A Deep-Learning Model of a Shell Ring

A shell ring may look like an ordinary pile of trash, but these middens tell us a lot about people and the natural environment they inhabited. They’re a crucial link to the past and provide insight into the changing relationships between early Native American communities as well as their increasing efficiency in extracting natural resources from the tidal creeks and hammocks of Southeastern North America.

Located in coastal Georgia and South Carolina, the earliest known shell rings date back to 4200 cal B.P., establishing them as both ceremonial centers and year-round village communities. They also reveal the geographic range of this distinctive building tradition.

To document these sites, researchers at the University of Florida used photogrammetry—a technique that uses digital images to create a 3D model of an archaeological site—to map out the locations of all known shell rings and mounds within a three-county area in South Carolina, Georgia, and North Carolina. Then, they “taught” a deep learning algorithm—an advanced machine-learning method that uses neural networks to recognize patterns and structures in images—to identify these features.

The new model identified a total of 120 shell rings and mounds across the three counties, including one in Georgetown County that had never been documented before. Its validation rate of 75% suggests that additional potential shell rings in the region could be detected with increased training data. Moreover, the identification of these new sites opens up the possibility for comparative analyses of resource exploitation patterns across a large portion of the American Southeast, with potential implications for sustainable shell-fishing in the future. shell ring

Leave a Reply

Your email address will not be published. Required fields are marked *