Machine learning research: Convolutional Neural Networks and oil saturation in core images.
Researchers: Daniel Torres, Salvador Vargas
Researchers: Daniel Torres, Salvador Vargas
Core is a piece of cylindrical rock that oil companies recover from the underground using special diamond tip fitted drills.
Core is used to estimate the amount of oil in a reservoir and the quality of the rock.
According to our research, using a technique called transfer learning, CNN's are able to classify oil saturation values.
To train, we obtained core images under ultra-violet light and the associated data associated with the 1 foot intervals on the image. To label, we used a python script to automatically label 167 x 167 sized images with the corresponding oil saturation. These images were used to train the neural networks.
The results below show testing accuracy as a function of an epoch and conclude that the neural networks hover around 86% accuracy.
