Abstract:We propose a new method to tackle the mapping challenge from time-series data to spatial image in the field of seismic exploration, i.e., reconstructing the velocity model directly from seismic data by deep neural networks (DNNs). The conventional way of addressing this ill-posed inversion problem is through iterative algorithms, which suffer from poor nonlinear mapping and strong nonuniqueness. Other attempts may either import human intervention errors or underuse seismic data. The challenge for DNNs mainly lies in the weak spatial correspondence, the uncertain reflection-reception relationship between seismic data and velocity model, as well as the time-varying property of seismic data. To tackle these challenges, we propose end-to-end seismic inversion networks (SeisInvNets) with novel components to make the best use of all seismic data. Specifically, we start with every seismic trace and enhance it with its neighborhood information, its observation setup, and the global context of its corresponding seismic profile. From the enhanced seismic traces, the spatially aligned feature maps can be learned and further concatenated to reconstruct a velocity model. In general, we let every seismic trace contribute to the reconstruction of the whole velocity model by finding spatial correspondence. The proposed SeisInvNet consistently produces improvements over the baselines and achieves promising performance on our synthesized and proposed SeisInv data set according to various evaluation metrics. The inversion results are more consistent with the target from the aspects of velocity values, subsurface structures, and geological interfaces. Moreover, the mechanism and the generalization of the proposed method are discussed and verified. Nevertheless, the generalization of deep-learning-based inversion methods on real data is still challenging and considering physics may be one potential solution.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the mapping challenge from time - series data (i.e., seismic data) to spatial images (i.e., velocity models) in the field of seismic exploration. Specifically, the paper proposes a new method to directly reconstruct the velocity model from seismic data through deep neural networks (DNNs) to address the difficulties encountered by traditional iterative algorithms in solving this nonlinear inversion problem, such as poor nonlinear mapping, strong non - uniqueness, and possible human - intervention errors or insufficient use of seismic data. ### Core Problems of the Paper 1. **Nonlinear Mapping and Strong Non - uniqueness**: When processing seismic data, traditional iterative algorithms are prone to fall into local minima due to strong nonlinearity and non - uniqueness of solutions, and they have high requirements for the initial model. 2. **Human - Intervention Errors**: Some traditional methods require human intervention when processing seismic data, which may lead to errors. 3. **Spatio - Temporal Characteristics of Seismic Data**: - **Weak Spatial Correspondence**: There is a weak spatial correspondence between the reflection signals in seismic data and the interfaces in the velocity model. - **Uncertain Reflection - Reception Relationship**: The reflection signals received by the same receiver may come from different interfaces, and the signals received by different receivers may be the same. - **Time - Varying Characteristics**: Seismic waves gradually weaken over time, making it difficult to capture patterns. ### Solutions To solve the above problems, the paper proposes an end - to - end deep - learning framework - Seismic Inversion Network (SeisInvNet), whose main features include: 1. **Embedding Encoder**: - **Neighborhood Information**: Extract the neighborhood information of each seismic profile through a shallow CNN. - **Observation Settings**: Convert the positions of receivers and sources into one - hot vectors. - **Global Context**: Extract the global context information of the seismic profile through the encoder. 2. **Spatially Aligned Feature Generator**: - Use multi - layer perceptrons (MLPs) to further compress the embedding vectors into feature maps, and the information of these feature maps is aligned with the spatial positions of the velocity model. 3. **Velocity Model Decoder**: - Gather all feature maps and regress the velocity model through CNNs. During the training process, the dropout technique is used to improve the robustness and generalization ability of the model. 4. **Loss Function**: - In addition to the commonly used L2 norm, multi - scale structural similarity (MSSIM) is also introduced to optimize the local structure and details of the model. ### Main Contributions 1. Conduct in - depth analysis of DNN - based seismic inversion problems. 2. Design SeisInvNet to make full use of all seismic data through novel and efficient components. 3. Demonstrate performance superior to the baseline model on the self - built SeisInv dataset. 4. Provide comprehensive mechanism research, including visualization and statistical analysis. Through these methods, the paper effectively solves the mapping problem from seismic data to velocity models and improves the accuracy and consistency of inversion results.

Deep-Learning Inversion of Seismic Data

Deep-learning inversion: A next-generation seismic velocity model building method

A Physics-Based Neural-Network Way To Perform Seismic Full Waveform Inversion

RETRACTED: Seismic Impedance Inversion in Depth Domain Based on Deep Learning

Deep-Learning Full-Waveform Inversion Using Seismic Migration Images

Deep Learning with Adaptive Attention for Seismic Velocity Inversion

Joint Inversion of Seismic and Resistivity Data Powered by Deep-learning

Seismic Impedance Inversion Based on Deep Learning with Geophysical Constraints

Deep learning for high-resolution multichannel seismic impedance inversion

Joint Inversion of Audio-Magnetotelluric and Seismic Travel Time Data With Deep Learning Constraint

Seismic Inversion Based on Acoustic Wave Equations Using Physics-Informed Neural Network

Evolution Inversion: Co-evolution of Model and Data for Seismic Reservoir Parameters Inversion

Deep-learning seismic full-waveform inversion for realistic structural models

Deep Learning Inversion of Rayleigh-wave Dispersion Curves with Geological Constraints for Near-Surface Investigations

Building Complex Seismic Velocity Models for Deep Learning Inversion

Deep Learning Inversion of Electrical Resistivity Data

Research on Elastic Parameter Inversion Method Based on Seismic Facies-Controlled Deep Learning Network

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

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

High-resolution Seismic Impedance Inversion Integrating the Closed-Loop Convolutional Neural Network and Geostatistics: an Application to the Thin Interbedded Reservoir

A Deep Learning Gravity Inversion Method Based on a Self-Constrained Network and Its Application