960化工网/ 文献/ 论文详情

Segmentation of digital rock images using texture analysis and deep network

ShahzadManzoor,TehreemQasim,NaeemBhatti,MuhammadZia
Arabian Journal of Geosciences Pub Date : 06/27/2023 00:00:00 , DOI:10.1007/s12517-023-11549-0
Abstract
Image segmentation is an important part of the standard digital rock physics (DRP) workflows. In this paper, we present ground-truthing of digital rock images using texture analysis. We propose a deep learning–based approach for automated segmentation which is validated using the extracted ground-truth. To generate the ground-truth, we utilize the a priori knowledge of minerals and pores. We perform texture based iterative intensity thresholding to calculate the porosity and mineral composition. The ground-truth is generated for the berea sandstone image dataset and the grosmont carbonate image dataset. Results obtained from the experiments conform to the porosity and composition ranges provided in the literature. For the purpose of automated segmentation, we use a SegNet deep network which performs pixel-wise segmentation of digital rock images into pores and minerals. We train and test the SegNet using 1024 images of berea sandstone dataset ground-truthed using iterative intensity thresholding. We utilize a split of 90% and 10% for training and testing, respectively. Experimental outcomes show that the SegNet-based segmentation results for the minerals such as clay, ankerite, and zircon are better than the existing methods.
引用文献
推荐文献
微信二维码
  • 微信公众号二维码
  • 关注官方微信公众号
  • 微信二维码
  • 微信扫码联系客服
平台客服