Table of Contents: :: Library Catalog

Saved in:

Bibliographic Details
Main Authors:	Ivanovici, Mihai, Popa, Stefan, Adrian, Rujoi, Rujoi-Nan, Adriana-Georgiana, Mosut, Camelia
Format:	Recurso digital
Language:	English
Published:	Zenodo 2026
Subjects:	soil roughness nadir view RGB images heights profile
Online Access:	https://doi.org/10.5281/zenodo.19679743
Tags:	Add Tag No Tags, Be the first to tag this record!

Table of Contents:

The SoRoDa25k dataset was produced within the framework of the AI4AGRI European project GA 101079136. SoRoDa25k stands for Soil Roughness Dataset, comprising 25,195 annotated, top-view, 24-bit-per-pixel RGB images of bare soil. Each image has a spatial resolution of 256 × 256 pixels and utilizes the PNG lossless compression format. The images are processed frames extracted from Full-HD (i.e. with a resolution of 1920 × 1080 pixels), 25 frames per second video sequences captured with a Canon EOS 5D Mk II DSLR camera placed on top of a cart, which was manually pushed over the soil surface in two scenarios: 1) in a controlled environment with a more stabilized travel of the cart and 2) in an actual agricultural field in the North of Brasov City, Romania. The processing of each frame involved the following steps: i) the determination of the vertical height profile, with a resolution of about 0.27 mm, along the projection of a red line laser on the soil surface relative to a reference on the cart using a method similar to that presented in [1]; ii) the determination of the soil roughness value from the vertical profile as the standard deviation of the heights, ii) the cropping of the central square with an edge length of 1080 pixels and iii) the subsampling to a square image with an edge of 256 pixels. The height profiles are not continuous due to interruptions in the lateral-view laser projection visibility. The interruptions are larger and more frequent in the field data. Only the non-stationary video sequences were used and only the frames for which the soil roughness value was determined with a reasonable degree of confidence (i.e. using at least 14% of the profile). In all images, the red line laser projection appears as a horizontal trace at the top.  The directory controlled_environment contains 14,325 top-view PNG images of the soil from the controlled environment, a roughness.csv file which associates each image with the determined soil roughness value and a vertical_profiles.csv file which associates each image to its heights profile. The profile heights are expressed in mm and the unknown values are listed as NaN. The names follow the format frame_x_5D[…]-y.png, where x represents the video sequence number and y is a 5-digit number representing the frame number. The directory field contains 10,870 top-view PNG images of the soil from the agricultural field, a roughness.csv file which associates each image with the determined soil roughness value and a vertical_profiles.csv file which associates each image to its heights profile. The profile heights are expressed in mm and the unknown values are listed as NaN. The names follow the format clipx-y.png, where x represents the video sequence number and y is a 5-digit number representing the frame number. References [1] M. Ivanovici, S. Popa, K. Marandskiy, and C. Florea, “Deep automatic soil roughness estimation from digital images,” European Journal of Remote Sensing, vol. 57, no. 1, p. 2342955, 2024. Available: <a href="https://doi.org/10.1080/22797254.2024.2342955">https://doi.org/10.1080/22797254.2024.2342955</a>  Funded by the European Union. The AI4AGRI project entitled “Romanian Excellence Center on Artificial Intelligence on Earth Observation Data for Agriculture” received funding from the European Union’s Horizon Europe research and innovation programme under the grant agreement no. 101079136. Views and opinions expressed are however those of the authors only and do not necessarily reflect those of the European Union. Neither the European Union nor the granting authority can be held responsible for them.

Similar Items