Precision livestock farming: an overview on the application in extensive systems

Italian Journal of Animal Science, Journal Year: 2025, Volume and Issue: 24(1), P. 859 - 884

Published: March 24, 2025

Language: Английский

---

A New Method for Non-Destructive Identification and Tracking of Multi-Object Behaviors in Beef Cattle Based on Deep Learning

Animals, Journal Year: 2024, Volume and Issue: 14(17), P. 2464 - 2464

Published: Aug. 24, 2024

The method proposed in this paper provides theoretical and practical support for the intelligent recognition management of beef cattle. Accurate identification tracking cattle behaviors are essential components production management. Traditional methods time-consuming labor-intensive, which hinders precise farming. This utilizes deep learning algorithms to achieve multi-object cattle, as follows: (1) behavior detection module is based on YOLOv8n algorithm. Initially, a dynamic snake convolution introduced enhance ability extract key features expand model's receptive field. Subsequently, BiFormer attention mechanism incorporated integrate high-level low-level feature information, dynamically sparsely behavioral improved YOLOv8n_BiF_DSC algorithm achieves an accuracy 93.6% nine behaviors, including standing, lying, mounting, fighting, licking, eating, drinking, working, searching, with average 50 50:95 precisions 96.5% 71.5%, showing improvement 5.3%, 5.2%, 7.1% over original YOLOv8n. (2) Deep SORT detector replaced accuracy. re-identification network model switched ResNet18 algorithm's capability gather appearance information. Finally, trajectory generation matching process optimized secondary IOU reduce ID mismatching errors during tracking. Experimentation five different complexity levels test video sequences shows improvements IDF1, IDS, MOTA, MOTP, among other metrics, IDS reduced by 65.8% MOTA increased 2%. These enhancements address issues omission misidentification sparse long-range dense environments, thereby facilitating better group-raised laying foundation

Language: Английский

---

A Novel Fusion Perception Algorithm of Tree Branch/Trunk and Apple for Harvesting Robot Based on Improved YOLOv8s

Agronomy, Journal Year: 2024, Volume and Issue: 14(9), P. 1895 - 1895

Published: Aug. 24, 2024

Aiming to accurately identify apple targets and achieve segmentation the extraction of branch trunk areas trees, providing visual guidance for a picking robot actively adjust its posture avoid trunks obstacle avoidance fruit picking, spindle-shaped which are widely planted in standard modern orchards, were focused on, an algorithm tree detection robots was proposed based on improved YOLOv8s model design. Firstly, image data trees orchards collected, annotations object pixel-level conducted data. Training set then augmented improve generalization performance algorithm. Secondly, original network architecture’s design by embedding SE module attention mechanism after C2f Backbone architecture. Finally, dynamic snake convolution embedded into Neck structure architecture better extract feature information different branches. The experimental results showed that can effectively recognize images segment branches trunks. For recognition, precision 99.6%, recall 96.8%, mAP value 98.3%. 81.6%. compared with YOLOv8s, YOLOv8n, YOLOv5s algorithms recognition test images. other three algorithms, increased 1.5%, 2.3%, 6%, respectively. 3.7%, 15.4%, 24.4%, fruits, branches, is great significance ensuring success rate harvesting, provide technical support development intelligent harvesting robot.

Language: Английский

---

FireNet: A Lightweight and Efficient Multi-Scenario Fire Object Detector

Remote Sensing, Journal Year: 2024, Volume and Issue: 16(21), P. 4112 - 4112

Published: Nov. 4, 2024

Fire and smoke detection technologies face challenges in complex dynamic environments. Traditional detectors are vulnerable to background noise, lighting changes, similar objects (e.g., clouds, steam, dust), leading high false alarm rates. Additionally, they struggle with detecting small objects, limiting their effectiveness early fire warnings rapid responses. As real-time monitoring demands grow, traditional methods often fall short smart city drone applications. To address these issues, we propose FireNet, integrating a simplified Vision Transformer (RepViT) enhance global feature learning while reducing computational overhead. Dynamic snake convolution (DSConv) captures fine boundary details of flames smoke, especially curved edges. A lightweight decoupled head optimizes classification localization, ideal for inter-class similarity targets. FireNet outperforms YOLOv8 on the Scene dataset (FSD) [email protected] 80.2%, recall 78.4%, precision 82.6%, an inference time 26.7 ms. It also excels FSD dataset, addressing current challenges.

Language: Английский

---

Research on Cattle Behavior Recognition and Multi-Object Tracking Algorithm Based on YOLO-BoT

Animals, Journal Year: 2024, Volume and Issue: 14(20), P. 2993 - 2993

Published: Oct. 17, 2024

In smart ranch management, cattle behavior recognition and tracking play a crucial role in evaluating animal welfare. To address the issues of missed false detections caused by inter-cow occlusions infrastructure obstructions barn environment, this paper proposes multi-object method called YOLO-BoT. Built upon YOLOv8, first integrates dynamic convolution (DyConv) to enable adaptive weight adjustments, enhancing detection accuracy complex environments. The C2f-iRMB structure is then employed improve feature extraction efficiency, ensuring capture essential features even under or lighting variations. Additionally, Adown downsampling module incorporated strengthen multi-scale information fusion, head (DyHead) used robustness boxes, precise identification rapidly changing target positions. further enhance performance, DIoU distance calculation, confidence-based bounding box reclassification, virtual trajectory update mechanism are introduced, accurate matching occlusion minimizing identity switches. Experimental results demonstrate that YOLO-BoT achieves mean average precision (mAP) 91.7% detection, with recall increased 4.4% 1%, respectively. Moreover, proposed improves higher order (HOTA), (MOTA), (MOTP), IDF1 4.4%, 7%, 1.7%, 4.3%, respectively, while reducing switch rate (IDS) 30.9%. tracker operates real-time at an speed 31.2 fps, significantly performance scenarios providing strong support for long-term analysis contactless automated monitoring.

Language: Английский

---

A Review of You Only Look Once Algorithms in Animal Phenotyping Applications

Animals, Journal Year: 2025, Volume and Issue: 15(8), P. 1126 - 1126

Published: April 13, 2025

Animal phenotyping recognition is a pivotal component of precision livestock management, holding significant importance for intelligent farming practices and animal welfare assurance. In recent years, with the rapid advancement deep learning technologies, YOLO algorithm—as pioneering single-stage detection framework—has revolutionized field object through its efficient approach has been widely applied across various agricultural domains. This review focuses on as research target structured around four key aspects: (1) evolution algorithms, (2) datasets preprocessing methodologies, (3) application domains (4) future directions. paper aims to offer readers fresh perspectives insights into research.

Language: Английский

---

AI-Based Smart Monitoring Framework for Livestock Farms

Applied Sciences, Journal Year: 2025, Volume and Issue: 15(10), P. 5638 - 5638

Published: May 18, 2025

Smart farms refer to spaces and technologies that utilize networks automation monitor manage the environment livestock without constraints of time space. As various devices installed on are connected a network automated, farm conditions can be observed remotely anytime anywhere via smartphones or computers. These smart have evolved into farming, which involves collecting, analyzing, sharing data across entire process from production growth post-shipment distribution consumption. This data-driven approach aids decision-making creates new value. However, in evolving technology challenges remain essential requirements collection intelligence. Many face difficulties applying intelligent technologies. In this paper, we propose an monitoring system framework for using artificial intelligence implement deep learning-based monitoring. To detect cattle lesions inactive individuals within barn, apply RT-DETR method instead traditional YOLO model. YOLOv5 YOLOv8 representative models series, both Non-Maximum Suppression (NMS). NMS is postprocessing technique used eliminate redundant bounding boxes by calculating Intersection over Union (IoU) between all predicted boxes. computationally intensive may negatively impact speed accuracy object detection tasks. contrast, (Real-Time Detection Transformer) Transformer-based real-time model does not require achieves higher compared models. Given environments where large-scale datasets obtained CCTV, methods like expected outperform approaches terms performance. reduces computational costs optimizes query initialization, making it more suitable maintenance behaviors related abnormal behavior detection. Comparative analysis with existing verifies confirms shows performance than YOLOv8. research contributes resolving low high redundancy recognition, increasing efficiency management, improving productivity learning farms.

Language: Английский

---

Recognition of Foal Nursing Behavior Based on an Improved RT-DETR Model

Animals, Journal Year: 2025, Volume and Issue: 15(3), P. 340 - 340

Published: Jan. 24, 2025

Foal nursing behavior is a crucial indicator of healthy growth. The mare being in standing posture and the foal suckling are important markers for behavior. To enable recognition mare’s its foal’s stalls, this paper proposes an RT-DETR-Foalnursing model based on RT-DETR. employs SACGNet as backbone to enhance efficiency image feature extraction. Furthermore, by incorporating multiscale multihead attention module channel into Adaptive Instance Feature Integration (AIFI), strengthens utilization integration capabilities, thereby improving accuracy. Experimental results demonstrate that improved RT-DETR achieves best mAP@50 98.5%, increasing 1.8% compared Additionally, study real-time statistical analysis duration posture, which one indicators determining whether suckling. This has significant implications growth foals.

Language: Английский

---

Chinese Named Entity Recognition for Dairy Cow Diseases by Fusion of Multi-Semantic Features Using Self-Attention-Based Deep Learning

Animals, Journal Year: 2025, Volume and Issue: 15(6), P. 822 - 822

Published: March 13, 2025

Named entity recognition (NER) is the basic task of constructing a high-quality knowledge graph, which can provide reliable in auxiliary diagnosis dairy cow disease, thus alleviating problems missed and misdiagnosis due to lack professional veterinarians China. Targeting characteristics Chinese diseases corpus, we propose an ensemble NER model incorporating character-level, pinyin-level, glyph-level, lexical-level features characters. These multi-level were concatenated fed into bidirectional long short-term memory (Bi-LSTM) network based on multi-head self-attention mechanism learn long-distance dependencies while focusing important features. Finally, globally optimal label sequence was obtained by conditional random field (CRF) model. Experimental results showed that our proposed outperformed baselines related works with F1 score 92.18%, suitable effective for named disease corpus.

Language: Английский

---

Semi-automated annotation for video-based beef cattle behavior recognition

Scientific Reports, Journal Year: 2025, Volume and Issue: 15(1)

Published: May 17, 2025

Language: Английский

---