Generating and Annotating Thin Section Images of Tight Oil Reservoirs Using Deep Learning
Author Information
Author(s): Liu Tao, Liu Zongbao, Zhang Kejia, Li Chunsheng, Zhang Yan, Mu Zihao, Mu Mengning, Xu Mengting, Zhang Yue, Li Xue
Primary Institution: Northeast Petroleum University
Hypothesis
Can deep learning methods effectively generate and annotate thin section images of tight oil reservoirs?
Conclusion
The proposed method significantly improves the quality and efficiency of generating and annotating thin section images.
Supporting Evidence
- The CA-StyleGAN method achieved the highest Inception Score, indicating better image diversity.
- The method demonstrated lower Frechet Inception Distance, suggesting higher similarity to original images.
- SALM achieved annotation accuracy between 61% and 72%, showing practical applicability.
Takeaway
This study shows how computers can help create and label pictures of rocks to make it easier for scientists to study them.
Methodology
The study used a deep learning model called SACA-StyleGAN to generate and semi-automatically annotate thin section images.
Limitations
The method still has issues with long training times and the accuracy of annotations needs improvement.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website