Seismic Image Embedding Comparison for Perceptual Difference Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Comparing seismic images taken at different times or under different conditions to determine subtle differences is challenging due to variations in interpretation processes and noise attenuation.
Innovation Solution
A method involving neural networks to partition seismic images into windowed images, generate embeddings, determine differences based on these embeddings, and interpolate similarity tiles to display perceptual differences using a decoder.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple seismic images are compared to determine subtle differences, then the ability to detect subsurface changes is improved, but the difficulty of detecting and measuring increases due to variations in interpretation processes and noise attenuation
Solution Approach 1:
The patent segments the seismic images into multiple patches, which are then processed independently through the neural network. This segmentation approach breaks down the complex task of comparing entire seismic images into manageable smaller units, reducing the overall computational complexity while maintaining the ability to detect subtle differences across the full image domain
Solution Approach 2:
The patent introduces embeddings as an intermediary representation between the input seismic images and the final difference maps. The neural network transforms the raw image data into embedding spaces where comparisons can be performed more effectively, serving as a mediator that simplifies the detection of subtle differences while accounting for variations in interpretation processes
2Measurement precision
If neural networks are used to generate embeddings and determine differences, then the accuracy of difference detection is improved, but the device complexity increases
Solution Approach 1:
The patent uses the encoder neural network to create embedding copies of the input seismic image patches. These embeddings are simplified representations that capture the essential features needed for comparison. By working with these copied embedding representations rather than the original complex image data, the system achieves accurate difference detection while managing computational complexity
Solution Approach 2:
The patent transforms the spatial domain seismic image data into an embedding space through the neural network encoder. This dimensionality change allows the system to perform comparisons in a transformed feature space where subtle differences are more apparent and can be detected with higher accuracy, while the structure of the transformation manages the complexity
3Loss of information
If similarity tiles are generated and interpolated to create output seismic images, then the visualization of perceptual differences is improved, but the processing time increases
Solution Approach 1:
The patent processes the seismic images in segmented patches rather than as complete images. This segmentation allows the system to generate similarity tiles and perform interpolations on smaller, more manageable units, reducing the overall processing time while ensuring that perceptual differences are preserved through systematic coverage of the entire image domain
Solution Approach 2:
The patent generates similarity tiles for representative patches and uses interpolation to extend these results to the full image domain. This partial action approach processes only essential portions in detail while using mathematical interpolation to fill in the remaining areas, reducing processing time while maintaining the preservation of perceptual differences across the complete image
Data Source
AI summary
A method for displaying seismic images includes receiving a first seismic image and a second seismic image, partitioning the first seismic image into a first windowed image, and the second seismic image into a second windowed image, generating embeddings based at least in part on the first and second windowed images using an encoder comprising a neural network, determining differences between in the first and second seismic images based at least in part on the embeddings, generating similarity tiles representing at least some of the differences, generating output seismic images by interpolating the similarity tiles using a decoder comprising a neural network, and displaying perceptual differences in the first and second seismic images generated based at least in part on the output seismic images.