Automated extraction of mining-induced ground fissures using deep learning and object-based image classification

Abstract

Accurate extraction of ground fissures caused by intense coal mining has the potential to significantly improve the efficiency of environmental monitoring in mining areas. However, the extraction results using previous methods have often exhibited issues of discontinuity and substantial deviation from ground truth data, resulting in low extraction accuracy. In this study, a novel approach, ENVINet5-OBIC, for extracting ground fissures in mining areas is proposed, which integrates object-based image classification (OBIC) with the pixel-based deep learning model ENVINet5. ENVINet5-OBIC uses OBIC to segment high-resolution unmanned aerial vehicle (UAV) images across different scales, effectively considering shape, texture and correlative information between adjacent pixels. Furthermore, by utilizing homogeneous objects as building blocks, it establishes a deep learning model for the automated extraction of ground fissures. Experimental results show that ENVINet5-OBIC performs better when compared with OBIC, U-Net, PSPNet and ENVINet5 methods in terms of continuity, accuracy and error reduction. In addition, the ground fissure area extracted by ENVINet5-OBIC closely aligns with ground truth data. This study provides a more effective method for automatic extraction of ground fissures, which improves the efficiency of environmental monitoring in mining areas.