Microfossil classification is an important discipline in subsurface exploration for 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 is not enough labelled data to use the standard procedure of training a deep learning classifier.

To overcome these challenges, scientists at the University of Tromsø (UiT) machine learning group at The Arctic University of Norway have proposed an efficient pipeline for processing and grouping fossils by genus or species from microscope slides using self-supervised learning.

Firstly, they demonstrated how efficiently crop extraction from whole slide images can be done by adapting previously trained object detection algorithms. Then, they provided a comparison of a range of self-supervised learning methods to classify and identify microfossils from very few labels.

Great potential

"This work shows that there is great potential in utilizing AI in this field," says researcher Iver Martinsen, first and co-corresponding author of the study. "By using AI to automatically detect and recognize fossils, geologists might have a tool that can help them better utilize the enormous amount of information that wellbore samples provide."

Microfossils are found in vast amounts everywhere, but the time and expertise required to analyze the data mean that only a fraction of the available fossils are analyzed. The researchers used state-of-the-art AI methodology—training an AI model completely without annotations, utilizing the large pool of raw data provided by the Norwegian Offshore Directorate.

According to Martinsen, they used AI to detect fossils from one selected well on the Norwegian continental shelf. In turn, 100,000 of the detected fossils were used to train a model for image recognition. To evaluate the model's performance, the researchers tested the model by classifying several hundred labelled fossils from the same well.

From the study, they obtained excellent results with both convolutional neural networks and vision transformers fine-tuned by self-supervision. According to the researchers, their approach is fast and computationally light, providing a handy tool for geologists working with microfossils.

As per analysis, the AI model exceeds previous benchmarks available out there. The researchers hope that the present work will benefit geologists both in industry and academia.

Full report