Deep Learning Method for Lithology Identification from Borehole Images

Lithology identification is one of the keys to understand the nature of hydrocarbon reservoir. Deep learning has become a popular and reliable method for image classification and in other fields. Instead of using ordinary neural networks and conventional logging curves, this paper developed deep learning methods and showed that it is possible to identify lithology, using results from borehole image logs. In this work, a Convolutional Neural Network (CNN), which consists of two convolutional layers, two pooling layers and one fully-connected layer, is employed to identify lithology. Training is performed through back-propagation using the stochastic gradient descent algorithm with Nesterov Momentum. The trained CNN can be applied to new wells and provide accurate output (about 95%) of lithology types.