Digital Geoscience: Smartphone App for detecting fossil ammonites

I’ve created an App to detect fossil ammonites for children and the visually impaired on a phone. It might also be a good way to get more young people interested in applying AI in the Geosciences. It is relatively easy to build your own models.

It uses over 800 annotations of ammonites, many in-situ within different rock formations, at various states of weathering, and loose in shingle. Combined with deep learning techniques, I used these to teach a neural network what an Ammonite looks like.

Whilst Geological / Fossil Smartphone Apps have been created before, they tend to focus on either showing people where fossil localities are, or help identify fossils once they have already been found, rather than actually help a person find fossils.

I tested the model on 100 unseen images, which gave an accuracy of 69%. Early simulation testing is promising, but it needs more refinement which I’ll test with a field trip to Lyme Regis next month.

I’ll release the machine vision CoreML and YOLO models freely for anyone to use on their Smartphone or Tablets later this year once I’ve finished the refinements.

A very big thank you to Joao Domingos and Datature for helping the model conversion for an Apple iPhone.