期刊
APPLIED SCIENCES-BASEL
卷 11, 期 16, 页码 -出版社
MDPI
DOI: 10.3390/app11167736
关键词
ichnology; deep learning; transfer learning; artificial intelligence; trace fossils
类别
资金
- Natural Sciences and Engineering Research Council (NSERC) Discovery Grant
- College of Petroleum and Geosciences, KFUPM
The automated technique proposed in this study uses deep convolutional neural networks to determine bioturbation index in cores and outcrops, providing a faster and more robust solution compared to manual classification by experts. By compiling a large dataset and training the DCNN model, high classification accuracies were achieved, marking a pioneering work in the field of ichnological analysis.
Ichnological analysis, particularly assessing bioturbation index, provides critical parameters for characterizing many oil and gas reservoirs. It provides information on reservoir quality, paleodepositional conditions, redox conditions, and more. However, accurately characterizing ichnological characteristics requires long hours of training and practice, and many marine or marginal marine reservoirs require these specialized expertise. This adds more load to geoscientists and may cause distraction, errors, and bias, particularly when continuously logging long sedimentary successions. In order to alleviate this issue, we propose an automated technique to determine the bioturbation index in cores and outcrops by harnessing the capabilities of deep convolutional neural networks (DCNNs) as image classifiers. In order to find a fast and robust solution, we utilize ideas from deep learning. We compiled and labeled a large data set (1303 images) composed of images spanning the full range (BI 0-6) of bioturbation indices. We divided these images into groups based on their bioturbation indices in order to prepare training data for the DCNN. Finally, we analyzed the trained DCNN model on images and obtained high classification accuracies. This is a pioneering work in the field of ichnological analysis, as the current practice is to perform classification tasks manually by experts in the field.
作者
我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。
推荐
暂无数据