Excretion Detection in Pigsties Using Convolutional and Transformerbased Deep Neural Networks

Anthony Stein; Simon Mielke

arxiv: 2412.00256 · v3 · submitted 2024-11-29 · 💻 cs.CV

Excretion Detection in Pigsties Using Convolutional and Transformerbased Deep Neural Networks

Simon Mielke , Anthony Stein This is my paper

Pith reviewed 2026-05-23 07:45 UTC · model grok-4.3

classification 💻 cs.CV

keywords excretion detectionpigstiesdeep learningobject detectionthermal imagesconvolutional networkstransformerslivestock farming

0 comments

The pith

Deep learning models detect animal excretions in pig barns with average precision above 90 percent.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper evaluates four deep learning models on detecting urine and feces in thermal images from pigsties. All models achieve over 90 percent average precision and show robustness when tested on images from different conditions than the training data. This matters because manual detection is slow and simple threshold methods fail when animals and excretions have similar temperatures. The results indicate that deep learning can automate soiled floor detection in barns despite variations in housing and environment.

Core claim

This work is the first to compare convolutional and transformer-based detection models for excretions in pigsties. The models Faster R-CNN, YOLOv8, DETR and DAB-DETR are evaluated on two datasets from different pig houses using nested cross-validation. They demonstrate suitability for reliable detection with average precision over 90 percent and robustness on out-of-distribution data with only slight performance decreases.

What carries the argument

Comparison of object detection architectures (Faster R-CNN, YOLOv8, DETR, DAB-DETR) on thermal images for identifying animal excretions.

If this is right

Automated detection supports improved barn management processes.
Information from detections can help model emission dynamics from livestock.
The approach handles variations due to housing type, animal factors, and weather.
Models maintain performance on data with differences from training conditions.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

Similar models could be applied to detect excretions in other livestock settings like cattle barns.
Real-time implementation could enable proactive cleaning schedules.
The method might reduce the need for human monitoring in large-scale farming operations.
Further testing on more diverse farms could confirm broader applicability.

Load-bearing premise

The created datasets sufficiently represent the range of conditions in pigsties so that performance generalizes to new barns.

What would settle it

If a new set of thermal images from an unseen pig barn yields average precision significantly below 90 percent, the claim of reliable and robust detection would be challenged.

Figures

Figures reproduced from arXiv: 2412.00256 by Anthony Stein, Simon Mielke.

**Figure 2.** Figure 2: Differences between datasets and challenges identified during annotation. [PITH_FULL_IMAGE:figures/full_fig_p012_2.png] view at source ↗

**Figure 3.** Figure 3: Significant differences of the models on the first dataset in the respective metrics [PITH_FULL_IMAGE:figures/full_fig_p028_3.png] view at source ↗

**Figure 4.** Figure 4: Significant differences of the models on the second dataset in the respective [PITH_FULL_IMAGE:figures/full_fig_p029_4.png] view at source ↗

read the original abstract

Animal excretions in form of urine puddles and feces are a significant source of emissions in livestock farming. Automated detection of soiled floor in barns can contribute to improved management processes but also the derived information can be used to model emission dynamics. Previous research approaches to determine the puddle area require manual detection of the puddle in the barn. While humans can detect animal excretions on thermal images of a livestock barn, automated approaches using thresholds fail due to other objects of the same temperature, such as the animals themselves. In addition, various parameters such as the type of housing, animal species, age, sex, weather and unknown factors can influence the type and shape of excretions. Due to this heterogeneity, a method for automated detection of excretions must therefore be not only be accurate but also robust to varying conditions. These requirements can be met by using contemporary deep learning models from the field of artificial intelligence. This work is the first to investigate the suitability of different deep learning models for the detection of excretions in pigsties, thereby comparing established convolutional architectures with recent transformer-based approaches. The detection models Faster R-CNN, YOLOv8, DETR and DAB-DETR are compared and statistically assessed on two created training datasets representing two pig houses. We apply a method derived from nested cross-validation and report on the results in terms of eight common detection metrics. Our work demonstrates that all investigated deep learning models are generally suitable for reliably detecting excretions with an average precision of over 90%. The models also show robustness on out of distribution data that possesses differences from the conditions in the training data, however, with expected slight decreases in the overall detection performance.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

This applies four off-the-shelf detectors to thermal pig-barn images and reports AP over 90 percent, but the two-house dataset leaves the generalization claim under-supported.

read the letter

The paper's core contribution is a head-to-head test of Faster R-CNN, YOLOv8, DETR, and DAB-DETR on thermal images of urine and feces in pigsties. They built two new datasets from separate houses, ran a nested cross-validation style evaluation, and measured eight standard detection metrics. All four models clear 90 percent average precision on the held-out portions, with the expected modest drop on a separate out-of-distribution set. That direct CNN-versus-transformer comparison on this imagery is new, and the motivation around handling housing and animal variation is clearly stated up front. The work also ships concrete numbers rather than just architecture claims, which is useful for anyone who needs a starting point for barn monitoring hardware. The main weakness is the data scope. Two houses cannot cover the full range of housing types, animal ages, sexes, weather, and unknown factors the abstract itself lists as sources of heterogeneity. Without reported image counts, annotation protocols, or split details, it is difficult to judge whether the OOD test actually stresses those variables or just measures intra-house shifts. No statistical tests on the metric differences are mentioned either, so the claim that all models are “generally suitable” rests on point estimates alone. This is the kind of paper that belongs in an applied computer-vision or precision-agriculture venue rather than a general ML conference. A reader building a barn sensor system could extract practical baseline numbers and model choices from it. Anyone needing strong evidence of robustness across real farm diversity would need the missing dataset statistics and a broader collection before treating the results as settled. I would send it to referees once the authors add the dataset sizes, annotation process, and any significance checks; the empirical comparison is worth a look even if the generalization story needs tightening.

Referee Report

3 major / 2 minor

Summary. The paper claims that deep learning models (Faster R-CNN, YOLOv8, DETR, DAB-DETR) are suitable for reliable automated detection of animal excretions (urine and feces) in pigsties, achieving average precision above 90% on two newly created datasets from distinct pig houses. It further claims robustness to out-of-distribution images with only slight performance drops, using a nested cross-validation approach and eight standard detection metrics, positioning this as the first such comparison of convolutional and transformer architectures for the task motivated by heterogeneity in housing, species, age, sex, weather, and other factors.

Significance. If the performance claims hold under proper documentation, the work offers a practical DL-based alternative to manual or threshold-based detection of soiled areas, with potential applications in emission modeling and barn management. It provides a direct empirical head-to-head comparison of established CNN detectors against recent transformer detectors on a real-world agricultural vision task, which is a useful contribution even if generalization claims require further validation.

major comments (3)

[Abstract] Abstract: The central claim of average precision >90% and robustness on OOD data is load-bearing for the paper's conclusions, yet the abstract (and presumably the results section) supplies no dataset sizes, number of annotated images, annotation protocol details, or explicit train-test split ratios. Without these, the reported metrics cannot be assessed for statistical reliability or overfitting risk.
[Abstract] Abstract / Methods (nested cross-validation description): The claim that the two training datasets adequately represent heterogeneity across housing type, animal species, age, sex, weather, and unknown factors (explicitly listed as motivation for DL) is not supported by any quantitative characterization of inter-house or OOD variability; the OOD robustness is described only qualitatively as 'slight decreases,' leaving open whether the test truly probes the motivating sources of variation or only minor intra-house shifts.
[Abstract] Abstract: The statement that models were 'statistically assessed' on eight metrics lacks any mention of specific tests (e.g., significance testing of AP differences across models or periods), which is required to substantiate the comparative suitability claim.

minor comments (2)

[Title] Title: 'Transformerbased' should be 'Transformer-based'.
[Abstract] Abstract: Minor phrasing issues such as 'not only be accurate' (duplicate 'be') and 'in form of' (should be 'in the form of').

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive feedback on the abstract and methods description. We address each major comment below and will make revisions to improve clarity and support for the claims.

read point-by-point responses

Referee: [Abstract] Abstract: The central claim of average precision >90% and robustness on OOD data is load-bearing for the paper's conclusions, yet the abstract (and presumably the results section) supplies no dataset sizes, number of annotated images, annotation protocol details, or explicit train-test split ratios. Without these, the reported metrics cannot be assessed for statistical reliability or overfitting risk.

Authors: We agree that the abstract would benefit from including these details to allow independent assessment of the results. The Methods section already describes the two datasets (including image counts from each pig house), the annotation process by domain experts, and the nested cross-validation procedure with explicit train/validation/test ratios. In the revised manuscript, we will condense and add the key figures (e.g., total annotated images, split ratios) into the abstract. revision: yes
Referee: [Abstract] Abstract / Methods (nested cross-validation description): The claim that the two training datasets adequately represent heterogeneity across housing type, animal species, age, sex, weather, and unknown factors (explicitly listed as motivation for DL) is not supported by any quantitative characterization of inter-house or OOD variability; the OOD robustness is described only qualitatively as 'slight decreases,' leaving open whether the test truly probes the motivating sources of variation or only minor intra-house shifts.

Authors: The two datasets come from distinct commercial pig houses that differ in housing design, ventilation, and animal management, which we describe qualitatively as capturing the listed sources of variation. However, we acknowledge that no quantitative metrics (such as feature distribution statistics or inter-house image similarity measures) are provided to characterize the degree of heterogeneity. The OOD test uses images from a third barn with documented differences, and performance drops are reported numerically per metric rather than only qualitatively. We will add a quantitative comparison of variability (e.g., temperature histograms or bounding-box statistics) between the datasets in the revision. revision: partial
Referee: [Abstract] Abstract: The statement that models were 'statistically assessed' on eight metrics lacks any mention of specific tests (e.g., significance testing of AP differences across models or periods), which is required to substantiate the comparative suitability claim.

Authors: The phrase 'statistically assessed' refers to the nested cross-validation protocol, which yields mean and standard deviation values for each of the eight metrics across multiple folds and thereby quantifies variability due to data partitioning. No additional hypothesis tests (such as paired t-tests on AP scores between models) were performed. We believe the cross-validation results already provide a robust basis for comparing the models; however, if the referee considers pairwise significance tests necessary, we can include them in a revision. revision: partial

Circularity Check

0 steps flagged

No circularity: empirical model comparison on held-out image data

full rationale

The paper reports an empirical evaluation of four object detection architectures (Faster R-CNN, YOLOv8, DETR, DAB-DETR) trained and tested on image datasets collected from two pig houses, with additional OOD testing. All reported figures are direct performance measurements (average precision, etc.) on held-out images; no equations, parameter fits, or predictions are claimed to derive from first principles or to reduce to the training data by construction. No self-citation chain is invoked to justify a uniqueness result or ansatz. The work is therefore self-contained against external benchmarks and receives the default non-circularity finding.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

The work rests on standard deep-learning assumptions that the chosen architectures can learn the visual features of excretions when trained on representative thermal images; no new entities or ad-hoc axioms are introduced.

pith-pipeline@v0.9.0 · 5836 in / 1243 out tokens · 29594 ms · 2026-05-23T07:45:38.278855+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

14 extracted references · 14 canonical work pages · 1 internal anchor

[1]

Jour- nal of Agricultural Engineering Research 66, 93–102

Effect of Type of Slatted Floor and Degree of Fouling of Solid Floor on Ammonia Emission Rates from Fattening Piggeries. Jour- nal of Agricultural Engineering Research 66, 93–102. doi: 10.1006/ jaer.1996.0121. https://www.sciencedirect.com/science/article/ pii/S0021863496901213. Abadi, M., Barham, P., Chen, J., Chen, Z., Davis, A., Dean, J., Devin, M., Gh...

work page doi:10.1016/j.applanim.2019.104888 1996
[2]

https://www.sciencedirect.com/ science/article/pii/S0021863497902158

doi: 10.1006/jaer.1997.0215. https://www.sciencedirect.com/ science/article/pii/S0021863497902158. Carion, N., Massa, F., Synnaeve, G., Usunier, N., Kirillov, A., Zagoruyko, S., 2020. End-to-End Object Detection with Transformers, in: Vedaldi, A., Bischof, H., Brox, T., Frahm, J.M. (Eds.), Computer Vision – ECCV

work page doi:10.1006/jaer.1997.0215 1997
[3]

volume 12346, pp

Springer International Publishing, Cham. volume 12346, pp. 213–

work page
[4]

MMDetection: Open MMLab Detection Toolbox and Benchmark

doi: 10.1007/978-3-030-58452-8_13 . https://link.springer. com/10.1007/978-3-030-58452-8_13 . Caro, D., Davis, S.J., Bastianoni, S., Caldeira, K., 2014. Global and regional trends in greenhouse gas emissions from livestock. Climatic Change 126, 203–216. doi:10.1007/s10584-014-1197-x . https://doi.org/10.1007/ s10584-014-1197-x. Chen, K., Wang, J., Pang, J...

work page internal anchor Pith review Pith/arXiv arXiv doi:10.1007/978-3-030-58452-8_13 2014
[5]

, author Dong, W

doi: 10.1109/CVPR.2009.5206848. https://ieeexplore.ieee.org/ document/5206848. Eurich-Menden, B., Wolf, U., Horlacher, D., Dehler, G., Smirnov, A., 2024. Ermittlung von Emissionsdaten f¨ ur die Beurteilung der Umweltwirkun- gen der Nutztierhaltung (EmiDaT). Abschlussbericht. Technical Re- port. Kuratorium f¨ ur Technik und Bauwesen in der Landwirtschaft e...

work page doi:10.1109/cvpr.2009.5206848 2009
[6]

Environ- mental Pollution 86, 43–82

Effects of atmospheric ammonia on vegetation—A review. Environ- mental Pollution 86, 43–82. doi:10.1016/0269-7491(94)90008-6. https: //www.sciencedirect.com/science/article/pii/0269749194900086. G¨ abel, G., Loeffler, K., Pfannkuche, H., 2018. Anatomie und Physiologie der Haustiere. UTB. 15., vollst¨ andig ¨ uberarbeitete und erweiterte Auflage ed., Verla...

work page doi:10.1016/0269-7491(94)90008-6 2018
[7]

https://openreview.net/forum?id=H1oyRlYgg

On Large-Batch Training for Deep Learning: Generalization Gap and Sharp Minima, in: International Conference on Learning Representa- tions. https://openreview.net/forum?id=H1oyRlYgg. Kim, J., Baek, J., Choi, H., Kim, E., 2016. Wet area and puddle detection for Advanced Driver Assistance Systems (ADAS) using a stereo camera. International Journal of Contro...

work page doi:10.1007/s12555-015-0024-0 2016
[8]

Microsoft COCO: common objects in context,

doi: 10.1007/978-3-319-10602-1_48 . Liu, S., Li, F., Zhang, H., Yang, X., Qi, X., Su, H., Zhu, J., Zhang, L.,

work page doi:10.1007/978-3-319-10602-1_48
[9]

Dab-detr: Dynamic anchor boxes are better queries for detr.arXiv preprint arXiv:2201.12329, 2022

DAB-DETR: Dynamic Anchor Boxes are Better Queries for DETR, in: DAB-DETR: Dynamic Anchor Boxes Are Better Queries for DETR. doi:10.48550/arXiv.2201.12329. https://openreview.net/forum?id= oMI9PjOb9Jl. Mao, R., Shen, D., Wang, R., Cui, Y., Hu, Y., Li, M., Wang, M., 2024. An Integrated Gather-and-Distribute Mechanism and Attention-Enhanced Deformable Convol...

work page doi:10.48550/arxiv.2201.12329 2024
[10]

https://www.mdpi.com/2076-2615/ 14/9/1316

doi: 10.3390/ani14091316. https://www.mdpi.com/2076-2615/ 14/9/1316. McCulloch, W.S., Pitts, W., 1943. A logical calculus of the ideas immanent in nervous activity. The bulletin of mathematical biophysics 5, 115–133. doi:10.1007/BF02478259. https://doi.org/10.1007/BF02478259. Melfsen, A., Lepsien, A., Bosselmann, J., Koschmider, A., Hartung, E., 2023. Des...

work page doi:10.3390/ani14091316 2076
[11]

https://pubmed.ncbi.nlm

doi: 10.1128/mr.53.1.85-108.1989. https://pubmed.ncbi.nlm. nih.gov/2651866/. Optris GmbH & Co.KG, . Optris. Optris GmbH & Co.KG. https://www.optris.com/en/products/infrared-cameras/ software-pix-connect/. Pan, S.J., Yang, Q., 2010. A Survey on Transfer Learning. IEEE Transactions on Knowledge and Data Engineering 22, 1345–

work page doi:10.1128/mr.53.1.85-108.1989 1989
[12]

https://ieeexplore.ieee.org/ document/5288526/footnotes#footnotes-id-fn1

doi: 10.1109/TKDE.2009.191. https://ieeexplore.ieee.org/ document/5288526/footnotes#footnotes-id-fn1. Parmiggiani, A., Liu, D., Psota, E., Fitzgerald, R., Norton, T., 2023. Don’t get lost in the crowd: Graph convolutional network for online an- imal tracking in dense groups. Computers and Electronics in Agricul- ture 212, 108038. doi: 10.1016/j.compag.202...

work page doi:10.1109/tkde.2009.191 2009
[13]

2017, IEEE Transactions on Pattern Analysis and Machine Intelligence, 39, 1137, doi: 10.1109/TPAMI.2016.2577031

doi: 10.1109/TPAMI.2016.2577031. http://ieeexplore.ieee. org/document/7485869/. 45 Riekert, M., Opderbeck, S., Wild, A., Gallmann, E., 2021. Model selection for 24/7 pig position and posture detection by 2D camera imaging and deep learning. Computers and Electronics in Agricul- ture 187, 106213. doi: 10.1016/j.compag.2021.106213. https://www. sciencedirec...

work page doi:10.1109/tpami.2016.2577031 2016
[14]

https://www.sciencedirect

doi:10.1016/j.compag.2017.03.012. https://www.sciencedirect. com/science/article/pii/S0168169916301338. Snoek, D.J.W., Stigter, J.D., Ogink, N.W.M., Groot Koerkamp, P.W.G., 2014b. Sensitivity analysis of mechanistic models for estimating am- monia emission from dairy cow urine puddles. Biosystems Engineering 121, 12–24. doi: 10.1016/j.biosystemseng.2014.0...

work page doi:10.1016/j.compag.2017.03.012 2017

[1] [1]

Jour- nal of Agricultural Engineering Research 66, 93–102

Effect of Type of Slatted Floor and Degree of Fouling of Solid Floor on Ammonia Emission Rates from Fattening Piggeries. Jour- nal of Agricultural Engineering Research 66, 93–102. doi: 10.1006/ jaer.1996.0121. https://www.sciencedirect.com/science/article/ pii/S0021863496901213. Abadi, M., Barham, P., Chen, J., Chen, Z., Davis, A., Dean, J., Devin, M., Gh...

work page doi:10.1016/j.applanim.2019.104888 1996

[2] [2]

https://www.sciencedirect.com/ science/article/pii/S0021863497902158

doi: 10.1006/jaer.1997.0215. https://www.sciencedirect.com/ science/article/pii/S0021863497902158. Carion, N., Massa, F., Synnaeve, G., Usunier, N., Kirillov, A., Zagoruyko, S., 2020. End-to-End Object Detection with Transformers, in: Vedaldi, A., Bischof, H., Brox, T., Frahm, J.M. (Eds.), Computer Vision – ECCV

work page doi:10.1006/jaer.1997.0215 1997

[3] [3]

volume 12346, pp

Springer International Publishing, Cham. volume 12346, pp. 213–

work page

[4] [4]

MMDetection: Open MMLab Detection Toolbox and Benchmark

doi: 10.1007/978-3-030-58452-8_13 . https://link.springer. com/10.1007/978-3-030-58452-8_13 . Caro, D., Davis, S.J., Bastianoni, S., Caldeira, K., 2014. Global and regional trends in greenhouse gas emissions from livestock. Climatic Change 126, 203–216. doi:10.1007/s10584-014-1197-x . https://doi.org/10.1007/ s10584-014-1197-x. Chen, K., Wang, J., Pang, J...

work page internal anchor Pith review Pith/arXiv arXiv doi:10.1007/978-3-030-58452-8_13 2014

[5] [5]

, author Dong, W

doi: 10.1109/CVPR.2009.5206848. https://ieeexplore.ieee.org/ document/5206848. Eurich-Menden, B., Wolf, U., Horlacher, D., Dehler, G., Smirnov, A., 2024. Ermittlung von Emissionsdaten f¨ ur die Beurteilung der Umweltwirkun- gen der Nutztierhaltung (EmiDaT). Abschlussbericht. Technical Re- port. Kuratorium f¨ ur Technik und Bauwesen in der Landwirtschaft e...

work page doi:10.1109/cvpr.2009.5206848 2009

[6] [6]

Environ- mental Pollution 86, 43–82

Effects of atmospheric ammonia on vegetation—A review. Environ- mental Pollution 86, 43–82. doi:10.1016/0269-7491(94)90008-6. https: //www.sciencedirect.com/science/article/pii/0269749194900086. G¨ abel, G., Loeffler, K., Pfannkuche, H., 2018. Anatomie und Physiologie der Haustiere. UTB. 15., vollst¨ andig ¨ uberarbeitete und erweiterte Auflage ed., Verla...

work page doi:10.1016/0269-7491(94)90008-6 2018

[7] [7]

https://openreview.net/forum?id=H1oyRlYgg

On Large-Batch Training for Deep Learning: Generalization Gap and Sharp Minima, in: International Conference on Learning Representa- tions. https://openreview.net/forum?id=H1oyRlYgg. Kim, J., Baek, J., Choi, H., Kim, E., 2016. Wet area and puddle detection for Advanced Driver Assistance Systems (ADAS) using a stereo camera. International Journal of Contro...

work page doi:10.1007/s12555-015-0024-0 2016

[8] [8]

Microsoft COCO: common objects in context,

doi: 10.1007/978-3-319-10602-1_48 . Liu, S., Li, F., Zhang, H., Yang, X., Qi, X., Su, H., Zhu, J., Zhang, L.,

work page doi:10.1007/978-3-319-10602-1_48

[9] [9]

Dab-detr: Dynamic anchor boxes are better queries for detr.arXiv preprint arXiv:2201.12329, 2022

DAB-DETR: Dynamic Anchor Boxes are Better Queries for DETR, in: DAB-DETR: Dynamic Anchor Boxes Are Better Queries for DETR. doi:10.48550/arXiv.2201.12329. https://openreview.net/forum?id= oMI9PjOb9Jl. Mao, R., Shen, D., Wang, R., Cui, Y., Hu, Y., Li, M., Wang, M., 2024. An Integrated Gather-and-Distribute Mechanism and Attention-Enhanced Deformable Convol...

work page doi:10.48550/arxiv.2201.12329 2024

[10] [10]

https://www.mdpi.com/2076-2615/ 14/9/1316

doi: 10.3390/ani14091316. https://www.mdpi.com/2076-2615/ 14/9/1316. McCulloch, W.S., Pitts, W., 1943. A logical calculus of the ideas immanent in nervous activity. The bulletin of mathematical biophysics 5, 115–133. doi:10.1007/BF02478259. https://doi.org/10.1007/BF02478259. Melfsen, A., Lepsien, A., Bosselmann, J., Koschmider, A., Hartung, E., 2023. Des...

work page doi:10.3390/ani14091316 2076

[11] [11]

https://pubmed.ncbi.nlm

doi: 10.1128/mr.53.1.85-108.1989. https://pubmed.ncbi.nlm. nih.gov/2651866/. Optris GmbH & Co.KG, . Optris. Optris GmbH & Co.KG. https://www.optris.com/en/products/infrared-cameras/ software-pix-connect/. Pan, S.J., Yang, Q., 2010. A Survey on Transfer Learning. IEEE Transactions on Knowledge and Data Engineering 22, 1345–

work page doi:10.1128/mr.53.1.85-108.1989 1989

[12] [12]

https://ieeexplore.ieee.org/ document/5288526/footnotes#footnotes-id-fn1

doi: 10.1109/TKDE.2009.191. https://ieeexplore.ieee.org/ document/5288526/footnotes#footnotes-id-fn1. Parmiggiani, A., Liu, D., Psota, E., Fitzgerald, R., Norton, T., 2023. Don’t get lost in the crowd: Graph convolutional network for online an- imal tracking in dense groups. Computers and Electronics in Agricul- ture 212, 108038. doi: 10.1016/j.compag.202...

work page doi:10.1109/tkde.2009.191 2009

[13] [13]

2017, IEEE Transactions on Pattern Analysis and Machine Intelligence, 39, 1137, doi: 10.1109/TPAMI.2016.2577031

doi: 10.1109/TPAMI.2016.2577031. http://ieeexplore.ieee. org/document/7485869/. 45 Riekert, M., Opderbeck, S., Wild, A., Gallmann, E., 2021. Model selection for 24/7 pig position and posture detection by 2D camera imaging and deep learning. Computers and Electronics in Agricul- ture 187, 106213. doi: 10.1016/j.compag.2021.106213. https://www. sciencedirec...

work page doi:10.1109/tpami.2016.2577031 2016

[14] [14]

https://www.sciencedirect

doi:10.1016/j.compag.2017.03.012. https://www.sciencedirect. com/science/article/pii/S0168169916301338. Snoek, D.J.W., Stigter, J.D., Ogink, N.W.M., Groot Koerkamp, P.W.G., 2014b. Sensitivity analysis of mechanistic models for estimating am- monia emission from dairy cow urine puddles. Biosystems Engineering 121, 12–24. doi: 10.1016/j.biosystemseng.2014.0...

work page doi:10.1016/j.compag.2017.03.012 2017