Identification of rock pore structures and permeabilities using electron microscopy experiments and deep learning interpretations. (15th May 2020)

• Record Type: Journal Article
• Authors: Yu, Qingyang; Xiong, Ziwei; Du, Chao; Dai, Zhenxue; Soltanian, Mohamad Reza; Soltanian, Mojtaba; Yin, Shangxian; Liu, Wei; Liu, Chen; Wang, Chengbin; Song, Zeyu
• Is Part Of: Fuel. Volume 268(2020)
• Publication Date: 2020-05-15
• Subjects: Deep learning -- Fully convolutional neural network -- Permeability parameters -- SEM image; Fuel -- Periodicals; Coal -- Periodicals; Coal; Fuel; Periodicals; 662.6
• DOI: 10.1016/j.fuel.2020.117416
• Languages: English
• ISSNs: 0016-2361

Abstract

Abstract: Permeability is an important hydrogeological parameter for the quantitative evaluation of water resources and prediction of water inflow. In this study, we examine a typical water-bearing sandstone obtained in North China to explore the correlation between the microscopic pore characteristics and macroscopic permeability of the sandstone. In addition, pixel-level annotated data are generated from the images obtained in electron microscopy experiments for deep learning training. Using the deep learning framework, we analyse the pore characteristics through semantic image segmentation based on artificial intelligence and explore the relationship between the microscopic pore characteristics and the macroscopic permeability parameters of the sandstone. This method addresses the limitations of traditional image recognition methods, such as the inability to obtain the complete pore space characteristics in scanning electron microscopy (SEM) images as well as poor segmentation and low accuracy. Moreover, this method can be used to realise the full benefits of accurate image recognition, and it enables the automatic processing of microscopic images to significantly improve the accuracy of pore identification in rock samples.