Computer vision models for precision poultry farming: A narrative review of behavioral and welfare monitoring studies.

Computer vision models for precision poultry farming: A narrative review of behavioral and welfare monitoring studies - PMC Skip to main content An official website of the United States government Here's how you know Here's how you know Official websites use .gov A .gov website belongs to an official government organization in the United States. Secure .gov websites use HTTPS A lock ( Lock Locked padlock icon ) or https:// means you've safely connected to the .gov website. Share sensitive information only on official, secure websites. Search Log in Dashboard Publications Account settings Log out Search… Search NCBI Primary site navigation Search Logged in as: Dashboard Publications Account settings Log in Search PMC Full-Text Archive Search in PMC Journal List User Guide PERMALINK Copy As a library, NLM provides access to scientific literature. Inclusion in an NLM database does not imply endorsement of, or agreement with, the contents by NLM or the National Institutes of Health. Learn more: PMC Disclaimer | PMC Copyright Notice Poult Sci . 2026 Mar 31;105(7):106887. doi: 10.1016/j.psj.2026.106887 Search in PMC Search in PubMed View in NLM Catalog Add to search Computer vision models for precision poultry farming: A narrative review of behavioral and welfare monitoring studies Bidur Paneru Bidur Paneru a Department of Poultry Science, College of Agricultural & Environmental Sciences, University of Georgia, Athens, GA 30602, USA Find articles by Bidur Paneru a , Anjan Dhungana Anjan Dhungana a Department of Poultry Science, College of Agricultural & Environmental Sciences, University of Georgia, Athens, GA 30602, USA Find articles by Anjan Dhungana a , Samin Dahal Samin Dahal a Department of Poultry Science, College of Agricultural & Environmental Sciences, University of Georgia, Athens, GA 30602, USA Find articles by Samin Dahal a , Casey W Ritz Casey W Ritz a Department of Poultry Science, College of Agricultural & Environmental Sciences, University of Georgia, Athens, GA 30602, USA Find articles by Casey W Ritz a , Woo Kim Woo Kim a Department of Poultry Science, College of Agricultural & Environmental Sciences, University of Georgia, Athens, GA 30602, USA Find articles by Woo Kim a , Tianming Liu Tianming Liu b School of Computing, University of Georgia, Athens, GA 30602, USA Find articles by Tianming Liu b , Lilong Chai Lilong Chai a Department of Poultry Science, College of Agricultural & Environmental Sciences, University of Georgia, Athens, GA 30602, USA Find articles by Lilong Chai a, ⁎ Author information Article notes Copyright and License information a Department of Poultry Science, College of Agricultural & Environmental Sciences, University of Georgia, Athens, GA 30602, USA b School of Computing, University of Georgia, Athens, GA 30602, USA ⁎ Corresponding author. [email protected] Received 2026 Jan 2; Accepted 2026 Mar 30; Collection date 2026 Jul. © 2026 The Authors This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). PMC Copyright notice PMCID: PMC13091050  PMID: 41955872 Abstract This narrative review with structured literature screening combines comprehensive research on the rapid adoption of object detection computer vision models, particularly “You Only Look Once” (YOLO), used alone or in conjunction with other machine learning models, to advance Precision Poultry Farming (PPF), which refers to the application of data-driven and automated technologies to monitor, manage, and optimize poultry health, welfare, and production efficiency. A literature search across search engines, such as Google Scholar, was used because of its broad interdisciplinary coverage, allowing retrieval of literature spanning animal science, computer vision, and agricultural engineering, which are often indexed across different publications venues, on October 15 2024, which revealed 408 results when searching with search expression “YOLO + broilers + layers” and publications dated from 2015 to October 15, 2024. We removed 200 articles during screening, and 126 articles were excluded after eligibility evaluation, resulting in 82 eligible research papers to be included for this review. The YOLO object detection models have evolved from YOLOv1 to YOLO11 by 2024, progressively improving in model performance, speed, accuracy, and robustness through the refinement of key architectural components, including backbone networks, detection heads, and loss functions. This review highlights how YOLO models have been applied to broiler chickens and laying hens across diverse housing systems to support key tasks such as identification, behavior detection, counting, tracking, health and disease monitoring, flock distribution pattern, and calculating activity index, often in combination with other machine vision models. The analysis shows that it took 4 years to apply YOLO models for the object detection task in poultry since the release of the first version of the YOLO model in 2015. The application of YOLO models in poultry from 2019 to 2021 was very slow and sporadic while it took rapid growth in publications since 2021, led primarily by research groups in China and the USA, and mainly concentrated in journals such as Computers and Electronics in Agriculture (10), Institute of Electrical and Electronics Engineers (IEEE) Conference (10), Poultry Science (9), Animals (6), and AgriEngineering (5). Major opportunities and challenges are identified around deploying these models for reliable, real-time decision support on commercial farms, particularly for animal welfare assessment, disease and wild bird detection, and integration with complementary sensing and analytics frameworks. Keywords: Animal welfare, Deep learning, Poultry production, Precision poultry farming, Object detection Introduction Precision livestock farming (PLF) is defined as the management of individual animals by continuous, automated, and real-time monitoring of health, welfare, production/reproduction, and environmental impact ( Berckmans, 2017 ; Tullo et al., 2017 ; Riber et al., 2018 ). It encompasses a wide range of engineering principles, techniques, applications, and tools for data acquisition, pre-processing, modeling, post-processing, and analysis. Non-invasive techniques such as using wearable sensors, accelerometers, and radio frequency identification systems (RFID) are common in poultry and computer vision is gaining popularity these days ( Rowe et al., 2019 ; Li et al., 2020 ). The goal of PLF is to provide real-time, automatic, continuous monitoring of animal health and welfare, providing producers with early warning signs that have potential production implications. For the improvement of housing design and optimum resource allocation, monitoring the use of resources, chickens, and extracting information on how chickens use those resources is very important ( Ringgenberg et al., 2015 ). The use of wearable sensors, for instance, provides valuable information about the health conditions of the birds. Classification of specific broiler behavior, such as sitting, standing, feeding, drinking, walking, and preening, can be accomplished by recording 3D accelerometer data and analyzing acceleration data using machine learning ( Yang et al., 2021 ; Sakib et al., 2024 ). It can be used to determine the inactivity or death status of broiler chickens ( Dawson et al., 2007 ), activity, and lameness in conventional and slow-growing broiler breeds ( Pearce et al., 2023 ). However, these devices must be in contact with birds to collect that data, which may also affect birds’ movement patterns and behavior if the mounted sensors weigh more than 5% of the chicken’s body weight. Previous studies have highlighted that, presence of wearable devices with a weight under 5% of the chicken’s body weight do not affect the chicken’s behavior, leg health, cleanliness, and production in the long term, although there is usually an adaptation period of 1 to 2 days after the device is attached to the birds ( Anderson et al., 2025; Khan et al., 2025 ). However, the cost of individual sensors for each birds would be higher than cheaper alternatives, such as using a single camera to record videos and identify behaviors and movement patterns of multiple birds. Therefore, computer vision techniques have become increasingly common in the poultry industry to get that information, which is an economically viable option. Such techniques involve cameras to record videos of the birds and analyze production, behavior, and welfare indicators by analyzing captured videos and images. One of the effective and popular methods being used to detect objects in images of poultry in recent years involves the use of the “You Only Look Once” (YOLO) model. YOLO family of models is a widely used real-time object detection framework known for its speed and accuracy. Unlike traditional object detection methods, which employ region proposal networks or sliding window approaches, YOLO treats object detection as a single regression problem, directly predicting bounding boxes and class probabilities from an image in a single pass through the neural network ( Redmon et al., 2016 ). An extensive review of literature search across search engine such as Google Scholar was done on October 15, 2024, which revealed 408 results when searched with the search expression “YOLO + broilers + layers” and publications dated from 2015 to 2024. We removed 200 articles during screening, and 126 articles were excluded after eligibility evaluation, resulting in 82 eligible research papers to be included for this review in the updated manuscript. Using the “+” function in the search bar of Google Scholar helped refine results to papers that use poultry as an experimental animals and YOLO as an object detection computer vision model. While google scholar enabled identification of a wide range of relevant studies, we recognize that not systematically querying specialized databases such as Scopus, Web of Science, IEEE Xplore, or PubMed may have resulted in the omission of some relevant publications. The search strategy intentionally included highly targeted keywords such as “YOLO”, “broilers”, and “layers” to focus on studies specifically applying YOLO-based computer vision models within poultry production systems. However, we acknowledge that this approach may have excluded studies that applied YOLO models to poultry-related tasks without explicitly including these terms in titles or abstracts. To mitigate this potential bias, the screening process included manual review of reference lists from relevant articles, which helped identify additional studies that might not have been captured through keyword search alone. Nevertheless, we recognize that the selected keywords may have introduced a degree of selection bias toward explicitly labeled YOLO-based poultry studies. However, the resulting search results also showed research papers irrelevant to our interest, such as research papers focusing on plant science, livestock species other than broilers and layers (such as cattle, quail, goose, ducks, turkey, sheep, pig, goats, and heifers), repeated articles, book chapters, dissertations, review papers, and conference papers. For our review purposes, we manually filtered out the irrelevant papers that were out of the scope of this review. This included the removal of repeated articles, book chapters, dissertations, review papers, published in languages other than English, livestock species other than broilers and layers, and those papers for which the full text was not available. We also removed the conference paper whenever a full-text research article was published from that conference paper. However, based on the publication trend of the work, we found that most of the object detection task was published in the Institute of Electrical and Electronics Engineers (IEEE) conferences, and we kept that conference papers as these papers are relevant to our review topic. While most non-peer-reviewed conference abstracts were excluded, IEE conference proceedings were retained when they presented full-peer-reviewed technical papers, which is common in computer vision and engineering research. In fields such as artificial and computer vision, conference publications often represent the primary dissemination channel for novel methods, particularly for architectures such as YOLO. Therefore, including selected IEEE conference papers allowed the review to capture important methodological developments relevant to poultry monitoring applications. After applying all the selection criteria and reviewing the literature search results, we included a summary of 82 eligible research papers that met the selection criteria for this review. History and evolution of YOLO models (YOLOv1 to YOLO11) until 2024 The “You Only Look Once” (YOLO) object detection algorithm was first developed by Redmon et al. (2016) in 2015, which has contributed significantly to the development of real-time object detection by combining region proposal and classification into a single neural network, reducing significant computation time. YOLOv1, was introduced in 2016 by Redmon et al. (2016) and offered a significant speed advantage over previous methods, such as Faster Region-based Convolutional Neural Network (Faster R-CNN), a two-stage object detection model. Since then, multiple versions have been developed to improve accuracy, generalization, and computational efficiency. YOLOv2 and YOLOv3 were introduced in 2017 and 2018 by Redmon and Farhadi, (2017) ; 2018) with anchor boxes, multi-scale predictions, and darknet-based architecture. YOLOv2 introduced Darknet-19 as the backbone architecture, enhancing feature extraction and refining anchor boxes through K-means clustering. YOLOv3 utilized a deeper network, Darknet-53 architecture, integrating multi-scale detection with residual connections. It introduced multi-scale detection using a Feature Pyramid Network (FPN). It balanced accuracy and speed better than earlier versions. YOLOv4 was introduced by Bochkovskiy et al. (2020) and integrated advanced techniques, including Cross Stage Partial Darknet-53 (CPSDarknet-53), Path Aggregation Network (PANet), and Mish activation. It enhanced data augmentation using Mosaic augmentation and self-adversarial training. YOLOv5 was developed by Ultralytics in 2020 ( Jocher, Glenn, 2020 ), diverging from the original YOLO series. It improved ease of use, training speed, and deployment with PyTorch. It introduced four model sizes: small (S), medium (M), large (L), and extra-large (XL), for different applications. YOLOv6, YOLOv7, and YOLOv8 were introduced in 2022 and 2023 ( Wang et al., 2022 ; 2023 ). YOLOv6 was introduced in 2022, focused on efficiency and edge-device compatibility. It introduced new architectural optimizations for better trade-offs in speed and accuracy. YOLOv7 was also introduced in 2022 and claimed to be the fastest and most efficient object detection model at the time of introduction. It introduced model scaling techniques to improve performance. YOLOv8 was introduced in 2023 and added support for instance segmentation along with detection. It was customized for real-world applications, integrating multiple learning paradigms. YOLOv9 was introduced in 2023–2024, and further refinements in network structure led to improved accuracy and were also designed to be more efficient for large-scale datasets. YOLOv10 was introduced in 2024 and enhanced real-time detection capabilities with better handling of occlusions and complex scenes. It focused on improving robustness in various environments. YOLO11 was developed by Ultralytics and officially released on September 30, 2024, at the company’s annual YOLO Vision 2024 (YV24) event. YOLO11 introduced architectural enhancements, building on earlier versions. The C3 block with k = 2 convolution group (C3k2) block replaces the CSP with a 2-layer Feature-Fusion (C2f) block from YOLOv8, offering a more computationally efficient CSP block with smaller convolutions for faster processing. The C2 block with Pyramid Spatial Attention (C2PSA) block was introduced, incorporating spatial attention to enhance detection accuracy, particularly for smaller objects. These changes improved feature extraction for more precise object detection. Overall, key features of YOLO Models include (1) single-stage detection, unlike two-stage detectors like Faster R-CNN, YOLO predicts object locations and classes in a single forward pass, achieving real-time inference, (2) Grid-based approach, (YOLO divides an image into an S × S grid, where each grid cell predicts bounding boxes and confidence scores, (3) Anchor boxes, Later versions of YOLO, YOLOv2, and after that use anchor boxes to improve detection accuracy for varying object sizes, (4) tradeoff between speed and accuracy, newer versions of YOLO balance computational efficiency and precision, making them suitable for applications like autonomous driving, surveillance, and animal behavior analysis. The evolution of YOLO models, along with a summary of key innovations and technical milestones in YOLO's development from YOLOv1 to YOLO11, is depicted in Fig. 1 and Table 1 below. Fig. 1. Open in a new tab The evolution of YOLO models with changes in Architecture, Optimization, Efficiency, and a Unified framework across years. Table 1. Summary of Key Innovations and Technical Milestones in YOLO Evolution from YOLOv1 to YOLOv11. YOLO Version Year Key Innovations Benefits / Impact YOLOv1 2015 Introduced single-stage end-to-end object detection network (Darknet). Unified bounding box regression and classification in one CNN Foundation of real-time detection; high speed compared to multi-stage detectors YOLOv2 2016 Added anchor boxes, dimension clustering, batch normalization, passthrough layer, multi-scale training, and Darknet-19 backbone. Combined classification and detection datasets Improved localization, stability, and small-object detection; scalable to 9000+ classes YOLOv3 2018 Introduced Feature Pyramid Network (FPN) for multi-scale detection; Darknet-53 backbone with residuals; sigmoid abjectness; multi-label prediction More correct detection across object sizes; better balance between recall and precision YOLOv4 2020 Implemented CSPDarknet-53, Weighted Residual Connections (WRC), Cross-Stage Partial Networks (CSP), Mish activation, DropBlock regularization, and Mosaic data augmentation Boosted mAP while keeping real-time inference; improved gradient flow and regularization YOLOv5 2020 Fully in PyTorch; introduced auto-anchor, adaptive image scaling, Mosaic & MixUp augmentation, and Focus module. Multiple model sizes (s, m, l, x) Simplified training/deployment; modular and highly adaptable; edge-device compatible YOLOX 2021 Converted YOLO to an anchor-free detector; added decoupled head, SimOTA label assignment, and EMA model averaging Enhanced convergence stability; reduced anchor computation overhead; improved speed YOLOv6 2022 Industrial-grade redesign with EfficientRep backbone, RepOptimizer, RepVGG block, and SimOTA; improved for edge devices Superior real-time inference on embedded hardware; simplified architecture for deployment YOLOv7 2022 Introduced E-ELAN, model re-parametrization, and planned gradient path aggregation; optimized training of concatenated models State-of-the-art accuracy–speed trade-off; improved feature utilization YOLOv8 2023 Unified framework supporting detection, segmentation, pose estimation, and classification; introduced anchor-free detection and new NMS Highly versatile across multiple CV tasks; plug-and-play usability YOLOv9 2024 Introduced Programmable Gradient Information (PGI) and Generalized Efficient Layer Aggregation Network (GELAN) Significantly improved gradient flow and feature aggregation; enhanced mAP without latency penalty YOLOv10 2024 Improved for end-to-end efficiency; replaced NMS with single-label assignment; focused on speed–accuracy balance Ideal for industrial real-time deployment with minimal computation cost YOLOv11 2024 Expected enhancements: Transformer-CNN hybrid backbone, Context-aware attention modules, and lightweight vision transformer fusion Further improvements in small-object and occluded-object detection; superior contextual awareness Open in a new tab Training, validation, testing, and performance evaluations of YOLO models To facilitate understanding among poultry and animal scientists, this section provides a brief and intuitive explanation of how YOLO-based object detection models are trained, validated, and tested, as well as how model performance is evaluated. YOLO is a deep-learning object detection framework that simultaneously identifies what an object is and where it is located in an image. In poultry research, YOLO is commonly used to detect birds, body parts, or specific behaviors such as feeding, drinking, perching, and dustbathing from images or videos. Model development generally involves three sequential datasets i.e., training, validation, and testing. Annotated images are required for the training and validation process, in which each object of interest (e.g., chicken) is manually or semi-automatically labeled with bounding boxes and corresponding class names. During the model training, the YOLO model learns to recognize poultry-related objects by iteratively adjusting millions of internal parameters (weights) in a neural network. Each training image includes the class labels (e.g., chicken, feeder, drinker, etc.) and the bounding box location (x, y position, width, and height). The model predicts bounding boxes and class probabilities, which are compared against the annotated ground truth. Errors are quantified using a loss function, which typically includes localization loss (how accurately the box fits the object), classification loss (whether the object is correctly classified), and confidence loss (whether an object is present). These losses are minimized using gradient-based optimization over multiple training cycles, known as epochs. The validation dataset is used during training to monitor how well the model generalizes to unseen data. Validation serves three main purposes, detect overfitting (model memorizes training data but performs poorly on new data), optimize hyperparameters (learning rate, confidence threshold), and to select the best-performing model checkpoint. Performance metrics reported during the validation guide decisions on when to stop training. Testing datasets is used only after training is complete and represent fully unseen data. This dataset provides an unbiased estimate of real-world performance, such as detecting hens or behaviors. YOLO performance is quantified using standard object detection metrics, which are briefly explained below in biologically intuitive terms. Precision measures the fraction of detected objects that are correct. High precision indicates that most detected chickens or behaviors are real, while lower precision suggests many false detections. For example, if the model detects 100 chickens and 87 are chickens, the precision of the model is 0.87. P r e c i s i o n = T r u e P o s i t u v e T r u e p o s i t i v e P + F a l s e P o s i t i v e Recall measures the number of actual objects successfully detected. High recall indicates most chickens present in the image are detected, while low recall means many chickens are missed in detection. If 100 chickens are present in the scene but only 80 chickens are detected, recall is 0.80. R e c a l l = T r u e P o s i t i v e T r u e P o s i t i v e + F a l s e N e g a t i v e Average precision summarizes the trade- off between precision and recall across different detection confidence thresholds. It reflects how well the model balances object detection with false detection rates. mAP@50 means a detection is considered correct if the predicted bounding box overlaps the ground-truth box by at least 50%, measured using intersection over union (IoU). F1-score is a harmonic mean between precision and recall. It provides a single metric that balances precision and recall. Importance of behavior analysis and welfare monitoring in poultry Poultry welfare is a critical concern in modern animal agriculture, affecting productivity, ethical concerns, public concerns, and perceptions of consumers on how animals are being raised. The topic of poultry welfare has a long history and controversy from the domestication of chickens until today. A good state of welfare includes birds having a healthy body, a positive affective state, which means chickens experiencing good feelings or emotions, not just the absence of stress or illness, and the opportunity to express natural behaviors ( Sih et al., 2004 ). Behavior analysis and welfare monitoring offer essential insights into poultry health, stress levels, and environmental adaptability. Recent advances in sensor technology, machine learning, and real-time monitoring have enabled more effective and objective welfare assessment ( Neethirajan and Kemp, 2021 ; Manikandan and Neethirajan, 2025 ) However, there are various guides for the inclusive assessment of animal welfare, such as behavioral indicators, physiological stress, and proxies of productivity. Many of them interact with each other, making evaluation difficult and time-consuming, especially at the commercial level. Among these welfare indices, behavior is the most easily understood and commonly used welfare assessment index ( Welfare Quality®, 2009 ; De Jong et al., 2016 ), because behaviors that may be influenced by living conditions, and diseases of animals can reflect physiological conditions. Poultry behavior serves as a primary indicator of health and welfare. Normal behaviors such as foraging, perching, and dustbathing are strongly motivated and associated with positive welfare, while abnormal behaviors such as feather pecking and aggression can be indicative of welfare problems. Exploratory behavior is also an important motivation in chickens ( Newberry, 1999 ; Janczak and Riber, 2015 ; Campbell et al., 2019 ). Stress-related behaviors, including excessive preening and panting, can result from inadequate housing conditions, high stocking densities, or poor environmental enrichment ( Estevez et al., 2007 ). Therefore, monitoring animal behaviors and formulating feeding, drinking, and management measures accordingly can not only promote the growth of animals but also help to improve the production performance and protect animal welfare. Welfare monitoring is a vital component of modern farming that ensures proper poultry health and productivity. For example, birds expressing chronic stress show compromised immune function, reduced growth rates, and lower egg production ( Dawkins, 2017 ). By analyzing behavior patterns, farmers can implement proactive management strategies to mitigate welfare issues, optimize feeding strategies, improve environmental conditions, and intervene when necessary. Traditional welfare assessment relies on manual observations, which can be labor-intensive, erroneous, and subject to cognitive bias. Recent advancements, such as wearable sensors, automated video monitoring, and deep learning-based models, a type of artificial intelligence model that learns patterns from data using neural networks with many layers, for image and video processing, eliminate issues such as manual observation and assessment, which include non-invasive methods, and furthermore could detect issues that the human eye couldn’t detect within frames of videos. Object detection is a critical part of computer vision systems, which allows automated systems to identify and localize objects of interest within images or video frames. The output of an object detector is a set of bounding boxes that enclose the objects in the image, along with class labels and confidence scores for each box. It is important to note that many computer vision studies on poultry behavior detection rely on models trained using isolated images rather than continuous video sequences. I/n these approaches, behaviors are typically inferred from static body posture or spatial relationships within a single frame, such as defining drinking based on the proximity of a bird’s beak to a drinker or inferring pecking from beak proximity to another bird. While such approaches can provide useful indicators of behaviors, they differ conceptually from traditional ethological methods, which characterize behavior based on temporal sequences of movement and interaction over time. Therefore, in many studies reviewed here, the term “behavior detection “more precisely refers to posture-based inference from individual frames rather than dynamic behavioral observation. Bibliographic information of the articles included in this review The detailed bibliographic information of the articles included in this review is presented below ( Fig. 2 , Fig. 3 , Fig. 4 , Fig. 5 ). Country-wise analysis reveals that more than 50% of the publications came from China and the USA, while the remaining publications came from the other 17 countries. A small number of countries account for about half of the publications, suggesting that research activity is geographically distributed but concentrated almost equally between China and the USA, with China leading the publications ( Fig. 2 ). Fig. 2. Open in a new tab Analysis of publication count of research articles using YOLO-based models in poultry by country. Fig. 3. Open in a new tab Analysis of Top journals of research articles using YOLO-based models in poultry by publication count (≥2 publications). Fig. 4. Open in a new tab Analysis of publication count of research articles using YOLO-based models in poultry by category by yearly distribution showing the trend line over the years (2015–2024). Fig. 5. Open in a new tab Analysis of the word cloud of research titles included in this review (83 research articles). Words that appear more often in the research titles are displayed in larger fonts. Journal-wise results show that a limited number of journals serve as the primary outlets for this research area. The bar graph below indicates clear publishing preferences, suggesting where researchers most often disseminate their work. Top 5 journal articles publishing YOLO-based models used in poultry include Computer and Electronics in Agriculture, Poultry Science, IEEE Conference, Animals, and AgriEngineering ( Fig. 3 ). The distribution of research topics across publication years ( Fig. 4 ) highlights a clear and accelerating growth in the use of computer vision-based methods, particularly YOLO architecture, for poultry behavior and welfare assessment. Fig. 4 shows the publication trend among research categories and publication year. For the purpose of this review, the research category includes poultry behavior detection, wild birds, and pathogen detection, identification and counting, integrating YOLO models with other computer vision models, and others. For the purpose of this review, wild birds refer to non-domesticated avian species that come in contact with poultry facilities and potentially act as carriers of infectious agents. Pathogens refer to disease-causing microorganisms affecting poultry, including viruses, bacteria, and parasites, reported in the reviewed studies. A particularly notable trend appears from 2023 onward, where the number of publications rises sharply across nearly all categories. Research on behavior detection, such as dustbathing, feeding, aggression, and activity monitoring, became especially prominent, mirroring the industry’s priority to quantify welfare-relevant behaviors in Cage-Free (CF) systems. At the same time, identification/counting/tracking studies expanded dramatically, driven by the adoption of advanced tracking frameworks (e.g., DeepSORT, ByteTrack) and increasing emphasis on monitoring individual hens within large flocks. The emergence of YOLO-based integration studies in 2023 further underscores the field’s rapid technical progress, as researchers began combining YOLO models with additional modalities and pipelines to enable more complex, multi-stage analyses. By 2024, the field reached its peak activity in the dataset, with sustained growth across all categories. We only included the papers published until October 2024 and therefore could not reflect the true growth of papers in that year. However, we are aware of the fact that the growth in this sector is rapidly increasing, and more published papers can be expected in the coming days. This period is marked by the highest number of publications in identification/tracking and disease detection, emphasizing a transition toward more robust, automated systems capable of supporting Precision Livestock Farming (PLF). Although we did not include articles from 2025, the ongoing presence of behavior-focused work suggests the field remains active and continues to evolve. Overall, the temporal distribution illustrates a maturing research landscape, moving from foundational model testing to diversified, application-oriented systems that integrate detection, tracking, and behavior recognition. This progression reflects a broader shift toward scalable, automated approaches for monitoring poultry health and welfare in commercial poultry farming. The analysis shows that it took 4 years to apply YOLO models for the object detection task in poultry since the release of the first version of the YOLO model in 2015. The application of YOLO models in poultry from 2019 to 2021 was very slow and sporadic, while it took rapid growth in publications since 202. Early work between 2019 and 2021 was sparse and largely exploratory, with only isolated studies focusing on general identification or miscellaneous applications. Beginning in 2022, however, the field experienced a measurable expansion, with an increase in publications addressing behavior detection, disease and pathogen monitoring, and bird identification/counting/tracking. This shift reflects both improved accessibility of deep-learning tools and growing interest in data-driven animal welfare research ( Fig. 4 ). The number of publications started to accelerate from the year 2021 through 2024, and we believe that it is still accelerating as the adoption of YOLO-based models in poultry is increasing day by day. The word cloud of research titles emphasizes recurring concepts and key themes, helping to visually capture the most common research focuses on literature. Words that appear more often in the research titles are displayed in larger fonts ( Fig. 5 ). Application of YOLO models in poultry research Behavior detection YOLO models have been used extensively in various real-life scenarios, such as vehicle counting, traffic light recognition, and fruit counting on conveyor belts. However, the application of YOLO models in poultry has mainly focused on recognizing and detecting different behaviors, as well as counting the frequency of these behaviors. This section of the review summarizes the use of earlier versions of the YOLO models, such as YOLOv3, YOLOv4, and YOLOv5, for behavior detection in poultry. Results from numerous studies following the release of YOLOv1 show that YOLO models performed better than pre-existing deep learning models. A study demonstrated that the YOLOv3 model can accurately detect and recognize six behaviors of egg breeders with a mean average precision (mAP) of 0.9209, achieving high recognition rates for mating (0.9472), standing (0.9457), and feeding (0.9310), among others ( Wang et al., 2020 ). Additionally, the study highlighted the use of mating frequency as a welfare indicator and provides a method for judging abnormal behaviors based on the number of fights and overall activity levels ( Wang et al., 2020 ). Another study compared four deep learning methods, such as EfficientNet-YOLOv3, YOLOv4-Tiny, YOLOv5, and Faster-RCNN, for real-time and accurate recognition of four behaviors of laying hens: standing, lying, feeding, and grooming. YOLOv5 emerged as the optimal algorithm, achieving high accuracy, mAP values of 0.9724 to 0.9861, and faster detection speed (55.55 frames per second, FPS), making it suitable for real-time behavior detection in poultry farming ( Wang et al., 2023 ). Researchers have used YOLO models in automatic detection of feeding and aggressive behavior detection, and some researchers have proposed an improved YOLOv5 for detecting the health status of the chicken. A study that utilized the YOLOv4 algorithm to automatically detect feeding and aggressive behaviors in broiler chickens achieved a high mAP value of 0.9998 for feeding behavior and 0.994 for aggressive behavior, showing its effectiveness for real-time monitoring and improving poultry welfare ( Wahjuni et al., 2024 ). The authors hope that this approach offers a promising solution for automated behavior detection in smart coop monitoring systems that provide notifications if any unexpected condition occurs in a chicken coop. Another study proposed an improved YOLOv5 model for detecting the health status of broilers, incorporating the lightweight network model for mobile (EMO) and Decoupled Detect module, where improved YOLOv5 achieved a precision of 0.9560, a recall rate of 0.9590, and an average mAP of 0.9780, outperforming the original YOLOv5, YOLOv7, and YOLOv8 models in terms of accuracy and efficiency ( Cui et al., 2023 ), revealing the potential for real-time health monitoring in broilers. Behavior detection of CF laying hens Various studies have applied YOLO-based models in a CF facility to detect and monitor applied as well as problematic behaviors. This section of the review summarizes the use of YOLO models, mainly, YOLOv5, YOLOv6, YOLOv7, and YOLOv8, for the detection of natural/applied behavior (dustbathing, perching) and problematic behavior or health issues (feather pecking, mislaying, piling, bumble foot, footpad dermatitis, huddling, and inactivity) in the CF facility. Paneru et al. (2024a) demonstrated that the YOLOv8x-PB model outperformed other models in detecting perching behavior, achieving a precision of 0.9480, a recall of 0.9510, and [email protected] of 0.9760. The study highlighted the model’s effectiveness across different age groups of Hy-Line W-36, a genetically selected layer strain for high egg laying rate, efficient feed conversion, small body size, and white eggs, with detection precision ranging from 0.8880 for the starter phase to 0.9740 for the peaking phase, providing a valuable tool for monitoring perching behavior in CF housing systems. Another study by Paneru et al. (2024b) demonstrated that the YOLOv8x-DB model outperformed other models in detecting dustbathing behavior, achieving a precision of 0.9340, a recall of 0.9120, and [email protected] of 0.9370. The study highlights the model’s effectiveness across different growth phases, with the highest detection precision during the grower phase (0.9680), providing a valuable tool for monitoring dustbathing behavior in a CF housing system. Feather pecking, mislaying, and piling are significant behavioral and management problems in a CF facility, leading to physical injuries, stress, and reduced productivity among hens. Researchers have tried to address these issues by early detection of these issues with YOLO-based machine learning models. Subedi et al. (2023a) developed and evaluated two YOLOv5-based deep learning models, YOLOv5s-pecking and YOLOv5x-pecking, for tracking feather pecking (FP) behaviors in a CF laying hens. YOLOv5x-pecking outperformed YOLOv5s-pecking in precision (0.8830 vs. 0.8520), recall (0.6880 vs. 0.6020), and mAP (0.7870 vs. 0.7330). However, YOLOv5s-pecking was more efficient, using 75% less GPU memory and 80% less training time, making it a more practical choice for real-time monitoring. Bist et al. (2023a) developed a deep learning-based approach for detecting floor egg-laying behavior (FELB) in CF housing using YOLOv5 models, with YOLOv5m (medium) and YOLOv5x (extra-large) achieving the highest precision (0.9990), recall (0.9920), and F1-score (0.9960). While these models provided superior accuracy, YOLOv5s was identified as the most efficient for real-time applications due to its faster processing speed and lower computational requirements, making it a promising candidate for future integration into autonomous robotic monitoring systems. Bist et al. (2023b) utilized a novel YOLOv6 object detector to detect piling behavior of CF laying hens in a dataset of 9000 images. The YOLOv6l relu-PB model demonstrated superior performance with high average recall (0.7060), [email protected] (0.9890), and [email protected] (0.6370). These findings provide a foundation for developing automated tracking systems to improve poultry welfare in commercial CF systems while highlighting the importance of camera placement and environmental conditions in improving detection accuracy. Bumble foot and footpad dermatitis is another leg health issue in the CF facility, and researchers have addressed the issues using machine learning models to detect these issues in laying hens in the CF facility. Bist et al. (2024a) find out that the YOLOv5m-BFD model outperformed other models in detecting bumblefoot, achieving a precision of 0.9370, a recall of 0.8460, and [email protected] of 0.9090. The study highlights the model’s effectiveness under various settings, including different camera heights, epochs, and batch sizes, providing a valuable tool for early detection and improving animal welfare in a CF housing system. Footpad dermatitis is a serious leg health issue among CF laying hens, with around 40% of hens in CF flocks showing FPD in a large study, with significant variation between flocks. Bist et al. (2024b) developed an automated method for scoring poultry footpad dermatitis (FPD) using deep learning and thermal imaging, with the YOLOv8l (large) model achieving the highest accuracy ([email protected] of 0.9700) across different scoring levels. Results show that thermal imaging slightly improves FPD detection over RGB imaging, and the proposed technique offers a non-invasive approach to enhance automation and animal welfare in the egg industry. Another study by Ehsan and Mohtavipour (2024) presents a real-time framework for detecting abnormal behaviors, achieving a mAP of 0.9000 for chicken detection, and a high precision, recall, and F1 scores for huddling (0.9300, 0.8800, 0.8500) and inactivity detection (0.9200, 0.9000, 0.9100). The framework effectively finds and tracks broilers, offering valuable insights for prompt interventions to support chicken health and enhance productivity on poultry farms. In a study by Yang et al. (2022) YOLOv5x-hen model achieved over 0.9500 accuracy in real-time detection of hens under various lighting intensities, angles, and ages, but faced challenges with high flock density and occlusions from equipment. The study highlights the model’s potential for improving precision poultry farming by providing a technical basis for tracking individual birds and evaluating their behavior and welfare in a commercial CF house. Comfort/preference behavior detection Besides applied and problematic behavior detection, YOLO models have also been used for detecting comfort behaviors, as these behaviors indicate a positive state of the birds. Sozzi et al. (2022) studied the use of deep learning tools, specifically YOLOv4 and YOLOv4-tiny models, to measure comfort behaviors in laying hens, such as dustbathing, in an experimental aviary housing system. Although both models successfully identified hens on the floor with high precision (around 0.9400), they struggled to accurately classify dust-bathing hens, achieving precisions of 0.2820 and 0.3160 for YOLOv4-tiny and YOLOv4, respectively. The findings suggest that while machine learning can effectively check certain behaviors, additional tools, such as models like Long Short-Term Memory network (LSTM), handle sequential data well and can learn temporal patterns, while transformers can model long-range dependencies and often perform better on video-based behavior classification tasks. Either approach can be trained to recognize subtle or complex actions, such as dustbathing , grooming, pecking, etc., which may be needed to accurately assess more complex comfort behaviors like dustbathing. Kodaira et al. (2023) investigated the preference behavior of layer hens under different light colors (white, green, and red) and temperature conditions (cold, comfort, and heat stress) using a computer vision system. The study found that hens prefer white and red light under thermoneutral conditions but show no clear preference under heat stress. The detection model used in the study was based on the YOLOv4 architecture, achieving a mean average precision of 0.9990 and an accuracy of 0.9880. This section of the review focused on summarizing the use of YOLO-based models on the detection of comfort and preference behavior of the chicken and shows that there is a limited study on this topic within the poultry industry, and it needs further investigation to promote a positive affective state of the birds. The models employed for this task was mainly YOLOv4. Behavior detection of caged & free-range chicken Researchers have utilized YOLO models not only in detecting broiler behavior and behavior of CF laying hens but also to detect chickens in caged housing conditions. This section of the review summarizes the use of computer vision models to detect the behavior of caged-housed chickens and free-range chickens. The models utilized include YOLOv5, YOLOv6, YOLOv7, CAE, and VAE, which shows that researchers started to use other models, such as CAE and VAE, along with YOLO models. Liu et al. (2024) compared various object detection models for detecting caged chickens, with an improved YOLOv5s (small) model demonstrating superior performance. The improved-YOLOv5s achieved mAP of 0.9828, surpassing Single Shot MultiBox Detector (SSD), Faster R-CNN, YOLOv3, YOLOv4, YOLOv5s, and YOLO-X in accuracy, computational efficiency, and inference speed. Key improvements included modifications to the residual structure, the incorporation of a convolutional block attention module (CBAM), and the use of depthwise separable convolutions, significantly enhancing detection accuracy while reducing model complexity and computational costs. Xia et al. (2023) proposed an improved object detection algorithm for recognizing behaviors of caged white-feather broilers. The algorithm integrates a multi-scale detail feature fusion module and an object relationship inference module, enhancing detection accuracy in complex farming environments. Experimental results show high recognition accuracies for key behaviors such as feeding (0.9960), drinking (0.9960), moving (0.9920), and mouth opening (0.9830). The model used for this study was based on the YOLOv7 framework, with significant improvements in detection accuracy. Hao et al. (2022b) developed an improved Faster R-CNN model for detecting the feeding behavior of laying hens in stacked cages, achieving a precision of 0.9012, a recall of 0.7914, and an F1-score of 0.843. The model’s enhancements, including the Path Aggregation Network and IoU loss function, significantly improved detection accuracy and efficiency compared to baseline models. Use of YOLO models is not limited to caged housing; researchers are using YOLO models to detect multiple chickens in a free-range environment as well. Wu et al. (2023) proposed a Super-resolution Chicken Detection (SRCD) method that combines image super-resolution with YOLOX object detection to accurately detect multiple chickens in free-range environments. By enhancing low-resolution images using a generative adversarial network (GAN) and applying YOLOX, the method significantly improved detection performance, especially in challenging scenarios with occlusion and small object size, achieving up to 6.3% and 4.1% increase in AP50:90 for two different datasets. This approach demonstrates a solution for detecting tiny objects like chickens in free-range. Maybe future uses of this methodology could be integrating GAN with YOLO for detecting floor eggs with higher detection precision, as they cover small pixel areas. However, balancing detection precision with speed remains a key area for future improvement. Nguyen et al. (2023) employed a convolutional autoencoder (CAE) and a variational autoencoder (VAE) to detect anomalies in poultry behavior using video data. Both models were trained to reconstruct normal behavior, with deviations in reconstruction error indicating potential abnormal activities. The VAE outperformed the CAE in identifying anomalies, showing better generalization and clear separation between normal and abnormal behavior through lower reconstruction loss and more distinguishable error patterns. The study uses YOLOv6 as part of the behavior analysis pipeline. YOLOv6 was employed specifically for object detection to detect and localize individual hens in the video frames before feeding the data into CAE and VAE for anomaly detection. Detection of injured, or dead birds Identifying sick, injured, or dead birds is crucial for preventing the spread of diseases to the rest of the flock, protecting public health, and preventing problematic behavior such as aggressive feather pecking leading to cannibalism. Promptly removing sick birds helps to stop the spread of contagious disease, while removing dead or injured birds prevents other birds from pecking them. In addition, some avian diseases can be transmitted to humans, and early identification allows for proper public health measures and veterinary diagnostics. Therefore, researchers have utilized YOLO-based object detection models to find sick, injured, or dead birds automatically. Hao et al. (2022a) developed an autonomous inspection platform to detect dead broilers in large-scale breeding farms using an improved YOLOv3 model, achieving a mean mAP of 0.9860 and processing images at 0.007 FPS. Compared to other models like YOLOv3 and an SVM classifier, the improved YOLOv3 model showed higher precision, recall, and faster convergence, making it robust to different lighting conditions and broiler ages. Bist et al. (2023b) looked at the detection of dead hens in a CF housing system reported a higher mAP score of 0.9950, precision of 0.9840, and recall of 1.0 at a camera height of 0.5 m using YOLOv5s-MD model than ( Hao et al., 2022a ) study. The study highlighted the model’s usefulness under various CF housing conditions, including various levels of feather and litter coverage, providing a valuable tool for prompt mortality detection, and improving poultry health. Liu et al. (2021) presented a chicken removal system consisting of a robotic arm for Taiwanese poultry houses, integrating deep learning with the YOLOv4 algorithm for dead chicken detection. The system achieved a precision of 0.9524, accuracy of 0.9750, and recall of 1.0, outperforming YOLOv3 and Tiny YOLOv4 in terms of mAP at both 0.5 and 0.75 intersection over union (IoU) thresholds. The system effectively reduces human-poultry contact, enhancing biological safety and operational efficacy, where a robotic arm collects a dead chicken, and the system can remove two dead chickens in one operation at a speed of 3.3 cm per second. Luo et al. (2023) proposed a deep learning-based method to detect dead hens on commercial farms using thermal infrared (TIR), near-infrared (NIR), and depth images. Among the tested models, Deformable DETR with NIR-depth images achieved the highest performance, with an average precision (AP) of 0.9970 and a recall of 0.990 at IoU 0.5. The dual-source model (TIR-NIR-depth) offered only marginal improvements, indicating that more data sources do not always guarantee better results. Yang et al. (2024d) developed an improved YOLOv7 model for detecting dead caged hens, addressing challenges like occlusion and low brightness in caged environments. The model, enhanced with Convolutional Block Attention Module (CBMA), repulsion loss, and Distance Intersection over Non-maximum Suppression (DIoU-NMS), achieved a precision of 0.9570, a recall of 0.8680, and a mAP of 0.8620. Compared to the original YOLOv7, the improved model showed significant performance gains, including a 13.4% increase in mAP and a 43 FPS increase in detection speed. The review above highlights that there is limited research being done in the areas of detection of injured or dead birds in poultry with one research focusing on detection of dead hens in a CF facility while other research focusing on developing autonomous inspection platform for dead bird detection on large-scale breeding farms and a chicken removal system combined with a robotic arm as well as using multi-source images such as using thermal infrared (TIR), near-infrared (NIR), and depth images. YOLO models deployed for these tasks include YOLOv3, YOLOv4, YOLOv5, and YOLOv7. Disease, pathogen, and wild bird detection Disease, pathogen, and wild birds pose a significant threat to the poultry industry, leading to not only economic losses but also animal welfare issues. Early detection of disease, pathogen, and wild birds in chicken houses is especially important for preventing economic losses, protecting public health, and protecting other flocks. Wild birds can function as vectors for pathogens like avian influenza and Salmonella, creating pathways for disease transmission between wild and domestic flocks and even to humans. Traditional methods of disease detection, identification of a pathogen, and wild bird detection rely on human labor, are time-consuming, and subject to cognitive biases. To solve these challenges, researchers have recently been utilizing machine learning models to automatically detect and identify symptoms of diseases and pathogens from various sources, such as fecal images, abnormal droppings, and coughs that could be associated with specific illnesses. This section of the review summarizes the use of YOLO-based deep learning models along with other deep learning models for the detection of symptoms of diseases, pathological phenomena, and wild birds at poultry farms. Digestive diseases are one of the common diseases that significantly affect production and animal welfare in broiler breeding. Wang et al. (2019) developed a deep convolutional neural network (CNN)-based system for the classification of broiler droppings that could be associated with digestive diseases. Two deep learning-based object detection models, Faster R-CNN and YOLOV3, were evaluated for their effectiveness in detecting abnormal droppings. Faster R-CNN demonstrated superior accuracy with a recall of 0.9910 and a mAP of 0.9300, while YOLOv3 showed a lower accuracy (recall of 0.8870, mAP of 0.8430) but achieved significantly faster detection speed (43.5 FPS) compared to Faster R-CNN (5.7 FPS). These findings highlight a trade-off between accuracy and computational efficiency, where YOLOv3 may be preferred for real-time monitoring. Qin et al. (2024) proposed a deep learning model combining YOLOv5 and SuperPoint-SuperGlue for early warning and traceability of digestive diseases in caged hens based on manure images. The YOLOv5 model achieved high precision (0.9570), recall (0.9540), and [email protected] (0.9810) in detecting abnormal manure, while the SuperPoint-SuperGlue model demonstrated robust performance in monitoring manure belt movement with MAE, RMSE, and MAPE values of 0.19 m, 0.14 m, and 0.34%, respectively. The combined system achieved a correct tracking rate of 0.916 in field tests. Coccidiosis, Salmonella, and Newcastle disease are very common diseases in chickens, and various researchers have utilized YOLO-based models to detect symptoms seen in fecal images to associate the possibility of confirming these diseases. Hewawasam et al. (2023b) presented a system for identifying poultry diseases (Coccidiosis, Salmonella, Newcastle) using object detection techniques on fecal images. The YOLOv8 model was employed, achieving high accuracy in detecting diseases, with notable [email protected] scores for Salmonella (0.9120) and healthy (0.8430) categories. Compared to a baseline CNN model, YOLOv8 demonstrated superior performance in disease detection and localization. In contrast, Machuve et al. (2022) developed a deep learning model to diagnose poultry diseases like Coccidiosis, Salmonella, and Newcastle by classifying fecal images. MobileNetV2 emerged as the best model for deployment on smartphones due to its high accuracy of 0.9802 and lightweight nature, outperforming VGG16, InceptionV3, and Exception, which also showed high accuracies but were less efficient for mobile deployment. Bharti and Yogi (2024) proposed a Grey Wolf Optimized Deep Convolutional Neural Network (GWO-Deep CNN) for detecting poultry diseases, achieving high accuracy (0.9520), sensitivity (0.9620), and specificity (0.9400). This model outperformed traditional methods like CNN, DNN, DenseNet, and SVM, offering improved efficiency and reduced computational time. Uddin et al. (2023) introduced ChickenNet21, a hybrid deep neural network combining CNN and fine-tuned VGG19, achieving 0.9883 accuracy in detecting chicken diseases from fecal images. This model outperformed earlier methods like MobileNetV2 (0.9802) and Exception (0.9824), offering enhanced. Accuracy and robustness. Leg diseases are also very common in chickens, and researchers have utilized YOLO-based models and even modified YOLO models to detect leg diseases such as joint effusion, which appears as a semi-transparent grey shadow around the joint, and tibial dyschondroplasia, characterized by the overlap of tibial cartilage at the tibiotarsal joint, in broiler chickens. Zhang et al. (2024) proposed an improved YOLOv8 model to detect leg diseases in broiler chickens using X-ray images, addressing challenges like low contrast and small lesion areas. Key enhancements included Partial Convolution (PConv) for efficient feature extraction, channel Prior Convolutional Attention (CPCA) for improved feature representation, and Gather-Distribute (GD) mechanism for better feature fusion. The modified YOLOv8 outperformed standard YOLO versions, achieving a 7.2% increase in average precision while maintaining a real-time processing speed of 66.8 FPS. Compared to YOLOv5 and other object detection models, the improved YOLOv8 demonstrated superior accuracy and efficiency in detecting joint effusion and tibial dyschondroplasia in broiler chickens. The study by Sun et al. (2024) developed Tibia-YOLO, a deep learning -based assisted detection system for identifying broiler leg diseases using industrial CT and digital radiography (DR). The Tibia-YOLO model, an improved version of YOLOv8, incorporates Content-Aware ReAssembly of Features (CARAFE), Efficient Multi-scale Attention (EMA), and Parallel Network Attention (ParNet) to enhance detection accuracy and generalization. The model achieved 0.908 mean average precision (mAP) for tibia detection and a root mean square (RMSE) of 3.37 mm in tibia length estimation. Regression analysis using Support Vector Regression (SVR) demonstrated superior accuracy over alternative models, making Tibia-YOLO a highly effective tool for early detection of leg disorders in poultry farming. Besides diseases, deep learning-based models have been utilized in detecting broiler pathological phenomena using both thermal and visual images. The study by Elmessery et al. (2023) developed a YOLO-based deep learning model for detecting broiler pathological phenomena using thermal and visual images in intensive poultry houses. A dataset of 10000 images with 50000 annotations was created, classifying broilers as healthy, lethargic, suffering from slipped tendons, diseased eyes, stressed (beak open), or having a pendulous crop. Three YOLO versions (YOLOv5, YOLOv7, and YOLOv8) were evaluated, with YOLOv8 trained on thermal images, achieving the best results (mAP50:0.9880, F1 score:0.9720). The mosaic augmentation techniques significantly improved detection performance, particularly for thermal images, making the model highly reliable for automated broiler health monitoring in complex lighting conditions. Tong et al. (2023) developed a modified YOLOv5 detector to identify the health status of chickens in real-time, incorporating a receptive field enhancement module and a sliding loss function to improve detection accuracy. The modified YOLOv5 achieved a superior performance compared to the original YOLOv5 and other detectors, with [email protected] and [email protected] values of 0.9490 and 0.660, respectively. This approach offers a robust solution for automated poultry health monitoring, enhancing both detection precision and real-time processing capabilities. Along with object detection models, researchers have used a pose estimation model to detect abnormal chicken behavior that helps to detect diseases. Elbarrany et al. (2023) explored the use of pose estimation for monitoring abnormal behaviors in poultry farms, specifically focusing on detecting wry neck disease in chickens. The study employs the YOLOv8 pose model for key-point extraction and a 1D Convolutional Neural Network (CNN) for behavior classification, achieving an accuracy of 0.9972. Compared to other models like DeepLabCut with ResNet-50, which achieved varying accuracies for different behaviors (e.g., 0.9300 for eating, 0.6200 for running), the proposed system demonstrates superior accuracy and efficiency in classifying chicken behaviors. Harshitha et al. (2024) proposed a deep learning-based method for detecting Coccidiosis in poultry using the VGGNet convolutional neural network architecture. The CGGNet model, trained on microscopic images of fecal samples, achieved a detection accuracy of 0.9650, outperforming other models such as YOLOv5 (0.8920), ResNet-FPN (0.9370), and XceptionNet (0.9400). Leveraging transfer learning, the VGGNet model demonstrated high precision, robustness, and suitability for early disease diagnosis in poultry health monitoring. Depuru et al. (2024) utilized the YOLOv5s model for object detection to monitor individual chickens for early disease detection and achieved 0.9600 accuracy in detecting and categorizing chickens by age, enabling early disease detection, and improving overall farm productivity. Zhou et al. (2024) developed an improved YOLOv8 model, named YOLOv8-BCE, to detect open-beak behavior in chickens, which is an indicator of respiratory disease and heat stress. The model achieved a precision of 0.8640, a recall of 0.8520, and mAP of 0.9370, outperforming other models like Faster R-CNN, SSD, and YOLOv5. This model offers a robust and efficient method for early warning of health issues in poultry farms. Bai et al. (2023) proposed a Cascade R-CNN model based on visual technology to identify the behavior of yellow-feathered broilers, especially heat stress behaviors in broilers, achieving an average accuracy of 0.8840. The heat stress evaluation model, optimized using the PLSR method, reached a test accuracy of 0.8580, demonstrating its feasibility for regulating broiler chambers. Yu et al. (2023) used the SOLOv2 model, enhanced with DenseNet-169, Efficient Channel Attention (ECA), and DropBlock regularization, to detect heat stress in poultry. The improved model, FPN-DenseNet-SOLO, achieved a recall of 0.9540 and mAP of 0.9090, outperforming Mask R-CNN, Faster R-CNN, and the original SOLOv2 in both accuracy and efficiency. The study demonstrates the model’s ability to accurately segment and classify poultry under heat stress, providing valuable support for precise poultry breeding and welfare management. Deo et al. (2024) compared five deep learning models for detecting poultry diseases through fecal image analysis, finding that EfficientNetB3 achieved the highest accuracy (0.9887) and balanced precision and recall compared to a study by Syafaah et al. (2024) , where YOLOv8 achieved accuracy of 0.9825 for sick chickens based on symptoms of disease and movement pattern of the chicken, and 0.7125 for dead chickens. YOLOv8 (thermal) demonstrated exceptional performance with an [email protected] of 0.988 and an F1 score of 0.972 ( Elmessery et al., 2023 ) EfficientNetB3’s superior performance suggests it is the most effective model for non-intrusive poultry disease detection. In addition to diseases, YOLO models have been used in detecting wild birds at poultry farms. Yang et al. (2024c) compared three object detection models- YOLOv5, YOLOv7, and YOLOv8 on a wild bird image dataset to assess their performance in detecting wild birds. YOLOv8 outperformed the other models with a mean average precision (mAP) of 0.79, followed by YOLOv5 at 0.76 and YOLOv7 at 0.72. The findings suggest that YOLOv8 is the most suitable for real-time wild bird detection due to its superior accuracy and robust performance. Li et al. (2023b) developed a deep-learning-based epidemic surveillance system for live poultry transport, using the YOLOv5 algorithm to monitor poultry health through excreta analysis. The system achieved high accuracy (overall mAP of 0.900) and recall rates, significantly improving disease detection and prevention during transportation compared to traditional methods. The YOLOv5 model’s real-time capabilities and high precision in multi-object detection make it particularly effective for this application. In addition, researchers have developed a smartphone-based system for detecting and classifying poultry diseases from fecal images. Degu and Simegn (2023) developed a smartphone-based system for detecting and classifying poultry diseases from chicken fecal images using YOLOv3 for object detection and ResNet50 for image classification. The YOLov3 model achieved a mAP of 0.8748 for detecting regions of interest, while the ResNet50 model demonstrated a classification accuracy of 0.987. This system can accurately identify three prevalent poultry diseases: Salmonella, Coccidiosis, and Newcastle Disease, providing a valuable tool for poultry farmers and veterinarians. The use of YOLO models was progressive, utilizing YOLOv3 in earlier research and progressing towards YOLOv5 and mostly utilizing YOLOv8, while a few researchers also compared the performances of YOLOv5, YOLOv7, and YOLOv8 for disease detection. Most of the researchers focused on three diseases, such as detection of Coccidiosis, Salmonella, and Newcastle disease through fecal images, while limited research has been conducted on digestive diseases, leg disorders, wry neck disease, pathological phenomenon, and respiratory disease and heat stress. While most of the researcher was focused on improving the YOLO model for the detection of diseases, one study focused on developing a smartphone-based system for classifying and detecting poultry disease. This review shows that there needs to be more focus on other diseases that are also critical in poultry besides Coccidiosis, Salmonella, and Newcastle disease. Overall, the above section of the review provided information on how researchers have applied YOLO-based models in various research settings to identify and detect behavior of their interest with high precision, mostly over 0.90. The trend shows that researchers utilized YOLOv3, YOLOv4, YOLOv5, YOLOv6, YOLOv7, YOLOv8, and YOLOX for these purposes as the models evolve. These models have been used in egg breeders, caged housing, CF housing, coop systems, broiler houses, and free-range housing systems, with most of the research focused on CF housing. Behaviors that were detected utilizing YOLO models in the above section could be categorized into five categories, i.e., (1) natural/normal behavior (feeding, drinking, standing, lying,), (2) applied/comfort behavior (perching, dustbathing, grooming, mouth opening), (3) problematic behavior (feather pecking, mislaying, piling, huddling), (4) leg issues (footpad dermatitis, bumble foot, claw detection) and (5) mating behavior, and (6) behavior of broilers and lying hens in various housing system. This section also dissects the use of YOLO-based models in terms of detecting injured or dead birds as well as the detection of diseases, pathological phenomena, and wild birds at a poultry farm. Identification, counting, and tracking activity of chickens Abdalshefie Abuhussein et al. (2024) explored the use of thermal imaging and deep learning models, YOLOv7 and YOLOv8, for automated broiler detection and counting. YOLOv8 outperformed YOLOv7, achieving a higher mAP of 0.9500 compared to 0.8500, and faster convergence within 20 epochs versus 60 epochs for YOLOv7. Additionally, YOLOv8 exhibited a comparatively lower error rate of 0.2 versus 0.5 for the counting tasks, showing its effectiveness for real-time poultry monitoring and precision agriculture applications. Researchers have not only used YOLO based models but also modified them to carry out different tasks. For example, Siriani et al. (2022) evaluated a modified YOLOv4 model integrated with a Kalman filter for detecting and tracking chickens in low-light, low-resolution video conditions. The YOLOv4 model achieved an impressive 0.9900 accuracy in detecting chickens, while the Kalman filter successfully tracked individual birds, assigning unique identities and handling movement even in crowded environments. These results demonstrate the effectiveness of deep learning and tracking algorithms in precision livestock farming, providing a practical solution for real-time poultry monitoring under challenging conditions. Subedi et al. (2023b) developed and compared three deep learning models (YOLOv5s-egg, YOLOv5x-egg, and YOLOv7-egg) for tracking floor eggs in CF hen houses. The YOLOv5x-egg model performed best, detecting floor eggs with a precision of 0.90, a recall of 0.879, and a mean average precision (mAP) of 0.921. This study demonstrated the potential for automated monitoring of floor eggs to improve efficiency in CF egg production. Yang et al. (2024a) introduced and enhanced the Track Anything Model (TAM) for tracking the locomotion of individual chickens in diverse settings, providing a non-intuitive and highly accurate method for analyzing poultry movement. TAM achieved a mean Intersection over Union (mIOU) of 0.9312, outperforming YOLOv5 (0.8563) and YOLOv8 (0.8772) in segmentation accuracy, while its modified TAM-speed model demonstrated precise speed detection with a root mean square error (RMSE) of 0.02m/s. Compared to tracking models like YOLOv5+DeepSORT (MOTA: 0.9213) and YOLOv8+StrongSORT (MOTA: 0.9456), TAM-speed achieved the highest tracking accuracy (MOTA: 0.9745) with the fewest identity switches, making it the most effective model for real-time chicken locomotion monitoring. Yang et al. (2024b) developed convolutional neural network (CNN) models to monitor the activity index of CF hens, using a dataset of 1500 top-view images. Among the tested models, YOLOv8 combined with DeepSORT achieved the highest performance, with a multi-object tracking accuracy (MOTA) of 0.94. This approach effectively facilitated the detection of abnormal behaviors, such as smothering and piling, and enabled the quantification of flock activity into low, medium, and high levels to assess footpad health status. Compared to other YOLO-based models, YOLOv8+DeepSORT outperformed alternatives like YOLOv5 and ByteTrack in terms of tracking accuracy and identification robustness while maintaining real-time processing capabilities. Neethirajan (2022) presented a Quantitative Tracking Tool for monitoring Chicken Activity based on YOLOv5 for detecting, counting, and tracking chickens in various environments. The ChickTrack model addresses challenges such as complex backgrounds, varying lighting conditions, and occlusions. The model uses a Kalman filter to track multiple chickens simultaneously, providing real-time and online applications. The ChickTrack model, utilizing YOLOv5, demonstrated high accuracy and tracking chickens, with precision and recall values stabilizing after 25-30 epochs. The model effectively managed occlusions and varying lighting conditions, outperforming earlier methods in real-time detection and tracking of chicken activity. Li et al. (2023a) utilized a combination of deep learning techniques, including a CNN) for perching status classification and the YOLOv5m-oriented object detection model for detecting birds and calculating interindividual distances and orientations. The CNN achieved over 0.990 accuracy in classifying perching status, while the YOLOv5m model achieved 0.9090 precision, 0.9320 recall, and 0.92 F1 score in detecting birds. Li et al. (2022) proposed a novel multi-object tracking framework called YOLOX-Birth Growth Death (Y-BGD) for automatic broiler counting, using YOLOX for detection and a GBD data association strategy to reduce identity-switching errors. Evaluated on the ChickenRun-2022 dataset, Y-BGD achieved 0.98131 accuracy, outperforming state-of-the-art MOT models in both accuracy and speed (58.98 FPS) compared to other methods like QDTrack (0.9539) and DEFT (0.9703), Y-BGD demonstrated superior tracking stability and efficiency, making it highly effective for real-time broiler counting in CF environments. Mehdizadeh et al. (2024) compared different deep-learning models for chicken tracking and detection in poultry environments. The modified YOLOv8 model outperformed other approaches, achieving over 0.9800 accuracy and a loss of less than 0.1, demonstrating robustness in three challenging scenarios: temporary invisibility, occlusions, and coalescence of chickens using morphological and movement-based tracking, significantly enhancing identification and tracking accuracy compared to conventional YOLO and tracking methods. Doornweerd et al. (2024) explored tracking individual broilers in group-housed settings using video technology to assess their locomotion phenotypes. The YOLOv7-tiny model used in the study demonstrated excellent performance in detecting broilers within video frames. It achieved a precision, recall, and average precision at an IoU threshold of 0.5 of 0.9900. The AP at an IoU threshold of 0.75 was 0.9800. These metrics show that the model was highly accurate in identifying broilers, with minimal false positives and false negatives. The tracking performance using the SORT algorithm was evaluated based on ground-truth trajectories of 13 broilers. The number of ID-switches per ground-truth trajectory varied from 5 to 20 (mean 9.92). Tracking time ranged from 1 to 51 min (mean 12.36 min), and tracking distances ranged from 0.02 to 17.07 m (mean 1.89 m). Tracking errors mainly occurred when broilers were occluded by objects. Yang et al. (2023a) developed an improved YOLOv5 deep learning model to monitor the spatial distribution of CF hens across perching, feeding, drinking, and nesting, zones throughout their growth. The model achieved high detection precision of 0.8700-0.9400 across zones and age groups, with performance affected by bird age and occlusion from equipment. Specifically, detection precision was 0.891 for baby chicks, 0.942 for older birds in perching zones, and 0.87 (baby chicks) and 0.932 (older birds) in feeding/drinking zones. The model outperformed prior methods and supports automated, non-intrusive monitoring for poultry welfare. However, it is very important to note that although high-performance metrics are shown in these studies, there remains a significant challenge, such as issues of identity switches and loss of tracking due to occlusions. This section of the review summarizes the use of YOLO models along with combination of other models to accomplish tasks such as identification, counting, and tracking activity in different research settings. Integrating YOLO with other machine learning models Integrating YOLO models with additional tracking algorithms is necessary because YOLO alone can only detect objects in individual frames and cannot maintain consistent identities over time, which is essential for monitoring poultry behavior. In dense and dynamic broiler environments, birds often overlap, look visually similar, and move unpredictably, making it difficult for a detector to distinguish individuals across consecutive frames. By combining YOLO with multi-object tracking methods such as DeepSORT, StrongSORT, or ByteTrack, researchers can ensure persistent identification, generate movement trajectories, and accurately quantify behaviors. This integration enables reliable assessment of activity levels, welfare indicators, and early signs of abnormal behavior tasks that require continuous tracking rather than single-frame detection. Therefore, integrating YOLO with tracking models is crucial for transforming static detections into meaningful behavior monitoring systems suitable for precision livestock farming. Triyanto et al. (2023) designed an automated broiler tracking system using YOLOv4 with DeepSORT, outperforming other configurations like YOLOv3, SSD, and YOLOv4-tiny in detection accuracy. Their system effectively identified 17 broilers within dense flocks, demonstrating the potential of YOLO-based approaches for precision livestock monitoring despite longer processing times (702.35 seconds), while SSD with MobileNet and Caffe Model Weights achieved the fastest detection time (32.45 seconds) but identified only five broilers. These results demonstrate that YOLOv4 with DeepSORT is highly effective for poultry behavior monitoring, offering potential applications in precision livestock farming for health and welfare assessment. Similarly, Zou et al. (2023) presented a novel method for tracking yellow feather broilers using an improved YOLOv3 algorithm combined with DeepSORT. The improved YOLOv3 achieved mAP of 0.9320 and 29 FPS, while the tracking accuracy (MOTA) increased from 0.51 to 0.54, and identity switches (IDSW) decreased by 62% compared to traditional YOLOv3-based detectors. Other works have expanded deep learning applications toward behavioral and health assessment. Nasiri et al. (2023) developed an algorithm using a convolutional neural network and image processing to estimate the feeding time of individual broilers. The algorithm achieved an overall accuracy of 0.8730 in estimating feeding time per visit to the feeding pan, demonstrating its potential as a real-time tool for monitoring broiler feeding behavior on commercial farms. Nasiri et al. (2024) developed an automated video recognition system to detect and quantify preening and stretching behaviors in broilers, which are key welfare indicators. Using a deep learning pipeline consisting of U-Net, YOLO, and MoviNet-A4 models, along with three filtering mechanisms, the system achieved high accuracy (0.9670), precision (0.8810), and sensitivity (0.8896). The results show the feasibility of using this approach for real-time behavior checking in commercial poultry farms, providing a valuable tool for improving broiler welfare management. Chemme and Alitappeh (2024) developed an automated system for detecting and counting chickens in densely populated environments by using of a segmentation module combined with the YOLOv8 model for detection. The SAM (Segment Anything Model) was used for image segmentation, and the outputs were converted to YOLO format for training. The proposed model achieved a 0.9300 accuracy in chicken detection and counting, showing a 0.3 improvement over the previous method by ( Neethirajan, 2022 ), in which the author achieved 0.9000 accuracy with a slower bounding box creation speed (60 seconds compared to 30 seconds for the new model). More recent research has integrated spatiotemporal learning and attention mechanisms for comprehensive behavior analysis. Hu et al. (2024) proposed an end-to-end broiler behavior recognition system (BBRS) combining YOLOv8s, BytTrack, and 3D-ResNet50-TSAM model, achieving near-perfect detection ([email protected] of 0.9950), and high the ByteTrack tracker attained a mean MOTA of 0.9389 across different occlusion levels, and the 3D-ResNet50-TSAM model reached an accuracy of 0.9784. Compared to other deep learning models like TSM, 3D, and TSN, the BBRS demonstrated superior performance in recognizing multiple simultaneous behaviors of CF broilers, making it a highly effective tool for automated behavior analysis. Guo et al. (2023) further refined YOLOv5 through attention (C3CBAM-BiFPN), to enhance detection in occluded conditions, which achieved a precision (0.9820), recall (0.9290), and mAP (0. 9670), outperforming other models like YOLOv5 and YOLOv5-CBAM while Yang et al. (2023b) and Subedi et al. (2025) developed YOLOv5-based classifiers for multi-behavior detection in CF laying hens. Yang et al. (2023b) developed a six-behavior classifier for monitoring CF laying hens using YOLOv5-cls and EfficientNetV2 models. The YOLOv5-cls-m model achieved the highest accuracy of 0.9533, outperforming EfficientNetV2-l by 0.501. The classifier effectively monitors behaviors such as feeding, drinking, walking, perching, dustbathing, and nesting, with drinking behavior detection reaching 0.978 accuracy. Subedi et al. (2025) developed three YOLO-based models (YOLOv5s_BH, YOLOv5x_BH, and YOLOv7_BH) for classifying multiple behaviors of CF laying hens. The YOLOv5s_BH model achieved the highest performance with a precision of 0.781, recall of 0.717, and mean average precision (mAP) of 0.753, outperforming YOLOv5x_BH and YOLOv7_BH by 0.19 and 0.22 in precision, respectively. The study highlighted the potential for automated monitoring of poultry behaviors, offering valuable insight for producers. Other hybrid frameworks have also demonstrated promise. Joo et al. (2022) used Mask R-CNN and YOLOv4-ResNet50 to detect chicken posture and behaviors. Mask R-CNN achieved a weighted F1 score of 0.8846 for posture and behavior detection, while the YOLOv4-ResNet50 pipeline reached up to 0.9100 behavior classification accuracy and 0.8650 for posture detection. Despite data imbalance affecting performance for less frequent behaviors, the models demonstrated strong potential for non-invasive, automated welfare assessment in poultry farming. Nakrosis et al. (2023) applied YOLOv5 for classification and K-means for segmentation for classifying poultry droppings, achieving high accuracy rates of 0.9178 and 0.88 dice coefficient, respectively. The findings highlight the potential of these techniques to monitor poultry health effectively by analyzing droppings for abnormalities. Khairunissa et al. (2021) used a Multi-Object tracking algorithm combined with a pre-trained SSD-based object detection model to analyze poultry behavior from surveillance video, achieving a precision of 0.604. The system successfully extracted movement data and object trajectories but faced limitations in identity retention during object intersections. Future work aims to enhance tracking accuracy by incorporating identity labeling and improved assignment methods. Besides object detection, identification, and tracking, recent work also focuses on combining YOLO with other models to detect the severity and accuracy of the broiler stunned state. Electrical stunning is the first step towards the humane slaughtering process of broilers. In a study by Ye et al. (2020) to investigate the effects of different dataset construction and data augmentation methods on the detection sensitivity and accuracy of broiler stunned state recognition. The author compared the performances of YOLO+MRM, YOLO, and BP-NN and found that YOLO YOLO+MRM algorithm achieved a superior performance with an accuracy of 0.9677 compared to the BP-NN classifier (0.9011) and YOLO (0.9474). Together, these studies underscore the versatility and efficiency of YOLO-driven frameworks, often combined with advanced tracking, segmentation, or attention models, in automating poultry monitoring and welfare assessment in modern smart poultry farming systems. This section of the review highlighted the integration of other machine learning models with YOLO models to accomplish a more complex task than object detection in poultry. Challenges and limitations Implementing YOLO-based models for PPF presents several challenges that impede widespread adoption and scalability. Infrastructure constraints, high initial costs, and limited literacy, especially among smallholder farmers, create substantial barriers to deploying advanced machine learning technologies in real settings. The need for specialized hardware, reliable internet connectivity, and continuous system maintenance further complicates the integration of AI solutions into routine farm management. These socioeconomic factors mean that even high-performing YOLO models may be beyond the reach of resource-limited poultry producers. Occlusion and real-time data processing represent persistent technical hurdles in poultry farm environments. YOLO models often struggle with occlusion caused by objects like feeders, perches, nestboxes, and other poultry, which can obscure target birds and lead to missed or inaccurate detections as reported in many studies ( Guo et al., 2020 , 2023b ; Subedi et al., 2023a , 2023b , 2023c , 2023d ; Yang et al., 2023b , 2023c ; Bist et al., 2023a ; 2023b , 2023c , 2023d , 2024a , 2024b ). This issue is exacerbated in densely populated housing systems, where birds repeatedly block each other from view, and fluctuating lighting conditions degrade image quality further. Although recent model enhancements, such as multi-scale feature extractors and context fusion modules, have improved robustness against occlusions, reliable identification of partially obscured animals remains a significant challenge. Another critical area is the dataset and annotation limitations. Creating diverse, adequately labeled datasets for training YOLO detectors is still highly time-consuming and labor-intensive, with most available datasets derived from controlled research conditions that do not fully reflect real farm environments in both small-scale and large-scale farms. This restricts the generalizability of the models, their ability to perform well across varying breeds and housing conditions. The lack of scalable, annotated data from commercial farms means that the YOLO models often fail to account for the myriad variables present in operational environments, such as environmental changes, bird behavior variations, and equipment layouts. Generalization across diverse settings is a further concern, as models that perform well in one environment frequently experience performance degradation when deployed elsewhere due to differences in background, lighting, or bird type. This needs more retraining and adoption, increasing operational costs and complexity. Finally, ethical and implementation concerns must be considered when introducing AI-driven systems into commercial poultry farming. There are ongoing debates about whether intensive automation reduces animal welfare by transforming animals into mere data points and potentially entrenching stressful or crowded living conditions. Ensuring responsible, welfare-oriented deployment calls for transparent guidelines, robust legal protections, and frequent stakeholder consultation to balance technological efficiency with ethical obligations to animal health and well-being. Overall, three categories: technical challenges (e.g., occlusion, lighting variability, dataset limitations), operational challenges (e.g., deployment in commercial poultry houses, hardware constraints, and ethical considerations related to animal monitoring systems. Segmentation and generative adversarial network (GAN) for improving detection accuracy Besides Object detection models, researchers have tried segmentation and generative adversarial networks to improve the detection accuracy in occluded environments. ( Yang et al., 2023d ) proposed a deep learning-based defense algorithm (CCD) that combines U-Net and pix2pixHD to automatically remove cage wire mesh and significantly improve the detection accuracy of caged chickens. After defencing, detection precision and recall increased across all YOLOv5 models (YOLOv5s, YOLOv5m, YOLOv5l, YOLOv5x), with the most notable improvements being up to 0.161 in precision and 0.291 in recall. This method demonstrates the effectiveness of using semantic segmentation and generative adversarial networks to enhance object detection in occluded environments. ( Schiele et al., 2023 ) proposed simple dropout-based methods for training U-Net segmentation models using sparse annotations like points and scribbles, significantly reducing labeling effort. Models trained on point annotations achieved performance nearly equivalent to fully labelled data, with only 1% of labelled pixels, while scribble-based models underperformed due to less uniform coverage. The findings highlight that sparse, well-distributed labels can be a highly efficient alternative for semantic segmentation in domains like poultry farming. Saeidifar et al. (2024) developed and optimized a zero-shot image segmentation pipeline using Segment Anything Model (SAM) to monitor the thermal conditions of individual CF laying hens in thermal images. The modified SAM, enhanced with automatic point selection (pre-processing) and optimal mask selection via a Decision Tree Classifier (post-processing), achieved superior performance with a success rate of 0.844, F1-score of 0.923, recall of 0.910, and IoU of 0.855, outperforming other zero-shot and instance segmentation models, including YOLOv8, Mask R-CNN, U2-Net, ISNet, FastSAM, and MobileSAM. Wu et al. (2024b) developed a YOLO-Claw, a chicken claw detection algorithm, demonstrated that it outperformed other YOLO models (YOLOv5, YOLOv6, YOLOv7, YOLOv8) and six state-of-the-art methods in terms of precision, recall, and mAP with values of 0.953, 0.936, and 0.971, respectively. The study introduced the Tanh-aided Channel Attention (TCA and AFPN-Neck modules, which enhanced feature extraction and fusion, contributing to the superior performance of YOLO-Claw. Additionally, YOLO-Claw maintained high detection accuracy under varying illumination conditions and claw size variations, proving its robustness in commercial farm scenarios. This section of the review included the highlights of the research papers that go beyond applying YOLO models to overcome some of the challenges faced by YOLO models by applying segmentation and GAN to improve detection accuracy. Potential for automated poultry assessment systems Recent advancements in artificial intelligence and computer vision have propelled the development of fully automated assessment systems. For instance, Xin et al. (2024) engineered a mobile device equipped with YOLOv6 to identify and grasp dead yellow-feather broilers, achieving an identification accuracy of 0.8610. This device, integrated with a robotic arm, demonstrated an average grasping success rate of 0.8130, outperforming SSD and Faster-RCNN. YOLOv6 showed competitive performance with a precision of 0.8000, a recall of 0.8100, and an F1 score of 0.8000, while Faster R-CNN achieved slightly higher precision and recall (0.8100 each) but a slower detection speed. The use of a lightweight YOLO architecture integrated with a mobile robotic platform suggests potential scalability for deployment across commercial houses; however, the study preliminarily validated the system at the device level and did not explicitly evaluate large-scale farm deployment. Similarly, Ty et al. (2024) developed an intelligent video surveillance system with color-calibrated cameras and the YOLOv4 algorithm to monitor broiler chickens’ comb color, which changes as they grow, which is a critical indicator of growth and health. The YOLOv4 model used in this study for comb detection achieved a precision of 0.9000 and a recall of 0.8900. The model was highly correct in finding chicken combs and had a strong ability to correctly detect combs under various lighting conditions and zoom levels. However, it faced challenges in distinguishing small combs or those positioned too far away in the images. Although the approach supports continuous automated monitoring, the study focused on detection performance rather than explicitly testing scalability across multiple barns or large commercial flocks. Hewawasam et al. (2023a) presented a comprehensive AI-driven poultry management system, employing computer vision, machine learning, and IoT technologies to automate bloodstain detection, meat quality assessment, breed identification, and disease diagnosis. Models such as CNN, YOLOv5, and YOLOv8 were employed, achieving high accuracy rates across various tasks. Overall, GoogleNet performed best for chicken quality assessment, while YOLOv5 showed superior results for breed identification using RGB images. YOLOv8 demonstrated effective disease detection capabilities. Collectively, these models not only enhanced operational efficiency but also set a new benchmark for accuracy, underscoring a transformative leap from labor-intensive manual methods toward digitized, intelligent poultry farming. The integration with IoT infrastructure suggests a scalable framework capable of handling multiple monitoring tasks within a digital poultry management system; however, scalability testing across large production system was not explicitly reported. Among the reviewed approaches, Gong et al. (2024) addressed the industry need for a low-cost, scalable fertilized egg detection system by proposing FEDM, an enhanced YOLOv5-based model with SENet, C3_DCNv3, Dyhead, and a simplified MPDIoU loss function. FEDM achieved a 0.9670 average precision, 5.5% higher than YOLOv5s, and demonstrated superior performance across varied angles and early-stage egg detection, offering tangible benefits in cost, space utilization, and practical deployment. In parallel, Ali et al. (2024) demonstrated a cost-effective and sustainable approach to poultry disease management by integrating Internet of Things (IoT) with advanced imaging technologies using YOLOv5 and ResNet50. The system achieved 0.9900 accuracy in disease detection and improved poultry health by 35%, highlighting the potential of smart technologies to enhance farm management and animal welfare. The focus on low computational cost and practical deployment highlights its suitability for scalable implementation in hatchery environments. Likewise, Wu et al. (2024a) developed an efficient and lightweight egg-counting system for commercial laying cages by combining the Strong Sort tracking algorithm with a customized, optimized version of YOLOv5s, called EGG-YOLO-GD. By integrating modules such as GAM, RFB, and WIoU for accuracy, and GhostNet, DSC, and model pruning for efficiency, the final model achieved 0.9930 mean average precision, 128.7 FPS, and a model size of only 5.1 MB, outperforming the baseline YOLOv5s and other models like Faster R-CNN, SSD, and TOLOv8s. Field validation on 200 cages showed 100% localization accuracy, 98.5% counting accuracy, and halved counting time compared to manual methods. The EGG-YOLO-GD model offered the best balance of accuracy, speed, and model size, making it highly suitable for real-time applications in smart agriculture, confirming the system’s suitability for real-time, large-scale commercial deployment. The convergence of deep learning, computer vision, and IoT technologies is revolutionizing poultry farming. Systems leveraging advanced YOLO detectors, innovative tracking, and model optimization are achieving unprecedented levels of accuracy and operational efficiency. Automated solutions for birds’ identification, health monitoring, fertilized egg detection, disease management, and egg counting now offer scalable, cost-effective, and real-time alternatives to manual labor, positioning smart poultry farming as the future standard for PPF. The IoT-enabled architecture facilitates distributed data collection and centralized monitoring, which supports scalability across multiple barns or production facilities. Collectively, these studies demonstrate that lightweight YOLO-based models, particularly those optimized for computational efficiency and integrated with IoT or edge devices, offer strong potential for scalable implementation in smart poultry farming systems. Conclusions YOLO models have evolved from YOLOv1 to YOLOv11 over the past decade, with architectural innovations in backbone design, detection heads, loss functions, and attention mechanisms progressively improving the balance between speed, accuracy, and robustness for real-time, non-invasive monitoring in precision poultry farming. This review synthesized 82 studies from 408 records (2015 to 2024), showing that modern YOLO-based systems, often integrated with tracking and other machine vision models, are now widely used for identification, counting, tracking, behavior detection, mortality surveillance, disease and pathogen detection, wild bird detection, flock distribution mapping, and activity index calculation in broilers and layers across diverse housing systems, with rapid publication growth since 2021 driven mainly by China and the USA. These applications are catalyzing a shift from manual observation to continuous, data-driven welfare and health management, with clear benefits for productivity, early disease detection, and animal welfare, yet practical challenges persist, including occlusion, variable farm infrastructure, limited dataset generalizability, and ethical considerations around monitoring. Fully realizing scalable, real-time, and humane YOLO-powered systems for commercial farms will require continued advances in dataset diversity and quality, edge-device compatibility, robust performance under commercial conditions, and explicit integration of animal welfare safeguards into system design and deployment. CRediT authorship contribution statement Bidur Paneru: Writing – original draft, Validation, Methodology, Investigation, Formal analysis, Data curation. Anjan Dhungana: Writing – original draft, Investigation. Samin Dahal: Writing – original draft, Investigation. Casey W. Ritz: Writing – original draft, Supervision, Investigation. Woo Kim: Writing – original draft, Supervision, Investigation. Tianming Liu: Writing – original draft, Supervision, Investigation. Lilong Chai: Writing – original draft, Supervision, Resources, Project administration, Investigation, Funding acquisition, Conceptualization. Disclosures The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. Acknowledgements The study was sponsored by USDA-NIFA AFRI (2023-68008-39853; Precision farming practices for sustainable egg productions), Georgia Research Alliance, UGA CAES-CVM collaboration grant, and UGA Institute for Integrative Precision Agriculture (IIPA) seed grant. References Abdalshefie Abuhussein M.F., El-Soaly I..S., Elmessery W.M., Abd El-Wahhab G.G. Thermal imaging and advanced deep learning for automated broiler detection and counting. Al-Azhar J. Agric. Eng. 2024;7 https://azeng.journals.ekb.eg/article_362167.html 0–0 Available at. (verified 6 November 2024) [ Google Scholar ] Ali W., Din I.U., Almogren A., Rodrigues J.J.P.C. Poultry health monitoring with advanced imaging: towards next-generation agricultural applications in consumer electronics. IEEE Trans. Consum. Electron. 2024 https://ieeexplore.ieee.org/document/10547190/ 1–1 Available at. verified 8 November 2024. [ Google Scholar ] Anderson G., Johnson A., Arguelles-Ramos M., Ali A. Impact of body-worn sensors on broiler chicken behavior and agonistic interactions. J. Appl. Anim. Welf. Sci. 2025.;28:1–10. doi: 10.1080/10888705.2023.2186788. https://www.tandfonline.com/doi/full/10.1080/10888705.2023.2186788 Available at. (verified 11 December 2025) [ DOI ] [ PubMed ] [ Google Scholar ] Bai Y., Zhang J., Chen Y., Yao H., Xin C., Wang S., Yu J., Chen C., Xiao M., Zou X. Research into heat stress behavior recognition and evaluation index for yellow-feathered broilers, based on improved Cascade region-based convolutional neural network. Agriculture. 2023;13:1114. https://www.mdpi.com/2077-0472/13/6/1114 Available at. (verified 6 November 2024) [ Google Scholar ] Berckmans D. General introduction to precision livestock farming. Anim. Front. 2017;7:6–11. https://academic.oup.com/af/article/7/1/6/4638786 Available at. (verified 1 January 2025) [ Google Scholar ] Bharti, V., and K. K. Yogi. 2024. Chicken disease detection in the poultry utilizing grey wolf optimized deep convolutional neural network. Available at https://www.researchsquare.com/article/rs-4635600/v1 (verified 8 November 2024). Bist R.B., Subedi S.., Yang X., Chai L. A novel YOLOv6 object detector for monitoring piling behavior of cage-free laying hens. AgriEngineering. 2023;5:905–923. https://www.mdpi.com/2624-7402/5/2/56 Available at. verified 23 January 2024. [ Google Scholar ] Bist R.B., Subedi S.., Yang X., Chai L. Automatic detection of cage-free dead hens with deep learning methods. AgriEngineering. 2023;5:1020–1038. https://www.mdpi.com/2624-7402/5/2/64 Available at. verified 1 July 2025. [ Google Scholar ] Bist R.B., Yang X.., Subedi S., Chai L. Mislaying behavior detection in cage-free hens with deep learning technologies. Poult. Sci. 2023;102 doi: 10.1016/j.psj.2023.102729. https://linkinghub.elsevier.com/retrieve/pii/S0032579123002511 Available at. verified 23 January 2024. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Bist R.B., Yang X.., Subedi S., Chai L. Mislaying behavior detection in cage-free hens with deep learning technologies. Poult. Sci. 2023;102 doi: 10.1016/j.psj.2023.102729. https://linkinghub.elsevier.com/retrieve/pii/S0032579123002511 Available at. verified 1 July 2025. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Bist R.B., Yang X.., Subedi S., Bist K., Paneru B., Li G., Chai L. An automatic method for scoring poultry footpad dermatitis with deep learning and thermal imaging. Comput. Electron. Agric. 2024;226 https://linkinghub.elsevier.com/retrieve/pii/S016816992400872X Available at. verified 13 March 2025. [ Google Scholar ] Bist R.B., Yang X.., Subedi S., Chai L. Automatic detection of bumblefoot in cage-free hens using computer vision technologies. Poult. Sci. 2024;103 doi: 10.1016/j.psj.2024.103780. https://linkinghub.elsevier.com/retrieve/pii/S0032579124003596 Available at. verified 9 July 2024. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Bochkovskiy, A., C.-Y. Wang, and H.-Y. M. Liao. 2020. YOLOv4: optimal speed and accuracy of object detection. Available at http://arxiv.org/abs/2004.10934 (verified 20 March 2025). Campbell D.L.M., De Haas E.N., Lee C. A review of environmental enrichment for laying hens during rearing in relation to their behavioral and physiological development. Poult. Sci. 2019;98:9–28. doi: 10.3382/ps/pey319. https://linkinghub.elsevier.com/retrieve/pii/S0032579119302822 Available at. verified 24 November 2025. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Chemme K.S., Alitappeh R.J. Proceedings of the 2024 13th Iranian/3rd International Machine Vision and Image Processing Conference (MVIP) IEEE; Tehran, Iran, Islamic Republic of: 2024. An end-to-end model for chicken detection in a cluttered environment; pp. 1–7. [ Google Scholar ] Cui Y., Kong X., Chen C., Li Y. Research on broiler health status recognition method based on improved YOLOv5. Smart Agric. Technol. 2023;6 https://linkinghub.elsevier.com/retrieve/pii/S2772375523001533 Available at. verified 6 November 2024. [ Google Scholar ] Dawkins M.S. Animal welfare and efficient farming: is conflict inevitable? Anim. Prod. Sci. 2017;57:201. http://www.publish.csiro.au/?paper=AN15383 Available at. verified 14 March 2025. [ Google Scholar ] Dawson M.D., Lombardi M..E., Benson E.R., Alphin R.L., Malone G.W. Using accelerometers to determine the cessation of activity of broilers. J. Appl. Poult. Res. 2007;16:583–591. https://linkinghub.elsevier.com/retrieve/pii/S1056617119316356 Available at. verified 6 January 2025. [ Google Scholar ] De Jong I.C., Hindle V..A., Butterworth A., Engel B., Ferrari P., Gunnink H., Moya T.P., Tuyttens F.A.M., Van Reenen C.G. Simplifying the Welfare Quality® assessment protocol for broiler chicken welfare. Animal. 2016;10:117–127. doi: 10.1017/S1751731115001706. https://linkinghub.elsevier.com/retrieve/pii/S1751731115001706 Available at. verified 19 February 2025. [ DOI ] [ PubMed ] [ Google Scholar ] Degu M.Z., Simegn G.L. Smartphone based detection and classification of poultry diseases from chicken fecal images using deep learning techniques. Smart Agric. Technol. 2023;4 https://linkinghub.elsevier.com/retrieve/pii/S2772375523000515 Available at. verified 8 November 2024. [ Google Scholar ] Deo N., Bakliwal A., Joshi A.D., Sawant S.T. Non-intrusive detection of poultry diseases through faecal analysis using deep learning. Proceedings of the 2024 5th International Conference on Innovative Trends in Information Technology (ICITIIT); Kottayam, India; IEEE; 2024. pp. 1–6. [ Google Scholar ] Depuru B.K., Putsala S.., Mishra P. Automating poultry farm management with artificial intelligence: real-time detection and tracking of broiler chickens for enhanced and efficient health monitoring. Trop. Anim. Health Prod. 2024;56:75. doi: 10.1007/s11250-024-03922-2. https://link.springer.com/10.1007/s11250-024-03922-2 Available at. verified 6 November 2024. [ DOI ] [ PubMed ] [ Google Scholar ] Doornweerd J.E., Veerkamp R..F., De Klerk B., Van Der Sluis M., Bouwman A.C., Ellen E.D., Kootstra G. Tracking individual broilers on video in terms of time and distance. Poult. Sci. 2024;103 doi: 10.1016/j.psj.2023.103185. https://linkinghub.elsevier.com/retrieve/pii/S0032579123007046 Available at. verified 6 November 2024. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Ehsan, T. Z., .and S. M. Mohtavipour. 2024. Broiler-Net: a deep convolutional framework for broiler behavior analysis in poultry houses. Available at https://arxiv.org/abs/2401.12176 (verified 6 November 2024). Elbarrany A.M., Mohialdin A.., Atia A. The use of pose estimation for abnormal behavior analysis in poultry farms. Proceedings of the 2023 5th Novel Intelligent and Leading Emerging Sciences Conference (NILES); Giza, Egypt; IEEE; 2023. pp. 33–36. [ Google Scholar ] Elmessery W.M., Gutiérrez J.., Abd El-Wahhab G.G., Elkhaiat I.A., El-Soaly I.S., Alhag S.K., Al-Shuraym L.A., Akela M.A., Moghanm F.S., Abdelshafie M.F. YOLO-based model for automatic detection of broiler pathological phenomena through visual and thermal images in intensive poultry houses. Agriculture. 2023;13:1527. https://www.mdpi.com/2077-0472/13/8/1527 Available at. verified 5 February 2024. [ Google Scholar ] Estevez I., Andersen I.-L., Nævdal E. Group size, density and social dynamics in farm animals. Appl. Anim. Behav. Sci. 2007;103:185–204. https://linkinghub.elsevier.com/retrieve/pii/S0168159106001936 Available at. verified 14 March 2025. [ Google Scholar ] Gong Z., Wang M., Song J. FEDM: a convolutional neural network based fertilised egg detection model. Br. Poult. Sci. 2024;65:546–558. doi: 10.1080/00071668.2024.2356656. https://www.tandfonline.com/doi/full/10.1080/00071668.2024.2356656 Available at. verified 8 November 2024. [ DOI ] [ PubMed ] [ Google Scholar ] Guo Y., Chai L., Aggrey S.E., Oladeinde A., Johnson J., Zock G. A machine vision-based method for monitoring broiler chicken floor distribution. Sensors. 2020;20:3179. doi: 10.3390/s20113179. https://www.mdpi.com/1424-8220/20/11/3179 Available at. verified 8 December 2023. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Guo Y., Regmi P., Ding Y., Bist R.B., Chai L. Automatic detection of brown hens in cage-free houses with deep learning methods. Poult. Sci. 2023;102 doi: 10.1016/j.psj.2023.102784. https://linkinghub.elsevier.com/retrieve/pii/S0032579123003036 Available at. verified 23 January 2024. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Guo Y., Regmi P., Ding Y., Bist R.B., Chai L. Automatic detection of brown hens in cage-free houses with deep learning methods. Poult. Sci. 2023;102 doi: 10.1016/j.psj.2023.102784. https://linkinghub.elsevier.com/retrieve/pii/S0032579123003036 Available at. verified 1 July 2025. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Hao H., Fang P., Duan E., Yang Z., Wang L., Wang H. A dead broiler inspection system for large-scale breeding farms based on deep learning. Agriculture. 2022;12:1176. https://www.mdpi.com/2077-0472/12/8/1176 Available at. verified 6 November 2024. [ Google Scholar ] Hao H., Fang P., Jiang W., Sun X., Wang L., Wang H. Research on laying hens feeding behavior detection and model visualization based on convolutional neural network. Agriculture. 2022;12:2141. https://www.mdpi.com/2077-0472/12/12/2141 Available at. verified 8 November 2024. [ Google Scholar ] Harshitha S., Likhitha K., Varshini K.R.G., Mahure S.J., Yasaswi A. Poultry pathogen detection: using deep learning for coccidiosis identification. Proceedings of the 2024 5th International Conference on Image Processing and Capsule Networks (ICIPCN); Dhulikhel, Nepal; IEEE; 2024. pp. 178–182. [ Google Scholar ] Hewawasam T., Gunasekara J., Jayarathna N., Indrawansha Y., Rathnayake S., Panduwawala P. Integrated AI-based system for comprehensive poultry management. Proceedings of the 2023 5th International Conference on Advancements in Computing (ICAC); Colombo, Sri Lanka; IEEE; 2023. pp. 382–387. [ Google Scholar ] Hewawasam T., Rathnayake S., Panduwawala P. Accurate poultry disease identification and risk assessment through object detection. Proceedings of the 2023 5th International Conference on Advancements in Computing (ICAC); Colombo, Sri Lanka; IEEE; 2023. pp. 591–596. [ Google Scholar ] Hu Y., Xiong J., Xu J., Gou Z., Ying Y., Pan J., Cui D. Behavior recognition of cage-free multi-broilers based on spatiotemporal feature learning. Poult. Sci. 2024;103 doi: 10.1016/j.psj.2024.104314. https://linkinghub.elsevier.com/retrieve/pii/S0032579124008939 Available at. verified 6 November 2024. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Janczak A.M., Riber A.B. Review of rearing-related factors affecting the welfare of laying hens. Poult. Sci. 2015;94:1454–1469. doi: 10.3382/ps/pev123. https://linkinghub.elsevier.com/retrieve/pii/S0032579119325076 Available at. verified 24 November 2025. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Jocher, G. 2020. Ultralytics YOLOv5. Available at https://github.com/ultralytics/yolov5 . Joo, K.H., Duan, S., Weimer, S.L., Teli, M.N. (2022). Birds' Eye View: Measuring Behavior and Posture of Chickens as a Metric for Their Well-Being. arXiv preprint arXiv: 2205.00069 . Khairunissa J., Wahjuni S., Soesanto I.R.H., Wulandari W. Detecting poultry movement for poultry behavioral analysis using the multi-object tracking (MOT) algorithm. Proceedings of the 2021 8th International Conference on Computer and Communication Engineering (ICCCE); Kuala Lumpur, Malaysia; IEEE; 2021. pp. 265–268. [ Google Scholar ] Khan I., Peralta D., Fontaine J., Soster De Carvalho P., Martinez-Caja A.M., Antonissen G., Tuyttens F., De Poorter E. Monitoring welfare of individual broiler chickens using ultra-wideband and inertial measurement unit wearables. Sensors. 2025;25:811. doi: 10.3390/s25030811. https://www.mdpi.com/1424-8220/25/3/811 Available at. verified 11 December 2025. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Kodaira V., Siriani A.L.R., Medeiros H.P., De Moura D.J., Pereira D.F. Assessment of preference behavior of layer hens under different light colors and temperature environments in long-time footage using a computer vision system. Animals. 2023;13:2426. doi: 10.3390/ani13152426. https://www.mdpi.com/2076-2615/13/15/2426 Available at. verified 8 November 2024. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Li N., Ren Z., Li D., Zeng L. Review: automated techniques for monitoring the behaviour and welfare of broilers and laying hens: towards the goal of precision livestock farming. Animal. 2020;14:617–625. doi: 10.1017/S1751731119002155. https://linkinghub.elsevier.com/retrieve/pii/S1751731119002155 Available at. verified 1 January 2025. [ DOI ] [ PubMed ] [ Google Scholar ] Li X., Zhao Z., Wu J., Huang Y., Wen J., Sun S., Xie H., Sun J., Gao Y. Y-BGD: broiler counting based on multi-object tracking. Comput. Electron. Agric. 2022;202 https://linkinghub.elsevier.com/retrieve/pii/S016816992200655X Available at. verified 6 November 2024. [ Google Scholar ] Li G., Li B., Shi Z., Lu G., Chai L., Rasheed K.M., Regmi P., Banakar A. Interindividual distances and orientations of laying hens under 8 stocking densities measured by integrative deep learning techniques. Poult. Sci. 2023;102 doi: 10.1016/j.psj.2023.103076. https://linkinghub.elsevier.com/retrieve/pii/S0032579123005953 Available at. verified 9 July 2024. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Li, Y., W. Xia, and X. Yu. 2023b. Research on a deep learning-based epidemic surveillance system for live poultry transport. 60:813–821. Liu H.-W., Chen C.-H., Tsai Y.-C., Hsieh K.-W., Lin H.-T. Identifying images of dead chickens with a chicken removal system integrated with a deep learning algorithm. Sensors. 2021;21:3579. doi: 10.3390/s21113579. https://www.mdpi.com/1424-8220/21/11/3579 Available at. verified 6 November 2024. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Liu, Y., H. Zhou, Z. Ni, Z. Jiang, and X. Wang. 2024. An accurate and lightweight algorithm for caged chickens detection based on deep learning. 61. Luo S., Ma Y., Jiang F., Wang H., Tong Q., Wang L. Dead laying hens detection using TIR-NIR-depth images and deep learning on a commercial farm. Animals. 2023;13:1861. doi: 10.3390/ani13111861. https://www.mdpi.com/2076-2615/13/11/1861 Available at. verified 8 November 2024. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Machuve D., Nwankwo E., Mduma N., Mbelwa J. Poultry diseases diagnostics models using deep learning. Front. Artif. Intell. 2022;5 doi: 10.3389/frai.2022.733345. https://www.frontiersin.org/articles/10.3389/frai.2022.733345/full Available at. verified 8 November 2024. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Manikandan V., Neethirajan S. AI-powered vocalization analysis in poultry: systematic review of health, behavior, and welfare monitoring. Sensors. 2025;25:4058. doi: 10.3390/s25134058. https://www.mdpi.com/1424-8220/25/13/4058 Available at. verified 29 September 2025. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Mehdizadeh S.A., Siriani A.L.R., Pereira D.F. Optimizing deep learning algorithms for effective chicken tracking through image processing. AgriEngineering. 2024;6:2749–2767. https://www.mdpi.com/2624-7402/6/3/160 Available at. verified 6 November 2024. [ Google Scholar ] Nakrosis A., Paulauskaite-Taraseviciene A., Raudonis V., Narusis I., Gruzauskas V., Gruzauskas R., Lagzdinyte-Budnike I. Towards early poultry health prediction through non-invasive and computer vision-based dropping classification. Animals. 2023;13:3041. doi: 10.3390/ani13193041. https://www.mdpi.com/2076-2615/13/19/3041 Available at. verified 9 July 2024. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Nasiri A., Amirivojdan A., Zhao Y., Gan H. Estimating the feeding time of individual broilers via convolutional neural network and image processing. Animals. 2023;13:2428. doi: 10.3390/ani13152428. https://www.mdpi.com/2076-2615/13/15/2428 Available at. verified 6 November 2024. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Nasiri A., Zhao Y., Gan H. Automated detection and counting of broiler behaviors using a video recognition system. Comput. Electron. Agric. 2024;221 https://linkinghub.elsevier.com/retrieve/pii/S0168169924003211 Available at. verified 6 November 2024. [ Google Scholar ] Neethirajan S., Kemp B. Digital livestock farming. Sens. Biosens. Res. 2021;32 https://linkinghub.elsevier.com/retrieve/pii/S2214180421000131 Available at. verified 28 September 2025. [ Google Scholar ] Neethirajan S. ChickTrack – A quantitative tracking tool for measuring chicken activity. Measurement. 2022;191 https://linkinghub.elsevier.com/retrieve/pii/S0263224122001154 Available at. verified 8 December 2023. [ Google Scholar ] Newberry R.C. Exploratory behaviour of young domestic fowl. Appl. Anim. Behav. Sci. 1999;63:311–321. doi: 10.1016/s0168-1591(01)00135-6. https://linkinghub.elsevier.com/retrieve/pii/S0168159199000167 Available at. verified 24 November 2025. [ DOI ] [ PubMed ] [ Google Scholar ] Nguyen K.H., Nguyen H.V.N., Tran H.N., Quach L.-D. Combining autoencoder and Yolov6 model for classification and disease detection in chickens. Proceedings of the 2023 8th International Conference on Intelligent Information Technology; Da Nang Vietnam; ACM; 2023. pp. 132–138. [ Google Scholar ] Paneru B., Bist R., Yang X., Chai L. Tracking perching behavior of cage-free laying hens with deep learning technologies. Poult. Sci. 2024 doi: 10.1016/j.psj.2024.104281. https://linkinghub.elsevier.com/retrieve/pii/S0032579124008605 Available at. verified 1 September 2024. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Paneru B., Bist R., Yang X., Chai L. Tracking dustbathing behavior of cage-free laying hens with machine vision technologies. Poult. Sci. 2024 doi: 10.1016/j.psj.2024.104289. https://linkinghub.elsevier.com/retrieve/pii/S003257912400868X Available at. verified 3 September 2024. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Pearce J., Chang Y.-M., Abeyesinghe S. Individual monitoring of activity and lameness in conventional and slower-growing breeds of broiler chickens using accelerometers. Animals. 2023;13:1432. doi: 10.3390/ani13091432. https://www.mdpi.com/2076-2615/13/9/1432 Available at. verified 6 January 2025. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Qin W., Yang X., Liu C., Zheng W. A deep learning method based on YOLOv5 and Super Point-Super Glue for digestive disease warning and cage location backtracking in stacked cage laying hen systems. Comput. Electron. Agric. 2024;222 https://linkinghub.elsevier.com/retrieve/pii/S0168169924003909 Available at. verified 8 November 2024. [ Google Scholar ] Redmon J., Farhadi A. YOLO9000: better, faster, stronger. Proceedings of the 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR); Honolulu, HI; IEEE; 2017. pp. 6517–6525. [ Google Scholar ] Redmon J., Divvala S., Girshick R., Farhadi A. You only look once: unified, real-time object detection. Proceedings of the 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR); Las Vegas, NV, USA; IEEE; 2016. pp. 779–788. [ Google Scholar ] Riber A.B., Van De Weerd H..A., De Jong I.C., Steenfeldt S. Review of environmental enrichment for broiler chickens. Poult. Sci. 2018;97:378–396. doi: 10.3382/ps/pex344. https://linkinghub.elsevier.com/retrieve/pii/S0032579119308879 Available at. verified 14 March 2025. [ DOI ] [ PubMed ] [ Google Scholar ] Ringgenberg N., Fröhlich E.K.F., Harlander-Matauschek A., Toscano M.J., Würbel H., Roth B.A. Effects of variation in nest curtain design on pre-laying behaviour of domestic hens. Appl. Anim. Behav. Sci. 2015;170:34–43. https://linkinghub.elsevier.com/retrieve/pii/S0168159115001732 Available at. verified 19 February 2025. [ Google Scholar ] Rowe E., Dawkins M.S., Gebhardt-Henrich S.G. A systematic review of precision livestock farming in the poultry sector: is technology focussed on improving bird welfare? Animals. 2019;9:614. doi: 10.3390/ani9090614. https://www.mdpi.com/2076-2615/9/9/614 Available at. verified 1 January 2025. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Saeidifar M., Li G., Chai L., Bist R., Rasheed K.M., Lu J., Banakar A., Liu T., Yang X. Zero-shot image segmentation for monitoring thermal conditions of individual cage-free laying hens. Comput. Electron. Agric. 2024;226 https://linkinghub.elsevier.com/retrieve/pii/S0168169924008275 Available at. verified 15 September 2024. [ Google Scholar ] Sakib M., Alam S.S., Atul S.M., Supto F.H., Hossain M.M. AIP Conference Proceedings. Vol. 3245. AIP Publishing LLC; 2024. Automated detection of broiler chicken behaviors through the integration of 3D accelerometer sensor and machine learning techniques; p. 020001. [ Google Scholar ] Schiele T., Kern D., Klauck U. Lazy labels for chicken segmentation. Procedia Comput. Sci. 2023;225:2664–2673. https://linkinghub.elsevier.com/retrieve/pii/S1877050923014163 Available at. verified 9 May 2025. [ Google Scholar ] Sih A., Bell A.M., Johnson J.C., Ziemba R.E. Behavioral syndromes: an integrative overview. Q. Rev. Biol. 2004;79:241–277. doi: 10.1086/422893. https://www.journals.uchicago.edu/doi/10.1086/422893 Available at. verified 19 February 2025. [ DOI ] [ PubMed ] [ Google Scholar ] Siriani A.L.R., Kodaira V., Mehdizadeh S.A., De Alencar Nääs I., De Moura D.J., Pereira D.F. Detection and tracking of chickens in low-light images using YOLO network and Kalman filter. Neural Comput. Appl. 2022;34:21987–21997. https://link.springer.com/10.1007/s00521-022-07664-w Available at. verified 6 November 2024. [ Google Scholar ] Sozzi M., Pillan G., Ciarelli C., Marinello F., Pirrone F., Bordignon F., Bordignon A., Xiccato G., Trocino A. Measuring comfort behaviours in laying hens using deep-learning tools. Animals. 2022;13:33. doi: 10.3390/ani13010033. https://www.mdpi.com/2076-2615/13/1/33 Available at. verified 23 January 2024. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Subedi S., Bist R., Yang X., Chai L. Tracking pecking behaviors and damages of cage-free laying hens with machine vision technologies. Comput. Electron. Agric. 2023;204 https://linkinghub.elsevier.com/retrieve/pii/S0168169922008535 Available at. verified 23 January 2024. [ Google Scholar ] Subedi S., Bist R., Yang X., Chai L. Tracking pecking behaviors and damages of cage-free laying hens with machine vision technologies. Comput. Electron. Agric. 2023;204 https://linkinghub.elsevier.com/retrieve/pii/S0168169922008535 Available at. verified 1 July 2025. [ Google Scholar ] Subedi S., Bist R., Yang X., Chai L. Tracking floor eggs with machine vision in cage-free hen houses. Poult. Sci. 2023;102 doi: 10.1016/j.psj.2023.102637. https://linkinghub.elsevier.com/retrieve/pii/S003257912300161X Available at. verified 23 January 2024. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Subedi S., Bist R., Yang X., Chai L. Tracking floor eggs with machine vision in cage-free hen houses. Poult. Sci. 2023;102 doi: 10.1016/j.psj.2023.102637. https://linkinghub.elsevier.com/retrieve/pii/S003257912300161X Available at. verified 1 July 2025. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Subedi S., Bist R.B., Yang X., Li G., Chai L. Advanced deep learning methods for multiple behavior classification of cage-free laying hens. AgriEngineering. 2025;7:24. https://www.mdpi.com/2624-7402/7/2/24 Available at. verified 22 March 2025. [ Google Scholar ] Sun S., Yang J., Chen Z., Li J., Sun R. Tibia-YOLO: an AssistedDetection system combined with industrial CT equipment for leg diseases in broilers. Appl. Sci. 2024;14:1005. https://www.mdpi.com/2076-3417/14/3/1005 Available at. verified 6 November 2024. [ Google Scholar ] Syafaah L., Faruq A., Setyawan N., Khair M.I. Sick and dead chicken detection system based on YOLO algorithm. ISI. 2024;29:1723–1729. https://www.iieta.org/journals/isi/paper/10.18280/isi.290506 Available at. verified 5 May 2025. [ Google Scholar ] Tong Q., Zhang E., Wu S., Xu K., Sun C. A real-time detector of chicken healthy status based on modified YOLO. SIViP. 2023;17:4199–4207. https://link.springer.com/10.1007/s11760-023-02652-6 Available at. verified 6 November 2024. [ Google Scholar ] Triyanto W.A., Adi K.., Suseno J.E. Detection and tracking of broiler flock movements in the chicken coop using YOLO (R Isnanto, Hadiyanto, and B Warsito, Eds.) E3S Web Conf. 2023;448 https://www.e3s-conferences.org/10.1051/e3sconf/202344802064 Available at. verified 6 November 2024. [ Google Scholar ] Tullo E., Fontana I., Diana A., Norton T., Berckmans D., Guarino M. Application note: labelling, a methodology to develop reliable algorithm in PLF. Comput. Electron. Agric. 2017;142:424–428. https://linkinghub.elsevier.com/retrieve/pii/S0168169917308256 Available at. verified 19 February 2025. [ Google Scholar ] Ty W., Th L., Yc T. Comb color analysis of broilers through the video surveillance system of a poultry house. Braz. J. Poult. Sci. 2024;26 http://www.scielo.br/scielo.php?script=sci_arttext&pid=S1516-635X2024000100305&tlng=en eRBCA-2023-1891 Available at. verified 13 November 2025. [ Google Scholar ] Uddin M.S., Islam M..F., Rahman S., Foysal N.U., Islam M.N., Razzaque M.d.A., Al Faisal F. A novel hybrid deep neural network for early detection and classification of chicken diseases. Proceedings of the 2023 5th International Conference on Sustainable Technologies for Industry 5.0 (STI); Dhaka, Bangladesh; IEEE; 2023. pp. 1–6. [ Google Scholar ] Wahjuni S., Wulandari W., Eknanda R.T.G., Susanto I.R.H., Akbar A.R. Automatic detection of broiler’s feeding and aggressive behavior using you only look once algorithm. IJ-AI. 2024;13:104. https://ijai.iaescore.com/index.php/IJAI/article/view/22664 Available at. verified 6 November 2024. [ Google Scholar ] Wang J., Shen M., Liu L., Xu Y., Okinda C. Recognition and classification of broiler droppings based on deep convolutional neural network. J. Sens. 2019;2019:1–10. https://www.hindawi.com/journals/js/2019/3823515/ Available at. verified 6 November 2024. [ Google Scholar ] Wang J., Wang N., Li L., Ren Z. Real-time behavior detection and judgment of egg breeders based on YOLO v3. Neural Comput. Appl. 2020;32:5471–5481. http://link.springer.com/10.1007/s00521-019-04645-4 Available at. verified 9 July 2024. [ Google Scholar ] Wang, C.-Y., A. Bochkovskiy, and H.-Y. M. Liao. 2022. YOLOv7: trainable bag-of-freebies sets new state-of-the-art for real-time object detectors. Available at http://arxiv.org/abs/2207.02696 (verified 23 January 2024). Wang F., Cui J., Xiong Y., Lu H. Application of deep learning methods in behavior recognition of laying hens. Front. Phys. 2023;11 https://www.frontiersin.org/articles/10.3389/fphy.2023.1139976/full Available at. verified 6 November 2024. [ Google Scholar ] Welfare Quality ® . Welfare Quality® Consortium; Lelystad, Netherlands: 2009. Welfare Quality® Assessment Protocol for Poultry (Broilers, Laying Hens) Internet] http://www.welfarequality.net/media/1293/poultry-protocol-watermark-6-2-2020.pdf updated 2009: cited 2026 March 11]. Available from: [ Google Scholar ] Wu Z., Zhang T., Fang C., Yang J., Ma C., Zheng H., Zhao H. Super-resolution fusion optimization for poultry detection: a multi-object chicken detection method. J. Anim. Sci. 2023;101:skad249. doi: 10.1093/jas/skad249. https://academic.oup.com/jas/article/doi/10.1093/jas/skad249/7230776 Available at. verified 8 November 2024. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Wu R., He P., He Y., Dou J., Di M., He S., Hayat K., Zhou Y., Yu L., Pan J., Lin H. Egg production monitoring in commercial laying cages via the StrongSort-EGG tracking-by-detection model. Comput. Electron. Agric. 2024;227 https://linkinghub.elsevier.com/retrieve/pii/S0168169924008998 Available at. verified 8 November 2024. [ Google Scholar ] Wu D., Ying Y., Zhou M., Pan J., Cui D. YOLO-claw: a fast and accurate method for chicken claw detection. Eng. Appl. Artif. Intell. 2024;136 https://linkinghub.elsevier.com/retrieve/pii/S0952197624010777 Available at. verified 6 November 2024. [ Google Scholar ] Xia Y., Xue H., Lu S., Wang L., Li L. Behavior detection algorithm of Caged White-feather broiler based on multi-scale detail feature fusion and object relation inference. Proceedings of the 2023 IEEE 35th International Conference on Tools with Artificial Intelligence (ICTAI); Atlanta, GA, USA; IEEE; 2023. pp. 1002–1006. [ Google Scholar ] Xin C., Li H., Li Y., Wang M., Lin W., Wang S., Zhang W., Xiao M., Zou X. Research on an identification and grasping device for dead yellow-feather broilers in flat houses based on deep learning. Agriculture. 2024;14:1614. https://www.mdpi.com/2077-0472/14/9/1614 Available at. verified 6 November 2024. [ Google Scholar ] Yang X., Zhao Y., Street G.M., Huang Y., Filip To S.D., Purswell J.L. Classification of broiler behaviours using triaxial accelerometer and machine learning. Animal. 2021;15 doi: 10.1016/j.animal.2021.100269. https://linkinghub.elsevier.com/retrieve/pii/S1751731121001117 Available at. verified 31 January 2025. [ DOI ] [ PubMed ] [ Google Scholar ] Yang X., Chai L., Bist R.B., Subedi S., Wu Z. A deep learning model for detecting cage-free hens on the litter floor. Animals. 2022;12:1983. doi: 10.3390/ani12151983. https://www.mdpi.com/2076-2615/12/15/1983 Available at. verified 23 January 2024. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Yang X., Bist R., Subedi S., Chai L. A deep learning method for monitoring spatial distribution of cage-free hens. Artif. Intell. Agric. 2023;8:20–29. https://linkinghub.elsevier.com/retrieve/pii/S2589721723000120 Available at. verified 23 January 2024. [ Google Scholar ] Yang X., Bist R., Subedi S., Chai L. A deep learning method for monitoring spatial distribution of cage-free hens. Artif. Intell. Agric. 2023;8:20–29. https://linkinghub.elsevier.com/retrieve/pii/S2589721723000120 Available at. verified 1 July 2025. [ Google Scholar ] Yang X., Bist R., Subedi S., Wu Z., Liu T., Chai L. An automatic classifier for monitoring applied behaviors of cage-free laying hens with deep learning. Eng. Appl. Artif. Intell. 2023;123 https://linkinghub.elsevier.com/retrieve/pii/S0952197623005614 Available at. verified 23 January 2024. [ Google Scholar ] Yang J., Zhang T., Fang C., Zheng H. A defencing algorithm based on deep learning improves the detection accuracy of caged chickens. Comput. Electron. Agric. 2023;204 https://linkinghub.elsevier.com/retrieve/pii/S0168169922008092 Available at. verified 8 November 2024. [ Google Scholar ] Yang X., Bist R.B., Paneru B., Chai L. Deep learning methods for tracking the locomotion of individual chickens. Animals. 2024;14:911. doi: 10.3390/ani14060911. https://www.mdpi.com/2076-2615/14/6/911 Available at. verified 9 July 2024. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Yang X., Bist R., Paneru B., Chai L. Monitoring activity index and behaviors of cage-free hens with advanced deep learning technologies. Poult. Sci. 2024;103 doi: 10.1016/j.psj.2024.104193. https://linkinghub.elsevier.com/retrieve/pii/S0032579124007727 Available at. verified 28 August 2024. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Yang X., Bist R.B., Subedi S., Wu Z., Liu T., Paneru B., Chai L. A machine vision system for monitoring wild birds on poultry farms to prevent avian influenza. AgriEngineering. 2024;6:3704–3718. https://www.mdpi.com/2624-7402/6/4/211 Available at. verified 8 November 2024. [ Google Scholar ] Yang J., Zhang T., Fang C., Zheng H., Ma C., Wu Z. A detection method for dead caged hens based on improved YOLOv7. Comput. Electron. Agric. 2024;226 https://linkinghub.elsevier.com/retrieve/pii/S0168169924007798 Available at. verified 8 November 2024. [ Google Scholar ] Ye C., Yu Z., Kang R., Yousaf K., Qi C., Chen K., Huang Y. An experimental study of stunned state detection for broiler chickens using an improved convolution neural network algorithm. Comput. Electron. Agric. 2020;170 https://linkinghub.elsevier.com/retrieve/pii/S0168169919313171 Available at. verified 6 November 2024. [ Google Scholar ] Yu Z., Liu L., Jiao H., Chen J., Chen Z., Song Z., Lin H., Tian F. Leveraging SOLOv2 model to detect heat stress of poultry in complex environments. Front. Vet. Sci. 2023;9 doi: 10.3389/fvets.2022.1062559. https://www.frontiersin.org/articles/10.3389/fvets.2022.1062559/full Available at. verified 8 November 2024. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Zhang X., Zhu R., Zheng W., Chen C. Detection of leg diseases in broiler chickens based on improved YOLOv8 X-ray images. IEEE Access. 2024;12:47385–47401. https://ieeexplore.ieee.org/document/10479196/ Available at. verified 6 November 2024. [ Google Scholar ] Zhou, Z., L. Li, H. Xue, Y. Jia, Y. Yu, Z. Xie, and Y. Gu. 2024. A method for detecting chicken open-beak behavior based on an improved Yolov8 model. Available at https://www.ssrn.com/abstract=4860824 (verified 8 November 2024). Zou X., Yin Z., Li Y., Gong F., Bai Y., Zhao Z., Zhang W., Qian Y., Xiao M. Novel multiple object tracking method for yellow feather broilers in a flat breeding chamber based on improved YOLOv3 and deep SORT. Int. J. Agric. Biol. Eng. 2023;16:44–55. http://www.ijabe.org/index.php/ijabe/article/view/7836 Available at. verified 6 November 2024. [ Google Scholar ] Articles from Poultry Science are provided here courtesy of Elsevier ACTIONS View on publisher site PDF (3.8 MB) Cite Collections Permalink PERMALINK Copy RESOURCES Similar articles Cited by other articles Links to NCBI Databases Cite Copy Download .nbib .nbib Format: AMA APA MLA NLM Add to Collections Create a new collection Add to an existing collection Name your collection * Choose a collection Unable to load your collection due to an error Please try again Add Cancel Follow NCBI NCBI on X (formerly known as Twitter) NCBI on Facebook NCBI on LinkedIn NCBI on GitHub NCBI RSS feed Connect with NLM NLM on X (formerly known as Twitter) NLM on Facebook NLM on YouTube National Library of Medicine 8600 Rockville Pike Bethesda, MD 20894 Web Policies FOIA HHS Vulnerability Disclosure Help Accessibility Careers NLM NIH HHS USA.gov Back to Top