Invited review: sensors to support health management on dairy farms

A scoping review of 'big data', 'informatics', and 'bioinformatics' in the animal health and veterinary medical literature

Research in big data, informatics, and bioinformatics has grown dramatically (Andreu-Perez J, et al., 2015, IEEE Journal of Biomedical and Health Informatics 19, 1193-1208). Advances in gene sequencing technologies, surveillance systems, and electronic medical records have increased the amount of health data available. Unconventional data sources such as social media, wearable sensors, and internet search engine activity have also contributed to the influx of health data. The purpose of this study was to describe how 'big data', 'informatics', and 'bioinformatics' have been used in the animal health and veterinary medical literature and to map and chart publications using these terms through time. A scoping review methodology was used. A literature search of the terms 'big data', 'informatics', and 'bioinformatics' was conducted in the context of animal health and veterinary medicine. Relevance screening on abstract and full-text was conducted sequentially. In order for articles to be relevant, they must have used the words 'big data', 'informatics', or 'bioinformatics' in the title or abstract and full-text and have dealt with one of the major animal species encountered in veterinary medicine. Data items collected for all relevant articles included species, geographic region, first author affiliation, and journal of publication. The study level, study type, and data sources were collected for primary studies. After relevance screening, 1093 were classified. While there was a steady increase in 'bioinformatics' articles between 1995 and the end of the study period, 'informatics' articles reached their peak in 2012, then declined. The first 'big data' publication in animal health and veterinary medicine was in 2012. While few articles used the term 'big data' (n = 14), recent growth in 'big data' articles was observed. All geographic regions produced publications in 'informatics' and 'bioinformatics' while only North America, Europe, Asia, and Australia/Oceania produced publications about 'big data'. 'Bioinformatics' primary studies tended to use genetic data and tended to be conducted at the genetic level. In contrast, 'informatics' primary studies tended to use non-genetic data sources and conducted at an organismal level. The rapidly evolving definition of 'big data' may lead to avoidance of the term.

Short Communication: Association between neonatal calf diarrhea and lying behaviors

The objective of this study was to determine the association of neonatal calf diarrhea (NCD) with step activity and lying behaviors in pre-weaned dairy calves. Calves were housed in individual hutches for the first 6 days of life, and then moved into a group pen. On the day of birth, calves (n = 30) were fitted with an accelerometer, and step activity and lying behaviors were recorded. Calves were assigned a fecal score (FS) twice daily using a 0 to 3 scale, and were diagnosed with NCD (n = 10) when the score was a 3. To ensure the only association noted was due to NCD, calves that had any other health complications were excluded from analyses (n = 1). Calves with NCD were pair matched by age, breed, and birthdate to a healthy calf. Day 0 was designated as the date of NCD diagnosis. Calves with NCD spent less time lying (P < 0.05) and displayed more lying bouts (P < 0.05) of a shorter duration (P < 0.01) than healthy calves. Specifically, calves with NCD displayed more lying bouts on days -7 (P < 0.05), -6 (P < 0.01), -5 (P < 0.01), -4 (P < 0.01), and -3 (P < 0.05). Similarly, lying bout duration was shorter for calves with NCD on days -6 (P < 0.05), -5 (P < 0.05), -4 (P < 0.01), and -3 (P < 0.01). Additional research is needed to examine if these tools can be used to identify diseased calves prospectively.

Grazing Cow Behavior's Association with Mild and Moderate Lameness

Accelerometer-based mobility scoring has focused on cow behaviors such as lying and walking. Accuracy levels as high as 91% have been previously reported. However, there has been limited replication of results. Here, measures previously identified as indicative of mobility, such as lying bouts and walking time, were examined. On a research farm and a commercial farm, 63 grazing cows' behavior was monitored in four trials (16, 16, 16, and 15 cows) using leg-worn accelerometers. Seventeen good mobility (score 0), 23 imperfect mobility (score 1), and 22 mildly impaired mobility (score 2) cows were monitored. Only modest associations with activity, standing, and lying events were found. Thus, behavior monitoring appears to be insufficient to discern mildly and moderately impaired mobility of grazing cows.

Cost benefit analysis of automatic lameness detection systems in dairy herds: A dynamic programming approach

Epidemiological data establish that lameness is second only to mastitis as the dairy industry's most prevalent and costly animal welfare issue. Using an automatic lameness detection (ALD) system in which continuous, accurate detection is coupled with proper treatment, is key for reducing economic losses due to lameness. It is reasonable to assume that the cost of lameness would vary with its severity. Therefore, our first objective was to estimate the cost of different lameness severity levels as a function of milk production, lameness risk, conception probability, and treatment cost using a dynamic programming (DP) model. Our second objective was to conduct a cost benefit analysis for ALD systems which can reduce production losses through early detection and treatment of lameness, when compared to visual-detection (VD; i.e., performed by humans) systems. The default production loss parameters for the VD system used as inputs to the DP model were either sourced from the literature or were estimated based on data from a field trial. The production loss parameters for the ALD system used as inputs to the DP model were based on extrapolations of parameter values used for the VD system. The profit per present cow per year under assumed expenses and revenues decreased from $426.05 (when lameness incidence was assumed to be 0%) to $389.69 when lameness incidence was 19.5 %. Out of the 19.5 % lameness incidence in our default scenario, 9.8 % were moderate cases and 9.7 % were severe cases. Average cost of lameness was $36.36 at 19.5 % incidence. Average cost of lameness increased with increased incidence and was respectively $82.05, $195.05, and $286.87 at the low, medium, and high incidence scenarios. We used an operational framework which compared the lameness costs between the VD and ALD systems with 25 %, 50 % and 75 % net avoided costs (NAC) for the 10 year lifespan of the ALD system, at default, low, medium and high lameness incidence scenarios. The net return per cow per year from using an ALD system over a VD system was $13, at low incidence and 25 % NAC. The net return per cow per year for the ALD system was as high as $99 at high incidence and 75 % NAC. Out of 351 (3 system prices, 3 system efficiencies, 3 levels of lameness incidence and 13 different herd sizes) scenarios tested, 295 resulted in a net profit within the system lifespan of 10 years, thus justifying the investment in ALD systems.

Combination of Sensor Data and Health Monitoring for Early Detection of Subclinical Ketosis in Dairy Cows

Subclinical ketosis is a metabolic disease in early lactation. It contributes to economic losses because of reduced milk yield and may promote the development of secondary diseases. Thus, an early detection seems desirable as it enables the farmer to initiate countermeasures. To support early detection, we examine different types of data recordings and use them to build a flexible algorithm that predicts the occurence of subclinical ketosis. This approach shows promising results and can be seen as a step toward automatic health monitoring in farm animals.

Symposium review: Considerations for the future of dairy cattle housing: An animal welfare perspective

Many contemporary dairy cattle housing practices are at odds with societal perceptions of positive animal welfare. The public (i.e., those external to the dairy industry) typically emphasizes the importance of naturalness for dairy cattle, such as through provision of pasture, freedom of movement, and the ability to interact socially with conspecifics. Yet, in the United States, the majority of lactating dairy cattle are reportedly housed without any access to pasture, and almost 39% of dairy farms use tiestalls, which restrict movement and social interactions. In addition to being in conflict with public expectations, a lack of pasture access and restrictive housing systems are also in conflict with the animals' own motivations, which can adversely affect their welfare. For example, dairy cattle are highly motivated to access pasture and show a reduction in oral stereotypies when allowed on pasture after periods of tethering. Calves housed without social contact have cognitive deficits and exhibit increased fear responses to novelty. We argue that the long-term sustainability of the dairy industry will depend on the extent to which housing systems reflect public concerns and the animals' priorities. The adoption of technologies, such as automated feeders and remote monitoring systems, may represent a means to practically promote the animals' natural behavior while simultaneously improving individualized care. Although older generations of the public may consider technological solutions to be a further deviation from naturalness and a departure from dairy farming's agrarian roots, the definition of "naturalness" for younger generations may well have expanded to include technology. As the buying power shifts to these younger generations, the adoption of technologies that promote natural cattle behaviors may be one means toward reconciling the disconnect between public perceptions of animal welfare and contemporary dairy farming practices.

Early pregnancy diagnoses based on physiological indexes of dairy cattle: a review

With the development of standardization and scaling in the dairy farming industry, timely and accurate pregnancy diagnosis is required to improve the benefits of breeding by shortening the calving interval. However, the current pregnancy diagnostic methods cannot meet the requirements of the industry. Here, we review changes in the physiological indexes and in the function and morphological status of the reproductive organs of dairy cows at the early stages of pregnancy. In addition, the corresponding pregnancy diagnostic methods based on certain indexes are well development, and the pregnancy diagnostic approaches based on remote sensing and automation technology have become inevitable trends in the industry. These applications will reveal physiological regularities in pregnancy and benefit the detailed management of dairy cows during early pregnancy.

Boosted trees to predict pneumonia, growth, and meat percentage of growing-finishing pigs1

In pig production, efficiency is benefiting from uniform growth in pens resulting in single deliveries from a pen of possibly all animals in the targeted weight range. Abnormalities, like pneumonia or aberrant growth, reduce production efficiency as it reduces the uniformity and might cause multiple deliveries per batch and pigs delivered with a low meat yield or outside the targeted weight range. Early identification of pigs prone to develop these abnormalities, for example, at the onset of the growing-finishing phase, would help to prevent heterogeneous pens through management interventions. Data about previous production cycles at the farm combined with data from the piglet’s own history may help in identifying these abnormalities. The aim of this study, therefore, was to predict at the onset of the growing-finishing phase, that is, at 3 mo in advance, deviant pigs at slaughter with a machine-learning technique called boosted trees. The dataset used was extracted from the farm management system of a research center. It contained over 70,000 records of individual pigs born between 2004 and 2016, including information on, for example, offspring, litter size, transfer dates between production stages, their respective locations within the barns, and individual live-weights at several production stages. Results obtained on an independent test set showed that at a 90% specificity rate, the sensitivity was 16% for low meat percentage, 20% for pneumonia and 36% for low lifetime growth rate. For low lifetime growth rate, this meant an almost three times increase in positive predictive value compared to the current situation. From these results, it was concluded that routine performance information available at the onset of the growing-finishing phase combined with data about previous production cycles formed a moderate base to identify pigs prone to develop pneumonia (AUC > 0.60) and a good base to identify pigs prone to develop growth aberrations (AUC > 0.70) during the growing-finishing phase. The mentioned information, however, was not a sufficient base to identify pigs prone to develop low meat percentage (AUC < 0.60). The shown ability to identify growth aberrations and pneumonia can be considered a good first step towards the development of an early warning system for pigs in the growing-finishing phase.

New Zealand dairy farmers preference investments in automation technology over decision-support technology

Dairy farmers are adopting precision technologies to assist with milking and managing their cows due to increased herd sizes and a desire to improve labour efficiency, productivity and sustainability. In the present study, we evaluated the adoption of technologies installed at or near the dairy, and milking practices, on New Zealand dairy farms. These data quantify current use of technology for milking and labour efficiency, and decision-making, and provide insight into future technology adoption. A telephone survey of 500 farmers, randomly selected from a database of New Zealand dairy farms, was conducted in 2018. Adoption for all farms is indicated for six automation technologies, including automatic cup removers (39%), automatic drafting (24%), automatic teat spraying (29%), computer-controlled in-shed feeding (29%), automatic plant wash (18%) and automatic yard wash systems (27%). Five data-capture technologies also included in the survey were electronic milk meters (8%), automatic animal weighing (7%), in-line mastitis detection (7%), automatic heat detection (3%) and electronic animal-identification readers (23%). Analysis by dairy type indicated an adoption level for the automation technologies in rotary dairies of 36–77%, and 7–49% for data-capture technologies, with 10% having none of these 11 technologies installed. This compares with herringbone dairies at 4–21% and 2–11% for automation and data-capture technologies respectively, with 56% having none of these technologies. Rotary dairies, with a combination of automatic cup removers, automatic teat spraying, and automatic drafting, were associated with 43% higher labour efficiency (cows milked/h.person) and 14% higher milking efficiency (cows milked/h) than were rotary dairies without all three technologies. Dairy farmers will increasingly use technologies that deliver value, and the present study has provided information to guide investment decisions, product development and research in areas such as applying technology in new workplaces.

Getting value from artificial intelligence in agriculture

Artificial intelligence (AI) is beginning to live up to its promise of delivering real value, driven by recent advances in the availability of relevant data, computation and algorithms. In the present paper, I discuss the value to agriculture from AI over the next decade. The more immediate applications will be to improve precision information about what is happening on the farm by improving what is being detected and measured. A consequence of this are more accurate alerts to farmers. Another is an increased ability to understand why phenomena occur in farm systems, so as to improve their management. From improved data and understanding come improved predictions, enabling more optimal decisions about how to manage farm systems and stimulating the development of decision support and recommender systems. In many cases, robotics and automated systems will remove much of the need for human decision-making and improve farm efficiencies and farm health. Artificial intelligence will also be needed to enable organisations to harness the value of information distributed throughout supply chains, including farm data. Digital twins will also emerge as an important paradigm to improve how information about farm entities is organised to support decision-making. There are also likely to be negative impacts from AI, such as disruption to the roles and skills needed from farm workers, indicating the need to consider the social and ethical impacts of AI each time a new capability is introduced. I conclude that understanding these challenges more deeply tends to highlight new opportunities for positive change.

Identifying gram-negative and gram-positive clinical mastitis using daily milk component and behavioral sensor data

Opportunities exist for automated animal health monitoring and early detection of diseases such as mastitis with greater on-farm adoption of precision technologies. Our objective was to evaluate time series changes in individual milk component or behavioral variables for all clinical mastitis (CM) cases (ACM), for CM caused by gram-negative (GN) or gram-positive (GP) pathogens, or CM cases in which no pathogen was isolated (NPI). We developed algorithms using a combination of milk and activity parameters for predicting each of these infection types. Milk and activity data were collated for the 14 d preceding a CM event (n = 170) and for controls (n = 166) matched for breed, parity, and days in milk. Explanatory variables in the univariate and multiple regression models were the slope change in milk (milk yield, conductivity, somatic cell count, lactose percentage, protein percentage, and fat percentage) and activity parameters (steps, lying time, lying bout duration, and number of lying bouts) over 7 d. Slopes were estimated using linear regression between d -7 and -5, d -7 and -4, d -7 and -3, d -7 and -2, and d -7 and -1 relative to CM detection for all parameters. Univariate analyses determined significant slope ranges for explanatory variables against the 4 responses: ACM, GN, GP, and NPI. Next, all slope ranges were offered into the multivariate models for the same 4 responses using 3 baselines: d -10, -7, and -3 relative to CM detection. In the univariate analysis, no explanatory variables were significant indicators of ACM, whereas at least 1 parameter was significant for each of GN, GP, and NPI models. Superior sensitivity (Se) and specificity (Sp) estimates were observed for the best GP (Se = 82%, Sp = 87%) and NPI (Se = 80%, Sp = 94%) multiple regression models compared with the best ACM (Se = 73%, Sp = 75%) and GN (Se = 71%, Sp = 74%) models. Sensitivity for the GN model was greater at the baseline closest to the day of CM detection (d -3), whereas the opposite was observed for the GP and NPI model as Se was maximized at the d -10 baseline. Based on this screening of relationships, milk and activity sensor data could be used in CM detection systems.

A Vision for Development and Utilization of High-Throughput Phenotyping and Big Data Analytics in Livestock

Automated high-throughput phenotyping with sensors, imaging, and other on-farm technologies has resulted in a flood of data that are largely under-utilized. Drastic cost reductions in sequencing and other omics technology have also facilitated the ability for deep phenotyping of livestock at the molecular level. These advances have brought the animal sciences to a cross-roads in data science where increased training is needed to manage, record, and analyze data to generate knowledge and advances in Agriscience related disciplines. This paper describes the opportunities and challenges in using high-throughput phenotyping, “big data,” analytics, and related technologies in the livestock industry based on discussions at the Livestock High-Throughput Phenotyping and Big Data Analytics meeting, held in November 2017 (see: https://www.animalgenome.org/bioinfo/community/workshops/2017/). Critical needs for investments in infrastructure for people (e.g., “big data” training), data (e.g., data transfer, management, and analytics), and technology (e.g., development of low cost sensors) were defined by this group. Though some subgroups of animal science have extensive experience in predictive modeling, cross-training in computer science, statistics, and related disciplines are needed to use big data for diverse applications in the field. Extensive opportunities exist for public and private entities to harness big data to develop valuable research knowledge and products to the benefit of society under the increased demands for food in a rapidly growing population.

Key animal welfare issues in commercially raised dairy calves: social environment, nutrition, and painful procedures

Dairy calf welfare concerns are growing and new evidence suggests that the early life environment influences appropriate physical, behavioral, and cognitive development lasting into adulthood. This review highlights key evidence for the impacts of housing, diets, and painful procedures on calf welfare. We argue that these topics are currently critical welfare concerns, but are not the only points of concern. In addition to environmental requirements to maintain optimal health, dairy calves experience other challenges including social and nutritional restrictions. Individual housing is associated with impaired behavioral development and cognitive ability. Pair and group housing can mitigate some of these negative effects and should be encouraged. Restrictive milk allowances (<15% of body weight) lead to poor growth and hunger; these welfare concerns can be addressed with proper enhanced milk allowances and gradual weaning programs. Finally, dehorning is a critical animal welfare issue when pain control is withheld; calves show negative behavioral, physiological, and emotional responses during and after dehorning. The combined use of local anaesthetics and analgesics can mitigate these effects. An industry shift toward providing social companionship, enhanced milk allowances, and pain control during painful procedures would help to improve the welfare of dairy calves in intensive commercial rearing facilities.

On-farm use of disease alerts generated by precision dairy technology

Wearable precision dairy monitoring (PDM) technologies currently used to detect estrus may provide insight into disease detection. However, the incorporation of PDM into farm management and its perceived usefulness for dairy producers have not been explored. As the targeted end users of these products, information is needed on how producers use generated disease alerts as well as barriers to adoption and usefulness. The objective of this research was to assess the perceived usefulness producers attributed to alerts from a daily generated alert list designed to identify sick or injured cows or cows that experienced a major management change. Data from 1,171 cows on 4 commercial farms in Kentucky were collected from October 2015 to October 2016. Each cow was equipped with 2 PDM technologies: a leg tag (measuring activity in steps/d and lying time in h/d) and a neck collar (measuring eating time in h/d). Alerts were generated based on an individual cow's decrease of ≥30% in activity, lying, and eating time compared with each cow's 10-d moving mean. Producers sorted alerts into 3 overall categories: (1) the cow alert was perceived to be true and the cow was visually checked, (2) the cow alert was perceived to be true, but the cow was not visually checked, and (3) the cow alert behavior change was doubted by the producer and the cow was not visually checked. Further subdivisions were also provided to explain the choice for an overall category. Over the year, 24,012 cow alerts were generated (eating time, n = 9,543; lying time, n = 9,777; activity, n = 1,590; or a combination of behaviors, n = 3,102). Only 8% of the alerts were doubted by the producer. Although 55% of alerts were perceived to be true, producers visually assessed cows based on only 21% of the alerts with a large variation between farms (2 to 45% of the alerts visually assessed). Producers were more likely to completely ignore alerts over time. Producers were more likely to perceive cow alerts to be true and visually check cows when ≤20 alerts occurred per day, cows were fresh or in early lactation, alerts occurred during the work week, or when cow alerts were for eating time, activity, or a combination of multiple behaviors. Behavioral disease alerts must be improved and correspond to an actionable change for producers to use them. Incorporating herd management software, creating and managing alerts by lactation stage, and focusing on behaviors producers already find useful could improve future alert utilization.

The Use of an Activity Monitoring System for the Early Detection of Health Disorders in Young Bulls

Simple Summary

In large intensive beef production systems, the identification of sick animals is difficult. We hypothesized that sick bulls would change daily activities when sick. Thus, the use of activity monitoring devices might allow for the early identification of sick bulls. The device used measured steps counts, lying time, lying bouts, and frequency and time at the feed bunk. Sick bulls started to behave differently from healthy bulls at least 10 days before the appearance of clinical signs. The prediction model identified bulls at risk of becoming sick 9 days before the visual diagnostic based on the time attending to the feed bunk, the time lying, and the frequency of lying bouts. The validation indicated that the prediction resulted in 50% false positives and 7% false negatives. Activity monitoring systems may be useful tools to identify bulls at risk of becoming sick.

Abstract

Bulls (n = 770, average age = 127 days, SD = 53 days of age) were fitted with an activity monitoring device for three months to study if behavior could be used for early detection of diseases. The device measured the number of steps, lying time, lying bouts, and frequency and time of attendance at the feed bunk. All healthy bulls (n = 699) throughout the trial were used to describe the normal behavior. A match-pair test was used to assign healthy bulls for the comparison vs. sick bulls. The model was developed with 70% of the data, and the remaining 30% was used for the validation. Healthy bulls did 2422 ± 128 steps/day, had 28 ± 1 lying bouts/day, spent 889 ± 12 min/day lying, and attended the feed bunk 8 ± 0.2 times/d for a total of 95 ± 8 min/day. From the total of bulls enrolled in the study, 71 (9.2%) were diagnosed sick. Their activities changed at least 10 days before the clinical signs of disease. Bulls at risk of becoming sick were predicted 9 days before clinical signs with a sensitivity and specificity of 79% and 81%, respectively. The validation of the model resulted in a sensitivity, specificity, and accuracy of 92%, 42%, and 82 %, respectively, and a 50% false positive and 12.5% false negative rates. Results suggest that activity-monitoring systems may be useful in the early identification of sick bulls. However, the high false positive rate may require further refinement.

Evaluation of lameness detection using radar sensing in ruminants

BACKGROUND

Lameness is a major health, welfare and production-limiting condition for the livestock industries. The current 'gold-standard' method of assessing lameness by visual locomotion scoring is subjective and time consuming, whereas recent technological advancements have enabled the development of alternative and more objective methods for its detection.

METHODS

This study evaluated a novel lameness detection method using micro-Doppler radar signatures to categorise animals as lame or non-lame. Animals were visually scored by veterinarian and radar data were collected for the same animals.

RESULTS

A machine learning algorithm was developed to interpret the radar signatures and provide automatic classification of the animals. Using veterinary scoring as a standard method, the classification by radar signature provided 85 per cent sensitivity and 81 per cent specificity for cattle and 96 per cent sensitivity and 94 per cent specificity for sheep.

CONCLUSION

This radar sensing method shows promise for the development of a highly functional, rapid and reliable recognition tool of lame animals, which could be integrated into automatic, on-farm systems for sheep and cattle.

Prediction of metabolic status of dairy cows in early lactation with on-farm cow data and machine learning algorithms

Metabolic status of dairy cows in early lactation can be evaluated using the concentrations of plasma β-hydroxybutyrate (BHB), free fatty acids (FFA), glucose, insulin, and insulin-like growth factor 1 (IGF-1). These plasma metabolites and metabolic hormones, however, are difficult to measure on farm. Instead, easily obtained on-farm cow data, such as milk production traits, have the potential to predict metabolic status. Here we aimed (1) to investigate whether metabolic status of individual cows in early lactation could be clustered based on their plasma values and (2) to evaluate machine learning algorithms to predict metabolic status using on-farm cow data. Through lactation wk 1 to 7, plasma metabolites and metabolic hormones of 334 cows were measured weekly and used to cluster each cow into 1 of 3 clusters per week. The cluster with the greatest plasma BHB and FFA and the lowest plasma glucose, insulin, and IGF-1 was defined as poor metabolic status; the cluster with the lowest plasma BHB and FFA and the greatest plasma glucose, insulin, and IGF-1 was defined as good metabolic status; and the intermediate cluster was defined as average metabolic status. Most dairy cows were classified as having average or good metabolic status, and a limited number of cows had poor metabolic status (10-50 cows per lactation week). On-farm cow data, including dry period length, parity, milk production traits, and body weight, were used to predict good or average metabolic status with 8 machine learning algorithms. Random Forest (error rate ranging from 12.4 to 22.6%) and Support Vector Machine (SVM; error rate ranging from 12.4 to 20.9%) were the top 2 best-performing algorithms to predict metabolic status using on-farm cow data. Random Forest had a higher sensitivity (range: 67.8-82.9% during wk 1 to 7) and negative predictive value (range: 89.5-93.8%) but lower specificity (range: 76.7-88.5%) and positive predictive value (range: 58.1-78.4%) than SVM. In Random Forest, milk yield, fat yield, protein percentage, and lactose yield had important roles in prediction, but their rank of importance differed across lactation weeks. In conclusion, dairy cows could be clustered for metabolic status based on plasma metabolites and metabolic hormones. Moreover, on-farm cow data can predict cows in good or average metabolic status, with Random Forest and SVM performing best of all algorithms.

Automatic recording of individual oestrus vocalisation in group-housed dairy cattle: development of a cattle call monitor

Oestrus detection remains a problem in the dairy cattle industry. Therefore, automatic detection systems have been developed to detect specific behavioural changes at oestrus. Vocal behaviour has not been considered in such automatic oestrus detection systems in cattle, though the vocalisation rate is known to increase during oestrus. The main challenge in using vocalisation to detect oestrus is correctly identifying the calling individual when animals are moving freely in large groups, as oestrus needs to be detected at an individual level. Therefore, we aimed to automate vocalisation recording and caller identification in group-housed dairy cows. This paper first presents the details of such a system and then presents the results of a pilot study validating its functionality, in which the automatic detection of calls from individual heifers was compared to video-based assessment of these calls by a trained human observer, a technique that has, until now, been considered the 'gold standard'. We developed a collar-based cattle call monitor (CCM) with structure-borne and airborne sound microphones and a recording unit and developed a postprocessing algorithm to identify the caller by matching the information from both microphones. Five group-housed heifers, each in the perioestrus or oestrus period, were equipped with a CCM prototype for 5 days. The recorded audio data were subsequently analysed and compared with audiovisual recordings. Overall, 1404 vocalisations from the focus heifers and 721 vocalisations from group mates were obtained. Vocalisations during collar changes or malfunctions of the CCM were omitted from the evaluation. The results showed that the CCM had a sensitivity of 87% and a specificity of 94%. The negative and positive predictive values were 80% and 96%, respectively. These results show that the detection of individual vocalisations and the correct identification of callers are possible, even in freely moving group-housed cattle. The results are promising for the future use of vocalisation in automatic oestrus detection systems.

SmartHerd management: A microservices-based fog computing-assisted IoT platform towards data-driven smart dairy farming

Internet of Things (IoT), fog computing, cloud computing, and data-driven techniques together offer a great opportunity for verticals such as dairy industry to increase productivity by getting actionable insights to improve farming practices, thereby increasing efficiency and yield. In this paper, we present SmartHerd, a fog computing-assisted end-to-end IoT platform for animal behavior analysis and health monitoring in a dairy farming scenario. The platform follows a microservices-oriented design to assist the distributed computing paradigm and addresses the major issue of constrained Internet connectivity in remote farm locations. We present the implementation of the designed software system in a 6-month mature real-world deployment, wherein the data from wearables on cows is sent to a fog-based platform for data classification and analysis, which includes decision-making capabilities and provides actionable insights to farmer towards the welfare of animals. With fog-based computational assistance in the SmartHerd setup, we see an 84% reduction in amount of data transferred to the cloud as compared with the conventional cloud-based approach.

Remote Identification of Sheep with Flystrike Using Behavioural Observations

Simple Summary

Flystrike in sheep is a common condition in Australia where parasitic flies lay eggs on soiled wool or open wounds; and the resulting maggots feed off the flesh. Identification of ‘flystruck’ individuals is crucial for treatment; but requires labour-intensive physical examination of every animal. The aim of this study was to investigate the behaviour of sheep; while they remained in the paddock; to try and visually distinguish those suffering from flystrike. Observers who were blinded to the flystrike status of the sheep were asked to score the animal’s body language from video footage. These scores were then compared with the condition of the wool and whether the sheep were flystruck. The observers found that the flystruck sheep exhibited behavioural characteristics that corresponded to the flystrike severity and the condition of the wool around the tail (breech) of the sheep. We therefore conclude that behavioural monitoring of sheep in the paddock could be used to identify animals that had flystrike.

Abstract

Flystrike is a major problem affecting sheep in Australia. Identification of ‘flystruck’ individuals is crucial for treatment; but requires labour-intensive physical examination. As the industry moves toward more low-input systems; there is a need for remote methods to identify flystruck individuals. The aim of this study was to investigate the behaviour of sheep with breech flystrike within a paddock setting. Video footage of sixteen Merino sheep; eight later confirmed with flystrike and eight without; was collected as they moved freely within the paddock with conspecifics. Quantitative behavioural measurements and a qualitative behavioural assessment (QBA) were conducted and compared to their breech conditions (i.e., faecal/urine staining; flystrike severity). Both qualitative and quantitative assessments indicated behavioural differences between flystruck and non-flystruck animals. Flystruck sheep had a behavioural profile characterised by restless behaviour; abnormal postures and reduced grazing time (p < 0.05). Furthermore; flystruck sheep were scored to have a more ‘exhausted/irritated’ demeanour using QBA (p < 0.05). The behavioural responses also corresponded to the flystrike severity scores and condition of the breech area. We conclude that remotely assessed behaviour of flystruck sheep diverges markedly from non-flystruck sheep; and thus could be a low-input method for identifying and treating affected animals.

A Survey of Italian Dairy Farmers' Propensity for Precision Livestock Farming Tools

A targeted survey was designed with the aim of describing the diffusion of precision livestock farming (PLF) tools in one of the most intensive dairy farming provinces in Italy. Technicians at the Provincial Breeder Association of Cremona interviewed 490 dairy farmers and obtained data regarding the role and age of the respondents; the land owned by the farmers; their herd sizes (HS, lactating plus dry cows; small HS < 101, medium HS 101-200, large HS > 200 cows/herd); their average 305 day milk yield (low MY < 9501, medium MY 9501-10,500, high MY > 10,500 kg/head); the cow to employed worker ratio (low CW < 33, medium CW 33-47, high CW > 47 cows/worker); the use of PLF tools to monitor production, reproduction, and health; and the criteria and motivations for investing in PLF tools. The use of automated MY recording and estrus detection systems was primarily associated with HS (more present in larger farms), followed by MY (more present in more productive farms), and then CW (more present with a high cow: worker ratio). Concern about the time required to manage data was the most common subjective issue identified as negatively affecting the purchase of these tools. The future of PLF use in this region will depend upon the availability of an effective selection of tools on the market.

Livestock Genomics for Developing Countries - African Examples in Practice

African livestock breeds are numerous and diverse, and typically well adapted to the harsh environment conditions under which they perform. They have been used over centuries to provide livelihoods as well as food and nutritional security. However, African livestock systems are dynamic, with many small- and medium-scale systems transforming, to varying degrees, to become more profitable. In these systems the women and men livestock keepers are often seeking new livestock breeds or genotypes - typically those that increase household income through having enhanced productivity in comparison to traditional breeds while maintaining adaptedness. In recent years genomic approaches have started to be utilized in the identification and development of such breeds, and in this article we describe a number of examples to this end from sub-Saharan Africa. These comprise case studies on: (a) dairy cattle in Kenya and Senegal, as well as sheep in Ethiopia, where genomic approaches aided the identification of the most appropriate breed-type for the local productions systems; (b) a cross-breeding program for dairy cattle in East Africa incorporating genomic selection as well as other applications of genomics; (c) ongoing work toward creating a new cattle breed for East Africa that is both productive and resistant to trypanosomiasis; and (d) the use of African cattle as resource populations to identify genomic variants of economic or ecological significance, including a specific case where the discovery data was from a community based breeding program for small ruminants in Ethiopia. Lessons learnt from the various case studies are highlighted, and the concluding section of the paper gives recommendations for African livestock systems to increasingly capitalize on genomic technologies.

Automatic weaning based on individual solid feed intake: Effects on behavior and performance of dairy calves

Calves are typically weaned from milk to solids once they reach a predetermined age or when they are consuming a predetermined intake of solids. The first aim of this study was to compare feeding behavior and performance of calves weaned based on age versus starter intake. The latter method can result in considerable variation in the age at which calves are weaned, so a secondary aim was to compare calves that weaned early or late when weaned based on starter intake. In experiment 1, dairy calves were randomly assigned to be either (1) weaned by age at d 70 (n = 16), or (2) weaned by intake, where calves were weaned based on starter intake (n = 16). All calves were fed using an automatic milk feeder and offered 12 L/d of milk until 30 d of age. On d 31, all calves had their milk rations reduced. Calves weaned by age were reduced to 6 L/d of milk over 5 d and received 6 L/d milk from d 35 until d 63, when milk was reduced over 7 d until complete weaning at d 70. For calves weaned by intake, the milk ration was reduced on d 31 to 75% of that calf's previous milk intake (3-d average) and further reduced by 25% when the calf met each of 3 targets for starter intake: 225, 675, and 1,300 g/d. Calves that failed to reach the final target by d 63 (failed-intake group; n = 6) were weaned over 7 d to complete weaning at d 70. Ten calves met all 3 starter intake targets (successful-intake group). In experiment 2, all calves were assigned to the weaned-by-intake treatment (n = 48). The weaning strategy was identical to that described for experiment 1, but calves were permitted up to d 84 to reach the final starter intake target. Forty-three calves met all 3 targets and were retrospectively divided into early-weaning (weaned before d 63; n = 31) and late-weaning (weaned on or after d 63; n = 12) categories. In both experiments, the weaning period was considered from the time of initial milk reduction at d 31 until complete weaning at d 70 (weaned by age) or when consuming 1,300 g/d (weaned by intake). Postweaning growth was monitored from weaning until final weight in the calf-rearing period at d 98 (experiment 1) and d 105 (experiment 2). Final weight in the grower period was measured at d 134 (experiment 1) and d 145 (experiment 2). In experiment 1, successful-intake calves (vs. calves weaned by age) consumed 125.3 ± 16.4 L less milk and 41.3 ± 9.3 kg more starter over the experimental period, engaged in more unrewarded visits to the milk feeder during weaning (11.1 ± 1.5 vs. 5.0 ± 1.3 visits/d), and achieved similar weights at the end of the grower period (188.2 ± 6.6 vs. 195.2 ± 5.7 kg). In experiment 2, calves that weaned by intake early (vs. late) consumed 93.3 ± 26.0 L less milk and 57.2 ± 12.2 kg more starter, engaged in a similar number of unrewarded visits during weaning (7.0 ± 0.6 vs. 7.6 ± 1.0 visits/d), had greater average daily gain during weaning (1.08 ± 0.02 vs. 0.94 ± 0.03 kg/d), and achieved greater final weights at the end of the grower period (203.2 ± 2.9 vs. 192.6 ± 4.2 kg). These results indicate that calves weaned based on starter intake can achieve similar weights to those weaned by age, despite consuming less milk. However, some calves will fail to meet starter intake targets unless given sufficient time to do so. Variation in preweaning feed intake provides an opportunity for individualized management of calves.

Farm-specific failure costs of production disorders in European organic dairy herds

On-farm decision support in animal health management requires a tailor-made failure costs (FCs) assessment of production disorders for the individual farm. In our study we defined a generic framework to estimate the FC of production disorders in dairy cows. We converted the framework to a practical tool in which the farm-specific FC of mastitis, ketosis, lameness and metritis were estimated for 162 organic dairy farms in four European countries. Along with the structure of the framework, the FC estimation required three distinct types of model input: performance input (related to herd performance parameters), consequential input (related to the consequences of the disorders) and economic input (related to price levels). Input was derived from official herd recordings (e.g. test-day records and animal health recordings) and farmers' responses (e.g. questionnaire replies). The average FC of mastitis, ketosis, lameness and metritis amounted to € 96, € 21, € 43 and € 10 per cow per year, respectively. The variation in FC outcomes was high among farmers and countries. Overall ranking of the disorders based on absolute values was the same for all countries, with mastitis being the costliest disorder followed in order by lameness, ketosis, and metritis. Farm specific estimates can be used to rank production related disorders in terms of their associated failure costs and thus provide valuable insights for herd health management. The practical calculation tool developed in this study should be considered by farmers or herd health advisors to support their animal health practices or advice.

Symposium review: Challenges and opportunities for evaluating and using the genetic potential of dairy cattle in the new era of sensor data from automation

Sensor data from automation are becoming available on an increasingly large scale, and associated research is slowly starting to appear. This new era of sensor data from automation leads to many challenges but also new opportunities for assessing and maximizing the genetic potential of dairy cattle. The first challenge is data quality, because all uses of sensor data require careful data quality validation, potentially using external references. The second issue is data accessibility. Indeed, sensor data generated from automation are often designed to be available on-farm in a given system. However, to make these data useful-for genetic improvement for example-the data must also be made available off-farm. By nature, sensor data often are very complex and diverse; therefore, a data consolidation and integration layer is required. Moreover, the traits we want to select have to be defined precisely when generated from these raw data. This approach is obviously also beneficial to limit the challenge of extremely high data volumes generated by sensors. An additional challenge is that sensors will always be deployed in a context of herd management; therefore, any efforts to make them useful should focus on both breeding and management. However, this challenge also leads to opportunities to use genomic predictions based on these novel data for breeding and management. Access to relevant phenotypes is crucial for every genomic evaluation system. The automatic generation of training data, on both the phenotypic and genomic levels, is a major opportunity to access novel, precise, continuously updated, and relevant data. If the challenges of bidirectional data transfer between farms and external databases can be solved, new opportunities for continuous genomic evaluations integrating genotypes and the most current local phenotypes can be expected to appear. Novel concepts such as federated learning may help to limit exchange of raw data and, therefore, data ownership issues, which is another important element limiting access to sensor data. Accurate genome-guided decision-making and genome-guided management of dairy cattle should be the ultimate way to add value to sensor data from automation. This could also be the major driving force to improve the cost-benefit relationship for sensor-based technologies, which is currently one of the major obstacles for large-scale use of available technologies.

Evaluation of an ear-attached accelerometer for detecting estrus events in indoor housed dairy cows

Precision dairy farming technologies have tremendous potential to improve and support farmers in herd management decisions, particularly in reproductive management. Nowadays, estrus detection in cows is challenging and several supporting tools are available. In this study, a 3D-accelerometer integrated into an ear-tag (SMARTBOW, Smartbow GmbH, Weibern, Austria) was used for the detection of cows in estrus. Movement pattern based on accelerometer data were analyzed and processed by algorithms and machine learning, resulting in estrus alerts. For the evaluation of the system, reproductive performance data of 579 estrus events of multiparous cows were used to retrospectively evaluate the accuracy of estrus alerts generated by the accelerometer-based system and the overall performance of the system. Estrus events were classified as 'gold standard' events, if an estrus followed by AI resulted in pregnancy, and as 'recorded estrus' events, if two estrus events with an interval of 18-25 d were in the herd records, independent of whether estrus was followed by AI or pregnancy. In total, 316 'gold standard' events were matched with estrus alerts generated by the accelerometer-based system, resulting in a sensitivity of 97%. Furthermore, 263 'recorded estrus' events were compared with correct or incorrect estrus alerts by the system. Sensitivity, specificity, positive and negative predictive values, accuracy, and error rate for 'recorded estrus' events were 97%, 98%, 96%, 94%, 96%, and 2%, respectively. In summary, the SMARTBOW system is suitable for an automated detection of estrus events of multiparous cows in indoor housed dairy cows.

Automatic lameness detection in cattle

There is an increasing demand for health and welfare monitoring in modern dairy farming. The development of various innovative techniques aims at improving animal behaviour monitoring and thus animal welfare indicators on-farm. Automated lameness detection systems have to be valid, reliable and practicable to be applied in veterinary practice or under farm conditions. The objective of this literature review was to describe the current automated systems for detection of lameness in cattle, which have been recently developed and investigated for application in dairy research and practice. The automatic methods of lameness detection broadly fall into three categories: kinematic, kinetic and indirect methods. The performance of the methods were compared with the reference standard (locomotion score and/or lesion score) and evaluated based on level-based scheme defining the degree of development (level I, sensor technique; level II, validation of algorithm; level III, performance for detection of lameness and/or lesion; level IV, decision support with early warning system). Many scientific studies have been performed on levels I-III, but there are no studies of level IV technology. The adoption rate of automated lameness detection systems by herd managers mainly yields returns on investment by the early identification of lame cows. Long-term studies, using validated automated lameness detection systems aiming at early lameness detection, are still needed in order to improve welfare and production under field conditions.

Invited review: Big Data in precision dairy farming

Insight into current scientific applications of Big Data in the precision dairy farming area may help us to understand the inflated expectations around Big Data. The objective of this invited review paper is to give that scientific background and determine whether Big Data has overcome the peak of inflated expectations. A conceptual model was created, and a literature search in Scopus resulted in 1442 scientific peer reviewed papers. After thorough screening on relevance and classification by the authors, 142 papers remained for further analysis. The area of precision dairy farming (with classes in the primary chain (dairy farm, feed, breed, health, food, retail, consumer) and levels for object of interest (animal, farm, network)), the Big Data-V area (with categories on Volume, Velocity, Variety and other V’s) and the data analytics area (with categories in analysis methods (supervised learning, unsupervised learning, semi-supervised classification, reinforcement learning) and data characteristics (time-series, streaming, sequence, graph, spatial, multimedia)) were analysed. The animal sublevel, with 83% of the papers, exceeds the farm sublevel and network sublevel. Within the animal sublevel, topics within the dairy farm level prevailed with 58% over the health level (33%). Within the Big Data category, the Volume category was most favoured with 59% of the papers, followed by 37% of papers that included the Variety category. None of the papers included the Velocity category. Supervised learning, representing 87% of the papers, exceeds unsupervised learning (12%). Within supervised learning, 64% of the papers dealt with classification issues and exceeds the regression methods (36%). Time-series were used in 61% of the papers and were mostly dealing with animal-based farm data. Multimedia data appeared in a greater number of recent papers. Based on these results, it can be concluded that Big Data is a relevant topic of research within the precision dairy farming area, but that the full potential of Big Data in this precision dairy farming area is not utilised yet. However, the present authors expect the full potential of Big Data, within the precision dairy farming area, will be reached when multiple Big Data characteristics (Volume, Variety and other V’s) and sources (animal, groups, farms and chain parts) are used simultaneously, adding value to operational and strategic decision.

Automatically monitoring of dietary effects on rumination and activity of finishing heifers

Rumination and activity behaviours are important welfare indexes in beef-cattle housing. The main objective of the present study was to assess whether the automatic collars are able to reliably assess the rumination and activity patterns of beef heifers. For this purpose, individual rumination time and activity (RT and AT respectively) were continuously recorded using an automatic neck-collar system (Hr-Tag, SCR Engineers Ltd, Israel) on the three trials performed. For Experiment 1, 60 Italian crossbred heifers were randomly assigned to one of two experimental diets for 9 months: the corn–silage diet (CS), which included 50% forage on a dry-matter (DM) basis (43% corn–silage, 7% wheat straw) and the hay diet (HAY), with 57% forage on DM basis (28.5% grass hay, 28.5% alfalfa hay). Heifers consuming HAY diet showed greater (P &lt; 0.05) RT (min/day) and AT (bits/day) than did those on CS diet. Rumination time per kilogram of dry matter and per kilogram of amylase- and sodium sulfite-treated neutral detergent fibre corrected for ash residue (aNDFom) intake were similar in the two experimental groups, while RT per kilogram of physical effective NDF (peNDF) intake was greater (P &lt; 0.05) in the CS group than in the HAY one. Daily rumination and activity patterns (min/2 h and bits/2 h respectively) were significantly different between CS and HAY groups. In the second experiment, 32 beef heifers were randomly allocated in two homogeneous pens, each containing eight animals, and two non-homogeneous ones, in which animals were added at different periods. The AT of non-homogeneous pens was significantly higher than that of the homogeneous ones, suggesting a distress condition for values higher than 309 bits/day AT. In the third experiment, RT and AT of three animals with respiratory disease were collected using the automatic collar system and compared with AT and RT of the healthy animals. Sick animals presented a significantly lower RT and higher AT than did the healthy ones. The cut-off to distinguish sick from healthy heifers was set to 537 bits/day AT and 381 min/day RT. In conclusion, the present study demonstrated that automatic collars can reliably monitor different rumination and activity behaviours of beef animals in various management conditions and different health statuses.

Smart Animal Agriculture: Application of Real-Time Sensors to Improve Animal Well-Being and Production

Consumption of animal products such as meat, milk, and eggs in first-world countries has leveled off, but it is rising precipitously in developing countries. Agriculture will have to increase its output to meet demand, opening the door to increased automation and technological innovation; intensified, sustainable farming; and precision livestock farming (PLF) applications. Early indicators of medical problems, which use sensors to alert cattle farmers early concerning individual animals that need special care, are proliferating. Wearable technologies dominate the market. In less-value-per-animal systems like sheep, goat, pig, poultry, and fish, one sensor, like a camera or robot per herd/flock/school, rather than one sensor per animal, will become common. PLF sensors generate huge amounts of data, and many actors benefit from PLF data. No standards currently exist for sharing sensor-generated data, limiting the use of commercial sensors. Technologies providing accurate data can enhance a well-managed farm. Development of methods to turn the data into actionable solutions is critical.

Comparison of different applications of automatic herd control systems on dairy farms - a review

Recent years have seen the rapid development of different devices which can be helpful in the daily work of livestock farmers. The growing size of livestock herds has led farmers to lose individual contact with their animals, while behavioral studies show that breeders can effectively and precisely monitor a herd of up to 100 cows. This was the main motivation for this study, which aims to identify and test various electronic devices which provide useful herd management data, including estrus detection, individual activity and body temperature measurement, monitoring rumen pH levels, milk quality and content as well as milk temperature and somatic cell count measurements. Some devices can detect the metabolic status of animals with a reasonable level of precision. Contemporary animal farms are offered a large number of systems for monitoring the behavior of the animals in the herd and helping to identify those that are intended for insemination or are too active or excessively apathetic. Monitoring devices support herd management and help to reduce costs through the early detection of animal diseases and nutritional problems. This review aims to compile and summarize the information currently available on the use of automatic herd control systems on dairy farms, as well as to discuss the interpretation of the results, providing a useful diagnostic tool in nutritional evaluations of dairy herds. © 2018 Society of Chemical Industry.

A parsimonious software sensor for estimating the individual dynamic pattern of methane emissions from cattle

Large efforts have been deployed in developing methods to estimate methane emissions from cattle. For large scale applications, accurate and inexpensive methane predictors are required. Within a livestock precision farming context, the objective of this work was to integrate real-time data on animal feeding behaviour with an in silico model for predicting the individual dynamic pattern of methane emission in cattle. The integration of real-time data with a mathematical model to predict variables that are not directly measured constitutes a software sensor. We developed a dynamic parsimonious grey-box model that uses as predictor variables either dry matter intake (DMI) or the intake time (IT). The model is described by ordinary differential equations.Model building was supported by experimental data of methane emissions from respiration chambers. The data set comes from a study with finishing beef steers (cross-bred Charolais and purebred Luing finishing). Dry matter intake and IT were recorded using feed bins. For research purposes, in this work, our software sensor operated off-line. That is, the predictor variables (DMI, IT) were extracted from the recorded data (rather than from an on-line sensor). A total of 37 individual dynamic patterns of methane production were analyzed. Model performance was assessed by concordance analysis between the predicted methane output and the methane measured in respiration chambers. The model predictors DMI and IT performed similarly with a Lin's concordance correlation coefficient (CCC) of 0.78 on average. When predicting the daily methane production, the CCC was 0.99 for both DMI and IT predictors. Consequently, on the basis of concordance analysis, our model performs very well compared with reported literature results for methane proxies and predictive models. As IT measurements are easier to obtain than DMI measurements, this study suggests that a software sensor that integrates our in silico model with a real-time sensor providing accurate IT measurements is a viable solution for predicting methane output in a large scale context.

Review: Precision nutrition of ruminants: approaches, challenges and potential gains

A plethora of sensors and information technologies with applications to the precision nutrition of herbivores have been developed and continue to be developed. The nutritional processes start outside of the animal body with the available feed (quantity and quality) and continue inside it once the feed is consumed, degraded in the gastrointestinal tract and metabolised by organs and tissues. Finally, some nutrients are wasted via urination, defecation and gaseous emissions through breathing and belching whereas remaining nutrients ensure maintenance and production. Nowadays, several processes can be monitored in real-time using new technologies, but although these provide valuable data 'as is', further gains could be obtained using this information as inputs to nutrition simulation models to predict unmeasurable variables in real-time and to forecast outcomes of interest. Data provided by sensors can create synergies with simulation models and this approach has the potential to expand current applications. In addition, data provided by sensors could be used with advanced analytical techniques such as data fusion, optimisation techniques and machine learning to improve their value for applications in precision animal nutrition. The present paper reviews technologies that can monitor different nutritional processes relevant to animal production, profitability, environmental management and welfare. We discussed the model-data fusion approach in which data provided by sensor technologies can be used as input of nutrition simulation models in near-real time to produce more accurate, certain and timely predictions. We also discuss some examples that have taken this model-data fusion approach to complement the capabilities of both models and sensor data, and provided examples such as predicting feed intake and methane emissions. Challenges with automatising the nutritional management of individual animals include monitoring and predicting of the flow of nutrients including nutrient intake, quantity and composition of body growth and milk production, gestation, maintenance and physical activities at the individual animal level. We concluded that the livestock industries are already seeing benefits from the development of sensor and information technologies, and this benefit is expected to grow exponentially soon with the integration of nutrition simulation models and techniques for big data analysis. However, this approach may need re-evaluating or performing new empirical research in both fields of animal nutrition and simulation modelling to accommodate a new type of data provided by the sensor technologies.

Now You See Me: Convolutional Neural Network Based Tracker for Dairy Cows

To maintain dairy cattle health and welfare at commensurable levels, analysis of the behaviors occurring between cows should be performed. This type of behavioral analysis is highly dependent on reliable and robust tracking of individuals, for it to be viable and applicable on-site. In this article, we introduce a novel method for continuous tracking and data-marker based identification of individual cows based on convolutional neural networks (CNNs). The methodology for data acquisition and overall implementation of tracking/identification is described. The Region of Interest (ROI) for the recordings was limited to a waiting area with free entrances to four automatic milking stations and a total size of 6 × 18 meters. There were 252 Swedish Holstein cows during the time of study that had access to the waiting area at a conventional dairy barn with varying conditions and illumination. Three Axis M3006-V cameras placed in the ceiling at 3.6 meters height and providing top-down view were used for recordings. The total amount of video data collected was 4 months, containing 500 million frames. To evaluate the system two 1-h recordings were chosen. The exit time and gate-id found by the tracker for each cow were compared with the exit times produced by the gates. In total there were 26 tracks considered, and 23 were correctly tracked. Given those 26 starting points, the tracker was able to maintain the correct position in a total of 101.29 min or 225 s in average per starting point/individual cow. Experiments indicate that a cow could be tracked close to 4 min before failure cases emerge and that cows could be successfully tracked for over 20 min in mildly-crowded (< 10 cows) scenes. The proposed system is a crucial stepping stone toward a fully automated tool for continuous monitoring of cows and their interactions with other individuals and the farm-building environment.