Applications of deep learning to decorated ceramic typology and classification: A case study using Tusayan White Ware from Northeast Arizona

In this study, we present an alternate approach to archaeological typology, using deep learning to classify digital images of decorated pottery sherds into an existing typological framework. The study focuses on a specific kind of ancient painted pottery from the American Southwest, Tusayan White Ware, but we believe it has broader implications for a wide range of geographical settings and artifact types. Our results show that when properly trained, a deep learning model can assign types to digital images of decorated sherds with an accuracy comparable to, and sometimes higher than, four expert-level contemporary archaeologists. The technique also offers novel tools for visualizing both the importance of diagnostic design elements and overall design relationships between groups of pottery sherds. We demonstrate that this method can objectively match a specific unclassified sherd image to its most similar counterparts through a search of thousands of digital photos. This discovery has important archaeological implications for analyzing time relationships, monitoring stylistic trends, reconstructing fragmentary artifacts, identifying ancient artisans, and studying the evolution and spread of ancient technologies and styles. It also shows how deep learning models can potentially supplement or supplant traditional typologies in favor of more direct groupings and comparisons of artifacts.