Image Processing and Machine Learning Applied to Lithology Identification, Classification and Quantification of Thin Section Cutting Samples

Cuttings provide the opportunity to precisely look at the rock that has been drilled. A preliminary drill cuttings description is commontly performed by mudloggers and wellsite geologists using conventional binocular microscope at the drilling rig. After this preliminary description, often the bags of cuttings are stored in a warehouse and samples are seldom examined back again. Cuttings give the geologist information about the formation lithology needed for geologic correlation, understanding about reservoir quality, seals and source rocks, and can also be an input for the petrophysicist. In this study, we are testing a methodology to identify, classify and quantify lithologies present in cutting samples using thin section images. The method includes sample preparation (washing, drying and thin section cuttings preparation), image acquisition (to obtain whole thin section gigapixel high resolution microscopy images), virtual microscopy (to identify lithologies) and automatic image analysis (to perform supervised machine learning lithology clasiffication). Virtual microscopy allowed the identification of four main lithologies in all the studied thin sections: quartzites (including loose quartz grains), siltstones, claystones and carbonates. Image analysis allowed the classification and quantification of the identified lithologies in 16 drill cutting samples from two tight gas reservoirs. This innovative methodology allowed the fast identification of lithologies using virtual microscopy and their classification and quantification by image analysis and supervised machine learning. This approach is widely accessible as open source software was used for virtual microscopy and image analysis. Algorithm training and model generation was relativelly fast, and its performance or accuracy was qualititavely evaluated by virtual microscopy with good classification results.