Mariska van der Voort

Sensor systems have increasingly been explored as tools to support precision livestock farming, particularly in monitoring cow health and improving decision-making. This systematic literature review aims to evaluate advancements in sensor systems for detecting health conditions in dairy cows especially on mastitis, fertility, locomotion, and metabolic disorders. Relevant articles published between 2014 and 2024 were identified from Scopus. Each article was categorized by health condition and assigned to one of four development levels: sensor technique (Level I), data interpretation (Level II), integration of information (Level III), and decision making (Level IV). Relevant information from the articles was systematically reviewed and discussed. We identified 132 articles published in the past 10 years, describing a total of 151 sensor systems. Most sensor systems were aimed at mastitis and reproduction, followed by locomotion and metabolic disorders. The far majority of the articles were at level II (data interpretation) presenting research on (novel) algorithms to detect disease. A large number of different statistical, machine-learning or deep-learning models were described and evaluated, among others random forests. Level II systems applied statistical analysis or machine-learning/deep-learning models (e.g., random forests, you only look once, support vector machine, or convolutional neural network). These algorithms used a wide range of sensor data. Only a few articles aimed at level III research, integration of information and decision support. The Level III sensor systems integrated information from the sensor with economic information and other information (i.e., medication dosage, cost per disease, and supplier selection) and simulated various treatment scenarios. This review highlights the need for sensor systems research to be driven by real-world requirements for on-farm decision making. To move from proof-of-concept toward practical, future research must integrate sensor outputs with herd records and financial models, validate systems across multiple farms and at higher data frequencies, and embed economic evaluation alongside sensitivity and specificity metrics. Addressing these technical, integration, and economic challenges is essential before sensor systems can fully support automated, value-driven health management on commercial dairy farms.

Suboptimal mobility (SOM) is a costly health condition in dairy production. Current SOM management is based on visual SOM detection by farm staff. This often leads to cows with severe SOM being detected and promptly treated, whereas the detection and subsequent treatment of cows with mild SOM is delayed or nonexistent resulting in prolonged cases of mild SOM being treated only at half-year routine hoof trimming. Using automatic SOM detection sensors may improve early detection of mild SOM allowing for improved SOM management. However, the economic value of these sensors used for sensor-based SOM management are not well known. The objective of this study was to evaluate the added economic value of automatic SOM detection sensors. A recently developed bioeconomic simulation model was extended to simulate a farm without and with automatic SOM detection sensors and farm economic performance comparisons were drawn. Moreover, for the farm with sensors, novel sensor-based SOM management strategies were designed. Within these sensor based-management strategies multiple scenarios with different sensor performance in terms of sensitivity, specificity, and mobility score detection were simulated. A new alert prioritization method was also introduced. Results from this study provide insights on the economic tradeoffs in production losses and additional labor costs for the different sensor-based management strategies, sensor performances, and alert prioritization. Simulations show that the added economic value of automatic SOM detection sensors are sensitive to the sensor-based management strategies, sensor performance, and the introduced alert prioritization method. Thirty-nine of the 80 simulated scenarios obtained a positive mean net economic sensor effect: the highest was €6,360 per year (€51/cow per yr). Based on evidence from our scenarios we suggest that twice-yearly routine hoof trimming with the addition of automatic SOM detection sensors should be replaced with cow specific hoof trimmer treatments following SOM detection by the sensor. Earlier detection and subsequent treatment of mild SOM resulted in economic gains when the alert prioritization method was introduced. Implementing automatic SOM detection sensor systems allows for many options to alter SOM management where improvements in farm economic performance can be achieved in combination with improved cow mobility. The implications for future research are discussed.

Reduction of milk yield is one of the principal components in the cost of mastitis. However, past research into the association between milk yield and mastitis indicators is limited. Past research has not been based on online or in-line daily measurements and has not fully explored nonlinearity and the thresholds at which milk yield starts to decrease. In dairy herds with automated milking systems equipped with sensors, mastitis indicators of individual cows are measured on an intraday frequency, which provides unprecedented avenues to explore such effects in detail. The aim of this observational study was primarily to investigate the nonlinear associations of lactate dehydrogenase (LDH), electrical conductivity (EC), and somatic cell count (SCC) with milk yield at various stages of lactation, parity, and mastitis chronicity status (i.e., whether the cow had SCC ≥200,000 SCC/mL for the last 28 d). We also investigated thresholds at which mastitis indicators (LDH, EC, and SCC) started to be negatively associated with milk yield. We used data from 21 automated milking system herds measuring EC and online SCC. Of these herds, 7 of the 21 additionally measured online LDH. We operationalized milk yield as milk synthesis rate in kilograms per hour. Applying a generalized additive model, we estimated the milk synthesis rate as a function of the 3 mastitis indicators for 3 different subgroups based on parity, stage of lactation, and mastitis chronicity. Partial dependence plots of the mastitis indicators were used to evaluate the milk synthesis rate to study if the milk synthesis rate was associated with mastitis indicators at a specific level. Results showed that milk synthesis rate decreased with increasing SCC, LDH, and EC, but in a nonlinear fashion. The thresholds at which milk synthesis rate started to decrease were 2.5 LnSCC (12,000 SCC/mL) to 3.75 LnSCC (43,000 SCC/mL), 0 to 1 LnLDH (1-2.7 U/L), and 5.0 to 6.0 mS/cm for EC. Additionally, another substantial decrease of milk synthesis rate was observed at thresholds of 5.625 LnSCC (277,000 SCC/mL) and 3 LnLDH (20 LDH U/L) but not for EC. Having chronic mastitis decreased milk synthesis rate in all models. The identified nonlinearities between mastitis indicators and milk synthesis rate should be incorporated in statistical models for more accurate estimations of milk loss due to mastitis.

Sensor technologies for mastitis detection have resulted in the collection and availability of a large amount of data. As a result, scientific publications reporting mastitis detection research have become less driven by approaches based on biological assumptions and more by data-driven modeling. Most of these approaches try to predict mastitis events from (combinations of) raw sensor data to which a wide variety of methods are applied originating from machine learning and classical statistical approaches. However, an even wider variety in terminologies is used by researchers for methods that are similar in nature. This makes it difficult for readers from other disciplines to understand the specific methods that are used and how these differ from each other. The aim of this paper was to provide a framework (filtering, transformation, and classification) for describing the different methods applied in sensor data-based clinical mastitis detection research and use this framework to review and categorize the approaches and underlying methods described in the scientific literature on mastitis detection. We identified 40 scientific publications between 1992 and 2020 that applied methods to detect clinical mastitis from sensor data. Based on these publications, we developed and used the framework and categorized these scientific publications into the 2 data processing techniques of filtering and transformation. These data processing techniques make raw data more amendable to be used for the third step in our framework, that of classification, which is used to distinguish between healthy and nonhealthy (mastitis) cows. Most publications (n = 34) used filtering or transformation, or a combination of these 2, for data processing before classification, whereas the remaining publications (n = 6) classified the observations directly from raw data. Concerning classification, applying a simple threshold was the most used method (n = 19 publications). Our work identified that within approaches several different methods and terminologies for similar methods were used. Not all publications provided a clear description of the method used, and therefore it seemed that different methods were used between publications, whereas in fact just a different terminology was used, or the other way around. This paper is intended to serve as a reference for people from various research disciplines who need to collaborate and communicate efficiently about the topic of sensor-based mastitis detection and the methods used in this context. The framework used in this paper can support future research to correctly classify approaches and methods, which can improve the understanding of scientific publication. We encourage future research on sensor-based animal disease detection, including that of mastitis detection, to use a more coherent terminology for methods, and clearly state which technique (e.g., filtering) and approach (e.g., moving average) are used. This paper, therefore, can serve as a starting point and further stimulates the interdisciplinary cooperation in sensor-based mastitis research.

Bovine viral diarrhea virus (BVDV) infection has a major effect on the health of cows and consequently on herd performance. Many countries have implemented control or eradication programs to mitigate BVDV infection and its negative effects. These negative effects of BVDV infection on dairy herds are well documented, but there is much less information about the effects of new introduction of BVDV on dairy herds already participating in a BVDV control program. The objective of our study was to investigate the effect of a new BVDV introduction in BVDV-free herds participating in the Dutch BVDV-free program on herd performance. Longitudinal herd-level surveillance data were combined with herd information data to create 4 unique data sets, including a monthly test-day somatic cell count (SCC) data set, annual calving interval (CIV) and culling risk (CR) data sets, and a quarterly calf mortality rate (CMR) data set. Each database contained 2 types of herds: herds that remained BVDV free during the whole study period (defined as free herds), and herds that lost their BVDV-free status during the study period (defined as breakdown herds). The date of losing the BVDV-free status was defined as breakdown date. To compare breakdown herds with free herds, a random breakdown date was artificially generated for free herds by simple random sampling from the distribution of the breakdown month of the breakdown herds. The SCC and CIV before and after a new introduction of BVDV were compared through linear mixed-effects models with a Gaussian distribution, and the CR and CMR were modeled using a negative binomial distribution in generalized linear mixed-effects models. The explanatory variables for all models included herd type, BVDV status, year, and a random herd effect. Herd size was included as an explanatory variable in the SCC, CIV, and CMR model. Season was included as an explanatory variable in the SCC and CMR model. Results showed that free herds have lower SCC, CR, CMR, and shorter CIV than the breakdown herds. Within the breakdown herds, the new BVDV introduction affected the SCC and CMR. In the year after BVDV introduction, the SCC was higher than that in the year before BVDV introduction, with a factor of 1.011 [2.5th to 97.5th percentile (95% PCTL): 1.002, 1.020]. Compared with the year before BVDV breakdown, the CMR in the year of breakdown and the year after breakdown was higher, with factors of 1.170 (95% PCTL: 1.120; 1.218) and 1.096 (95% PCTL: 1.048; 1.153), respectively. This study reveals that a new introduction of BVDV had a negative but on average relatively small effect on herd performance in herds participating in a BVDV control program.

Machine learning, particularly deep learning (DL), has become a central and state-of-the-art method for several computer vision applications and remote sensing (RS) image processing. Researchers are continually trying to improve the performance of the DL methods by developing new architectural designs of the networks and/or developing new techniques, such as attention mechanisms. Since the attention mechanism has been proposed, regardless of its type, it has been increasingly used for diverse RS applications to improve the performances of the existing DL methods. However, these methods are scattered over different studies impeding the selection and application of the feasible approaches. This study provides an overview of the developed attention mechanisms and how to integrate them with different deep learning neural network architectures. In addition, it aims to investigate the effect of the attention mechanism on deep learning-based RS image processing. We identified and analyzed the advances in the corresponding attention mechanism-based deep learning (At-DL) methods. A systematic literature review was performed to identify the trends in publications, publishers, improved DL methods, data types used, attention types used, overall accuracies achieved using At-DL methods, and extracted the current research directions, weaknesses, and open problems to provide insights and recommendations for future studies. For this, five main research questions were formulated to extract the required data and information from the literature. Furthermore, we categorized the papers regarding the addressed RS image processing tasks (e.g., image classification, object detection, and change detection) and discussed the results within each group. In total, 270 papers were retrieved, of which 176 papers were selected according to the defined exclusion criteria for further analysis and detailed review. The results reveal that most of the papers reported an increase in overall accuracy when using the attention mechanism within the DL methods for image classification, image segmentation, change detection, and object detection using remote sensing images.

In automatic milking systems (AMS), sensors can measure cow behavior and milk composition at every milking. The aim of this observational study of previously collected data was to gain insight into the differences in dynamics of udder inflammation indicators between cows that recover and those that do not recover after detection of an initial inflammation. Milk diversion (milk separated from the bulk tank and thus indicating farmer intervention), conductivity, and somatic cell count (SCC) data from 4 wk before the initial inflammation to 12 wk after the initial inflammation were used to analyze 2,584 cases of udder inflammation. An udder inflammation case was defined as an initial observation of SCC ≥200,000 cells/mL as well as 1 additional SCC measurement >200,000 cells/mL within 10 d after the initial case, among other requirements. The data originated from 15 AMS herds in 6 countries. Four subsets of cows were created based on whether milk was diverted after the initial inflammation and whether the udder inflammation cases recovered, using a 10-d rolling average SCC threshold of 200,000 cells/mL and checking whether this rolling mean was below the threshold within 90 d after the initial inflammation as the indication of recovery. This formed the following subsets of cow lactations: milk diverted-recovered, milk diverted-not recovered, no milk diverted-not recovered, no milk diverted-recovered. Thresholds of 100,000 SCC/mL and 300,000 SCC/mL for the definition of case and recovery were also applied in a sensitivity analysis but with no substantial difference in results. Linear mixed models were used for each subset to study the variation in SCC (natural logarithm of SCC divided by 1,000) and σ-conductivity (natural logarithm of standard deviation of quarter conductivities). When observing the fraction of cows with SCC <200,000 cells/mL in the recovery subsets, most cows recovered within 20 d after the initial inflammation. In the recovery subsets, both σ-conductivity and SCC stabilized, mostly within 3 to 4 wk after the initial inflammation. σ-Conductivity stabilized above the pre-onset level in all subsets and did not show a clear increase in the no-milk-diverted subgroups, whereas SCC stabilized closer to the pre-onset level. Overall, this study indicated a cutoff point between nonchronic and chronic changes in indicators 3 to 4 wk after the initial inflammation for SCC and σ-conductivity.

Dairy cows are negatively affected by the introduction of bovine viral diarrhea virus (BVDV), and consequently, produce less milk. Existing literature on potential milk production losses is based on relatively outdated data and hardly evaluates milk production loss in relation to a new BVDV infection in a surveillance system. This study determined the annual and quarterly loss in milk production of BVDV introduction in 3,126 dairy herds participating in the Dutch BVDV-free program between 2007 and 2017. Among these herds, 640 were "breakdown-herds" that obtained and subsequently lost their BVDV-free status during the study period, and 2,486 herds obtained and retained their BVDV-free status during the study period. Milk yields before and after BVDV introduction were compared through annual and quarterly linear mixed models. The fixed variables for both models included herd type (breakdown-herd or free-herd), bovine viral diarrhea status (on an annual and quarterly basis), year, season, and a random herd effect. The dependent variable was the average daily milk yield on the test day. To define the possible BVDV-introduction dates, 4 scenarios were developed. In the default scenario, the date of breakdown (i.e., loss of the BVDV-free status) was assumed as the BVDV-introduction date. For the other 3 scenarios, the BVDV-introduction dates were set at 4, 6, and 9 mo before the date of breakdown, based on the estimated birth date of a persistently infected calf. In the default scenario, the loss in milk yield due to BVDV introduction occurred mainly in the first year after breakdown, with a reduction in yield of 0.08 kg/cow per day compared with the last year before breakdown. For the other 3 scenarios, the greatest yield reduction occurred in the second year after BVDV introduction, with a loss of 0.09, 0.09, and 0.1 kg/cow per day, respectively. For the first 4 quarters after BVDV introduction in the default scenario, milk yield loss was 0.14, 0.09, 0.02, and 0.08 kg/cow per day, respectively. These quarterly results indicated that milk yield loss was greatest in the first quarter after BVDV introduction. Overall, BVDV introduction had a negative, but on average a relatively small, effect on milk yield for herds participating in the BVDV-free program. This study will enable dairy farmers and policymakers to have a clearer understanding of the quantitative milk production effect of BVDV on dairy farms in a control program.