doi:10.12754/data-2026-0008
© Author(s) 2026. This work is distributed
under "Creative Commons Attribution 4.0 International"
Trained nnU-Net model weights for boulder segmentation from MBES bathymetry, southern Baltic Sea (Version 1.0)
Contributor(s): Papenmeier, Svenja [Supervisor]; Papenmeier, Svenja [ContactPerson]
Scientific purpose: The model is trained to do pixel-wise segmentation of boulders in multibeam echosounder (MBES) bathymetry data. It was trained to support large-scale mapping of boulder fields in the southern Baltic Sea.
Keywords: nnU-Net, deep learning, boulder detection, boulder segmentation, marine habitat mapping
Abstract. This resource contains the trained weights of an nnU-Net deep learning model for automated, pixel-wise segmentation of boulders in multibeam echosounder (MBES) bathymetry data. The model was trained and validated on manually annotated bathymetry tiles from multiple survey areas in the southern Baltic Sea, Mecklenburg Bay (see related dataset https://doi.iow.de/10.12754/data-2026-0007).
Citation
Download
Full Article or Dataset [737 MB]