Vanderbilt’s Steven Wernke, an Associate Professor of Archeology, is working with the Data Science Institute on a project using satellite imagery and deep learning to produce the largest image-based survey of archeological features in all of the Americas.
Archaeologists have faced perennial problems of scale and representation, struggling to match the scale of analysis to the scale of past social formations. Traditionally, field archaeology can only capture small samples of the past social formations. Syntheses of published studies will necessarily reproduce the sampling biases and path dependencies of prior research.
Professor Wernke is working with the DSI on a project that uses high resolution satellite imagery to seek a continuous distributional view of archaeological phenomena at scales previously impossible through traditional field research. Through the Geospatial Platform for Andean Culture, History, and Archaeology, the team has identified approximately 40,000 archaeological loci in a 175,000 km2 area in the central Andes.
The project will now use these human-identified features as training data for a deep learning model, which will be deployed over even larger areas. The goal is to cover the entirety of the central Andes. The team is developing CNN-based models (chip/object/segmentation) for detection of archaeological features. Once this autonomous survey is completed, the results will be distributed to teams of trained archaeology students led by regional experts, who will validate the data and produce a canonical dataset. The result will be the largest imagery based survey of archaeological features and sites in the Americas.