A Micromorphological Neural Network

We are currently beginning to develop a machine learning system based on a Convolutional Neural Network (CNN) to automatize the description of micromorphological thin sections. This is carried out in collaboration with ULL Computer Science Professor Rafael Arnay. Our latest funding comes from a project of which he is PI: “Neural network Image tagging-based system for micromorphological description in archaeology”.

Although in recent years neural network-based image identification systems have become widespread, Their applications in microstratigraphy and micromorphology are meager. Micromorphological description is carried out manually, which requires a considerable time investment and entails a degree of subjective observation. Thus, our goal is to develop a tool to speed up and normalize the descriptive process.

The idea is to train a CNN model to create a micromorphology image database. We work with high resolution thin section scans that are processed using a sliding window technique. This allows us to isolate individual micromorphological features such as different particle, porosity and microstructure types. In a second stage, we will also create a web-based app for image tagging and encourage outside users to add images to the database, leading to increasing CNN autonomy.