Broadscale Deep Learning for Maya Settlement Detection in G-LiHT Lidar (link.springer.com)

A new study has developed a broadscale deep learning model, named Q2000, for detecting ancient Maya settlements using Lidar data across an extensive area of the Yucatan Peninsula, totaling approximately 35,584 km². This innovative model overcomes limitations of smaller, area-specific models by achieving robust accuracy, with an F1 Score of 0.89, even when trained on a relatively small dataset. By integrating diverse geographical and cultural contexts into its training, the Q2000 model enhances the detection of a broader range of archeological features, facilitating the identification of previously undocumented sites in complex environments. This advancement is significant for the AI and archaeological communities as it demonstrates the potential of large-scale Lidar analyses combined with deep learning techniques to revolutionize archaeological research methodologies. By streamlining the detection process of ancient structures, Q2000 not only accelerates the pace of research—transforming decades of fieldwork into days of data analysis—but also enhances the robustness of findings by leveraging a Human-in-the-Loop framework. This integration of domain expertise ensures higher accuracy and enriches the model's applicability across diverse environments, paving the way for collaborative projects that enhance our understanding of ancient Maya civilization through innovative technology.