Machine Learning-Ready Remote Sensing Data for Maya Archaeology in Chactún

Five types of multimodal annotated data cover the area around Chactún, a major Maya urban centre in the Yucatán peninsula. The dataset includes ALS visualisations, a canopy height model, Sentinel-1 SAR, Sentinel-2 optical data, and manual annotations for three structure types. It was published by Kokalj et al. in Scientific Data in 2023 and used for a computer vision competition.

Use Cases

• Semantic segmentation of Maya archaeological structures based on manual binary mask annotations.
• Object detection of ancient Maya buildings, platforms, and reservoirs using multimodal satellite and lidar data.
• Training convolutional neural networks for object recognition in remote sensing imagery.
• Improving existing computer vision models for Maya archaeology investigations.

Strengths

• Includes five distinct data types: ALS visualisations, ALS-derived canopy height model, Sentinel-1 SAR, Sentinel-2 optical data, and manual annotations.
• Manual annotations provide precise locations and boundaries for three specific Maya structure classes: buildings, platforms, and aguadas.
• Dataset is explicitly designed and ready for use with convolutional neural networks.
• Has been validated through use in the 'Discover the Mysteries of the Maya' computer vision competition.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count and dataset size are unknown, which may limit suitability assessment.
• Data is geographically focused on one Maya centre, Chactún, which may limit generalizability.

Provenance

Source

Žiga Kokalj and co-authors, affiliated institutions unknown.

Collection Method

Likely compiled from high-resolution airborne laser scanning (ALS) and Sentinel satellite data, combined with manual archaeological annotations.

Geography

Area around Chactún, a major ancient Maya urban centre in the central Yucatán peninsula.