Fig 1 - available via license: Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International
Content may be subject to copyright.
Source publication
Cattle identification is crucial to be registered for breeding association, food quality tracing, disease prevention and control and fake insurance claims. Traditional non-biometrics methods for cattle identification is not really satisfactory in providing reliability due to theft, fraud, and duplication. In this study, a computer vision technique...
Context in source publication
Context 1
... experimental dataset collection was conducted in two housing farms of Hongxing Farm and Minle Farm, China. Fig. 1 depicts the experimental farm and the experimental equipment of the camera. A Sony FDR-AX 40 camera on a tripod was used to capture the cattle faces recorded in videos when they were fed from the cow standing in front of the trough to leaving the trough. The video for each cow is in the HD spatial resolution (3840 × 2160 pixels) with ...
Citations
... Cosine similarity is then used to capture directional and absolute value differences between cattle face vectors, enhancing network robustness [29]. Xu et al. introduced an innovative cattle face recognition framework called CattleFaceNet [30], which combines a lightweight RetinaFace-mobilenet model with an additive angular margin loss (ArcFace). RetinaFace-mobilenet, originally designed for human face detection and localization, was successfully applied to cattle face recognition, with ArcFace enhancing intra-class compactness and inter-class separation during training. ...
Cattle face recognition technology holds significant potential in applications such as livestock insurance, traceability, and real-time monitoring. However, existing methods predominantly focus on single-breed identification, limiting their applicability in multi-breed farms and complex scenarios. This paper proposes an innovative multi-breed cattle face recognition method leveraging image enhancement and deep learning. We introduce an iterative least squares pyramid (ILS-pyramid) algorithm to enhance cattle face images, effectively filtering out hair noise and redundant details. An efficient multi-scale attention-based detection model (EMA-YOLOv8) accurately detects cattle faces in complex backgrounds. Subsequently, an improved global attention residual network (IGAM-iResNet) extracts robust feature vectors for final recognition. Experimental results on a dataset of eight cattle breeds from Jilin Province, China, demonstrate our method’s exceptional performance, achieving a comprehensive accuracy rate of 99.846%, thus setting a new benchmark in multi-breed cattle face recognition. The open-source code and dataset facilitate replication and further research in this domain. The code is available at: https://github.com/xindw211/cattle-face-recognition. The datasets used in this article can be obtained from the following URL: https://pan.baidu.com/s/1wHKT-6R6B9OtxaYZ2N4MOw?pwd=6ckt.
... Video tracking consists of the use of optical sensors (especially RGB cameras) to monitor animal movement continuously. This technique has been gaining increasing interest recently [29], and it is made possible thanks to the advances in machine learning and image analysis [30]. The infrastructure of these systems consists of several cameras, normally placed on the roof, that can visually cover all the accessible areas of the building [31]. ...
Indoor Positioning Systems (IPS) are becoming increasingly popular
in livestock farming as possible tools for the implementation of Precision Agriculture principles, allowing precise tracking of animals, groups or operator within barns. Unlike GPS, which is common outdoors, a wide variety of technologies have been developed to address unique indoor challenges, each with specific strengths and limitations. This paper surveys current and emerging IPS technologies in livestock farming, evaluating their accuracy, latency, scalability, and energy efficiency. The use of IPS with other sensors could allow farmers to monitor animal health and behavior in real time, enhancing operational efficiency and empowering on-farm data visualization on portable devices like smartphones or immersive technologies as smart glasses. However, further research is needed to investigate scalability, cost-effectiveness, and integration of these systems with conventional farm management tools. IPS adoption could empower a more data-driven, sustainable approach to farming meeting modern demands for food security, animal welfare, and economic sustainability.
Keywords: Indoor Positioning System · Animal precise management ·
Precision livestock farming · Animal Monitoring
... ResNet introduced residual structures to effectively mitigate the vanishing gradient problem, enabling deeper network training with strong feature representation capabilities and widespread applicability (Islam et al., 2023;Yang et al., 2024). MobileNet, centered on depthwise separable convolution (DWConv), significantly reduces parameter and computational costs, making it a typical example of lightweight models (Xu et al., 2022). EfficientNet uses a compound scaling strategy to achieve an excellent balance between model depth, width, and resolution, though at a higher computational cost (Jin et al., 2023). ...
With the continuous advancement of modern agricultural technologies, the demand for precision fruit-picking techniques has been increasing. This study addresses the challenge of accurate recognition and harvesting of winter peaches by proposing a novel recognition model based on the residual network (ResNet) architecture—WinterPeachNet—aimed at enhancing the accuracy and efficiency of winter peach detection, even in resource-constrained environments. The WinterPeachNet model achieves a comprehensive improvement in network performance by integrating depthwise separable inverted bottleneck ResNet (DIBResNet), bidirectional feature pyramid network (BiFPN) structure, GhostConv module, and the YOLOv11 detection head (v11detect). The DIBResNet module, based on the ResNet architecture, introduces an inverted bottleneck structure and depthwise separable convolution technology, enhancing the depth and quality of feature extraction while effectively reducing the model’s computational complexity. The GhostConv module further improves detection accuracy by reducing the number of convolution kernels. Additionally, the BiFPN structure strengthens the model’s ability to detect objects of different sizes by fusing multi-scale feature information. The introduction of v11detect further optimizes object localization accuracy. The results show that the WinterPeachNet model achieves excellent performance in the winter peach detection task, with P = 0.996, R = 0.996, mAP50 = 0.995, and mAP50-95 = 0.964, demonstrating the model’s efficiency and accuracy in the winter peach detection task. The high efficiency of the WinterPeachNet model makes it highly adaptable in resource-constrained environments, enabling effective object detection at a relatively low computational cost.
... Biometric systems using facial recognition offer a promising alternative that can improve both animal welfare and operational efficiency. Deep learning methods have been successfully applied to facial recognition in various species, including cattle [6,7], sheep [8], dogs [9], and birds [10]. Pig face recognition research has likewise advanced, with Hansen et al. [11] achieving around 83% accuracy using Convolutional Neural Networks, and Marsot et al. [12] noting that focusing on features like the eyes and snout can boost accuracy to about 91%. ...
Accurate animal face recognition is essential for effective health monitoring, behavior analysis, and productivity management in smart farming. However, environmental obstructions and animal behaviors complicate identification tasks. In pig farming, fences and frequent movements often occlude essential facial features, while high inter-class similarity makes distinguishing individuals even more challenging. To address these issues, we introduce the Pig Face Recognition and Inpainting System (PigFRIS). This integrated framework enhances recognition accuracy by removing occlusions and restoring missing facial features. PigFRIS employs state-of-the-art occlusion detection with the YOLOv11 segmentation model, a GAN-based inpainting reconstruction module using AOT-GAN, and a lightweight recognition module tailored for pig face classification. In doing so, our system detects occlusions, reconstructs obscured regions, and emphasizes key facial features, thereby improving overall performance. Experimental results validate the effectiveness of PigFRIS. For instance, YOLO11l achieves a recall of 94.92% and a AP50 of 96.28% for occlusion detection, AOTGAN records a FID of 51.48 and an SSIM of 91.50% for image restoration, and EfficientNet-B2 attains an accuracy of 91.62% with an F1 Score of 91.44% in classification. Additionally, heatmap analysis reveals that the system successfully focuses on relevant facial features rather than irrelevant occlusions, enhancing classification reliability. This work offers a novel and practical solution for animal face recognition in smart farming. It overcomes the limitations of existing methods and contributes to more effective livestock management and advancements in agricultural technology.
... The condition not only causes severe pain and discomfort [3,4], it also has ramifications that extend across a cow's productive and reproductive life. Chronic or untreated lameness can reduce longevity [5], compromise milk yields [6], interfere with normal fertility [7], and increase veterinary care expenses [8,9]. In many cases, recurring or advanced lameness necessitates premature culling [3,8], which undermines herd welfare, disrupts farm operations, and triggers significant economic losses. ...
... Better registration, traceability, and security of livestock/cattle Xu et al. [5] RetinaFace-mobilenet, ArcFace (CattleFaceNet) ...
Lameness remains a leading cause of economic loss in Canadian dairy herds while also compromising animal welfare. To address the urgent need for early detection, we introduce a novel bimodal artificial intelligence (AI) framework that leverages both facial biometric data and accelerometer-based movement metrics. Over a 21-day period, six Holstein cows were monitored to capture variations in facial expressions and locomotion, and a multimodal model was built by combining DenseNet-121 for image analysis with Long Short-Term Memory (LSTM) networks for time-series data. Crucially, our model employs a multi-head attention mechanism to fuse visual and movement features, enabling it to overcome confounding factors such as lighting conditions, barn environments, and individual behavioral differences. This approach achieved a 99.55 % accuracy-substantially exceeding single-modality baselines-and Grad-CAM interpretations revealed key facial cues (orbital tightening, ear posture, muzzle tension) linked to lameness. Lame cows also exhibited prolonged resting times, especially during peak activity hours, underscoring their discomfort. These findings illustrate how integrating facial and accelerometer data can promote timely interventions, significantly enhancing cow welfare and reducing medical expenditures and productivity losses. Moreover, our results highlight how tie-stall barn systems can exacerbate lameness by restricting natural movement, further supporting recommendations to transition toward more open, movement-friendly housing. In doing so, producers not only protect cow well-being but also safeguard vital economic returns.
... As one of the current state-of-the-art voiceprint recognition models, it significantly improves the accuracy and efficiency of voiceprint Table 3 The results of ω 1 values recognition. However, ECAPA-TDNN is sensitive to factors such as environmental noise, which may lead to performance degradation [23]. Addressing this issue, this paper proposes a dual-channel voiceprint recognition model based on ECAPA-TDNN, which achieves better recognition performance in complex environments with minimal increase in recognition time. ...
Voiceprint recognition technology, as a form of biometric identification, holds extensive potential applications in the realm of security authentication. However, practical implementations often encounter challenges in cross-scenario and cross-channel recognition, resulting in decreased recognition accuracy. To address this issue, a model based on deep learning and dual-channel voiceprint recognition is proposed in this paper. Firstly, a Distance-weighted Sub-center Arcface Loss (DWLoss) utilizing pole-zero distance weighting is designed to train the voiceprint recognition model, aimed at enhancing the model’s generalization capability and robustness. Secondly, an Equal Channel Attention (ECA) feature channel weighting module for deep feature extraction without dimension reduction is employed, combined with Probabilistic Linear Discriminant Analysis (PLDA) for channel compensation and speaker scoring to tackle the cross-channel recognition problem. Finally, based on the aforementioned designs, the SE-Res2Net-DWLoss-TDNN and ECA-Res2Net-TDNN-PLDA dual-channel voiceprint recognition models are presented. The weight matrix is trained to combine the scoring scores of the two channels, thereby obtaining a comprehensive speaker similarity score. Experimental validation demonstrates the Synergistic nature of the two channels during the recognition process, as well as the superiority of the model in improving recognition performance and robustness. Experimental results indicate that compared to traditional models, this model exhibits significant advantages in noise resistance and recognition accuracy. This study contributes beneficial advancements and enhancements to cross-scenario and cross-channel recognition within the realm of voiceprint recognition technology.
... The application of computer vision technology in animal husbandry encompasses body sizes analysis (7), behavior monitoring (8), appearance feature recognition (9), and health monitoring (10). This approach provides farmers with a more convenient and effective management tool, serving as a reference for practical applications while substantially reducing subjective errors and labor costs associated with classification (11). The YOLO (You Only Look Once) series (38) is a prominent representative of object detection methods, having undergone multiple iterations that have garnered significant attention in the field due to its high processing speed and accuracy. ...
Introduction
The facial coloration of sheep is not only a critical characteristic for breed and individual identification but also serves as a significant indicator for assessing genetic diversity and guiding selective breeding efforts.
Methods
In this study, 201 Ujumqin sheep were used as research objects and 1713 head image data were collected. We delineated feature points related to the facial coloration of Ujumqin sheep and successfully developed a head color recognition model (YOLOv8-CBAM) utilizing the YOLOv8 architecture in conjunction with the CBAM attention mechanism.
Results
The model demonstrated impressive performance in recognizing four head color categories, achieving an average precision (mAP) of 97.7% and an F1 score of 0.94. In comparison to YOLOv8n, YOLOv8l, YOLOv8m, YOLOv8s, and YOLOv8x, the YOLOv8-CBAM model enhances average accuracy by 0.5%, 1%, 0.7%, 0.7%, and 1.6%, respectively. Furthermore, when compared to YOLOv3, the improvement is 1%, while YOLOv5n and YOLOv10n show increases of 1.4% and 2.4%, respectively.
Discussion
The findings indicate that the smaller model exhibited superior performance in the facial color recognition task for Ujumqin sheep. Overall, the YOLOv8-CBAM model achieved high accuracy in the head color recognition task, providing reliable technical support for automated sheep management systems.
... As similar to identifying human emotions, classifying emotions from the facial expressions has been practiced by experienced animal researchers [29] [74]. They observed that adopting ICT technologies eases the process of emotion detection due to the inclusion of sophisticated learning algorithms to capture the facial emotion-related features of animals. ...
... Incorporating advanced augmentation techniques such as GANs not only increases the diversity of the training set but also enhances the model's ability to generalize across unseen data, making it more resilient in real-world applications. Although image augmentation is commonly performed with supervision, it has diverse applications (Xu et al. 2022). Rice et al. (2020) and Schmidt et al. (2018) have identified a range of strategies for image augmentation, which include simple techniques like horizontal flipping and random cropping, as well as more advanced methods that leverage unlabeled data for semi-supervised learning. ...
... The most commonly utilized activation function is SoftMax, which solely considers the accuracy of classification and disregards the inter-class distance. In this study, the researchers have selected the face detection model and the latest loss functions for face identification, as described by Xu et al. (2022), to be used specifically for identifying cattle faces. The selection of SoftMax as the activation function is due to its efficiency in handling multi-class classification problems, which is essential for differentiating among numerous cattle individuals. ...
Monitoring animal welfare, disease prevention, vaccination administration, production supply, and ownership management all depend on accurate cattle identification. Ear tag-based cattle identification techniques are frequently used in livestock farm management. These are not used to identify specific cattle on large-scale farms. However, ear tags can come off, which makes it challenging to identify a particular individual. Ear tags are susceptible to fraud, can be copied, and run the risk of being damaged. Long-term animal identification is impossible with lost tags. For this purpose, a data set was created by taking images of cattle in their natural environment. The dataset, which contains 15,000 records from 30 different cattle, was divided into three sections: training, validation, and testing. Deep learning algorithms InceptionResNetV2, MobilenetV2, DenseNet201, Xception, and NasNetLarge were used in this study to identify specific cattle faces. The DenseNet201 algorithm achieved the highest test accuracy of 99.53% with a validation accuracy of 99.83%.
... Traditional identification methods, such as ear cutting, hot iron branding, and tagging with ear tags [4], cause permanent physical and mental damage to domestic animals [5]. Current radio frequency identification (RFID) technology [6] is relatively expensive, susceptible to interference by ferrous products, could be artificially tampered with [7], and cannot be applied over long distances [8]. Machine vision technology can rapidly and accurately recognize animals' identity through their biological features such as nose prints [9], iris patterns [10], retinal images [11], and facial characteristics [12], which are becoming a hot topic in current research [13]. ...
... The confidence vector of each type of appearance image is multiplied by the accuracy of the single appearance model whose appearance corresponds (i.e., the accuracy value metrics from the corresponding base model above) to obtain the weight vector. The average value of the five weight vectors is calculated to be the predicted probability vector for the identity of the goat individual under the fusion of the five appearances (the length is 54), as shown in Equation (8). The index of the largest value in this probability vector corresponds to the goat's identity number predicted by the model. ...
... In contrast to previous studies on network improvement for recognition [8,20,21,28], this study is not limited to improving the accuracy and speed of goat individual identity recognition from a single viewpoint appearance by improving the network structure. It aims to utilize the contribution of different appearance views to goat individual identity recognition. ...
Accurate identification of individual goat identity is necessary for precision farming. Previous studies have primarily focused on using front face images for goat identification, leaving the potential of other appearances and multi-source appearance fusion unexplored. In this study, we used a self-developed multi-view appearance image acquisition platform to capture five different appearances (left face, right face, front face, back body, and side body) from 54 Wanlin white goats. The recognition ability of different goat appearance images and its multi-source appearance fusion for its identity recognition was then systematically examined based on the four basic network models, namely, MobileNetV3, MobileViT, ResNet18, and VGG16, and the best combination of goat appearance and network was screened. When only one kind of goat appearance image was used, the combination of side body image and MobileViT was the best, with an accuracy of 99.63%; under identity recognition based on multi-source image appearance fusion, all recognition models after outlook fusion of two viewpoints generally outperformed single viewpoint appearance identity recognition models in recognizing the identity of individual goats; when three or more kinds of goat appearance images were utilized for fusion, any of the four models were capable of identifying the identity of an individual goat with 100% accuracy. Based on these results, a goat individual identity recognition strategy was proposed that balances accuracy, computation, and time, providing new ideas for goat individual identity recognition in complex farming contexts.
