AI-assisted identification of stromatoporoids

Stromatoporoid sponge fossils were major diverse reef-builders in the Palaeozoic Era; their taxonomic identification relies on thin sections examined under transmit-ted light microscopy, where vertical and transverse skeletal elements reveal diag-nostic architectural features that vary with taxa. These elements typically appear darker than the cement-filled internal spaces, allowing stromatoporoid taxa to be distinguished. However, stromatoporoid architecture is variable, so that judgement of identification is not always unequivocal. Therefore, this study investigates the application of artificial intelligence (AI) to automate stromatoporoid identification, introducing a novel approach to streamline and standardize palaeontological tax-onomy. For the first time, both vertical and transverse sections have been simulta-neously analysed and integrated into an automated framework. High-resolution im-ages of thin sections from four well-established Silurian genera, collected from the West Midlands and Shropshire counties, UK, were used to train supervised ma-chine learning models. The images, captured using plane-polarised transmitted light microscopy on thin sections, were digitally enhanced to increase contrast and eliminate background noise, ensuring that only skeletal features were used to in-form the models. Despite variations in fossil preservation, section orientation, and image quality, the AI models achieved classification accuracies of up to 96%. This demonstrates that stromatoporoid skeletal architecture is highly amenable to auto-mated analysis, even under suboptimal conditions. The results represent a signifi-cant step forward in the application of AI to palaeontology, reducing reliance on manual identification and accelerating the classification process. Ultimately, this approach lays the groundwork for fully automated taxonomic workflows that improve research efficiency and accessibility.