Enhancing Petrographic Analysis with Convolutional Neural Networks

Summary

This is an abstract from the "Practice, Theory, and Ethics of Machine Learning in Archaeology" session, at the 90th annual meeting of the Society for American Archaeology.

Archaeological research has highlighted the role of mollusks in coastal communities' foodways, construction practices, and cultural traditions, but its use within pottery production has received less attention. Key morphological and chemical signatures are altered during pottery manufacture, impeding identification of shell taxon. Increased systematic examination of fracture pattern variability among mollusks has presented a new avenue for visual identification and the application of machine-learning techniques. To increase the reliability and accuracy of this approach, we created an experimental dataset composed of various types of shell temper available to Mississippian communities living along the Northern Gulf and trained a convolutional neural network (CNN) to identify variability in mollusk fracture patterns. We review the basics steps for creating a dataset, techniques for capturing cross section images, and the process for training an image classification model. We apply this model to archaeological samples from the Pensacola region to show how training a CNN can be used to identify patterns that reflect raw material selection, environmental availability, and functional properties considered by indigenous potters. While the focus here is on pottery identification, the methods discussed are broadly applicable to other archaeological questions, as we aim to demonstrate the advantages of using image classification models.

This Resource is Part of the Following Collections

• Society for American Archaeology 90th Annual Meeting, Denver, CO (2025)   •
• Practice, Theory, and Ethics of Machine Learning in Archaeology

Keywords

Record Identifiers

Abstract Id(s): 52584