Prospecting Architectural Features Using UAV-LiDAR, Deep Neural Networks, and Visualization Techniques: A Case Study of Kuélap and Cambolín (NW Peru)

High-resolution and accurate synoptic images of terrestrial topography, even in densely forested areas, have proven valuable for archaeology by enabling the identification and characterization of relief patterns associated with ancient human activities. This study presents a novel approach that integrates Digital Terrain Models (DTMs) obtained through Airborne Laser Scanning (ALS) from a drone, along with advanced visualization techniques (VT) based on computer vision algorithms, evaluated using objective performance metrics. The research was conducted at the archaeological sites of Kuélap and Cambolín, belonging to the Chachapoyas culture in the Amazonas region, north-western Peru. Seventeen visualization techniques were applied to a DTM derived from ALS with a resolution of 0.5 m. Additionally, the Mask Region-Convolutional Neural Network (Mask R-CNN) model in ArcGIS Pro was used for the automatic detection and segmentation of architectural features. The results indicate that the Color Relief Image Map (CRIM) visualization technique achieved the highest average precision score, reaching 71.89% in Kuélap and 43.54% in Cambolín. The model detected a total of 137 out of 185 reference structures in Kuélap and 53 out of 73 in Cambolín. The combination of visualization techniques and deep learning supports archaeological prospection in areas with dense vegetation and complex topography, serving as a complementary tool to manual interpretation in the study of Chachapoya settlements