Zero-Shot Self-Consistency Learning for Seismic Irregular Spatial Sampling Reconstruction

TL;DR

This paper proposed a zero-shot self-consistency learning strategy and employed an extremely lightweight network for seismic data reconstruction, which does not require additional datasets and utilizes the correlations among different parts of the data to design a self-consistency learning loss function.

Abstract

Seismic exploration is currently the most important method for understanding subsurface structures. However, due to surface conditions, seismic receivers may not be uniformly distributed along the measurement line, making the entire exploration work difficult to carry out. Previous deep learning methods for reconstructing seismic data often relied on additional datasets for training. While some existing methods do not require extra data, they lack constraints on the reconstruction data, leading to unstable reconstruction performance. In this paper, we proposed a zero-shot self-consistency learning strategy and employed an extremely lightweight network for seismic data reconstruction. Our method does not require additional datasets and utilizes the correlations among different parts of the data to design a self-consistency learning loss function, driving a network with only 90,609 learnable parameters. We applied this method to experiments on the USGS National Petroleum Reserve-Alaska public dataset and the results indicate that our proposed approach achieved good reconstruction results. Additionally, our method also demonstrates a certain degree of noise suppression, which is highly beneficial for large and complex seismic exploration tasks.

Figures (8)

• Figure 1: The first column, from top to bottom, represents the irregularly sampled seismic data with 50$\%$ randomly missing, the seismic data reconstructed using ZS-SCL, the seismic data reconstructed using traditional learning, and the fully sampled field data (ground truth). The second column shows an enlarged view of the red boxes in the first column, the third column shows an enlarged view of the yellow boxes in the first column, and the forth column shows an enlarged view of the blue boxes in the first column.
• Figure 2: The location of our research area and lines.
• Figure 3: The first column represents the irregularly sampled data with 50$\%$ randomly missing, the second column shows the reconstructed data from ZS-SCL, and the third column displays the complete field data (ground truth).
• Figure 4: a), b), c) and d) represent the four seismic traces randomly extracted from Line 27; e), f), g) and h) represent the four seismic traces randomly extracted from Line 28.
• Figure 5: From left to right, they are: ground truth, irregularly sampled data, ZS-SCL processing results, and traditional learning processing results.