WeedMap: Large-scale Multispectral Weed Mapping Dataset

Year: 2018
DOI: 10.3929/ethz-c-000788571

Multispectral UAV dataset over sugar beet fields in Eschikon (Switzerland) and Rheinbach (Germany) with orthomosaic maps and tiled crops for weed/crop segmentation. Captured with MicaSense RedEdge-M and Sequoia cameras.

Highlights

• 8 datasets (indices 000–007), 18,746 images across 129 directories (~5.36 GB)
• Orthomosaics and tiles with RGB/CIR/NDVI composites, masks, and pixel-level labels
• Ground truth: background/crop/weed (color and indexed label maps)
• Flight paths provide ~1 cm GSD; includes interactive viewers for orthomosaics
• Camera specs and folder structure docs included; MATLAB scripts for tile/orthomosaic conversion

Folder Structure

• Orthomosaic/RedEdge and Orthomosaic/Sequoia: composite PNG, groundtruth labels, reflectance TIFFs
• Tiles/RedEdge and Tiles/Sequoia: per-channel tiles, masks, and groundtruth labels
• Naming: XXX_frameYYYY_GroundTruth_color.png / ..._iMap.png, masks frameYYYY, tiles per channel

Downloads

• Full dataset (5.36 GB): https://doi.org/10.3929/ethz-c-000788571
• Orthomosaics:
  • RedEdge bundle (1.9 GB): https://doi.org/10.3929/ethz-c-000788571
  • Sequoia bundle (1.49 GB): https://doi.org/10.3929/ethz-c-000788571
  • Per-dataset archives 000–007 available (links to be updated)
• Tiles:
  • RedEdge bundle (637 MB): https://doi.org/10.3929/ethz-c-000788571
  • Sequoia bundle (444 MB): https://doi.org/10.3929/ethz-c-000788571
  • Per-dataset archives 000–007 available (links to be updated)

Citation

I. Sa, M. Popovic, R. Khanna, Z. Chen, P. Lottes, F. Liebisch, J. Nieto, C. Stachniss, A. Walter, R. Siegwart,
“WeedMap: A large-scale semantic weed mapping framework using aerial multispectral imaging and deep neural network for precision farming”, Remote Sensing, 2018. DOI: 10.3390/rs10091423.

Tags: agriculture, multispectral, semantic-segmentation, orthomosaic, uav