Shale Core Fracture Extraction Method Based on Edge Detection and Hierarchical Semantic Fusion Network

Shale gas is an emerging clean, stable and efficient energy. Counts of Shale core fracture distribution density, length and growth direction are necessary to establish a basic knowledge of total organic carbon content in the region and assess shale gas reservoirs. As fractures exhibit different orientations and scales with distribution density, it is difficult for professionals to label all fractures in core scanning electron microscopy images. This paper proposes a semantic segmentation network, HEF-Net (Hierarchical Extraction and Fusion Networks), to remedy these shortcomings by automatically extracting microfractures from scanning images of shale cores. It contains two branches, edge detection and convolutional structure, for detecting the complete contours of pores and fractures and filtering out mineral and background interference to extract the main content of large-scale fractures, which are subsequently fused into one by a feature fusion module for complete and accurate extraction. The comparison validates that our method leads the way in fracture boundary clarity and overall detection rate, achieving 84.26% for IoU and 90.57% & 91.05% for Dice and Pix-acc, respectively, on a shale sample dataset from the Fuling, Chongqing. Useful for digital core reconstruction studies and regional resource abundance assessments.