Abstract:Due to the influence of natural and human factors, the actually collected seismic data often have the situation that the seismic trace data is missing, which will affect the subsequent seismic data processing and interpretation. Data processing and interpretation. However, traditional seismic data reconstruction methods usually have the problems that the seismic data reconstruction effect is constrained by prior conditions, the selection of hyperparameters requires manual intervention, and the degree of automation is low. This is an unavoidable drawback of current geophysical methods. Since its inception, the deep learning method has the advantages of high degree of automation, completed training, fast processing of target tasks, wide application fields, and flexibility. Therefore, people turn their attention to the rapidly developing field of deep learning. As of today, it has been There are many deep learning methods applied to the field of seismic data reconstruction to solve the above-mentioned problems in the seismic data reconstruction process. This paper will focus on the analysis of representative deep learning seismic data reconstruction methods, which are based on Convolutional Neural Network（CNN）, Recurrent Neural Network（RNN）, convolutional autoencoder, and generative adversarial network. The above three methods are selected from the following considerations. The CNN is the representative of the classic visual neural network and super-resolution image reconstruction, and it is very representative; the RNN is a significant attempt based on the characteristics of the seismic data sequence, because it gets rid of the long-term perception of deep learning methods. The thinking inertia of processing through the visual neural network; the generative adversarial network uses the network game theory, and the deep learning network is constantly adjusted in the continuous learning process. In this paper, the advantages and disadvantagesof the deep learning seismic data reconstruction method are discussed in depth through the reconstruction results residual comparison chart and the reconstruction result signal-to-noise ratio analysis. And further expound the research status, method advantages, existing problems and future development trends of deep learning seismic data reconstruction methods, and summarize and prospect the current deep learning reconstruction methods.

Unsupervised Deep Learning for Seismic Data Reconstruction

Intelligent Reconstruction for Spatially Irregular Seismic Data by Combining Compressed Sensing with Deep Learning

Unsupervised 3D seismic data reconstruction using a weighted-attentive deep learning framework

Deep Learning for Irregularly and Regularly Missing 3-D Data Reconstruction

Reconstruct 3-D Seismic Data With Randomly Missing Traces via Fast Self-Supervised Deep Learning

Reconstruct 3D Seismic Data with Randomly Missing Traces Via Fast Self-Supervised Deep Learning

Seismic Data Reconstruction Based on Multiscale Attention Deep Learning.

Seismic Data Reconstruction via Wavelet-Based Residual Deep Learning

Deep-seismic-prior-based reconstruction of seismic data using convolutional neural networks

Simultaneous Denoising and Reconstruction of Distributed Acoustic Sensing Seismic Data Via a Multicascade Deep-Learning Method

Deep learning for high-resolution seismic imaging

Sparse Seismic Data Reconstruction Based on a Convolutional Neural Network Algorithm

ADDCNN: an Attention-Based Deep Dilated Convolutional Neural Network for Seismic Facies Analysis with Interpretable Spatial–Spectral Maps

Ground-truth-free Deep Learning for 3-D Seismic Denoising and Reconstruction with Channel Attention Mechanism

Self-Supervised Deep Learning to Reconstruct Seismic Data With Consecutively Missing Traces

Self-Supervised Multitask 3-D Partial Convolutional Neural Network for Random Noise Attenuation and Reconstruction in 3-D Seismic Data

Review of deep learning seismic data reconstruction methods

Self-Supervised Learning for Seismic Data Reconstruction and Denoising

Unsupervised Deep Learning for Random Noise Attenuation of Seismic Data

Reconstructing missing seismic data using Deep Learning

Automatic Seismic Fault Interpretation with the Aid of Data-, Physics-, and Math-assisted Synthetic Data Generation