SFN-YOLO: Towards Free-Range Poultry Detection via Scale-aware Fusion Networks

Hao Wang; Hongtao Chen; Jiancong Bai; Jie Chen; Mingzhe Liu; Tao Dai; Yuhong Feng; Zhaoxi He

arxiv: 2509.17086 · v2 · pith:XA4MRGCJnew · submitted 2025-09-21 · 💻 cs.CV

SFN-YOLO: Towards Free-Range Poultry Detection via Scale-aware Fusion Networks

Jie Chen , Yuhong Feng , Tao Dai , Hao Wang , Hongtao Chen , Zhaoxi He , Mingzhe Liu , Jiancong Bai This is my paper

Pith reviewed 2026-05-22 12:24 UTC · model grok-4.3

classification 💻 cs.CV

keywords poultry detectionfree-range farmingscale-aware fusionobject detectionYOLOsmart agricultureM-SCOPE datasetcomputer vision

0 comments

The pith

SFN-YOLO detects free-range poultry at 80.7% mAP with 7.2 million parameters by fusing scale-aware features.

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

The paper presents SFN-YOLO as a detection approach built for free-range poultry where birds vary in size, often sit behind obstacles, and appear against shifting or cluttered backgrounds. It relies on scale-aware fusion to blend fine local details with wider scene context so the network can handle these variations without needing heavy computation. The authors also release the M-SCOPE dataset to cover diverse free-range conditions for training and testing. Experiments report 80.7% mean average precision at 7.2 million parameters, 35.1% fewer than the reference model, while performance holds across different settings. This combination of accuracy and efficiency is positioned to support real-time automated systems in smart poultry farming.

Core claim

SFN-YOLO employs scale-aware fusion to combine detailed local features with broader global context, enabling robust detection of multiscale poultry targets amid obstructions and complex or dynamic backgrounds in free-range environments. Supported by the newly introduced M-SCOPE dataset, the model reaches an mAP of 80.7% using only 7.2M parameters—a 35.1% reduction relative to the benchmark—while preserving strong generalization across domains.

What carries the argument

Scale-aware fusion that merges detailed local features with broader global context to manage varying target sizes and environmental complexity.

If this is right

Supports real-time automated monitoring on devices with limited computing resources.
Improves localization accuracy for poultry in cluttered outdoor scenes compared with standard single-scale detectors.
Provides a compact baseline that can be fine-tuned for related agricultural detection tasks.
Enables deployment in varied farm domains without retraining from scratch due to retained cross-domain performance.

Where Pith is reading between the lines

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

The same fusion approach could be tested on other free-roaming livestock such as sheep or cattle where scale variation is also common.
Pairing the detector with continuous video streams would allow measurement of bird movement patterns for welfare monitoring.
Extending the method to aerial imagery from drones could cover larger pasture areas while keeping parameter count low.

Load-bearing premise

The M-SCOPE dataset and chosen evaluation protocol sufficiently capture the multiscale targets, obstructions, and complex or dynamic backgrounds that occur in actual free-range poultry operations.

What would settle it

Evaluating SFN-YOLO on a fresh dataset collected from a commercial free-range poultry operation outside the M-SCOPE collection and checking whether mAP remains near 80.7% or drops substantially.

read the original abstract

Detecting and localizing poultry is essential for advancing smart poultry farming. Despite the progress of detection-centric methods, challenges persist in free-range settings due to multiscale targets, obstructions, and complex or dynamic backgrounds. To tackle these challenges, we introduce an innovative poultry detection approach named SFN-YOLO that utilizes scale-aware fusion. This approach combines detailed local features with broader global context to improve detection in intricate environments. Furthermore, we have developed a new expansive dataset (M-SCOPE) tailored for varied free-range conditions. Comprehensive experiments demonstrate our model achieves an mAP of 80.7% with just 7.2M parameters, which is 35.1% fewer than the benchmark, while retaining strong generalization capability across different domains. The efficient and real-time detection capabilities of SFN-YOLO support automated smart poultry farming.

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

SFN-YOLO adds a scale-aware fusion step to YOLO and releases a farm-specific poultry dataset, delivering lighter models with usable accuracy but thin support for the generalization claims.

read the letter

The main point to take away is that this work adapts YOLO for poultry detection in free-range conditions by adding a scale-aware fusion network and creates a dedicated dataset called M-SCOPE. The reported result is an mAP of 80.7 percent using 7.2 million parameters, which is notably smaller than the benchmark they compare against. They handle the problem description reasonably. Free-range poultry detection involves birds of varying sizes, obstructions from other birds or structures, and backgrounds that shift with lighting or movement. The scale-aware fusion combines local feature details with global context, which is a direct response to those multiscale and complex scene issues. This kind of targeted modification can lead to models that run efficiently on edge devices common in farming setups. Releasing the M-SCOPE dataset adds value because it focuses on the specific conditions of free-range operations rather than relying solely on generic datasets. That can help others working in similar agricultural vision tasks. Where things are softer is the supporting evidence for the claims. The performance numbers are given, but there is limited information on the experimental setup, such as how the train and test splits were made or whether there are breakdowns by difficulty factors like scale or occlusion. The generalization across different domains is stated, yet without quantitative validation like per-subset results or cross-farm evaluations, it is difficult to separate the effect of the fusion module from characteristics of the particular data collected. Ablation experiments would help clarify what the new component contributes. This kind of paper appeals to researchers and engineers in precision agriculture and computer vision for livestock monitoring. Someone developing real-time detection systems for farms would likely find the architecture description and the dataset useful as a reference or starting point. I would recommend putting it through peer review. The idea is practical and the efficiency improvement is worth a closer look from referees, who can ask for the missing experimental details to make the results more robust.

Referee Report

3 major / 1 minor

Summary. The manuscript proposes SFN-YOLO, a scale-aware fusion network based on YOLO for detecting poultry in free-range settings. It introduces the M-SCOPE dataset to handle challenges such as multiscale targets, obstructions, and dynamic backgrounds. The model is claimed to achieve 80.7% mAP with 7.2M parameters, representing a 35.1% reduction compared to the benchmark, while demonstrating strong generalization across different domains.

Significance. If validated, this work could contribute to efficient object detection in agricultural applications, particularly for smart poultry farming. The emphasis on parameter efficiency and real-time capability is promising for practical deployment. However, the significance is tempered by the need for more rigorous experimental validation of the generalization claims and the specific contributions of the scale-aware fusion mechanism.

major comments (3)

[4. Experiments] The reported mAP of 80.7% is presented without details on the evaluation protocol, including train/test split ratios, dataset statistics such as scale distributions or occlusion rates, or statistical significance measures like error bars. This undermines the ability to verify the claimed performance and generalization.
[3. Methodology] The scale-aware fusion is described conceptually but lacks a precise mathematical definition or diagram showing how local detailed features are combined with global context, making it difficult to assess novelty relative to existing feature fusion techniques in YOLO variants.
[5. Results] No ablation studies are provided to isolate the impact of the scale-aware fusion network on the performance gains versus the baseline YOLO model or the M-SCOPE dataset alone.

minor comments (1)

[Abstract] The abstract mentions 'comprehensive experiments' but does not reference specific tables or figures in the main text.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for their constructive and detailed feedback. We address each major comment below and commit to revisions that enhance the manuscript's clarity, reproducibility, and rigor without altering the core claims.

read point-by-point responses

Referee: [4. Experiments] The reported mAP of 80.7% is presented without details on the evaluation protocol, including train/test split ratios, dataset statistics such as scale distributions or occlusion rates, or statistical significance measures like error bars. This undermines the ability to verify the claimed performance and generalization.

Authors: We agree that additional experimental details are necessary to support reproducibility and verification. In the revised manuscript, we will expand the Experiments section to specify the train/test split ratios, provide dataset statistics on scale distributions and occlusion rates within M-SCOPE, and include error bars or standard deviations from repeated runs to demonstrate statistical significance of the 80.7% mAP result. revision: yes
Referee: [3. Methodology] The scale-aware fusion is described conceptually but lacks a precise mathematical definition or diagram showing how local detailed features are combined with global context, making it difficult to assess novelty relative to existing feature fusion techniques in YOLO variants.

Authors: We concur that a formal mathematical definition and visual aid would strengthen the presentation of novelty. We will add an explicit mathematical formulation of the scale-aware fusion operation (e.g., the adaptive weighting of local and global feature maps) along with a new diagram in the Methodology section to illustrate the combination process and facilitate direct comparison with prior YOLO fusion approaches. revision: yes
Referee: [5. Results] No ablation studies are provided to isolate the impact of the scale-aware fusion network on the performance gains versus the baseline YOLO model or the M-SCOPE dataset alone.

Authors: We acknowledge the importance of ablation studies for isolating contributions. In the revised manuscript, we will include new ablation experiments that compare the baseline YOLO model, the baseline trained on M-SCOPE, and the full SFN-YOLO incorporating scale-aware fusion, thereby quantifying the specific performance impact of the fusion mechanism. revision: yes

Circularity Check

0 steps flagged

No circularity in derivation chain; results are empirical measurements

full rationale

The paper introduces SFN-YOLO as a scale-aware fusion network for poultry detection and reports empirical mAP of 80.7% along with parameter counts on the newly created M-SCOPE dataset. No equations, derivations, or first-principles predictions appear in the provided text or abstract. Performance figures are presented as direct experimental outcomes rather than quantities computed from fitted constants or self-referential definitions inside the paper. The work is therefore self-contained against external benchmarks with no load-bearing steps that reduce to inputs by construction.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on standard assumptions of convolutional feature learning and on the representativeness of the newly collected M-SCOPE images; no new physical entities or ad-hoc constants are introduced.

axioms (1)

domain assumption Convolutional networks can extract useful multiscale features when local and global context are fused.
Implicit in the scale-aware fusion description.

pith-pipeline@v0.9.0 · 5692 in / 1278 out tokens · 50286 ms · 2026-05-22T12:24:10.651371+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

The SFM integrates the local feature extraction capability of convolutional operations with the global contextual modeling ability of self-attention mechanisms... bidirectional guidance mechanism... Eq. (1) Attention(Q,K,V)=Softmax(N(Q)N(K)^T / γ)V
IndisputableMonolith/Foundation/RealityFromDistinction.lean reality_from_one_distinction unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

SFN-YOLO achieves an mAP of 80.7% with just 7.2M parameters... strong generalization capability across different domains

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

28 extracted references · 28 canonical work pages · 1 internal anchor

[1]

INTRODUCTION Object detection, a fundamental computer vision technique for iden- tifying and localizing objects, is crucial to AI-driven smart poultry farming. It enables critical tasks, including flock counting, behavior analysis, and abnormality detection [1, 2], thus reducing labor costs and significantly boosting operational efficiency. The deployment...

work page
[2]

SFN-YOLO: Towards Free-Range Poultry Detection via Scale-aware Fusion Networks

METHOD 2.1. The SFN-YOLO Architecture Fig. 1 illustrates the overall architecture of SFN-YOLO, which intro- duces a novel SFM to enhance the backbone feature representation through the fusion of fine-grained local details and global contextual information. The SFM replaces the original cross stage partial with 2 convolutions-fusion (C2f) blocks to overcom...

work page internal anchor Pith review Pith/arXiv arXiv 2026
[3]

The dataset includes videos captured across 14 scenes, covering both indoor scenes and outdoor free-range scenes

EXPERIMENTS M-SCOPE Dataset.We introduce the M-SCOPE (Multiscale Clut- tered & Occluded Poultry Environment) dataset, designed for object detection in cluttered free-range environments, with annotations for the “chicken” class. The dataset includes videos captured across 14 scenes, covering both indoor scenes and outdoor free-range scenes. Outdoor scenes ...

work page 2025
[4]

The model fuses local details with global context to significantly enhance object perception amid challenges like scale variation, occlusion, and cluttered backgrounds

CONCLUSION We introduce SFN-YOLO, an efficient real-time detector for poul- try. The model fuses local details with global context to significantly enhance object perception amid challenges like scale variation, occlusion, and cluttered backgrounds. Evaluated on our novel M- SCOPE dataset, SFN-YOLO achieves state-of-the-art performance and strong generali...

work page
[5]

A review on development of an IoT-based system for enhancing yield in poultry farming,

J. G. Ambafi, Y . H. Muhammad, A. S. Ahmad, H. O. Ohize, U. S. Dauda, and J. J. Musa, “A review on development of an IoT-based system for enhancing yield in poultry farming,” Confluence University Journal of Science and Technology, vol. 2, no. 1, pp. 116–125, 2025

work page 2025
[6]

Automation and smart poul- try farm management: A review,

Priya, S.K. Bhanja, and V . Sahu, “Automation and smart poul- try farm management: A review,”Indian Journal of Poultry Science, vol. 56, no. 2, pp. 89–99, 2021

work page 2021
[7]

Faster R-CNN: Towards real-time object detection with region proposal net- works,

S. Ren, K. He, R. Girshick, and J. Sun, “Faster R-CNN: Towards real-time object detection with region proposal net- works,”Transactions on Pattern Analysis and Machine Intelli- gence, vol. 39, no. 6, pp. 1137–1149, 2017

work page 2017
[8]

Focal loss for dense object detection,

T.-Y . Lin, P. Goyal, R. Girshick, K. He, and P. Doll´ar, “Focal loss for dense object detection,” inProceedings of IEEE In- ternational Conference on Computer Vision (ICCV), 2017, pp. 2999–3007

work page 2017
[9]

End-to-end object detection with transform- ers,

N. Carion, F. Massa, G. Synnaeve, N. Usunier, A. Kirillov, and S. Zagoruyko, “End-to-end object detection with transform- ers,” inProceedings of European Conference on Computer Vision, 2020, pp. 213–229

work page 2020
[10]

DETRs beat YOLOs on real-time object detection,

Y . Zhao, W. Lv, S. Xu, J. Wei, G. Wang, Q. Dang, Y . Liu, and J. Chen, “DETRs beat YOLOs on real-time object detection,” inProceedings of the IEEE/CVF conference on computer vi- sion and pattern recognition, 2024, pp. 16965–16974

work page 2024
[11]

Ultralytics YOLOv5,

G. Jocher, “Ultralytics YOLOv5,” 2020

work page 2020
[12]

Ultralytics YOLOv8,

G. Jocher, A. Chaurasia, and J. Qiu, “Ultralytics YOLOv8,” 2023

work page 2023
[13]

Ultralytics YOLO11,

G. Jocher and J. Qiu, “Ultralytics YOLO11,” 2024

work page 2024
[14]

Mamba YOLO: A simple baseline for object detection with state space model,

Z. Wang, C. Li, H. Xu, X. Zhu, and H. Li, “Mamba YOLO: A simple baseline for object detection with state space model,” in Proceedings of the AAAI Conference on Artificial Intelligence, 2025, vol. 39, pp. 8205–8213

work page 2025
[15]

CAF-YOLO: A robust framework for multi-scale lesion detection in biomedical imagery,

Z. Chen and S. Lu, “CAF-YOLO: A robust framework for multi-scale lesion detection in biomedical imagery,” in Proceedings of IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2025, pp. 1–5

work page 2025
[16]

FBRT-YOLO: Faster and better for real-time aerial image detection,

Y . Xiao, T. Xu, Y . Xin, and J. Li, “FBRT-YOLO: Faster and better for real-time aerial image detection,” inProceedings of the AAAI Conference on Artificial Intelligence, 2025, vol. 39, pp. 8673–8681

work page 2025
[17]

YOLOX: Exceeding YOLO series in 2021,

Z. Ge, S. Liu, F. Wang, Z. Li, and J. Sun, “YOLOX: Exceeding YOLO series in 2021,” 2021

work page 2021
[18]

An efficient transformer network for detecting multi- scale chicken in complex free-range farming environments via improved RT-DETR,

X. Li, M. Cai, X. Tan, C. Yin, W. Chen, Z. Liu, J. Wen, and Y . Han, “An efficient transformer network for detecting multi- scale chicken in complex free-range farming environments via improved RT-DETR,”Computers and Electronics in Agricul- ture, vol. 224, pp. 109160, 2024

work page 2024
[19]

Chicktrack – a quantitative tracking tool for measuring chicken activity,

S. Neethirajan, “Chicktrack – a quantitative tracking tool for measuring chicken activity,”Measurement, vol. 191, pp. 110819, 2022

work page 2022
[20]

Only detect broilers once (ODBO): A method for monitoring and tracking individual behavior of cage-free broilers,

C. Yin, X. Tan, X. Li, M. Cai, and W. Chen, “Only detect broilers once (ODBO): A method for monitoring and tracking individual behavior of cage-free broilers,”Agriculture, vol. 15, no. 7, 2025

work page 2025
[21]

A survey of open-access datasets for computer vision in precision poultry farming computer vision datasets for poul- try,

G. Li, “A survey of open-access datasets for computer vision in precision poultry farming computer vision datasets for poul- try,”Poultry Science, vol. 104, no. 2, pp. 104784, 2025

work page 2025
[22]

A review of machine learn- ing models for disease prediction in poultry chickens,

D. Verma, N. Goel, and V .K. Garg, “A review of machine learn- ing models for disease prediction in poultry chickens,” inPro- ceedings of International Conference on Paradigms of Com- munication, Computing and Data Analytics, 2023, pp. 723– 737

work page 2023
[23]

Real-time behavior detection and judgment of egg breeders based on YOLO v3,

J. Wang, N. Wang, L. Li, and Z. Ren, “Real-time behavior detection and judgment of egg breeders based on YOLO v3,” Neural Computing and Applications, vol. 32, no. 10, pp. 5471– 5481, 2020

work page 2020
[24]

Detection of sick broilers by digital image processing and deep learning,

X. Zhuang and T. Zhang, “Detection of sick broilers by digital image processing and deep learning,”Biosystems Engineering, vol. 179, pp. 106–116, 2019

work page 2019
[25]

A deep learning model for detecting cage-free hens on the litter floor,

X. Yang, L. Chai, R.B. Bist, S. Subedi, and Z. Wu, “A deep learning model for detecting cage-free hens on the litter floor,” Animals, vol. 12, no. 15, pp. 1983, 2022

work page 1983
[26]

Squeeze-and-excitation net- works,

J. Hu, L. Shen, and G. Sun, “Squeeze-and-excitation net- works,” inProceedings of IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2018, pp. 7132–7141

work page 2018
[27]

T-rex label: Online ai data labeling plat- form,

T-Rex Label Team, “T-rex label: Online ai data labeling plat- form,” 2025, Accessed: 2025-08-26

work page 2025
[28]

The pascal visual object classes challenge 2012 (voc2012) development kit,

M. Everingham and J. Winn, “The pascal visual object classes challenge 2012 (voc2012) development kit,”Pattern Analysis, Statistical Modelling and Computational Learning, Tech. Rep, vol. 8, no. 5, pp. 2–5, 2011. 5

work page 2012

[1] [1]

INTRODUCTION Object detection, a fundamental computer vision technique for iden- tifying and localizing objects, is crucial to AI-driven smart poultry farming. It enables critical tasks, including flock counting, behavior analysis, and abnormality detection [1, 2], thus reducing labor costs and significantly boosting operational efficiency. The deployment...

work page

[2] [2]

SFN-YOLO: Towards Free-Range Poultry Detection via Scale-aware Fusion Networks

METHOD 2.1. The SFN-YOLO Architecture Fig. 1 illustrates the overall architecture of SFN-YOLO, which intro- duces a novel SFM to enhance the backbone feature representation through the fusion of fine-grained local details and global contextual information. The SFM replaces the original cross stage partial with 2 convolutions-fusion (C2f) blocks to overcom...

work page internal anchor Pith review Pith/arXiv arXiv 2026

[3] [3]

The dataset includes videos captured across 14 scenes, covering both indoor scenes and outdoor free-range scenes

EXPERIMENTS M-SCOPE Dataset.We introduce the M-SCOPE (Multiscale Clut- tered & Occluded Poultry Environment) dataset, designed for object detection in cluttered free-range environments, with annotations for the “chicken” class. The dataset includes videos captured across 14 scenes, covering both indoor scenes and outdoor free-range scenes. Outdoor scenes ...

work page 2025

[4] [4]

The model fuses local details with global context to significantly enhance object perception amid challenges like scale variation, occlusion, and cluttered backgrounds

CONCLUSION We introduce SFN-YOLO, an efficient real-time detector for poul- try. The model fuses local details with global context to significantly enhance object perception amid challenges like scale variation, occlusion, and cluttered backgrounds. Evaluated on our novel M- SCOPE dataset, SFN-YOLO achieves state-of-the-art performance and strong generali...

work page

[5] [5]

A review on development of an IoT-based system for enhancing yield in poultry farming,

J. G. Ambafi, Y . H. Muhammad, A. S. Ahmad, H. O. Ohize, U. S. Dauda, and J. J. Musa, “A review on development of an IoT-based system for enhancing yield in poultry farming,” Confluence University Journal of Science and Technology, vol. 2, no. 1, pp. 116–125, 2025

work page 2025

[6] [6]

Automation and smart poul- try farm management: A review,

Priya, S.K. Bhanja, and V . Sahu, “Automation and smart poul- try farm management: A review,”Indian Journal of Poultry Science, vol. 56, no. 2, pp. 89–99, 2021

work page 2021

[7] [7]

Faster R-CNN: Towards real-time object detection with region proposal net- works,

S. Ren, K. He, R. Girshick, and J. Sun, “Faster R-CNN: Towards real-time object detection with region proposal net- works,”Transactions on Pattern Analysis and Machine Intelli- gence, vol. 39, no. 6, pp. 1137–1149, 2017

work page 2017

[8] [8]

Focal loss for dense object detection,

T.-Y . Lin, P. Goyal, R. Girshick, K. He, and P. Doll´ar, “Focal loss for dense object detection,” inProceedings of IEEE In- ternational Conference on Computer Vision (ICCV), 2017, pp. 2999–3007

work page 2017

[9] [9]

End-to-end object detection with transform- ers,

N. Carion, F. Massa, G. Synnaeve, N. Usunier, A. Kirillov, and S. Zagoruyko, “End-to-end object detection with transform- ers,” inProceedings of European Conference on Computer Vision, 2020, pp. 213–229

work page 2020

[10] [10]

DETRs beat YOLOs on real-time object detection,

Y . Zhao, W. Lv, S. Xu, J. Wei, G. Wang, Q. Dang, Y . Liu, and J. Chen, “DETRs beat YOLOs on real-time object detection,” inProceedings of the IEEE/CVF conference on computer vi- sion and pattern recognition, 2024, pp. 16965–16974

work page 2024

[11] [11]

Ultralytics YOLOv5,

G. Jocher, “Ultralytics YOLOv5,” 2020

work page 2020

[12] [12]

Ultralytics YOLOv8,

G. Jocher, A. Chaurasia, and J. Qiu, “Ultralytics YOLOv8,” 2023

work page 2023

[13] [13]

Ultralytics YOLO11,

G. Jocher and J. Qiu, “Ultralytics YOLO11,” 2024

work page 2024

[14] [14]

Mamba YOLO: A simple baseline for object detection with state space model,

Z. Wang, C. Li, H. Xu, X. Zhu, and H. Li, “Mamba YOLO: A simple baseline for object detection with state space model,” in Proceedings of the AAAI Conference on Artificial Intelligence, 2025, vol. 39, pp. 8205–8213

work page 2025

[15] [15]

CAF-YOLO: A robust framework for multi-scale lesion detection in biomedical imagery,

Z. Chen and S. Lu, “CAF-YOLO: A robust framework for multi-scale lesion detection in biomedical imagery,” in Proceedings of IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2025, pp. 1–5

work page 2025

[16] [16]

FBRT-YOLO: Faster and better for real-time aerial image detection,

Y . Xiao, T. Xu, Y . Xin, and J. Li, “FBRT-YOLO: Faster and better for real-time aerial image detection,” inProceedings of the AAAI Conference on Artificial Intelligence, 2025, vol. 39, pp. 8673–8681

work page 2025

[17] [17]

YOLOX: Exceeding YOLO series in 2021,

Z. Ge, S. Liu, F. Wang, Z. Li, and J. Sun, “YOLOX: Exceeding YOLO series in 2021,” 2021

work page 2021

[18] [18]

An efficient transformer network for detecting multi- scale chicken in complex free-range farming environments via improved RT-DETR,

X. Li, M. Cai, X. Tan, C. Yin, W. Chen, Z. Liu, J. Wen, and Y . Han, “An efficient transformer network for detecting multi- scale chicken in complex free-range farming environments via improved RT-DETR,”Computers and Electronics in Agricul- ture, vol. 224, pp. 109160, 2024

work page 2024

[19] [19]

Chicktrack – a quantitative tracking tool for measuring chicken activity,

S. Neethirajan, “Chicktrack – a quantitative tracking tool for measuring chicken activity,”Measurement, vol. 191, pp. 110819, 2022

work page 2022

[20] [20]

Only detect broilers once (ODBO): A method for monitoring and tracking individual behavior of cage-free broilers,

C. Yin, X. Tan, X. Li, M. Cai, and W. Chen, “Only detect broilers once (ODBO): A method for monitoring and tracking individual behavior of cage-free broilers,”Agriculture, vol. 15, no. 7, 2025

work page 2025

[21] [21]

A survey of open-access datasets for computer vision in precision poultry farming computer vision datasets for poul- try,

G. Li, “A survey of open-access datasets for computer vision in precision poultry farming computer vision datasets for poul- try,”Poultry Science, vol. 104, no. 2, pp. 104784, 2025

work page 2025

[22] [22]

A review of machine learn- ing models for disease prediction in poultry chickens,

D. Verma, N. Goel, and V .K. Garg, “A review of machine learn- ing models for disease prediction in poultry chickens,” inPro- ceedings of International Conference on Paradigms of Com- munication, Computing and Data Analytics, 2023, pp. 723– 737

work page 2023

[23] [23]

Real-time behavior detection and judgment of egg breeders based on YOLO v3,

J. Wang, N. Wang, L. Li, and Z. Ren, “Real-time behavior detection and judgment of egg breeders based on YOLO v3,” Neural Computing and Applications, vol. 32, no. 10, pp. 5471– 5481, 2020

work page 2020

[24] [24]

Detection of sick broilers by digital image processing and deep learning,

X. Zhuang and T. Zhang, “Detection of sick broilers by digital image processing and deep learning,”Biosystems Engineering, vol. 179, pp. 106–116, 2019

work page 2019

[25] [25]

A deep learning model for detecting cage-free hens on the litter floor,

X. Yang, L. Chai, R.B. Bist, S. Subedi, and Z. Wu, “A deep learning model for detecting cage-free hens on the litter floor,” Animals, vol. 12, no. 15, pp. 1983, 2022

work page 1983

[26] [26]

Squeeze-and-excitation net- works,

J. Hu, L. Shen, and G. Sun, “Squeeze-and-excitation net- works,” inProceedings of IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2018, pp. 7132–7141

work page 2018

[27] [27]

T-rex label: Online ai data labeling plat- form,

T-Rex Label Team, “T-rex label: Online ai data labeling plat- form,” 2025, Accessed: 2025-08-26

work page 2025

[28] [28]

The pascal visual object classes challenge 2012 (voc2012) development kit,

M. Everingham and J. Winn, “The pascal visual object classes challenge 2012 (voc2012) development kit,”Pattern Analysis, Statistical Modelling and Computational Learning, Tech. Rep, vol. 8, no. 5, pp. 2–5, 2011. 5

work page 2012