Abstract: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.

What problem does this paper attempt to address?

The paper aims to address the issue of missing or inaccurate low-frequency information in full waveform inversion (FWI) for seismic data processing. Low-frequency information is crucial for improving the accuracy of FWI because it can reduce the scattering and absorption of reflected waves during propagation, thereby enhancing the accuracy of subsurface geological interpretation and helping to prevent FWI from converging to local minima. However, in actual collected seismic data, low-frequency information is often difficult to obtain directly or is suppressed due to noise pollution and limitations of acquisition equipment. To solve the above problem, the paper proposes a new network structure called the Low-Frequency Extrapolation Transformer (LFET), which is a deep learning method based on the transformer model, used to predict missing low-frequency information from band-limited seismic records. LFET constructs a nonlinear mapping relationship between band-limited data and low-frequency data through supervised learning. Specifically, LFET employs a Multi-Head Self-Attention (MSA) module to model the long-range dependencies of seismic data and introduces a window segmentation method to reduce computational cost. To train LFET, researchers used synthetic seismic records generated from submodels of the benchmark Marmousi model as training data, which have characteristics similar to actual field data. The input to LFET is a single time-domain band-limited seismic record, and its parameters are updated based on the error between the predicted record and the corresponding label. Experimental results show that LFET can effectively predict accurate low-frequency information from band-limited seismic records. Compared to traditional Convolutional Neural Network (CNN) methods, LFET not only achieves higher accuracy but also has lower computational cost. Additionally, LFET demonstrates good generalization ability, achieving accurate predictions on different models and different types of field data. Finally, applying the low-frequency information predicted by LFET to the FWI process proves the effectiveness and usability of this low-frequency information, significantly improving FWI results, mitigating cycle-skipping issues, and enhancing the accuracy of inversion results.

Transformer For Low-frequency Extrapolating of Seismic Data

Layered low-frequency extrapolation with deep learning in full-waveform inversion

Learning Seismic Low-Frequency Extrapolation in a Latent Space Using Limited Data

Extrapolated full-waveform inversion with deep learning

Deep Learning-Based Low-Frequency Extrapolation and Impedance Inversion of Seismic Data

Global-Feature-Fusion and Multiscale Network for Low-Frequency Extrapolation

A self-supervised learning framework for seismic low-frequency extrapolation

Prediction of Low-Frequency Seismic Data Without Relying on Wavelet Using Deep Learning

Irregularly Spatial Seismic Missing Data Reconstruction Using Transformer With Periodic Skip Connection

Efficient progressive transfer learning for full waveform inversion with extrapolated low-frequency reflection seismic data

Learning with real data without real labels: A strategy for extrapolated full-waveform inversion with field data

Deep Learning for Low-Frequency Extrapolation and Seismic Acoustic Impedance Inversion

Low-Frequency Extrapolation of Prestack Viscoacoustic Seismic Data Based on Dense Convolutional Network

Data-driven low-frequency signal recovery using deep-learning predictions in full-waveform inversion

LiteTransNet: an Interpretable Approach for Landslide Displacement Prediction Using Transformer Model with Attention Mechanism

Semi-Supervised Seismic Impedance Inversion With Convolutional Neural Network and Lightweight Transformer

A Multi-Task Fourier Transformer Network for Seismic Source Characterization Estimation from a Single-Station Waveform

Low-frequency constrained seismic impedance inversion combining large kernel attention and long short-term memory

Frequency Learning Via Multi-Scale Fourier Transformer for MRI Reconstruction.

FL-Net: A multi-scale cross-decomposition network with frequency external attention for long-term time series forecasting

Spatiotemporal wave forecast with transformer-based network: A case study for the northwestern Pacific Ocean