Related Experiment Video
Updated: Jan 7, 2026
08:27
Author Spotlight: Efficient Image Recognition Using Directional Gradient Histogram Technique and Support Vector Machines
Published on: January 5, 2024
1.6K
Wavelet-convolutional neural network for fault prediction in coal mine seismic data.
Guangui Zou1, Chengyang Han2, Hen-Geul Yeh3
1State Key Laboratory for Fine Exploration and Intelligent Development of Coal Resources, China University of Mining and Technology-Beijing, Beijing, 100083, China.
Scientific Reports
|December 26, 2025
Summary
A new Wavelet-Convolutional Neural Network (W-CNN) improves seismic fault identification by integrating discrete wavelet transforms with CNNs. This approach enhances accuracy and efficiency for geohazard prevention and resource exploration.
Area of Science:
- Geophysics
- Artificial Intelligence
- Signal Processing
Background:
- Seismic fault identification is crucial for resource exploration and geohazard prevention.
- Conventional methods are limited by subjective interpretation and inefficiency.
- Existing convolutional neural networks (CNNs) overlook multiscale frequency features, impacting accuracy.
Purpose of the Study:
- To introduce a novel Wavelet-Convolutional Neural Network (W-CNN) for enhanced seismic fault identification.
- To address the limitations of conventional methods and standard CNNs in capturing spatial-frequency information.
- To develop an accurate and efficient automated seismic fault detection system.
Main Methods:
- Architectural fusion of discrete wavelet transforms (DWT) with CNNs, creating a spatial-frequency learning paradigm.
- Embedding Haar wavelet filter banks with cross-scale residual connections within the W-CNN architecture.
- Development of W-CNN variants (W-CNN R1, R2, R3) for performance evaluation.
Main Results:
- W-CNN R3 achieved 90.0% accuracy and 90.3% F1-score on coal mine datasets, outperforming mainstream CNNs by up to 12.3%.
- The model demonstrated a 93.8% detection rate for complex micro-faults and superior recall (95.5%).
- W-CNN showed a 21% reduction in parameters, faster convergence, noise suppression, and accelerated 3D processing.
Conclusions:
- The proposed W-CNN framework offers an effective spatial-frequency learning paradigm for seismic fault identification.
- This approach significantly enhances accuracy, efficiency, and predictive capability for geological discontinuities, especially small-scale ones.
- The W-CNN provides an extensible solution for intelligent geological interpretation and mine safety monitoring.
