Abstract:Reconstruction of missing seismic traces is one of the key steps in seismic data processing. Deep neural network-based interpolation methods for seismic trace reconstruction have attracted much attention in recent years. However, the currently widely used supervised-based deep neural network interpolation models require the support of labeled data, which is difficult to obtain in real seismic exploration. In this study, we propose a novel unsupervised deep learning interpolation framework SSTI (Self-supervised Seismic Trace Interpolation), designed to address the limitations of existing supervised-based models by avoiding the need of labeled data. SSTI employs a unique combination of two semi-supervised models: temporal ensembling model and mean teacher model. The temporal ensembling model and mean teacher model jointly predict the missing traces by integrating the outputs and weights of each training epoch through Exponential Moving Average method, respectively. Furthermore, a joint loss function that combines consistency loss and self-supervised loss is designed to guide the training process of SSTI. We evaluated the interpolation performance of SSTI on synthetic and field seismic data. Comparative analyses demonstrate that our proposed method outperforms the traditional theory-guided methods in terms of mean square error, peak signal to noise ratio, structural similarity, and quality of the reconstructed traces. Notably, although it does not achieve the interpolation performance of supervised deep learning methods, SSTI has shown great potential since there is only a marginal performance difference between SSTI and supervised-based methods. Our findings suggest that SSTI presents a promising advancement in the field of unsupervised deep learning methods for seismic trace interpolation. The proposed architecture holds great potential for recovering the missing and corrupted traces at a low cost in field seismic exploration.

Consecutively Missing Seismic Data Interpolation Based on Coordinate Attention Unet

Seismic Data Consecutively Missing Trace Interpolation Based on Multistage Neural Network Training Process

Attention and Hybrid Loss Guided Deep Learning for Consecutively Missing Seismic Data Reconstruction

Seismic data interpolation based on U-net with texture loss

Depthwise separable convolution Unet for 3D seismic data interpolation

Seismic Interpolation Via Multi-Scale HU-Net

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

An Improved Multi-Task Approach for SHM Missing Data Reconstruction Using Attentive Neural Process and Meta-Learning

A self-supervised missing trace interpolation framework for seismic data reconstruction

Consecutively Missing Seismic Data Reconstruction Via Wavelet-Based Swin Residual Network.

Three-Dimensional Velocity Field Interpolation Based on Attention Mechanism

Hybrid Loss-Guided Coarse-to-Fine Model for Seismic Data Consecutively Missing Trace Reconstruction

Fully Connected U-Net and Its Application on Reconstructing Successively Sampled Seismic Data

Hybrid Loss Guided 2d Multi-Task Full Attention U-Net for Prestack Seismic Inversion

Deep learning based seismic data interpolation: a preliminary result

Attention and Hybrid Loss Guided 2-D Network for Seismic Impedance Inversion

Multitask Full Attention U-Net for Prestack Seismic Inversion

ContrAttNet: Contribution and attention approach to multivariate time-series data imputation

Interpolation and Denoising of Seismic Data using Convolutional Neural Networks

Seismic data interpolation with recurrent inference mechanism