su-fmi/sunflower-density-estimation-2025

Dataset Metadata

Identification Information

Citation

• Title: Sunflower density estimation dataset from May to June 2025
• Originator: Sofia University - Faculty of Mathematics and Informatics, SAP LABS Bulgaria
• Publication Date: 2025.07.07

Abstract

This dataset is focused on the sunflower plant density estimation problem. It is a continuation of our efforts and dataset from 2024. See: Sunflower density estimation dataset from April to July 2024 . In the current dataset we managed to significantly increase the studied field's area and even more importantly, to include validation data in the form of low-altitude orthophotos with full field coverage. The latter can help validate the accuracy of more scalable estimation algorithms that do not rely on exhaustive plant counting (e.g. spectral characterization from Sentinel-2 data).

DJI P4 Multispectral was again used to collect the dataset and DJI Terra was used for the post processing and reconstruction of orthophotos.

Purpose

This dataset was developed as part of a research project, investigating the capabilities and application of drones and multispectral cameras for the agricultural domain. The provided data can be used for the following scenarios:

1. Training models, relying on multispectral data sources.
2. Improving existing algorithms in the computer vision domain.
3. Developing and validating methods for sunflower density estimation.

Time Period of Content

• Single Date/Time: Start Date 2025-05-14 to End Date 2025-06-17

Data Quality Information

Composite images (orthophotos) have been generated with DJI Terra, with 75% frontal and 60% side overlap (50% front and 50% side overlap for the low-altitude orthophotos). Some of the surveys have been completed in suboptimal weather conditions (partly cloudy). This resulted in visible variation in color and reflectances in several regions of the orthophotos. The raw data is validated to be complete - representing the entirety of the observed field for every survey. An accompanying validation script is provided with the dataset.

Horizontal Coordinate System

• Geographic Coordinate System: EPSG:4326
  • Angular Unit: Decimal degrees
  • Datum: WGS 84
  • Prime Meridian: Greenwich
  • Domain: Raster

Entity and Attribute Information

Detailed Description

Entities

Data is organized into directories. Each directory corresponds to one survey and uses DD.MM.YYYY format.

Each survey directory contains the following subdirectories:

• aerial - raw aerial footage, used during the reconstruction of the orthophoto with DJI Terra. These photos were taken from an altitude of 50 m, because the field is close to an airport and the altitude of 120 m was not allowed without special permission.
• terra - resulting orthophotos. There are two subdirectories, default/map and lu/map. The former is a reconstruction with default settings, whereas in the latter, the light uniformity switch was activated.
  • There is a result.tif file, corresponding to the RGB orthophoto and 5 orthophotos for each band, following the result_<Blue, Green, NIR, Red, RedEdge>.tif naming pattern.
  • There are two subdirectories with 5 vegetation index orthophotos, calculated by DJI Terra (GNDVI, LCI, NDRE, NDVI, OSAVI).
    • index_map - these orthophotos contain the vegetation index values in float32 (range is -1:1)
    • index_map_color - these orthophotos contain a "false color" render of the vegetation index values, for the purposes of visualization.
  • In addition, there are .prj projection file and .tfw georeference file for each orthophoto.
• XXm - where XX is either 3, 5 or 10 contains the low-altitude images. For each of the 48 surveying points, there is one RGB image in JPEG and 5 images in TIFF format (corresponding to the 5 bands),

Additionally, there are three low-altitude orthophoto missions in the low_alt_aerial folder:

• 02.06.2025 - performed from an altitude of 14 m
• 05.06.2025 - performed from 10 m (with reconstruction gaps)
• 17.06.2025 - again performed from 10 m, but only covering the reconstruction gaps from the previous mission.

Reconstruction of these missions was done only once, with the light uniformity switch on. Raw photos from these missions were not published in this repository.

All images in this dataset are geo-referenced and contain timestamps, image quality, camera properties and other metadata.

Capture aperture

Drone surveys are executed with DJI Phantom 4 Multispectral drone. The drone uses the following sensors to capture data:

Sensors: Six 1/2.9” CMOS

Filters:

• Blue (B): 450 nm ± 16 nm
• Green (G): 560 nm ± 16 nm
• Red (R): 650 nm ± 16 nm
• Red edge (RE): 730 nm ± 16 nm
• Near-infrared (NIR): 840 nm ± 26 nm

Lenses:

• FOV (Field of View): 62.7°
• Focal Length: 5.74 mm
• Aperture: f/2.2

Software used for generating composite images: DJI Terra Agriculture 4.2.5.

Metadata Reference Information

• Metadata Contact:

  • Name: Pavel Genevski
  • Organization: SAP LABS Bulgaria
  • Position: Research expert
  • Email: pavel.genevski@sap.com

• Metadata Date: Date of creating this metadata (2025.07.14)

• Metadata Standard Name: FGDC Content Standard for Digital Geospatial Metadata

Additional Information

• Keywords: agriculture, multispectral, crop, sunflower
• Access Constraints: CC BY 4.0
• Use Constraints: CC BY 4.0