Core Image Classification using Deep Features

This poster demonstrates a machine learning workflow for whole core image classification. There have been significant advances in the field of image classification in recent years due to rise of AI technologies like deep learning. Convolutional neural networks (CNNs) are a popular deep architecture for image classification and computer vision tasks (Krizhevsky et al. 2012). However, CNNs require an enormous amount of training data to achieve accurate results. Labeled databases such as ImageNet (Russakovsky et al. 2015) contain millions of classified images (of general objects) that have been used to train networks to a high degree of accuracy (Krizhevsky et al. 2012). There aren’t currently geologic image databases of similar size. Here, we demonstrate that CNNs pre-trained on general image databases can be used to extract features from core images that can be used for geologic classification.