classification of reservoir rocks using deep learning

Porosity, permeability, and hydrocarbon in place in reservoirs are among the critical parameters in petroleum engineering. to find them, it is necessary first to determine and diagnose lithology. lithology can be determined by geological analysis of slabbed whole cores. this process is usually done by hand during macroscopic core studies, which is time-consuming and may be influenced by user bias. thus, using efficient and automatic methods for lithology detection is of prime interest. this study uses machine learning methods to identify lithology and classify rocks from whole core images. for this purpose, we compared three widely used network architectures (i.e., resnet-50, resnext-50, and a convolutional neural network). the architectures are coded in the pytorch library. 3000 meters of whole core images from 28 wells of sandstone reservoirs are used as a dataset. this approach is automatic and free of user bias. our result shows that resnext-50 could predict and classify the lithology of unseen whole core images with 96.78% accuracy.