Mustatil: An Open-Source AI and GIS Framework for Archaeological Feature Detection in Satellite Imagery

Resource information

Format

exe

Original repository

20800096

DOI

10.5281/zenodo.20800096

Credit basis

Creators in Zenodo record metadata

Access

Open files available from the original Zenodo deposit

Last checked

Nov 5, 2026

License

cc-by-4.0

Last record update

Nov 5, 2026

About this resource

Now with ADAF detection and RF-DETR, training and detection. This update expands Mustatil as a GIS-level AI vision workspace with improved support for geospatial processing, model experimentation, and multi-model remote-sensing workflows. The release focuses on stability and practical usability: the existing AI Pipeline tab remains unchanged, while the surrounding detection and GIS tooling has been extended. New and improved components include additional support for Faster R-CNN, Mask R-CNN, U-Net and SAM/SAM2-based workflows, optional video detection experiments, improved GIS/console tooling for Rasterio, GDAL/OGR and GMT, and better integration of model-based detection with geospatial export formats such as GeoPackage. These additions are intended to support large-image and satellite-map analysis, archaeological remote sensing, object detection, segmentation, review, and export in one desktop environment. Mustatil continues to combine annotation, YOLO training, large raster detection, satellite-map workflows, GIS export, and visual AI pipeline building in a single offline desktop application. This release extends Mustatil as a GIS-level AI vision workspace for remote sensing, archaeological detection, annotation, training, and geospatial AI pipelines. In addition to the existing YOLO workflow, the software now includes experimental support for several additional model families, including Google OWLv2, Grounding DINO, LAE-DINO, Faster R-CNN, Mask R-CNN, U-Net semantic segmentation, and SAM2-assisted segmentation. The new model integrations expand Mustatil from a YOLO-based detection and training environment into a broader multi-model AI workspace. Faster R-CNN and Mask R-CNN provide alternative region-based object detection workflows, U-Net enables semantic segmentation from project annotations, and SAM2 can be used as a segmentation refinement step from existing detection boxes. These functions can be accessed through dedicated tabs and integrated into the visual AI pipeline using selectable AI model blocks, while existing FormLearner and rule/IF logic blocks remain available. The update is designed for large-image and geospatial workflows, including satellite-map detection, GeoTIFF-based analysis, project-based training, visual review, and export of results to GIS-compatible formats such as GeoPackage. This makes Mustatil suitable for remote sensing workflows where object detection, segmentation, post-processing, and GIS export need to be combined in one offline desktop application. This release expands Mustatil with additional AI vision model support beyond YOLO, including Google OWL-ViT / OWLv2, Grounding DINO, and LAE-DINO. These models add open-vocabulary and text-guided detection capabilities, as well as project-based dataset preparation and experimental training workflows. Together, they extend Mustatil as an offline GIS-oriented AI workspace for annotation, detection, model training, and large-image analysis. Mustatil is an integrated GIS-level AI vision workspace for annotation, YOLO training, large-scale detection, satellite-map analysis, and visual pipeline building. It combines dataset creation, model training, geospatial inference, map-based review, and graphical AI pipelines in one desktop application — designed for images and map areas too large for conventional computer-vision tools. Mustatil is an open-source geospatial artificial intelligence platform designed for the detection, annotation, analysis, and mapping of a