CNN Training Dataset and Models for Agricultural Sub-Class Mapping from Multispectral UAV Imagery Using a Thematic Labeling System

This dataset contains multispectral UAV image samples, derived training patches, metadata, and trained CNN models for agricultural sub-class classification. The samples were collected using RGB and CIR UAV sensors with ground sampling distances (GSD) of 0.027 m, 0.053 m, and 0.064 m, and were resampled to spatial resolutions of 0.060 m and 0.080 m. Data was acquired over agricultural sites in Germany and structured by crop type - Winter Wheat (WW), Spring Barley (SG), Rapeseed (WR), and Corn (KM) - with labels assigned using a thematic sub-class labeling system based on the BBCH scale (vital/dry crop, vital/dry lodged crop, ripening/flowering crop, bare soil and weed infestation). 

The dataset includes raw (tif) and augmented image patches, patch matrices, and TensorFlow CNN models trained separately for each crop type.

Each ZIP file includes a dedicated README.txt. A complete project-wide README.md is provided with this record, detailing the sampling strategy, structure, licensing, and references. This dataset supports the study under review titled: *Toward Generalizable CNN-Based Classification of Agricultural Sub-Classes*.

Related resources:
- Code: https://github.com/Aranil/UAVSampleLab
- Publication [1]: https://doi.org/10.1016/j.atech.2025.100799
- IGARSS 2023 [2]: https://ieeexplore.ieee.org/document/10282406

The Datasets, Code and Publications were created in the frame of the Project Radar-Crop-Monitor funded by the Federal Ministry for Economic Affairs and Climate Action (BMWK), Germany, under the support code FKZ: 50EE1901.