Source rocks are composed of inorganic minerals, clay particles, and organic matter that are compacted to create a composite with texture and pore sizes that vary at the nano-and micro-scales [1-2]. These rocks exhibit chemical variations due to the dominant framework phases ( e.g . calcite, dolomite, or quartz) and the clay minerals ( e.g . illite, smectite, or kaolinite), while exhibiting variations in the organic components (kerogen, bitumen, and/or pyrobitumen) where the relative amount and molecular composition varies with thermal maturity [2-4]. Scanning Electron Microscopy (SEM) is commonly used for source rock characterization to quantify the fraction of various components, porosity, and pore size distribution, which are used to evaluate reservoir quality during hydrocarbon exploration. This variation in structural and chemical heterogeneity as well as imaging artifacts such as charging, surface contamination, and surface damage from sample preparation, create images with a broad multi-modal intensity histogram that is challenging to accurately segment [5]. Traditionally, image segmentation applies user-defined or mathematically selected gray-scale threshold cutoffs to separate the image components after pre-processing and has
Paper
Full text
Quantitative Image Analysis of Source Rocks Using Machine Learning Segmentation
Semantic Scholar · Computer Science · 2020
Abstract
Source rocks are composed of inorganic minerals, clay particles, and organic matter that are compacted to create a composite with texture and pore sizes that vary at the nano-and micro-scales [1-2]. These rocks exhibit chemical variations due to the dominant framework phases ( e.g . calcite, dolomite, or quartz) and the clay minerals ( e.g . illite, smectite, or kaolinite), while exhibiting variations in the organic components (kerogen, bitumen, and/or pyrobitumen) where the relative amount and molecular composition varies with thermal maturity [2-4]. Scanning Electron Microscopy (SEM) is commonly used for source rock characterization to quantify the fraction of various components, porosity, and pore size distribution, which are used to evaluate reservoir quality during hydrocarbon exploration. This variation in structural and chemical heterogeneity as well as imaging artifacts such as charging, surface contamination, and surface damage from sample preparation, create images with a broad multi-modal intensity histogram that is challenging to accurately segment [5]. Traditionally, image segmentation applies user-defined or mathematically selected gray-scale threshold cutoffs to separate the image components after pre-processing and has