Adversarial Neural Network for Seismic Fault Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing seismic fault detection methods using neural networks suffer from low accuracy and resolution due to a focus on local features and inadequate consideration of background information, leading to fuzzy boundaries and low reliability in fault identification.
Innovation Solution
A method and system employing an adversarial neural network with a segmentation module, feature fusion module, and discriminator module, where local fault features and seismic data are fused into a global feature map, using balanced and categorical cross-entropy loss functions, and regularization techniques to enhance prediction and generalization abilities.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional U-Net networks are used for fault detection, then the network structure is simple and easy to implement, but the detection accuracy and resolution are low with fuzzy boundaries
Solution Approach 1:
The patent combines the segmentation module (U-Net) with a discriminator module to form an adversarial neural network. The segmentation module extracts local fault features while the discriminator module evaluates global seismic patterns, merging local and global feature processing to improve detection accuracy and resolve fuzzy boundaries.
Solution Approach 2:
The patent introduces a new dimension of adversarial training by adding the discriminator module that operates in parallel to the segmentation module. This additional dimension enables the network to learn both local fault characteristics and global seismic context simultaneously, improving resolution without simply increasing network depth.
2Reliability
If image processing techniques and data enhancement are used, then training data diversity is increased, but fault features are highlighted by weakening background information resulting in low resolution
Solution Approach 1:
The patent applies local quality by using the segmentation module to focus on local fault features (edges, discontinuities) while the discriminator module simultaneously evaluates global seismic patterns. This allows the system to enhance fault features without completely suppressing background information, maintaining both reliability and precision.
Solution Approach 2:
The discriminator module acts as an intermediary that evaluates the segmentation output and provides feedback through adversarial training. It mediates between local fault feature extraction and global seismic context, ensuring that background information is not completely weakened but rather integrated appropriately to maintain boundary precision.
Data Source
AI summary
The present disclosure provides a method and a system for high-resolution seismic fault detection by means of an adversarial neural network, including following steps of: training a target adversarial neural network based on a preset training sample set, so as to obtain a trained target adversarial neural network, wherein the preset training sample set includes seismic data and fault labels, the target adversarial neural network includes: a segmentation module, a feature fusion module, and a discriminator module, the segmentation module is a module configured for obtaining a fault feature based on the preset training sample set, and the feature fusion module is a module configured for fusing the fault feature and the seismic data into a global feature map; and performing seismic fault detection on a target seismic image based on the trained target adversarial neural network.