Using image recognition techniques to classify different grain lithologies

For a global energy company we designed a machine learning model that first recognizes the different grain lithologies in digital images of ground samples and then calculates the surface of each lithology.

We used supervised recognition techniques. With a pattern recognition algorithm and a segmentation tool the different lithologies were detected. Then we determined a feature embedding for each lithology. With this feature embedding, the total surface of each lithologies in the ground samples was calculated.