Intelligent segmentation and digital reconstruction of rock fractures: Methodology and application

Abstract

Accurate extraction and quantification of rock fracture networks are critical for structural assessment and failure prediction in geotechnical and mining engineering. This study proposes a topology-enhanced fracture segmentation model, gated recurrent convolutional network (GRCNet), employing gated recurrent convolution blocks to adaptively fuse multi-directional dynamic snake convolution to improve the detection of slender and irregular fracture patterns in images. The GRCNet model achieves high performance on the DeepCrack road crack dataset (centerline dice coefficient = 90.22%) and generalizes well to the CS2025 rock fracture dataset (mean intersection over union = 86.18%, average symmetric surface distance = 3.038), demonstrating robustness across domains. To enable integration with simulation, the segmented fractures are digitally reconstructed as box or polyline sets for geometric characterization. A dedicated software system is developed to automate the pipeline—from image-based fracture segmentation to numerical data generation—bridging the gap between AI-powered interpretation and computational modeling. This pipeline is validated through two numerical examples: simulating uniaxial compression of fractured rock and tunnel collapse in jointed rock masses. The proposed framework offers a flexible and accurate tool for combining computer vision and simulation, contributing to next-generation design automation and computational intelligence in rock engineering.

Keywords

rock fractures / intelligent segmentation / digital reconstruction / gated recurrent convolution / numerical simulation

Rights & permissions

Higher Education Press 2026

Just Accepted

This article has successfully passed peer review and final editorial review, and will soon enter typesetting, proofreading and other publishing processes. The currently displayed version is the accepted final manuscript. The officially published version will be updated with format, DOI and citation information upon launch. We recommend that you pay attention to subsequent journal notifications and preferentially cite the officially published version. Thank you for your support and cooperation.