Using machine learning in Biostratigraphy

Using machine learning to cut the time for Biostratigraphical analysis from 3 weeks to 3 days. Very interesting presentation from David Wade at Equinor at the excellently organised GESGB (Geoscience Energy Society of Great Britain) conference on machine learning yesterday.

The presentation discussed machine vision techniques to scan slides prepared from borehole cuttings/cores and apply ML image detection techniques to automate the creation of biostratigraphical range charts in order to pick geological formations.

Latent-space clustering of microfossil image embeddings can be used to identify
meaningful groupings of microfossils. By extending this approach with a content-based image retrieval (CBIR) method a system called SCAMPI was developed.

In an experiment based on 27 samples in one borehole, significant time savings were shown which would most likely scale.

The ML model trained on 1.6 Million microfossils and supporting source code is available in Huggingface. Links in the comments. The abstract for the supporting published paper is included below, link to the paper also in the comments.

Abstract

“Microfossil classification is an important discipline in subsurface exploration, for both oil & gas and Carbon Capture and Storage (CCS). The abundance and distribution of species found in sedimentary rocks provide valuable information about the age and depositional environment. However, the analysis is difficult and time-consuming, as it is based on manual work by human experts. Attempts to automate this process face two key challenges: (1) the input data are very large – our dataset is projected to grow to 3 billion microfossils, and (2) there are not enough labeled data to use the standard procedure of training a deep learning classifier. We propose an efficient pipeline for processing and grouping fossils by genus, or even species, from microscope slides using self-supervised learning. First we show how to efficiently extract crops from whole slide images by adapting previously trained object detection algorithms. Second, we provide a comparison of a range of self-supervised learning methods to classify and identify microfossils from very few labels. We obtain excellent results with both convolutional neural networks and vision transformers fine-tuned by self-supervision. Our approach is fast and computationally light, providing a handy tool for geologists working with microfossils.”

Huggingface: https://huggingface.co/IverMartinsen/scampi-dino-vits16
GitHub: https://github.com/equinor/scampi-benchmark
Papers: https://www.sciencedirect.com/science/article/pii/S2666544124000212?ssrnid=4706163&dgcid=SSRN_redirect_SD

https://ina.tmsoc.org/meetings/INA19Conwy/abstracts/Stefanowicz%20et%20al%202024%20JNR%20abstract%20INA19.pdf