Advanced fracture segmentation from X-ray Tomography(XRT)images of reservoir core samples by machine

• 作 者: Chen Jin Jingsheng Ma
• 机 构: InstituteofPetroleumEngineering,Heriot-WattUniversity,Edinburgh,UK
• 出 处: EAGE-SCA International Symposium on Digital Rock Physics and
• 发表日期: 2016年3期

　　Extracting connected micro-fractures from 3D XRT images of carbonate and shale core samples is critically important to perform Digital Rock Property(DRP)analysis appropriately,but not trivial because of the complex intersections among the fractures and induced imaging artefacts.We consider a solution to this challenge by enforcing fracture connection as follows: fractures are traced on a training image to supress the effects appropriately,and the traces and that image are provided to the computer algorithms to learn rules,known as classifiers,to take into account the enforcements.The learnt classifiers can then be applied to other images to determine a point to belong to one or more fractures and the continuity of the fractures at that point.This supervised machine learning approach has been used in other disciplines to characterise connected curvilinear features.We demonstrate in this work that this approach is superior to a recent multiscale Hessian fracture filtering technique in capturing connectivity of the fractures.