Seismic Feature Segmentation via Deep Learning Dimensionality Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current backpropagation-enabled processes for identifying subsurface features in seismic data are inefficient and computationally intensive, especially when requiring detailed and refined classifications, leading to increased computational time and resource usage.
Innovation Solution
A method for training a backpropagation-enabled segmentation process that inputs multi-dimensional seismic data into a deep learning framework, where predictions are made on a reduced dimensionality grid, allowing for efficient identification of subsurface features by using convolutional neural networks and downscaling steps to reduce data dimensions while maintaining feature map content modification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If detailed and refined classification of subsurface features is performed using conventional backpropagation-enabled processes, then classification accuracy is improved, but computational time and resource usage increase significantly
Solution Approach 1:
The seismic data processing is divided into multiple passes or stages. A first pass performs coarser classification to identify potential features, and subsequent passes refine the classification in specific regions of interest. This segmentation allows the system to achieve detailed classification accuracy only where needed, rather than performing exhaustive analysis on the entire dataset, thereby reducing overall computational time while maintaining classification precision in critical areas.
2Reliability
If full multi-dimensional seismic data is processed for complete classification, then comprehensive feature identification is achieved, but computational resources and processing complexity increase
Solution Approach 1:
The system performs preliminary processing steps before full classification, such as preprocessing seismic data to enhance signal quality, applying initial filtering to remove noise, or performing quick assessments to identify regions containing subsurface features. These preliminary actions prepare the data in advance, allowing subsequent detailed classification to focus only on relevant portions, thereby reducing overall computational resource requirements while maintaining complete feature identification.
3Productivity
If conventional backpropagation processes are used for seismic data classification, then subsurface features can be identified, but processing efficiency decreases and resource consumption increases
Solution Approach 1:
The system applies classification processing selectively rather than uniformly across all seismic data. It performs detailed classification only on portions of the data where subsurface features are likely to be present, based on preliminary indicators or regions of interest. For other portions, simplified or skipped processing is applied. This partial action approach maintains productivity for critical feature identification while significantly reducing overall computational resource consumption and energy usage.
Data Source
AI summary
A method for training a backpropagation-enabled segmentation process is used for identifying an occurrence of a sub-surface feature. A multi-dimensional seismic data set with an input dimension of at least two is inputted into a backpropagation-enabled process. A prediction of the occurrence of the subsurface feature has a prediction dimension of at least 1 and is at least 1 dimension less than the input dimension.