ORCID
Uilson Ricardo Venancio Aires: https://orcid.org/0000-0002-6745-2787
Vitor Souza Martins: https://orcid.org/0000-0003-3802-0368
Dakota Hester: https://orcid.org/0000-0002-3143-1159
Thainara Lima: https://orcid.org/0000-0001-6492-0330
Lucas Borges Ferreira: https://orcid.org/0000-0001-6838-3114
MSU Affiliations
College of Agriculture and Life Sciences; Department of Agriculture and Biological Engineering; James Worth Bagley College of Engineering
Item Type
Research Data
Abstract
Open cattle feedlots are a major form of beef production infrastructure in the United States, characterized by outdoor confinement, high animal densities, and regulated feeding practices. Despite their economic importance, a comprehensive and spatially consistent national database of these facilities remains limited. This dataset provides labeled training data, trained deep learning models, and geospatial detection outputs developed to support automated identification of open cattle feedlots from aerial imagery. The dataset includes a manually curated shapefile of 11,746 open cattle feedlot facilities identified through visual interpretation of high-resolution aerial imagery in highly productive counties of Nebraska, Kansas, and Texas. These labels were used to construct training, validation, and test datasets for You Only Look Once (YOLO) object detection models. To reduce false positives, approximately 13,000 background image patches representing non-feedlot landscapes were included during model training. National Agriculture Imagery Program (NAIP) county-level imagery (approximately 43 TB of GeoTIFF data, resampled to 1 m resolution) acquired between 2019 and 2022 was processed into thousands of 640 × 640–pixel image patches for large-scale inference. The repository also provides five trained YOLO model variants optimized for open feedlot detection, as well as geospatial detection outputs generated from model inference. These outputs include a shapefile of detected feedlot pen polygons and a shapefile of centroid points representing the central location of each detected facility across the Contiguous United States. This dataset enables large-scale spatial analyses of beef production infrastructure and supports applications in agricultural monitoring, land-use mapping, environmental assessment, and epidemiological modeling.
Creation Date
10-5-2025
Publication Date
Fall 12-1-2025
Recommended Citation
Venancio Aires, Uilson Ricardo; Souza Martins, Vitor; Hester, Dakota; Lima, Thainara; and Borges Ferreira, Lucas, "National-Scale Open Cattle Feedlot Detection Using Deep Learning and High-Resolution Aerial Images: Spatial Distribution and Animal Welfare Analysis" (2025). Research Data. 8.
https://scholarsjunction.msstate.edu/research-data/8
Comments
OPEN CATTLE FEEDLOT DETECTION DATASET AND YOLO MODELS (CONUS – SELECTED STATES)
REPOSITORY OVERVIEW
This repository contains geospatial data layers and trained deep learning models developed to support the detection of open cattle feedlots in the United States using aerial imagery and object detection techniques.
The materials provided here were created as part of a research effort focused on developing a scalable and automated framework for identifying open cattle feedlot facilities using YOLO-based object detection models and high-resolution aerial imagery.
REPOSITORY CONTENTS
1. FEEDLOT LABEL DATA (TRAINING LABELS)
This repository includes a shapefile containing manually labeled open cattle feedlot facilities used for model training and validation.
Geographic coverage includes: - Texas - Nebraska - Kansas
Each feature represents the spatial footprint of an open cattle feedlot identified through visual interpretation of high-resolution aerial imagery.
Primary use: Training and evaluation of deep learning object detection models designed for open cattle feedlot identification.
2. TRAINED YOLO MODELS
The repository includes trained YOLO object detection models developed specifically to detect open cattle feedlots.
- Five trained YOLO model variants are provided. - Models were trained using tiled aerial imagery patches and associated feedlot labels. - Background (non-feedlot) imagery was incorporated during training to reduce false detections from visually similar features.
Primary use: Automated inference and large-scale detection of open cattle feedlots from aerial imagery.
3. FEEDLOT DETECTION OUTPUTS
a) Feedlot Pen Polygons
- Shapefile containing detected open cattle feedlot pen polygons. - Each polygon represents the spatial extent of feedlot pens detected by the trained models. - Polygons were derived through post-processing of model detection outputs.
b) Feedlot Facility Centroids
- Shapefile containing the central coordinates of detected feedlot facilities. - Each point represents the centroid of an identified feedlot operation. - Intended for spatial analysis, mapping, and integration with additional datasets.
COORDINATE REFERENCE SYSTEM (CRS)
All shapefiles include embedded spatial reference information through associated projection (.prj) files. Coordinate reference systems are appropriate for regional and national-scale geospatial analyses.
INTENDED USE
The data and models provided in this repository are intended to support: - Agricultural and environmental geospatial research - Development of regional or national inventories of open cattle feedlots - Epidemiological and disease spread modeling - Land-use and agricultural infrastructure analysis - Benchmarking of deep learning object detection methods in remote sensing
LIMITATIONS
- Training labels are limited to selected counties in Texas, Nebraska, and Kansas. - Model performance may vary in regions with different production systems or landscape characteristics. - Smaller feedlots and atypical facility layouts may be under-detected. - Outputs should be independently validated before use in regulatory, operational, or policy-related decision-making.
CITATION AND ACKNOWLEDGMENT
If these data or trained models are used in any publications, reports, or presentations, please cite the associated publication and acknowledge Scholars Junction, Mississippi State University, as the data repository.
CONTACT INFORMATION
Vitor S. Martins Agriculture and Biological Engineering Mississippi State University Email: vmartins@abe.msstate.edu