Analysing microfossils has always been a difficult task for several reasons. However, they are extremely important to study because they help map structures below the Earth's surface and provide insight into past geological times.
Currently, geologists spend a lot of time manually counting microfossils extracted from sedimentary rocks beneath the ocean floor to get accurate information.
However, advances in artificial intelligence (AI) have led to new solutions to address this challenge: Researchers are leveraging AI solutions to analyze microfossils directly from microscope images.
Applying computer vision to microfossil analysis in particular poses significant challenges: With an estimated 3 billion fossils to analyze, the sheer volume of data can be daunting, and the task is further complicated by a severe lack of labeled data to train machine learning models.
Advanced AI methods enhance microfossil detection and analysis
Recent research published in the journal Artificial Intelligence in Geosciences We introduced an advanced method for automated microfossil detection and analysis. The research team consisted of members of the Machine Learning Group at the Arctic University of Norway, University of Tromsø (UiT).
They developed a pipeline to extract fossil information from microscope slide images. They found that deep learning techniques outperform traditional image processing methods and that self-supervision can be effectively used for feature extraction.
Furthermore, models trained specifically on microfossils outperformed the benchmark baseline models.
Deep Learning Model Addresses Challenges of Microfossil Image Classification
The researchers note that using automated algorithms to classify or group images based on their content is a challenging task. Even seemingly simple image classification tasks with well-defined subjects can be challenging.
Moreover, images often contain redundant information stored in the pixels surrounding the object of interest. In all such cases, deep learning models have proven to be highly effective.
Deep learning methods excel at modeling complex relationships in data. To achieve accurate results, Convolutional Neural Networks (CNNs) are designed to extract meaningful information from images.
“CNNs have evolved significantly over the past 10-15 years and have until recently been the state-of-the-art technique in image modelling, with studies reporting excellent results in classifying labelled image data, including natural images,” the researchers explained.
They explained that in addition to CNNs, Vision Transformers (ViTs), inspired by large-scale language models, have become a promising candidate for image classification.
“Currently, both CNN and ViT are commonly used for image classification, but there is no clear winner or best architecture for this task. Therefore, it is standard to test and compare both architectures, and that is exactly what we do here.”
Research results highlight the effectiveness of AI in microfossil analysis
The findings demonstrate that the acquisition of self-supervised training is the most promising approach for further research. The researchers emphasized that AI can contribute significantly to the automatic detection and recognition of fossils.
In this study, we leveraged AI to detect fossils and achieved the expected results: by utilizing state-of-the-art deep learning techniques, we were able to efficiently extract the features required for tasks such as identifying, grouping, and counting microfossils.
“The results obtained on the labelled dataset demonstrate that self-supervised training of deep learning models on microfossils leads to significant improvements compared to existing benchmark models,” the report concludes.
The researchers say their approach can also be applied to other scenarios where large numbers of patterns and shapes need to be extracted and identified from large images, such as the analysis of foraminifera and other microfossils in the geological record.
The research team believes that their research will bring great benefits to the field of geology, both in industry and academia.
About the Editor
Gairika Mitra Geirika is a tech geek, an introvert and an avid reader. Lock her in a room full of books and she'll never complain.
