Fast Reservoir Characterization with AI-Based Lithology Prediction Using Drill Cuttings Images and Noisy Labels

In this paper, we considered one of the problems that arise during drilling automation, namely automation lithology identification from drill cuttings images. Usually, work is performed by experienced geologists, but a tedious and subjective process. Drill are cheapest source rock formation samples; therefore, reliable prediction can greatly reduce cost analysis drilling. To predict content images samples, used convolutional neural network (CNN). For training model with an acceptable generalization ability, applied dataset-cleaning techniques, which help to reveal bad as well samples uncertain labels. It was shown trained on cleaned dataset performs better in terms accuracy. Data cleaning using cross-validation technique, clustering embeddings, where it possible identify clusters distinctive visual characteristics visually similar rocks attributed different lithologies labeling