Accurate pig counting during herd transfers is fundamental to effective livestock management in large-scale swine production, yet existing methods struggle with bidirectional passages, boundary oscillations, and occlusion in real corridor environments. This study proposes an integrated system combining an improved YOLO-based detection model with a Hysteresis-based Multi-frame Temporal Confirmation Counting Strategy (HMTC). The YOLO11s baseline was enhanced using lightweight RepViT blocks, dynamic upsampling (DySample), and shape-aware bounding box regression (Shape-IoU). The resulting model achieves a mAP50 of 0.982 with a compact architecture of 8.28M parameters, representing a 12.3% reduction relative to the baseline while improving detection accuracy. To address bidirectional counting challenges, the HMTC strategy utilizes hysteresis-based region classification, temporal confirmation, and trajectory verification to suppress boundary jitter and ensure directional correctness. Evaluated on nine videos from a single transfer corridor, the proposed system achieves an overall counting accuracy of 99.21% on this test set and runs in real time on an embedded edge device at over 30 FPS without loss of counting accuracy. Together, the improved detection model and HMTC counting strategy provide a cohesive approach to pig passage counting, validated here under a single transfer-corridor condition; these results offer a promising basis for automated animal inventory management, pending further validation across more diverse farm environments.

Accurate multi-object tracking of group-housed pigs is essential for health monitoring in precision livestock farming, yet remains challenging due to similar pig appearances, frequent occlusions, dense interactions, and nonlinear motion. This paper proposes STA-PigTrack (Spatiotemporal Association Enhanced Pig Tracking), an appearance-free tracking method based on enhanced spatiotemporal association. The method introduces the Unscented Kalman Filter (UKF) to better handle nonlinear motion such as sudden accelerations and sharp turns, outperforming conventional Kalman-based models. To address identity loss from long-term occlusion and trajectory fragmentation, an AFLink module constructs a global cost matrix from spatiotemporal trajectory continuity, combined with the Hungarian algorithm to recover fragmented identities without relying on appearance embeddings. Evaluated on both a self-collected and a public dataset, STA-PigTrack achieved average HOTA, MOTA, and IDF1 scores of 83.82 %, 96.27 %, and 89.73 %, respectively, reducing identity switches by 70.2 % compared with the baseline SORT, while maintaining a real-time speed of 56 FPS. It outperformed DeepSORT, BotSORT, and OCSORT in identity consistency and robustness under heavy occlusion and high-density conditions. Using the obtained trajectories, group activity intensity and individual cumulative displacement were extracted as exploratory trajectory derived movement indicators, showing the feasibility of transforming tracking outputs into quantifiable behavioral features. STA-PigTrack provides a reliable real-time solution for individual tracking in complex farm environments and may provide a technical basis for future precision health monitoring and welfare related applications.

Detecting keypoints in dairy cows aims to locate and track the motion trajectories of the body's joints, which plays a crucial role in behavior analysis and lameness detection. However, real farming scenarios, characterized by occlusions and large variations in object scale may result in poor detection results. Therefore, we introduce the atrous spatial pyramid pooling (ASPP) module into the shallow layers network of ResNet101, designed to improve the multi-scale feature extraction capability of the model. The ASPP module enhances the robustness of recognition for different dimensional sizes and occluded keypoints using different dilatation rates in the parallel atrous convolutional layers to expand the model's receptive field. Furthermore, seven types of motion features, including tracking up, gait symmetry, step height balance, motion speed variability, head swing amplitude, head-neck slope and back curvature are extracted simultaneously by monitoring and tracking the motion trajectory of distinct keypoints. Several of these features represent innovative extraction models and attributes, first proposed in this study. Multiple models are trained and tested on datasets containing 2,385 frames for ablation experiments. The experiments show that, in comparison with the ResNet50, MobileNet_v2_1.0, and EfficientNet-b0 backbone networks, the training error and test error of ResNet101 are reduced by 4.04-30.12 pixels and 3.81-28.14 pixels. Therefore, ResNet101 is used as the benchmark for subsequent model improvement by adding the ASPP module. The training error and test error of the ResNet101-ASPP network are reduced by 0.27 and 0.24 pixels, respectively, compared to the benchmark network. The prediction confidence improves by 1.65-2.50% at three different dairy cow object scales. In addition, the keypoints under different occlusion conditions improve considerably, especially for small-scale keypoints, demonstrating the capability of the ASPP module for multi-scale feature extraction. By analyzing the distribution of the seven features and health, mild lameness, and severe lameness in dairy cows, it is shown that all the different features play an important role in distinguishing between different levels of lameness.

Cow pose estimation and real-time health monitoring are important for refined herd management, improved animal welfare, and reduced passive culling rates. However, existing multi-object pose estimation methods often struggle to adapt to multi-scale objects in complex environments and typically exhibit low accuracy in detecting occluded keypoints. To address these challenges, this study proposes a top-down deep neural network for multidairy cows pose estimation and lameness detection, which integrates lightweight object detection, multi-scale feature fusion, and comprehensive motion feature analysis to improve the robustness under complex farm conditions. First, the real-time object detector YOLOv8n is improved by introducing the Partial Convolution (PConv) and Slim-neck modules, which improve both the efficiency and accuracy of object bounding box predictions, providing a solid foundation for the subsequent pose estimation. Second, a Path Aggregation Feature Pyramid Network (PAFPN)-based multi-scale feature fusion module is introduced as the neck network within the Real-time Multi-person Pose Estimation (RTMPose). This is further supported by a transfer learning strategy to improve keypoint localization, particularly under-occlusion and scale variation conditions. The experimental results show that the improved model achieves a mean average precision (mAP) of 95.8 %, significantly outperforming the baseline model and other existing algorithms. Seven motion features, including gait symmetry, head swing amplitude, and back curvature, were extracted in real time through pose tracking and motion trajectory analysis. These features were normalized and input into a Random Forest classifier for lameness detection. The model was evaluated on a dataset of 418 dairy cows and achieved average accuracy, sensitivity, and specificity values of 93.8 %, 94.4 %, and 97.5 %, respectively. These results demonstrate that combining multiple motion features provides a more accurate assessment of lameness.

This study proposed an online early lameness detection method for dairy cow health management to overcome the inability of wearable sensor-based methods for online detection and low sensitivity to early lameness. Wearable IMU sensors collected acceleration data in stationary and moving states; a threshold discrimination module using variance of motion-direction acceleration was designed to distinguish states within 2 s, enabling rapid data screening. For moving-state windowed data, the InceptionTime network was modified with YOLOConv1D and SeparableConv1D modules plus Dropout, which significantly reduced model parameters and helped mitigate overfitting risk, enhancing generalization on the test set. Typical gait features were fused with deep features automatically learned by the network, enabling accurate discrimination among healthy, mild (early) lameness, and severe lameness. Results showed that the online detection model achieved 80.6% dairy cow health status detection accuracy with 0.8 ms single-decision latency. The recall and F1 score for lameness, including early and severe cases, reached 89.11% and 88.93%, demonstrating potential for early and progressive lameness detection. This study improves lameness detection efficiency and validates the feasibility and practical value of wearable sensor-based gait analysis for dairy cow health management, providing new approaches and technical support for monitoring and early intervention on large-scale farms.

With the development of precision livestock farming, in order to achieve the goal of fine management and improve the health and welfare of dairy cows, research on dairy cow motion monitoring has become particularly important. In this study, considering the problems surrounding a large amount of model parameters, the poor accuracy of multi-target tracking, and the nonlinear motion of dairy cows in dairy farming scenes, a lightweight detection model based on improved YOLO v11n was proposed and four tracking algorithms were compared. Firstly, the Ghost module was used to replace the standard convolutions in the YOLO v11n network and a more lightweight attention mechanism called ELA was replaced, which reduced the number of model parameters by 18.59%. Then, a loss function called SDIoU was used to solve the influence of different cow target sizes. With the above improvements, the improved model achieved an increase of 2.0 percentage points and 2.3 percentage points in mAP@75 and mAP@50-95, respectively. Secondly, the performance of four tracking algorithms, including ByteTrack, BoT-SORT, OC-SORT, and BoostTrack, was systematically compared. The results show that 97.02% MOTA and 89.81% HOTA could be achieved when combined with the OC-SORT tracking algorithm. Considering the demand of equipment in lightweight models, the improved object detection model in this paper reduces the number of model parameters while offering better performance. The OC-SORT tracking algorithm enables the tracking and localization of cows through video surveillance alone, creating the necessary conditions for the continuous monitoring of cows.

Under large-scale farming conditions, automated sow estrus detection is crucial for improving reproductive efficiency, optimizing breeding management, and reducing labor costs. Conventional estrus detection relies heavily on human expertise, a practice that introduces subjective variability and consequently diminishes both accuracy and efficiency. Failure to identify estrus promptly and pair animals effectively lowers breeding success rates and drives up overall husbandry costs. In response to the need for the automated detection of sows’ estrus states in large-scale pig farms, this study proposes a method for detecting sows’ vulvar status and estrus based on multi-dimensional feature crossing. The method adopts a dual optimization strategy: First, the Bi-directional Feature Pyramid Network—Selective Decoding Integration (BiFPN-SDI) module performs the bidirectional, weighted fusion of the backbone’s low-level texture and high-level semantic, retaining the multi-dimensional cues most relevant to vulvar morphology and producing a scale-aligned, minimally redundant feature map. Second, by embedding a Spatially Enhanced Attention Module head (SEAM-Head) channel attention mechanism into the detection head, the model further amplifies key hyperemia-related signals, while suppressing background noise, thereby enabling cooperative and more precise bounding box localization. To adapt the model for edge computing environments, Masked Generative Distillation (MGD) knowledge distillation is introduced to compress the model while maintaining the detection speed and accuracy. Based on the bounding box of the vulvar region, the aspect ratio of the target area and the red saturation features derived from a dual-threshold method in the HSV color space are used to construct a lightweight Multilayer Perceptron (MLP) classification model for estrus state determination. The network was trained on 1400 annotated samples, which were divided into training, testing, and validation sets in an 8:1:1 ratio. On-farm evaluations in commercial pig facilities show that the proposed system attains an 85% estrus detection success rate. Following lightweight optimization, inference latency fell from 24.29 ms to 18.87 ms, and the model footprint was compressed from 32.38 MB to 3.96 MB in the same machine, while maintaining a mean Average Precision (mAP) of 0.941; the accuracy penalty from model compression was kept below 1%. Moreover, the model demonstrates robust performance under complex lighting and occlusion conditions, enabling real-time processing from vulvar localization to estrus detection, and providing an efficient and reliable technical solution for automated estrus monitoring in large-scale pig farms.

Early lameness detection is crucial to ensure the welfare and productivity of dairy cows. However, current research on early lameness identification using wearable analysis relies on the limited robustness of indirect behavioral measures, which are susceptible to individual variations and imbalances in lameness samples. In this study, we propose a semi-supervised Long short-term memory (LSTM)-Autoencoder algorithm for early lameness detection in dairy cows through time series data reconstruction. We collected gait data from all four limbs of 30 dairy cows using four IMUs. A LSTM-Autoencoder with three LSTM hidden layers was trained to learn the time series features of healthy gaits. Each gait was reconstructed, and anomaly gaits exceeding a threshold were identified by comparing reconstructed gaits with actual gaits. The gait symmetry was measured by comparing the percentage of anomaly gait between opposite limbs as an indicator of lameness severity. With a high accuracy of 97.78% and a true negative rate of 98.33%, our integrated approach outperforms traditional methods in early lameness detection and lame limb identification, enabling real-time monitoring and timely identification of lameness. The study is the first attempt at using a time series anomaly detection framework with deep learningbased gait reconstruction for lameness detection. Wearable gait analysis offers portability and real-time capabilities, providing continuous, accurate, and comprehensive gait information unaffected by lighting and field-ofview limitations. This approach holds promise for enhancing animal welfare and optimizing management practices in the dairy industry through timely identification and continuous monitoring of lameness.

In order to analyze the application and development potential of pose estimation in animal behavior recognition and animal welfare research, this paper took the deep learning-based pose estimation method as a breakthrough, summarized the research progress and directions of animal pose estimation from the perspective of two-dimensional and three-dimensional space, and introduced the common data sets and evaluation indexes. Then integrated the research results related to animal behavior recognition based on pose estimation, focused on the algorithms of keypoint detection and behavior classification and their features. Pose estimation provides skeleton information and motion features for tasks such as action recognition and behavior analysis, and became a non-contact monitoring method for animal behavior recognition and abnormal information warning. However, due to the limitation of small training data set, the research and development of animal pose estimation was relatively slow compared to human pose estimation. Therefore, the used of cross-domain learning to further improve its performance has become an emerging tool in recent years. This review expands research ideas and research methods for researchers related to intelligent animal behavior recognition, animal welfare research, and smart farming.

Accurate wetland mapping is essential for their protection and management; however, it is difficult to accurately identify seasonal wetlands because of irregular rainfall and the potential lack of water inundation. In this study, we propose a novel method to generate reliable seasonal wetland maps with a spatial resolution of 20 m using a seasonal-rule-based method in the Zhalong and Momoge National Nature Reserves. This study used Sentinel-1 and Sentinel-2 data, along with a bi-weekly composition method to generate a 15-day image time series. The random forest algorithm was used to classify the images into vegetation, waterbodies, bare land, and wet bare land during each time period. Several rules were incorporated based on the intra-annual changes in the seasonal wetlands and annual wetland maps of the study regions were generated. Validation processes showed that the overall accuracy and kappa coefficient were above 89.8% and 0.87, respectively. The seasonal-rule-based method was able to identify seasonal marshes, flooded wetlands, and artificial wetlands (e.g., paddy fields). Zonal analysis indicated that seasonal wetland types, including flooded wetlands and seasonal marshes, accounted for over 50% of the total wetland area in both Zhalong and Momoge National Nature Reserves; and permanent wetlands, including permanent water and permanent marsh, only accounted for 11% and 12% in the two reserves, respectively. This study proposes a new method to generate reliable annual wetland maps that include seasonal wetlands, providing an accurate dataset for interannual change analyses and wetland protection decision-making.

针对奶牛行为判别自动化水平不足、准确率低的问题,采用惯性测量单元(IMU)和卷积神经网络(CNN),对细粒度奶牛行为判别进行研究.结果表明:1)在KNN、SVM、BPNN、CNN和LSTM 5个模型中,CNN模型在奶牛行为分类测试集上的准确率最高.2)含有三轴加速度计、陀螺仪和磁力计的IMU更加适用于奶牛行为分类,其分类效果优于含一种传感器的IMU.3)传感器频率与分类模型的性能相关,频率越高,正确率越高,当传感器频率设置为25 Hz时,奶牛行为判别效果最好.4)在1、2和4 s这3种时间窗中,使用4 s时间窗的奶牛行为分类模型性能最好.5)采用最优配置时,卷积神经网络模型能够有效的判别奶牛站立、躺卧2种状态,正确率为99％;可以对奶牛卷食、咀嚼、站立反刍、躺卧反刍、躺卧休息、站立休息6类行为进行判别,正确率为85％.采用IMU和卷积神经网络算法,可以有效的对细粒度奶牛行为进行判别,为奶牛养殖的自动化、智能化管理提供支撑.

Although it has been widely recognized that land fragmentation has increased chemical fertilizer application, little is known about the role of technology adoption in mitigating these adverse effects. To empirically examine the relationship between land fragmentation, technology adoption and chemical fertilizer application, we developed a mediation model. We applied our analysis to a survey data set encompassing 1388 farm-level samples collected in 14 Chinese provinces in 2019. Our study demonstrated that land fragmentation can not only directly increase chemical fertilizer application but also indirectly increase it by hindering the adoption of agricultural mechanization technologies (AMT's) and soil testing fertilization technologies (STFT's). Both are recognized as potent drivers of fertilizer use reductions. Moreover, the adoption of information and communications technologies (ICT's) can help mitigate the negative effects of land fragmentation on technology adoption, thus reducing chemical fertilizer application intensity (CFAI). However, the direct effects of land fragmentation on CAFI was unaffected by ICT's. Our findings suggest that ICT's have revolutionized farmer recognition, promotion and adoption of agricultural technologies by increasing awareness and diffusion of agricultural technology information.

Surface Enhanced Raman Spectroscopy (SERS) is a highly sensitive detection method, however, it is still challenging to directly detect florfenicol due to its low Raman activity. To solve this problem, we present a novel sensor for selective detection of florfenicol residue in milk using gap-enhance Raman tags (GERTs) competitive immunoassay. In this nanosensor, Raman reporters 4-mercaptobenzoic acid (4MBA) were embedded in the 1 nm gap of Au core-gap-Au shell nanoparticles, which can protect 4MBA molecules from unwanted external environments and provide a high enhancement. The parameters affecting the detection sensitivity were discussed and optimized. The limit detection of florfenicol in milk can reach 0.1 μg/kg. To evaluate the performance of GERTs immunoassay, Ag@4MBA tags immunoassay was conducted and the results were compared. Moreover, Raman mapping was investigated to improve the quantitative prediction model with R2 of 0.981 achieved, by eliminating the signal fluctuations caused by the inhomogeneous distribution of GERTs.

The spatial difference in agricultural product prices is a crucial factor affecting the benefits of related stakeholders. This study aims to analyze the mechanisms of spatial price transmission. In this paper, taking garlic as an example, we present a vector autoregression model analyzing relations of the price transmission between producing and selling cities. The partial Granger causality test is used to determine the direction and path of price transmission between the main producing areas and the main consuming areas. We find that the prices in different areas have a complex transmission network and fluctuate in correlation with each other. The results reveal the characteristics of agricultural product price transmission in China and provide reasons and evidence for market regulation.

The gait phase of dairy cows is an important indicator to reflect the severity of lameness. IThe accuracy of available gait segmentation methods was not enough for lameness detection. In this study, a gait phase recognition method based on Gaussian mixture model (GMM) and hidden Markov model (HMM) was proposed and tested. Firstly, wearable inertial sensors LPMS-B2 were used to collect the acceleration and angular velocity signals of cow hind limbs. In order to remove the noise of the system and restore the real dynamic data, Kalman filter was used for data preprocessing. The first-order difference of the angular velocity of the coronal axis was selected as the eigenvalue. Secondly, to analyze the long-term continuous recorded gait sequences of dairy cows, the processed data was clustered by GMM in the unsupervised way. The clustering results were taken as the input of the HMM, and the gait phase recognition of dairy cows was realized by decoding the observed data. Finally, the cow gait was segmented into 3 phases, including the stationary phase, standing phase and swing phase. At the same time, gait segmentation was achieved according to the standing phase and swing phase. The accuracy, recall rate and F1 of the stationary phase were 89.28%, 90.95% and 90.91%, respectively. The accuracy, recall rate and F1 of the standing phase recognition in continuous gait were 91.55%, 86.71% and 89.06%, respectively. The accuracy, recall rate and F1 of the swing phase recognition in continuous gait were 86.67%, 91.51% and 89.03%, respectively. The accuracy of cow gait segmentation was 91.67%, which was 4.23% and 1.1 % higher than that of the event-based peak detection method and dynamic time warping algorithm, respectively. The experimental results showed that the proposed method could overcome the influence of the cow's walking speed on gait phase recognition results, and recognize the gait phase accurately. This experiment provides a new method for the adaptive recognition of the cow gait phase in unconstrained environments. The degree of lameness of dairy cows can be judged by the gait features.

Monitoring the dynamics of wetland resources has practical value for wetland protection, restoration and sustainable utilization. Dongting Lake wetland reserves are well known for both their intra-annual and inter-annual dynamic changes due to the effects of natural or human factors. However, most wetland monitoring research has failed to consider the seasonal wetlands, which is the most fragile wetland type, requiring more attention. In this study, we used multi-source time series remote sensing data to monitor three Dongting Lake wetland reserves between 2000 and 2020, and the seasonal wetlands were separated from permanent wetlands. Multispectral and indices time series were generated at 30 m resolution using a two-month composition strategy; the optimal features were then selected using the extension of the Jeffries–Matusita distance (JBh) and random forest (RF) importance score; yearly wetland maps were identified using the optimal features and the RF classifier. Results showed that (1) the yearly wetland maps had good accuracy, and the overall accuracy and kappa coefficients of all wetland maps from 2000 to 2020 were above 89.6% and 0.86, respectively. Optimal features selected by JBh can improve both computational efficiency and classification accuracy. (2) The acreage of seasonal wetlands varies greatly among multiple years due to inter-annual differences in precipitation and evaporation. (3) Although the total wetland area of the three Dongting Lake wetland reserves remained relatively stable between 2000 and 2020, the acreage of the natural wetland types still decreased by 197.0 km2, and the change from natural wetland to human-made wetland (paddy field) contributed the most to this decrease. From the perspective of the ecological community, the human-made wetland has lower ecological function value than natural wetlands, so the balance between economic development and ecological protection in the three Dongting Lake wetland reserves requires further evaluation. The outcomes of this study could improve the understanding of the trends and driving mechanisms of wetland dynamics, which has important scientific significance and application value for the protection and restoration of Dongting Lake wetland reserves.

Livestock pollution is one of the main sources of agricultural pollution, which has a negative impact on the global environment. Monitoring, simulation and early warning of major pollutants emitted from livestock production is of great significance for reducing agricultural pollution. Especially, real-time comprehensive monitoring and early warning of the concentration and distribution of harmful gases could improve the harm of livestock production to people, livestock itself and the environment, and increase the safety level of livestock production. This study focused on the perception and monitoring of the discharge status of livestock farming simulation technics, mainly to carry out Internet of Things-based monitoring of the main components of livestock culture pollutants, and to use odor gas air dispersion software “ModOdor” to simulate the spread of pollutants. This study was aimed to determine the characteristics of pollutant diffusion in typical farms, which could provide decision reference to odor hygienic buffer zone and minimum shelter distance to achieve the ecological and safety objectives of livestock farming.

Climate change severely impacts agricultural production, which jeopardizes food security. China is the second largest maize producer in the world and also the largest consumer of maize. Analyzing the impact of climate change on maize yields can provide effective guidance to national and international economics and politics. Panel models are unable to determine the group-wise heteroscedasticity, cross-sectional correlation and autocorrelation of datasets, therefore we adopted the feasible generalized least square (FGLS) model to evaluate the impact of climate change on maize yields in China from 1979–2016 and got the following results: (1) During the 1979–2016 period, increases in temperature negatively impacted the maize yield of China. For every 1°C increase in temperature, the maize yield was reduced by 5.19 kg 667 m–2 (1.7%). Precipitation increased only marginally during this time, and therefore its impact on the maize yield was negligible. For every 1 mm increase in precipitation, the maize yield increased by an insignificant amount of 0.043 kg 667 m–2 (0.014%). (2) The impacts of climate change on maize yield differ spatially, with more significant impacts experienced in southern China. In this region, a 1°C increase in temperature resulted in a 7.49 kg 667 m–2 decrease in the maize yield, while the impact of temperature on the maize yield in northern China was insignificant. For every 1 mm increase in precipitation, the maize yield increased by 0.013 kg 667 m–2 in southern China and 0.066 kg 667 m–2 in northern China. (3) The resilience of the maize crop to climate change is strong. The marginal effect of temperature in both southern and northern China during the 1990–2016 period was smaller than that for the 1979–2016 period.

中国是养殖大国,传统的养殖存在生产成本高、技术落后、效率低下等问题,限制了养殖业的发展.随着物联网、大数据、人工智能、4/5G等现代信息技术的发展,养殖业已从传统养殖进入现代化养殖阶段.为理清养殖业信息化技术研究现状,本文阐述了信息化养殖的关键技术及其应用现状,提出了信息化养殖面临的挑战和发展建议,为中国信息化养殖技术的发展提供参考.

信息化水平不高是中国畜牧业生产效率低于国际平均水平的主要原因之一.针对现代信息技术与畜牧业深度融合难、畜禽养殖复杂环境多特征联动监测难等问题,创新提出了"物联牧场"理论方法与畜牧业物联网技术体系,研制了超声波风速风向、非侵入式肉牛维生素A缺乏诊断、奶牛发情监测、奶牛体征测量等畜禽"生态-生理-生长"多通量专用传感技术,并在畜牧养殖物联网标准建设、决策云平台构建、智能装备研发等方面提出了发展建议.未来,畜禽养殖物联网技术的发展与应用将促进信息技术与畜禽养殖的深度融合发展,助推畜牧业节本增效与转型升级.