arXiv subjects

Xintong Dong

Publications and source records attributed to Xintong Dong.

13 recordsLinked to original sources

arXiv2026

A Frequency-Aware Dynamic Knowledge Distillation Framework: An Effective Tool for Bridging Low- and High-Frequency Seismic Information

Seismic data contain rich information across different frequency bands, with low-frequency components primarily characterizing large-scale geological structures and high-frequency components preserving fine-scale seismic details. Effectively integrating these frequency-dependent components is essential for seismic feature learning to better preserve structural continuity and fine-scale details. Knowledge distillation provides an effective means for transferring informative representations from high-quality data. However, existing distillation-based frameworks usually treat seismic features in a full-band manner, ignoring relationships across frequency bands and thereby limiting the coordinated transfer of low- and high-frequency knowledge. To bridge low- and high-frequency seismic features through knowledge distillation, we propose a frequency-aware dynamic knowledge distillation framework (FADKD-Net), which establishes a teacher-student learning framework and performs frequency-aware knowledge transfer between low- and high-frequency bands. Specifically, FADKD-Net decomposes seismic features into low- and high-frequency components and performs targeted distillation to exploit their complementary information. Low-frequency distillation guides the student model to learn stable structural priors, thereby improving the overall continuity of seismic events. Meanwhile, high-frequency distillation enhances detailed feature modeling and improves the representational capability for complex and small-scale structures. Furthermore, a cross-domain feature alignment strategy is proposed to reduce distributional discrepancies across different surveys and enhance the transferability of the seismic representations learned by FADKD-Net.

physics.geo-ph↗

arXiv2026

Horizontal-Component Prior-based Framework for Adaptive Shear-wave Leakage Suppression in OBC Data

Shear-wave leakage in the vertical (Z) component of ocean-bottom cable (OBC) seismic data commonly results from the receiver tilt and poor seafloor coupling, introducing unwanted coherent noise that impacts the subsequent data processing and imaging. Traditional denoising methods are limited by manual parameter tuning and idealized model assumptions, while deep-learning (DL) approaches have shown significant potential in suppressing shear-wave leakage. However, supervised learning requires clean primary waves (P waves) as the label, which is generally impractical to obtain for field data. To address these challenges, we propose a framework based on horizontal-component priors for adaptive shear-wave leakage suppression (HPAS). Instead of relying on clean primary-wave (P-wave) data, HPAS generates input-label pairs directly from raw multi-component field data using an additive-subtractive noise strategy. Specifically, we extract shear-wave (S-wave) noise from the horizontal components and apply a linear transformation to match its first and second order moments with the S-wave leakage in the Z-component, and the statistically matched noise is then added to and subtracted from the original Z-component to create the input and label pairs. By allowing the denoising model to learn the S-wave features present in the differences between the input and the label, the adaptive denoising process approximates supervised learning. Evaluations on both synthetic and field data demonstrate that the proposed HPAS framework effectively and adaptively suppresses S-wave leakage while preserving the amplitude of the P-wave signals in the Z-component, offering a robust solution with strong generalization capabilities.

physics.geo-ph↗

arXiv2026

Transcending Classical Neural Network Boundaries: A Quantum-Classical Synergistic Paradigm for Seismic Data Processing

In recent years, a number of neural-network (NN) methods have exhibited good performance in seismic data processing, such as denoising, interpolation, and frequency-band extension. However, these methods rely on stacked perceptrons and standard activation functions, which imposes a bottleneck on the representational capacity of deep-learning models, making it difficult to capture the complex and non-stationary dynamics of seismic wavefields. Different from the classical perceptron-stacked NNs which are fundamentally confined to real-valued Euclidean spaces, the quantum NNs leverage the exponential state space of quantum mechanics to map the features into high-dimensional Hilbert spaces, transcending the representational boundary of classical NNs. Based on this insight, we propose a quantum-classical synergistic generative adversarial network (QC-GAN) for seismic data processing, serving as the first application of quantum NNs in seismic exploration. In QC-GAN, a quantum pathway is used to exploit the high-order feature correlations, while the convolutional pathway specializes in extracting the waveform structures of seismic wavefields. Furthermore, we design a QC feature complementarity loss to enforce the feature orthogonality in the proposed QC-GAN. This novel loss function can ensure that the two pathways encode non-overlapping information to enrich the capacity of feature representation. On the whole, by synergistically integrating the quantum and convolutional pathways, the proposed QC-GAN breaks the representational bottleneck inherent in classical GAN. Experimental results on denoising and interpolation tasks demonstrate that QC-GAN preserves wavefield continuity and amplitude-phase information under complex noise conditions.

cs.LG↗

arXiv2025

HistCAD: A Constraint-Aware Parametric History-Based CAD Representation, Dataset, and Benchmark with Industrial Complexity

Parametric CAD sequences are reusable because dimensional and geometric constraints govern how parameter changes propagate. Existing CAD generation datasets and benchmarks emphasize reconstruction fidelity, execution validity, or static shape similarity, leaving preservation of design intent under edits largely unmeasured. We introduce HistCAD, a representation standard, dataset, and benchmark for executable parametric CAD with explicit constraints. HistCAD defines an intermediate language independent of CAD software, recording sketch primitives, constraints, feature operations, and 3D point boundary references for operations such as fillet and chamfer. The dataset contains 170,236 executable sequences aligned with native CAD models, STEP files, rendered views, and text annotations, combining academic scale with professionally authored industrial complexity. Building on this representation, the Constraint-Aware Editability Benchmark applies parameter edits and reports Edit Reachability, conditional preserved constraint satisfaction, and Overall Editable Success, abbreviated ER, cPCSR, and OES; these metrics separate failures to reach a valid edited state from failures to preserve required constraints. Experiments show that explicit constraints are essential for preserving design intent after edits, and that HistCAD supports supervised CAD generation from text and direct LLM workflows. We argue that HistCAD reframes CAD generation from static shape imitation to the synthesis of reusable parametric sequences with explicit constraints.

cs.GR↗

arXiv2025

2.5D Transformer: An Efficient 3D Seismic Interpolation Method without Full 3D Training

Transformer has emerged as a powerful deep-learning technique for two-dimensional (2D) seismic data interpolation, owing to its global modeling ability. However, its core operation introduces heavy computational burden due to the quadratic complexity, hindering its further application to higher-dimensional data. To achieve Transformer-based three-dimensional (3D) seismic interpolation, we propose a 2.5-dimensional Transformer network (T-2.5D) that adopts a cross-dimensional transfer learning (TL) strategy, so as to adapt the 2D Transformer encoders to 3D seismic data. The proposed T-2.5D is mainly composed of 2D Transformer encoders and 3D seismic dimension adapters (SDAs). Each 3D SDA is placed before a Transformer encoder to learn spatial correlation information across seismic lines. The proposed cross-dimensional TL strategy comprises two stages: 2D pre-training and 3D fine-tuning. In the first stage, we optimize the 2D Transformer encoders using a large amount of 2D data patches. In the second stage, we freeze the 2D Transformer encoders and fine-tune the 3D SDAs using limited 3D data volumes. Extensive experiments on multiple datasets are conducted to assess the effectiveness and efficiency of T-2.5D. Experimental results demonstrate that the proposed method achieves comparable performance to that of full 3D Transformer at a significantly low cost.

physics.geo-ph↗

arXiv2025

Light-weighted foundation model for seismic data processing based on representative and non-redundant pre-training dataset

In the fields of computer vision (CV) and remote sensing (RS), foundational models typically follow the "big data + large model parameters" paradigm. However, the application of this strategy in seismic data processing faces several challenges: seismic data is difficult to obtain and the scarcity of publicly available datasets make it difficult to construct large-scale datasets. Additionally, the high computational cost associated with a large number of model parameters restricts widespread research in this domain. Therefore, we propose a lightweight seismic processing foundational model paradigm (SPFM), which aims to overcome the limitations of traditional methods by data engineering and network architecture innovation. Specifically, we propose an innovative dataset construction strategy that generates more seismic data by data augmentation techniques, including collecting publicly available field data and using generative diffusion models (GDM) for data enhancement. Furthermore, we optimize the data distribution by employing dimensionality reduction, cluster analysis, and stratified sampling methods, reducing redundant information while preserving important seismic features, thus constructing a comprehensive dataset. In terms of network architecture design, we introduce the selective structured state-space model (Mamba) structure, which effectively captures global features of seismic data and alleviates the quadratic growth of computational complexity inherent in Transformer-based models, thereby improving computational efficiency. This model, pre-trained with only four A800 GPUs, outperforms traditional methods across multiple tasks, including denoising, interpolation, frequency-band extrapolation, and resolution enhancement. The lightweight paradigm provides an solution for seismic data processing, advancing the generalization and accessibility of seismic data processing.

physics.geo-ph↗

arXiv2025

PreAdaptFWI: Pretrained-Based Adaptive Residual Learning for Full-Waveform Inversion Without Dataset Dependency

Full-waveform inversion (FWI) is a method that utilizes seismic data to invert the physical parameters of subsurface media by minimizing the difference between simulated and observed waveforms. Due to its ill-posed nature, FWI is susceptible to getting trapped in local minima. Consequently, various research efforts have attempted to combine neural networks with FWI to stabilize the inversion process. This study presents a simple yet effective training framework that is independent of dataset reliance and requires only moderate pre-training on a simple initial model to stabilize network outputs. During the transfer learning phase, the conventional FWI gradients will simultaneously update both the neural network and the proposed adaptive residual learning module, which learns the residual mapping of large-scale distribution features in the network's output, rather than directly fitting the target mapping. Through this synergistic training paradigm, the proposed algorithm effectively infers the physically-informed prior knowledge into a global representation of stratigraphic distribution, as well as capturing subtle variations in inter-layer velocities within local details, thereby escaping local optima. Evaluating the method on two benchmark models under various conditions, including absent low-frequency data, noise interference, and differing initial models, along with corresponding ablation experiments, consistently demonstrates the superiority of the proposed approach.

physics.geo-ph↗

arXiv2024

Transformer for seismic image super-resolution

Seismic images obtained by stacking or migration are usually characterized as low signal-to-noise ratio (SNR), low dominant frequency and sparse sampling both in depth (or time) and offset dimensions. For improving the resolution of seismic images, we proposed a deep learning-based method to achieve super-resolution (SR) in only one step, which means performing the denoising, interpolation and frequency extrapolation at the same time. We design a seismic image super-resolution Transformer (SIST) to extract and fuse local and global features, which focuses more on the energy and extension shapes of effective events (horizons, folds and faults, etc.) from noisy seismic images. We extract the edge images of input images by Canny algorithm as masks to generate the input data with double channels, which improves the amplitude preservation and reduces the interference of noises. The residual groups containing Swin-Transformer blocks and residual connections consist of the backbone of SIST, which extract the global features in a window with preset size and decrease computational cost meanwhile. The pixel shuffle layers are used to up-sample the output feature maps from the backbone to improve the edges, meanwhile up-sampling the input data through a skip connection to enhance the amplitude preservation of the final images especially for clarifying weak events. 3-dimensional synthetic seismic volumes with complex geological structures are created, and the amplitudes of half of the volumes are mixtures of strong and weak, then select 2-dimensional slices randomly to generate training datasets which fits field data well to perform supervised learning. Both numerical tests on synthetic and field data in different exploration regions demonstrate the feasibility of our method.

physics.geo-ph↗

arXiv2024

Self-Supervised Diffusion Model for 3-D Seismic Data Reconstruction

Seismic data reconstruction is an effective tool for compensating nonuniform and incomplete seismic geometry. Compared with methods for 2D seismic data, 3D reconstruction methods could consider more spatial structure correlation in seismic data. In the early studies, 3D reconstruction methods are mainly theory-driven and have some limitations due to their prior assumptions on the seismic data. To release these limitations, deep learning-based reconstruction methods rise and show potential in dealing with reconstruction problems. However, there are mainly two shortcomings in existing deep learning-methods. On the one hand, most of existing deep learning-based methods adopt the convolutional neural network, having some difficulties in dealing with data with complex or time-varying distributions. Recently, the diffusion model has been reported to possess the capability to solve data with complex distributions by gradually complicating the distribution of data to optimize the network. On the other hand, existing methods need enough paired-data to train the network, which are very hard to obtain especially for the starved 3D seismic data. Deep prior-based unsupervised and sampling-based self-supervised networks offer an available solution to this problem. In this paper, we develop a self-supervised diffusion model (S2DM) for 3D seismic data reconstruction. The proposed model mainly contains a diffusion restoration model and a variational time-spatial module. Extensive synthetic and field experiments demonstrate the superiority of the proposed S2DM algorithm.

physics.geo-ph↗

arXiv2024

MAE-GAN: A Novel Strategy for Simultaneous Super-resolution Reconstruction and Denoising of Post-stack Seismic Profile

Post-stack seismic profiles are images reflecting containing geological structures which provides a critical foundation for understanding the distribution of oil and gas resources. However, due to the limitations of seismic acquisition equipment and data collecting geometry, the post-stack profiles suffer from low resolution and strong noise issues, which severely affects subsequent seismic interpretation. To better enhance the spatial resolution and signal-to-noise ratio of post-seismic profiles, a multi-scale attention encoder-decoder network based on generative adversarial network (MAE-GAN) is proposed. This method improves the resolution of post-stack profiles, and effectively suppresses noises and recovers weak signals as well. A multi-scale residual module is proposed to extract geological features under different receptive fields. At the same time, an attention module is designed to further guide the network to focus on important feature information. Additionally, to better recover the global and local information of post-stack profiles, an adversarial network based on a Markov discriminator is proposed. Finally, by introducing an edge information preservation loss function, the conventional loss function of the Generative Adversarial Network is improved, which enables better recovery of the edge information of the original post-stack profiles. Experimental results on simulated and field post-stack profiles demonstrate that the proposed MAE-GAN method outperforms two advanced convolutional neural network-based methods in noise suppression and weak signal recovery. Furthermore, the profiles reconstructed by the MAE-GAN method preserve more geological structures.

physics.geo-ph↗

arXiv2024

Transformer For Low-frequency Extrapolating of Seismic Data

Full waveform inversion (FWI) is used to reconstruct the physical properties of subsurface media which plays an important role in seismic exploration. However, the precision of FWI is seriously affected by the absence or inaccuracy of low-frequency information. Therefore, reconstructing the low-frequency signals accurately is highly significant in seismic data processing. Low-frequency extrapolation of seismic records can be approached as a deep learning regression problem. Thus, to obtain low-frequency information from band-limited seismic records, a novel network structure called low-frequency extrapolation transformer (LFET) is proposed to construct the nonlinear mapping relationship between the data missing low-frequency and low-frequency data in a supervised learning approach, which is inspired by the transformer model widely used in natural language processing (NLP). We apply multi-head self-attention (MSA) modules to model the remote dependencies of seismic data. Based on this, we introduce a shifted window partitioning approach to reduce the calculating amount. Due to the field data are not suitable for supervised learning, we generate synthetic seismic records using submodels selected from the benchmark Marmousi model as training data whose characteristics are similar to that of the field data. A single trace of synthetic band-limited seismic data in the time domain is used as the input data, and the parameters of LFET are updated based on the errors between the predicted trace and the corresponding label. The experimental results on the data generated by different models, different wavelets, and different kinds of field marine data demonstrate the feasibility and generalization of the proposed method. Furthermore, the proposed method achieves higher accuracy with lower computational expense than the traditional CNN method.

physics.geo-ph↗

arXiv2024

Modeling Seismic Wave Propagation in TTI Media Using Residual Perfectly Matched Layer

The perfectly matched layer(PML) is commonly used in wave propagation, radiation and diffraction problems in unbounded space domains. A new implementation scheme of PML is presented. The PML formulation is pre-defined, and the wave field absorption is achieved by calculating the residual between the PML equation and original equation through backward induction. Two forms of the Residual PML (RPML) are presented: RPML-1, which defines the residual as the difference between the original and PML equations, and RPML-2, which defines the residual as the difference between the original and PML wave fields. RPML-2 is the simplest and easiest to extend, as it does not alter the original equation and only has one time partial derivative term in the residual equation. Additionally, since the residual equation has no spatial partial derivative term, high-order spatial difference discretization is unnecessary, which results in higher accuracy and computational efficiency. Furthermore, simulating a wave field in TTI media requires a high absorption effect and stability of PML. The numerical simulation demonstrates that RPML-2 provides better absorption performance and stability compared to ADEPML and NPML. To meet the needs of wave field simulation for complex media, a multiaxial complex frequency shifted RPML-2 (MCFS-RPML-2) is introduced, which employs double damping profiles and complex frequency shift technology to achieve higher stability and absorption effects.

physics.geo-ph↗

arXiv2024

Seismic Interpolation Transformer for Consecutively Missing Data: A Case Study in DAS-VSP Data

Distributed optical fiber acoustic sensing (DAS) is a rapidly-developed seismic acquisition technology with advantages of low cost, high resolution, high sensitivity, and small interval, etc. Nonetheless, consecutively missing cases often appear in real seismic data acquired by DAS system due to some factors, including optical fiber damage and inferior coupling between cable and well. Recently, some deep-learning seismic interpolation methods based on convolutional neural network (CNN) have shown impressive performance in regular and random missing cases but still remain the consecutively missing case as a challenging task. The main reason is that the weight sharing makes it difficult for CNN to capture enough comprehensive features. In this paper, we propose a transformer-based interpolation method, called seismic interpolation transformer (SIT), to deal with the consecutively missing case. This proposed SIT is an encoder-decoder structure connected by some U-shaped swin-transformer blocks. In encoder and decoder part, the multi-head self-attention (MSA) mechanism is used to capture global features which is essential for the reconstruction of consecutively missing traces. The U-shaped swin-transformer blocks are utilized to perform feature extraction operations on feature maps with different resolutions. Moreover, we combine the loss based on structural similarity index (SSIM) and L1 norm to propose a novel loss function for SIT. In experiments, this proposed SIT outperforms U-Net and swin-transformer. Moreover, ablation studies also demonstrate the advantages of new network architecture and loss function.

physics.geo-ph↗