Discovering a Historical Site Using Mustatil Yolo AI

Discussion
, , ,
1

I would like to share an example of how the open-source software Mustatil was used to identify a previously undocumented archaeological site in northwestern Arabia.

Case Study:
Zenodo Record 20531983

Mustatil Software:
Mustatil Zenodo Release

Project Website:
Mustatil.de

Workflow

The site was identified through a combination of satellite imagery analysis, GIS investigation, and AI-assisted object detection.

Mustatil allows researchers to:

  • annotate archaeological features,
  • train custom YOLO models,
  • perform large-scale satellite image detection,
  • export results directly to GIS-compatible formats,
  • build advanced processing workflows using graphical AI pipelines.

In this example, a custom model was applied to large satellite imagery datasets. The detection process highlighted a structure that differed from the surrounding natural landscape. Subsequent visual inspection and GIS verification indicated that the feature is likely an archaeological site that had not previously been documented.

Why This Matters

Large areas of Arabia remain difficult to survey on the ground due to their size and remoteness. While satellite imagery has become an essential tool for archaeological prospection, manually reviewing vast datasets remains extremely time-consuming.

AI-assisted workflows can dramatically accelerate this process. Instead of examining millions of image tiles manually, trained models can automatically highlight potential targets and direct researchers toward areas worthy of further investigation.

AI Does Not Replace Archaeologists

The AI did not independently discover the site. Human expertise remained essential throughout the process.

The final interpretation required:

  • archaeological knowledge,
  • visual inspection of imagery,
  • GIS analysis,
  • validation of the detected feature.

The AI served as a decision-support tool that reduced the search area and increased efficiency.

Conclusion

This case demonstrates how modern computer vision techniques can support archaeological prospection and landscape archaeology. By combining satellite imagery, GIS, and AI-based object detection, researchers can investigate large and remote regions more efficiently and identify archaeological features that might otherwise remain unnoticed.

Feedback and discussion are welcome.

2

Very nice example—thanks for sharing this. It’s a strong real-world case of how Ultralytics YOLO can accelerate archaeological prospection without replacing expert interpretation, very much in line with our write-up on AI in archaeology.

If you continue iterating on the pipeline, YOLO26 would be our recommended baseline for new training runs today. If you’d like, feel free to share a few validation details like dataset size, imgsz, and precision/recall so others in the community can better compare approaches.