
Full text loading...
A large high-resolution image dataset of rock chips obtained from offshore wells, the North West Shelf of Australia, is obtained. The dataset consists of images of 158 rock chip samples that comprise single-lithology samples of 21 carbonates, 57 sandstones, 46 mudstones, and 34 volcanics. The images are used to create a labelled dataset of rock chips of different lithologies using manual labelling using LabelMe software and a newly developed automatic approach. This procedure is implemented as a python code that uses JPEG images as input data and produces JSON files with grain boundaries and labels as an output. Cascade Mask R-CNN, an object detection algorithm, is used for grain detection and classification. Our study shows that recent progress in AI/ML approaches allows automation of manual and time-consuming tasks of lithology classification.