UAV-Based Multispectral Maize Dataset for Water Stress 2025 and Common Rust 2025 Detection: Full Dataset with Source Orthomosaics

This record provides the full version of a UAV-based multispectral maize dataset developed for 2025 water stress and common rust detection in maize under real field conditions.

This v2.0 release extends the representative subset released in v1.0 and includes the complete machine-learning-ready patch dataset together with the source UAV multispectral orthomosaics.

The dataset includes:

- source UAV multispectral orthomosaics for water stress and common rust mapping,
- six-channel multispectral image patches,
- pixel-wise semantic segmentation masks,
- RGB previews, mask visualizations, and overlay images,
- metadata files describing band information, orthomosaic properties, class definitions, and patch-level information,
- optional example scripts for loading and visualizing the dataset.

The harmonized semantic label space consists of five classes: soil/background, low water stress, high water stress, healthy maize, and common rust.

The GeoTIFF files contain six float32 bands. Band descriptions are not embedded in the GeoTIFF metadata. Based on the preprocessing pipeline used in the associated study, the vegetation-index calculations use Band 2 as Green, Band 3 as Red, Band 4 as NIR, and Band 5 as Red-edge. Band 6 is retained as an auxiliary/exported band and is not used in the reported NDVI, NDRE, SAVI, or GCI definitions unless explicitly stated.

This dataset is associated with the following peer-reviewed article:

Suiçmez, Ç., Yilmaz, C., & Kahraman, H. T. (2026). A Multi-Head UNet++ Framework with Fractional Differential Output Refinement for UAV Multispectral Crop Stress Mapping. Sensors, 26(10), 3228. https://doi.org/10.3390/s26103228