Open-source machine vision model for classifying geological images from documents

Ahead of Earth Science week I’ve openly released a free machine vision model which automatically classifies geological images from documents. Those that have large archives of geoscience documentation may find this helpful to discover, and potentially repurpose, old geological data for new knowledge.

Schools and researchers may also find the model helpful to spark their own projects around Geoscience, Information Science and Artificial Intelligence.

The simple classes I’ve started with are: SEM IMAGE, BOREHOLE LOG, THIN SECTION, GEOLOGY MAP, LOCATION MAP, MODEL, SEISMIC SECTION, CORE PHOTOGRAPH, REMOTE SENSING IMAGE, OUTCROP and CHARTS-CROSSPLOT.

You can run these types of YOLO machine vision models locally on your own laptop or infrastructure (e.g. using Python) without the need to connect to the Internet.

This type of approach avoids information security and data privacy issues which cloud hosted AI/LLM’s and API’s have, especially those which are marketed as ‘free’. They are not ‘free’, it’s at the cost of ‘your data’ which you need to upload and your activity tracked.

Whilst sharing public geoscience data is beneficial, researchers and organisations will have their own copyrighted or proprietary/confidential geological data, and these ‘free’ AI/LLM technologies can harvest and use this data without your explicit permission (always read the terms and conditions to be safe!).

You can download the model here: https://github.com/PCleverleyGeol/Geological-Image-Classifier