AI-Assisted Detection Dataset of a Previously Undocumented Necropolis in Central Saudi Arabia

AI-Assisted Detection Dataset of a Previously Undocumented Necropolis in Central Saudi Arabia This dataset contains geospatial detections associated with a previously undocumented necropolis identified during large-scale satellite imagery analysis in Central Saudi Arabia. The detections were generated using a custom YOLO object detection model within the program Mustatil 3. The study area was identified during systematic analysis of a georeferenced satellite dataset covering approximately 200 × 200 km. Following the initial detection, the area was re-examined using the same AI-assisted workflow to improve detection coverage and document the spatial extent of the archaeological features. The detections presented in this dataset are intended as indicators of potential archaeological features only and require independent verification before being interpreted as confirmed archaeological remains. The dataset includes: GeoPackage (GPKG) containing detected archaeological features Custom YOLO model used for detection Coordinate information for the detected necropolis Approximate location: J45P+549, Umm Ash Shubrum 19911, Saudi Arabia Version 2 contains a dataset generated with Faster R-CNN using the software Mustatil. Detection and GIS processing were performed using Mustatil 3, a desktop GeoAI platform for geospatial annotation, YOLO training, large-scale satellite image analysis, AI-assisted object detection, GIS export, and graphical AI workflows. Wasfy, T. (2026). Mustatil 3: GeoAI Workspace for Annotation, YOLO Training, Large-Scale Satellite Detection, GIS Export and Graphical AI Pipelines. Zenodo. https://doi.org/10.5281/zenodo.20481111 Project website: http://mustatil.de/ Source code: https://github.com/tarekwasfy01/Mustatil-YOLO-AI-Model-Trainer- AI Disclaimer The detections contained in this dataset were generated using artificial intelligence methods and subsequently reviewed by the author. AI-generated detections may contain false positives, false negatives, classification errors, and spatial inaccuracies. The dataset is provided as a research resource and should not be considered a definitive archaeological inventory. Independent verification, expert review, and where possible field validation are recommended before scientific, archaeological, or heritage management conclusions are drawn. Author Tarek Wasfy Year 2026