获取最新文章与专题资讯
