Deep Learning Extrapolation of Green's Function Beyond Receiver Grids
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing data-dependent re-datuming techniques for estimating Green's function in seismic imaging are unstable due to the presence of seabed tilt and sharp lateral variations, making multi-dimensional deconvolution (MDD) necessary but limited by receiver grid restrictions and computational challenges.
Innovation Solution
A method using generative adversarial networks (GANs) for deep learning is employed to transform upward-downward diffusion (UDD) seismic images into multi-dimensional deconvolution (MDD) images, iteratively training a generator and discriminator to enhance the estimation of Green's function beyond the receiver grid.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If up-down deconvolution (UDD) is used to estimate Green's function, then computational simplicity is improved, but stability deteriorates in the presence of seabed tilt and sharp lateral variations
Solution Approach 1:
The patent introduces a deep learning model as an intermediary between UDD and MDD. The model is trained to translate UDD results into MDD-equivalent results, combining the computational simplicity of UDD with the stability of MDD. This intermediary system resolves the contradiction by decoupling the computational process from the stability requirement.
Solution Approach 2:
The patent changes the parameter space by transforming the problem from direct MDD computation to a deep learning translation task. The model learns optimal parameter transformations that map UDD output characteristics to MDD output characteristics, achieving stability without the computational burden of traditional MDD.
2Reliability
If multi-dimensional deconvolution (MDD) is used to improve stability, then reliability is improved, but computational complexity and receiver grid restrictions worsen
Solution Approach 1:
The patent creates a computational copy of MDD's stabilizing effect through the deep learning model. Instead of performing actual MDD computations, the model learns to replicate MDD's output characteristics from UDD inputs, achieving the stability benefit without the computational cost.
Solution Approach 2:
The patent replaces the mechanical deconvolution process with a neural network-based system. The deep learning model substitutes the traditional signal processing mechanics with learned patterns, eliminating receiver grid restrictions and reducing computational complexity while maintaining stability.
3Measurement precision
If multi-dimensional deconvolution (MDD) is used to estimate Green's function, then accuracy is improved, but receiver grid limitations worsen
Solution Approach 1:
The deep learning model serves multiple functions: it maintains the accuracy benefits of MDD while simultaneously removing receiver grid limitations. The model can process inputs from various grid configurations and produce accurate outputs, making the system universally applicable across different acquisition geometries.
Data Source
AI summary
A method for transforming seismic images includes receiving input data. The input data includes an original upward-downward diffusion (UDD) seismic image and an original multi-dimensional deconvolution (MDD) seismic image. The method also includes training a generator and a discriminator based upon the input data to produce a trained generator and a trained discriminator.