Datasets:

Modalities:
Image
DOI:
License:
SeagrassMapper / README.md
greensinwater33's picture
Update README.md
298b7fc verified
metadata
language: '-en'
pretty_name: SeagrassMapper
license: cc-by-4.0
task_categories:
  - image-segmentation
tags:
  - biology
size_categories:
  - 1K<n<10K
The dataset contains images acquired by Unoccupied Aerial Vehicles (UAVs) and aircraft based imaging cameras along with their annotation masks. These images are collected over Danish coastal waters where Eelgrass (Zostera marina), a type of seagrass is found. The annotation masks contain 1 where eelgrass is present and 0 at absent. The images and annotation masks are 512x512 of height and width. The UAV images are obtained at approx. 3 cm ground sampling distance (GSD) while the aircraft based images are acquired at 12.5 cm GSD.

The dataset is aimed at performing domain adaptation studies where image segmentation models can be trained using the source data to be adapted for target dataset. This allows use of images of different origin and resolutions to be used for mapping seagrass. These image datasests are collected from multiple coastal water bodies of Denmark at various times. The target images are collected at three sites Horsens Fjord, Lovns Broad and Skive Fjord located in Danish coastal waters.

The dataset was used for domain adaptation expeirments described in this paper which can be cited as:

Pawar, S., Thomasberger, A., Bengtson, S. H., Pedersen, M., & Timmermann, K. (2025). Eye in the Sky for Sub-Tidal Seagrass Mapping: Leveraging Unsupervised Domain Adaptation with SegFormer for Multi-Source and Multi-Resolution Aerial Imagery. Remote Sensing, 17(14), 2518. https://doi.org/10.3390/rs17142518