A projection-onto-convex-sets network for 3D seismic data interpolation

• DOI: 10.1190/geo2022-0326.1
• Publication Date: 2023-02-15T13:32:23Z
• Authors: Yao Chen; Siwei Yu; Jianwei Ma
• Keywords: Interpolation

Seismic data interpolation is an essential procedure in seismic processing. However, conventional methods may generate inaccurate results due to the simplicity of assumptions, such as linear events or sparsity. In contrast, deep learning trains a neural network with large set without relying on predefined assumptions. lack physical priors traditional pure data-driven frameworks cause low generalization for different sampling patterns. Inspired by framework projection onto convex sets (POCS), new proposed interpolation, called POCS-Net. The forward Fourier transform, inverse and threshold parameter POCS are replaced networks that independent iterations. trainable POCS-Net rather than manually set. A nonnegative constraint imposed make it consistent POCS. essentially end-to-end pattern iterative framework. Numerical 3D synthetic field demonstrate superiority reconstruction accuracy method compared natural image-learned methods.