Detecting micro fractures: a comprehensive comparison of conventional and machine-learning-based segmentation methods

Abstract. Studying porous rocks with X-ray computed tomography (XRCT) has been established as a standard procedure for the non-destructive characterization of flow and transport in media. Despite recent advances field XRCT, various challenges still remain due to inherent noise imaging artifacts produced data. These issues become even more profound when objective is identification fractures and/or fracture networks. One challenge limited contrast between regions interest neighboring areas, which can mostly be attributed minute aperture fractures. In order overcome this challenge, it common approach apply digital image processing steps, such filtering, enhance signal-to-noise ratio. Additionally, segmentation methods based on threshold/morphology schemes have employed obtain enhanced information from features interest. However, workflow needs skillful operator fine-tune its input parameters, required computation time significantly increases complexity available large volume an XRCT dataset. study, dataset by successful visualization network Carrara marble micro (?XRCT), we present results five methods, three conventional two machine-learning-based ones. The provide interested reader comprehensive comparison existing approaches while presenting operating principles, advantages limitations, serve guide towards individualized workflow. all are compared each other terms quality efficiency. Due memory accomplish fair comparison, 2D scheme. output U-net model, one adopted shows best performance regarding time.