SYSTEMATIC REVIEW published: 14 May 2021 doi: 10.3389/fvets.2021.660565 A Systematic Review on Validated Precision Livestock Farming Technologies for Pig Production and Its Potential to Assess Animal Welfare Yaneth Gómez 1*, Anna H. Stygar 2 , Iris J. M. M. Boumans 3 , Eddie A. M. Bokkers 3 , Lene J. Pedersen 4 , Jarkko K. Niemi 2 , Matti Pastell 5 , Xavier Manteca 1 and Pol Llonch 1 1 Department of Animal and Food Science, Universitat Autònoma de Barcelona, Barcelona, Spain, 2 Bioeconomy and Environment, Natural Resources Institute Finland (Luke), Helsinki, Finland, 3 Animal Production Systems Group, Wageningen University and Research, Wageningen, Netherlands, 4 Department of Animal Science, Aarhus University, Tjele, Denmark, 5 Production Systems, Natural Resources Institute Finland (Luke), Helsinki, Finland Edited by: Edward Narayan, The University of Several precision livestock farming (PLF) technologies, conceived for optimizing farming Queensland, Australia processes, are developed to detect the physical and behavioral changes of animals Reviewed by: continuously and in real-time. The aim of this review was to explore the capacity of Giuliana Miguel-Pacheco, Independent Researcher, Nottingham, existing PLF technologies to contribute to the assessment of pig welfare. In a web search United Kingdom for commercially available PLF for pigs, 83 technologies were identified. A literature Johannes Baumgartner, search was conducted, following systematic review guidelines (PRISMA), to identify University of Veterinary Medicine Vienna, Austria studies on the validation of sensor technologies for assessing animal-based welfare *Correspondence: indicators. Two validation levels were defined: internal (evaluation during system building Yaneth Gómez within the same population that were used for system building) and external (evaluation yanethrocio.gomez@uab.cat on a different population than during system building). From 2,463 articles found, 111 Specialty section: were selected, which validated some PLF that could be applied to the assessment This article was submitted to of animal-based welfare indicators of pigs (7% classified as external, and 93% as Animal Behavior and Welfare, internal validation). From our list of commercially available PLF technologies, only 5% a section of the journal Frontiers in Veterinary Science had been externally validated. The more often validated technologies were vision-based Received: 29 January 2021 solutions (n = 45), followed by load-cells (n = 28; feeders and drinkers, force plates and Accepted: 19 April 2021 scales), accelerometers (n = 14) and microphones (n = 14), thermal cameras (n = 10), Published: 14 May 2021 photoelectric sensors (n = 5), radio-frequency identification (RFID) for tracking (n = 2), Citation: Gómez Y, Stygar AH, Boumans IJMM, infrared thermometers (n = 1), and pyrometer (n = 1). Externally validated technologies Bokkers EAM, Pedersen LJ, Niemi JK, were photoelectric sensors (n = 2), thermal cameras (n = 2), microphone (n = 1), Pastell M, Manteca X and Llonch P load-cells (n = 1), RFID (n = 1), and pyrometer (n = 1). Measured traits included activity (2021) A Systematic Review on Validated Precision Livestock Farming and posture-related behavior, feeding and drinking, other behavior, physical condition, Technologies for Pig Production and and health. In conclusion, existing PLF technologies are potential tools for on-farm Its Potential to Assess Animal Welfare. Front. Vet. Sci. 8:660565. animal welfare assessment in pig production. However, validation studies are lacking doi: 10.3389/fvets.2021.660565 for an important percentage of market available tools, and in particular research and Frontiers in Veterinary Science | www.frontiersin.org 1 May 2021 | Volume 8 | Article 660565 Gómez et al. PLF for Pig Welfare Assessment development need to focus on identifying the feature candidates of the measures (e.g., deviations from diurnal pattern, threshold levels) that are valid signals of either negative or positive animal welfare. An important gap identified are the lack of technologies to assess affective states (both positive and negative states). Keywords: PLF, sensor, validation, welfare, sows, piglets, fattening pigs INTRODUCTION based on human observation (14), which might imply some subjective judgements (15). This means that current protocols Animal welfare comprises three components (1): natural living, provide a limited picture of the welfare of animals throughout affective states, and basic health and functioning. Natural living their life, restricting the capacity for early detection welfare corresponds to the ability of animals to live according to their problems as well as overall life-time welfare. behavioral needs. An affective state refers to animal’s emotions The use of monitoring technology in animal production and moods, which can go from negative (e.g., depressed) to systems to optimize farming processes and reduce human positive (e.g., pleasure). Basic health deals with the normal workload, often called precision livestock farming (PLF), is biological functioning and fitness of animals. growing. According to Berckmans (15), the objective of PLF is These three components of animal welfare can be measured to provide the farmers with tools for online and continuous by indicators based, primarily on the animal, but the surrounding monitoring of the status of the animals and their environment. environment can also provide useful information. Animal-based These tools may therefore help in decision-making and indicators provide a more direct measure of the welfare of the management of the herd (16). Moreover, PLF could contribute animal compared with resource-based indicators. As an example, with relevant information related to animal welfare in an easier to assess the absence of prolonged hunger, Welfare Quality R and quicker manner, making continuous welfare assessments (WQ) (2), one of the most spread animal welfare assessment more feasible. protocols, uses the body-condition score as an animal-based Different sensors exist to measure features of individual indicator. However, in the absence of a reliable animal-based pig behavior, and/or physical conditions (e.g., accelerometers, indicator for assessing the absence of prolonged thirst, a resource- microphones, cameras) (17). PLF can add value for the welfare based indicator such as water supply, is used, which can only assessment of animals by (1) allowing individual or sub-group inform about an aspect of the environment animals live in. tracking, (2) avoiding stressful procedures involving an animal Knowledge on the welfare of pigs is important for producers handling during assessment (e.g., by body weight measurements (3) and consumers (4). As an example, for producers, poor using video cameras instead of manual weighing), and (3) health or the presence of damaging behavior such as tail biting allowing real-time monitoring. In addition, allows implementing negatively impact growth performance (5, 6). Diseases and early-warning signals of suboptimal status of the animals, to injuries might urge producers to increase the use of antibiotics prevent welfare problems (18). PLF technologies have some (7). Regarding consumers, animal welfare is considered as an limitations though. Technologies are created by humans, who important aspect of product quality (8), and studies indicate set limits for specific problem detection (e.g., tail biting), so their willingness to pay for pork produced with enhanced welfare could also be burdened with certain subjectivity (18, 19). Also, as (9–11). Goods produced under improved welfare conditions demonstrated in large-case studies for sensor profitability in dairy can be communicated to consumers by certification schemes farmers, investment in PLF technologies might not necessary lead and associated labeling. Most animal welfare labels related to to economic gain (20, 21). In addition, not all PLF tools have pig farming in Europe have requirements concerning resource- an automatic alert, making a gap between the time of problem based welfare indicators such as a space allowance, provision detection and the potential intervention of the staff. Reliability of of bedding and enrichment, and minimum transportation data management could be considered a further limitation, since time (12). However, animal-based indicators have gained more it is carried out by the PLF manufacturing company, which in fact attention, especially after the WQ protocols were published. are the data owners. To improve transparency, evaluation on the For example, most pig welfare labels consider mother-offspring PLF tools performance by external bodies is essential. interaction through setting a minimum weaning age (e.g., Mehr A procedure for validation in the real operation environment tierwohl in Germany, Beter Leven in Netherlands, and Bedre of a technology is required before it is transferred to the market Dyrevelfærd in Denmark). (22). Validation is the procedure for evaluating the performance At present, an adequate assessment of farm animal of a technology contrasted with a gold standard to know if it welfare requires a substantial amount of time and effort. achieves satisfactory prediction accuracy of a measured trait (23). Furthermore, current welfare assessment protocols have some For instance, how well a thermal camera detects fever, compared other limitations. To mention a few, they do not contain all with a standard thermometer, or how well an automatic feeding three components of animal welfare (1), often lack animal-based system can detect feeding behavior. This validation procedure indicators, focus on expressing the welfare status at group (farm) should be performed internally (on a sample of individuals level instead focusing on the individual (13), and are largely during the system building), but also externally (on different Frontiers in Veterinary Science | www.frontiersin.org 2 May 2021 | Volume 8 | Article 660565 Gómez et al. PLF for Pig Welfare Assessment individuals than those used during building phase) (24). For the second researcher (AS). The search was focused on finding sake of transparency, buyers (i.e., farmers) need to know the external validation studies on PLF technologies for pig welfare. exact features of the technology they are buying and how accurate In addition, the obtained data set (studies reporting different they monitor a given condition. It is preferable that the external validation levels) was used for checking internal validation to find validation need is carried out by independent bodies. potential technologies for pig welfare monitoring that are not yet To the best of our knowledge, an overview of existing externally validated. PLF technologies that potentially can be used for pig welfare The literature search was conducted through Web of Science assessments and the validity and reliability of these technologies, and Scopus databases, between the 1st of June and the 31st is still lacking. The aim of this review is to explore market of July 2020. Search terms included: different phases in the available PLF technologies that are potentially applicable in production cycle of pigs, terms regarding validation, types of commercial pig production, and to review (1) their ability to sensor or their commercial names. Besides, some animal-based contribute to longitudinal welfare assessment, and (2) their state welfare indicators were included as search terms, including of validation. This review focus on technologies that have been body temperature, body weight, and locomotion as physical validated (either internal or external) and which results have condition indicators; activity, feeding, drinking and vocalizations been published. as behavioral indicators; and cough and lameness as physiological indicators. Search terms related to individual recognition and MATERIALS AND METHODS animal location in the pen were also included. Search terms were applied to title, abstract and keywords Search for Commercially Available as follows: Technologies (pig OR sow OR weaner OR piglet OR fattenn∗ ) A web search to identify commercially available PLF systems AND for pigs was conducted by using Google search engine by (validat∗ OR evaluat∗ OR assess∗ OR test∗ ) one researcher (YG), between February and April 2020. Search AND (one of the following search combinations) terms included pigs (and related words such as sows, piglets), 1. (accelerometer), ((“activity sensor” OR “motion sensor” OR and different technologies known to monitor animal-based “locomotion sensor” OR “infrared motion” OR (activity welfare indicators for pigs. Technologies provided by a wide AND automat∗ )) range of suppliers were scanned. More specifically, the search 2. ((position∗ AND sensor) OR rfid OR “tracking system”) criteria included the following animal categories: (pig), (piglet), 3. ((vision AND camera) OR “image analysis”) (weaner), (fattener pig), (sow), and the technology using one of 4. ((thermistor OR infrared) OR (body temperature) AND the following terms: (automatic drinker OR automatic waterer), (monitor∗ OR detect∗ OR sensor) (automatic feeder), (electronic feeding station), (activity sensor 5. ((scale∗ AND weigh∗ ) AND automat∗ ) OR activity monitor), (RFID), (GPS), (thermal camera), (infrared 6. (“body condition scor∗ ” AND sensor OR automat∗ ) thermometer), (automatic weigh scale), (sorting scale), (weight 7. ((“feeding behavior∗ ” OR “feeding behavior∗ ”) AND sensor) camera), (body condition score sensor OR automatic body 8. (“feeding station” OR “feed∗ meter” OR “water meter” OR condition score), (body condition camera), (lameness sensor), “automatic feeder”) (automatic lameness detection), (pressure mat OR force sensor), 9. (“drinking behavio∗ ” AND monitoring) (automatic behavior analyzer), (image-based behavior analyzer), 10. ((sound AND sensor) OR (cough AND detect∗ )) (body-temperature sensor), (automatic sound analysis), (cough 11. (respiratory AND distress AND monitor) sensor OR cough monitor). No boolean operators were applied, 12. ((sound AND sensor) OR (vocali∗ AND detect∗ )) except OR boolean, as Google does not allow the use of ∗ to 13. ((gait OR lameness OR lame∗ ) AND (sensor OR “image automatically fill the search term to include related words. The analy∗ ” OR image OR automat∗ OR mat OR “pressure mat” example search looked as follows: pig automatic weigh scale OR OR “pressure sensor” OR “force plate∗ ”)) automatic weigher. The first five pages (50 hits) of results in each search NOT (review OR beef OR sheep OR survey OR goat∗ OR hors∗ were reviewed. Only commercially available technologies were OR pipeline OR genom∗ OR “wild boar” OR “swine model” OR selected for further review, excluding prototypes or devices “porcine model”). in the building phase. If required, technology providers were To make sure that all technologies identified in the first search approached to clarify the stage of development. Information on a were checked for validation, an additional search of literature sensor name, provider name, internet link, sensor type, aim, and using the name of identified commercial sensors in Google provider country were summarized. Information regarding the (Supplementary Table 1) was performed. An example of search production phase that the technology is applicable or designed criteria for “FLIR T300” technology was: pig OR piglet OR weaner for, was also specified. OR fattener OR sow FLIR T300. Literature Search Inclusion and Exclusion Criteria Following the Preferred Reporting Items for Systematic Reviews Only peer-reviewed articles, written in English and published and Meta-Analysis (PRISMA) guidelines (25), a literature search between January 2000 and July 2020 were considered. Articles was conducted by one researcher (YG), and verified by a related to welfare assessment in species other than domesticated Frontiers in Veterinary Science | www.frontiersin.org 3 May 2021 | Volume 8 | Article 660565 Gómez et al. PLF for Pig Welfare Assessment FIGURE 1 | Countries of origin of commercially available PLF technologies with potential use in pig welfare assessment. For companies with multiple locations, address of the headquarter was used. Some companies have operations in more than one country. TABLE 1 | Commercially available Precision Livestock Farming technologies categorized by the sensor type and measured trait. Type of technology Animal-based measure Number of identified % over total commercial technologies solutions (n = 83) Load cells and flow meters Force plates Gait attributes 2 Load cells with 22% 45% RFID Load cells Feed intake 3 18 Flow meter Water intake 2 Load cells/Flow meter 1 Feeder/drinker Feed/water intake 5 Scale Body weight 5 Feeder/drinker/RFID Feed/water intake/body weight 15 Load cells without 23% RFID 19 Scale/RFID Body weight 4 Cameras Body weight 14 22 26% Behavior and activity 8 Thermal cameras Body temperature 10 12% Microphones Cough 2 5 6% Animals sounds 3 Accelerometers Activity 4 5% Body temperature devices Contact-temperature device Body temperature 2 2% Pyrometer Body temperature Photoelectric sensors Lameness 2 2% GPS Location 1 1% RFID Individual identification and 1 1% tracking Frontiers in Veterinary Science | www.frontiersin.org 4 May 2021 | Volume 8 | Article 660565 Gómez et al. PLF for Pig Welfare Assessment FIGURE 2 | Modified PRISMA flow diagram (25) with the systematic review search strategy and study selection. pigs (Sus scrofa) were excluded. Studies not addressing Articles neither dealing with aspects directly related to animal technology development or validation, as well as studies using a welfare (such as estrus detection) nor with animal-based welfare PLF technology, but not testing its performance or validating it, indicators (e.g., environmental measurements such as climatic were also excluded. aspects) were excluded. Duplicates were also removed from the Only articles addressing automated and on-farm applicable data set. PLF technologies were included in this review. Studies testing on Selected studies were grouped based on the type of PLF pigs not meant for farm practices (e.g., minipigs) were excluded. technology [accelerometers, photoelectric sensors, RFID (Radio Frontiers in Veterinary Science | www.frontiersin.org 5 May 2021 | Volume 8 | Article 660565 Gómez et al. PLF for Pig Welfare Assessment Frequency Identification), load-cells, flow meters, microphones, cameras, thermal cameras, infrared (IR) thermometers, pyrometers]. The final data set included sample size, production phase, and the relevant animal-based indicator(s). Study Classification A gold standard is defined as a criterium by which given tool was evaluated (26, 27). In the conducted review, there were three possible options: FIGURE 3 | Temporal distribution of validation studies on PLF technologies 1) tool was validated against a human observer, included in this review, with potential use in pig welfare assessment. 2) tool was validated against other tool with well-defined performance record, 3) tool was validated based on its ability to detect change in animal behavior or physical condition during Location of providers was identified in a minor extent in other planned experiment. countries (including Spain, Australia, Slovakia, Scotland, Austria, Switzerland, Turkey, Sweden and England). As in Stygar et al. (28), a similar review but focusing on dairy As summarized in Table 1, load-cells based and vision-based cattle PLF technologies to monitor animal welfare, and based on technologies were the largest groups of identified technologies. Altman et al. (24), we defined the following levels of validation: Thermal-image technology was the third most common 1) External self-validation: studies where the system was type of sensor. Remaining identified technologies included evaluated using a fully independent data set, meaning microphones, accelerometers, body temperature devices, that data was collected from different herds not used for photoelectric sensors, GPS (Global Positioning System), and system development. Validation was conducted by either one RFID (for animal tracking). Most of the identified commercial scientist, at least, involved in the technology development tools can be used for different pig production phases, however, or by someone representing the company who owns some are targeted at a specific production phase. Of the the technology. commercially available technologies, 39% was used for fattening 2) External independent validation: studies where the technology pigs, 33% for sows, and 28% for piglets and weaned piglets. was validated using a fully independent data set, from different Load-cells based and vision-based body-weight tools are mainly herds than those used for technology development, and used for fattening pigs. No technologies exclusively developed or research was conducted by independent scientists with no adjusted for piglets and weaners were found. relationship with the company that owns the technology. 3) Internal validation: studies where the technology was Literature Search on Validation Trials The literature search through databases provided 2,463 results. validated using the same data set as for technology building, Nineteen studies used the commercial names of technologies or where the commercial name of the technology was not identified in the web search. After removing duplicates and specified, or the origin of the validation data set was unknown. applying the inclusion and exclusion criteria, 111 studies For determining the validation level within the literature search, remained. The PRISMA flow diagram in Figure 2 describes the the technology and the validation location were identified. The stages of studies selection process and reasons for exclusion. technologies were identified by looking for their commercial As illustrated in Figure 3, the number of publications on names or papers describing its development phase (prototypes). PLF internal and external validation increased over the last Studies where the specific location of herd was not mentioned decade. Neither the internal nor the external validation studies (for example due to privacy concerns), but clearly used different followed any particular pattern of temporal distribution of herds than for system building, were included as external the publications. Only eight (7%) of the 111 selected studies, validation level. fulfilled the external validation criteria, whereas 103 (93%) were classified as having an internal validation (Figure 4). Within RESULTS the internal validation studies, 23 (22%), did not meet the criteria for external validation, but could be included as internal Commercially Available Technologies validation. In 18 of those 23 studies, the name or origin of the All PLF technologies with a potential link to animal-based pig sensor was not provided; it was therefore impossible to identify welfare assessment are listed in the Supplementary Table 1. In its commercial availability or development stage. This applied total 83 technologies were found, based on 10 different types to nine studies with camera-based technologies (29–37), three of sensors, from 46 different providers whose headquarters studies on load-cells [a drinker (38), a scale (39), and a force plate are located in 17 countries. Figure 1 shows the origin of (40)], two on RFID (41, 42), two on accelerometer (including the commercially available technologies. Most of the providers one on accelerometer and microchip for body-temperature) are located in the United States of America (n = 22), the (43, 44), one study on microphone (45), and one on load Netherlands (n = 18), and Germany (n = 11), followed by cells with RFID (46). In the other five of those 23 studies, Belgium (n = 7), China (n = 5), and Canada (n = 4). the origin or location of the herds used, or the origin of the Frontiers in Veterinary Science | www.frontiersin.org 6 May 2021 | Volume 8 | Article 660565 Gómez et al. PLF for Pig Welfare Assessment FIGURE 4 | Number of studies classified as internal or external validation for different sensor categories. TABLE 2 | Number of peer-reviewed validation studies on sensor technologies used in pig production, categorized by sensor type and validation level (internal or external). Type of sensor Number of internal validation studies Number of external validation studies Total number of validation studies Camera 45 0 45 Load-cells With RFID- 8 (Feeders-9 Drinker-1) With RFID- 1 (Feeder-1) With RFID- 10 (Feeders-10 Drinker-1) Without RFID- 7 (Force plates-5 Scales-2) Without RFID- 0 Without RFID- 7 (Force plates-5 Scales-2) Accelerometer 14 0 14 Microphone 13 1 14 Thermal camera 8 2 10 Photoelectric 3 2 5 sensors Flow meters 2 0 2 RFID 1 1 2 Non-contact Infrared thermometer- 1 Pyrometer- 1 Infrared thermometer- 1 Pyrometer- 1 body-temperature sensors The bold numbers indicates the total sum of the number of internal and external validation studies on each type of sensor. In brackets: the specific sensors included in each sensor type category. sensor was not described [two studies on thermal cameras (47, and pyrometer). All validation studies, together with 48), one on load cells with RFID in a feeding station (49), performance indicators, are described in detail in one on cameras (50), and one on microphone (51)]. From Supplementary Table 2. the obtained list of commercially available PLF technologies, Regarding the productive phase of animals used for the 14% were validated in some identified papers of literature studies, the most frequently used pigs were fatteners (51 studies), search (12 of 83 technologies), of which 5% corresponded to followed by sows (28 studies), and weaners (21 studies). Sensors external validation (52–55). for piglets and gilts were less frequent (eight and five studies, An overview of internal and external validation studies respectively). In our results on commercial search, no PLF can be found in Table 2. Most internal validation studies solution developed or adapted exclusively for piglets or weaners concerned camera-based technologies, followed by load- was identified. However, research on PLF solutions for piglets cells based technologies. The next most frequent validated exists, as studies on cameras, thermal cameras, feeders with RFID, type of sensors were accelerometers and microphones, microphones, photoelectric sensors, pyrometers and RFID for followed by thermal-cameras, photoelectric sensors, tracking were identified using young pigs (from birth to 10 weeks flow meters, and RFID (for animal tracking). The old or up to 70 days old) as target animals. Five studies used less common validated technologies were non-contact pigs in general, not specifying the productive phase. Sample size body-temperature sensors (infrared thermometers, used in the selected studies are illustrated in Figure 5. Some Frontiers in Veterinary Science | www.frontiersin.org 7 May 2021 | Volume 8 | Article 660565 Gómez et al. PLF for Pig Welfare Assessment TABLE 3 | Studies on externally validated (independent or self-validated) sensor technologies with potential use in pig welfare assessment, specifying the sensor type, commercial name, the animal-based indicator assessed and its evaluation level (individual or group). Technology name Indicator Reason of use Evaluation Nr of Used sensors Independent Self- (monitored trait) level validation validationa validationb trials OPTEX RX-40QZ Activity and Active and/or passive Group 1 Photoelectric (56) posture-related (without distinguishing on behavior activity type) STREMODO Physical condition Stress vocalization (due to Group 1 Microphone (57) (commercially unavailable) handling) FLIR E5 thermal imaging Physical condition Body temperature Individual 2 Thermal camera (53) camera FLIR ThermoCAM S60 Physical condition Body temperature Individual (54) FIRE Physical condition Body weight Individual 1 Load cells and RFID (52) Feeding and Feed intake (kg) drinking behavior Pyrometer Optris Physical condition Body temperature Individual 1 Pyrometer (55) Prototype system Feeding and Feeding behavior, feeding Individual 1 RFID (58) drinking behavior time and/frequency Standing lying sensor Activity and Posture change (between Individual 1 Photoelectric (59) posture-related lying, standing and sitting) behavior a External independent validation—validated using independent data set (different animals and herd than for technology building) and co-authors were not involved in technology development. b External self-validation—validated using independent data set (different animals and herd than for technology building) and was developed and validated by at least one the same co-author (based on the authorship of papers) or have been validated by at least one co-author representing a company providing a technology. patterns were observed in relation to the size of the samples and validated technology. The smallest sample size (including samples of <10 animals) was used in studies validating cameras (eight studies), accelerometers (five studies), microphones (three studies), RFID for tracking (one study), and force plates (one study). However, most of the studies on accelerometers (11 out of 14 studies) and force plates (four out of five studies) were conducted using sample sizes smaller than 24 animals. Automatic feeders and drinkers, with or without RFID, and sorting scales systems, were mostly validated in studies using sample sizes from 55 to more than 1,000 animals. Three studies on drinkers with RFID validated the technology using samples between 25 and 30 animals. Studies validating accelerometers, force plates, cameras, FIGURE 5 | Sample size (the number of animals) used for external or internal microphones and RFID for tracking of animals used samples sizes validation in the reviewed studies. from 3 to more than 500 animals. Studies on thermal cameras and photoelectric sensors also were performed using varied ranges of sample sizes (from 11 to 297 animals). In the external validation studies sample sizes were between 20 and 63 animals. traits were grouped in following categories: activity and posture- related behavior, feeding and drinking behavior, other behaviors, physical condition, and health-related traits. Validation Studies and Technologies for Welfare Assessment in Pigs Activity and Posture-Related Behavior External Validation Studies We identified five sensor types (cameras, accelerometers, Table 3 summarizes the externally validated (self-validation and photoelectric sensors, thermal cameras, and RFID) that were independent validation) technologies with potential use for pig used for activity measurement (Table 4). The following traits welfare assessment. were monitored: general motion activity (active, inactive state), walking (number of steps, identified as separate behavior), Measured Traits and Technologies in Internal and tracking (identifying location or number of animals in this External Validation Studies location), postural state and transition between states (lying, Table 4 provides an overview of technologies tested to monitor standing and sitting), as well as general motion activity and different welfare indicators related to pig production. Validated tracking (studied in relation to thermal comfort). We identified Frontiers in Veterinary Science | www.frontiersin.org 8 May 2021 | Volume 8 | Article 660565 Gómez et al. PLF for Pig Welfare Assessment TABLE 4 | Summary of internally and externally validated technologies to monitor different pig welfare indicators, classified by monitored trait and sensor type. Indicator Reason of use (monitored trait) Technologies tested Activity and Active and/or passive (without distinguishing on activity type) Accelerometer (60–65) posture-related Photoelectric sensor (56)a , (66–68) behavior Camera (30, 69–71) Lying Camera (72–76) Accelerometer (43, 60–62, 77, 78) Standing Camera (72–74, 79) Accelerometer (43, 62, 77) Sitting Camera (72–74) Accelerometer (43, 77) Kneeling Camera (73, 74) Posture state and transitions between states (e.g., between lying and Photoelectric sensor (59)a standing) Accelerometer (78) Camera (74) General motion activity and tracking (related to thermal comfort) Camera (34, 76, 80) Accelerometer (78) Thermal camera (81) Walking (number of steps) Accelerometer (61) Tracking (identifying location or number of animals) Camera (32, 82–84) RFID (41) Feeding and drinking Feed intake (kg) Load cells with RFID (52)a , (49, 85) behavior Feeding time and/or frequency RFID (42, 58, 86)a , (87, 88) RFID and environment temperature and humidity sensors (46) Camera (37, 73, 74) Accelerometer (61, 64) Hunger stress identification Thermal camera (89) Microphone (90–92) Nursing, suckling Camera (37) Drinking time and/or frequency RFID (93) Accelerometer (64) Camera (37, 73, 74, 94–97) Thirst stress identification Thermal camera (89) Microphone (90, 92) Other behavior Nest- building behavior Accelerometer (98) Aggressive behavior Camera (99–104) Accelerometer (64) Cascade defense (freezing and startle duration) Camera (105) Rooting Accelerometer (61) Mounting behavior Camera (97, 106) Tail biting Camera (50, 107) Water flow meter and environment temperature sensor (108) Exploratory behavior Accelerometer (64) Playing behavior Camera (96)b Accelerometer (64)c Physical condition Gait attributes Load cells [force plates, (35, 40, 109–111)] Camera and accelerometer (112), Camera (113) Cough detection Microphone (45, 51, 114, 115) Body weight Camera (29, 33, 36, 116–122) Load cells (scales) with RFID (52)a Load cells (scales) (39, 123) Muscle score Camera (124) Body temperature Thermal camera (48, 53)a , (54)a , (125, 126) Pyrometer (55)a Stress (e.g., due to heat or cold, pain, fear) Microphone (57, 90)a , (91, 92, 127–130) Thermal camera (89) (Continued) Frontiers in Veterinary Science | www.frontiersin.org 9 May 2021 | Volume 8 | Article 660565 Gómez et al. PLF for Pig Welfare Assessment TABLE 4 | Continued Indicator Reason of use (monitored trait) Technologies tested Health-related traits Lameness and claw lesions detection Accelerometer (131) Camera (35) Thermal camera (132) African Swine Fever (sign: changes in activity level) Camera (31) Accelerometer and microchip for body temperature (44) Influenza A virus (signs: fever) and changes in activity level IR thermometer (133) Respiratory disease Thermal camera (47, 134) Microphone (135, 136) General health problems RFID (137) Diarrhea Water flow meter (38) a Externalvalidation study. b Water base play. c Use of manipulating material. five sensor types (cameras, accelerometers, photoelectric sensors, often tested for monitoring other behavior (n = 12), followed by thermal cameras, and RFID) that were used for activity accelerometers (n = 5). Cameras were used for assessing behavior measurement. Studies on accelerometers were mostly developed as a predictor of tail biting outbreak (restlessness) (107), as well for sows, to classify postures and activity. Several studies as for recognizing high and medium aggression events based validating the use of image analysis for postural states monitoring on image detection of motion and acceleration (as displacement were found. For activity traits related to tracking (individual in image) (100–103). Accelerometers were used for assessing recognition and pen location), two types of sensors were movement associated to nest-building (98) and aggression (64). identified: cameras and RFID. General motion activity and Water flow meters have been used in a study for predicting tracking related to thermal comfort (clustering behavior) was tail biting outbreaks by combining the frequency of use of assessed using thermal-imaging. water points and ambient temperature (108). Image analysis was also used for recognizing movement and location associated to Feeding and Drinking Behavior walking, running, exploring, playing, nursing, feeding, urinating Five types of technologies were identified for monitoring feeding and mounting. Image methods for analyzing low tail posture as and drinking behavior: RFID (feeders and drinkers), cameras, an early warning of tail biting have been studied (50). None of the accelerometers, thermal cameras, and microphones (Table 4). vision-based tools have been externally validated (see Table 2). Measured traits were: feed intake, feeding and drinking frequency and duration, stress related to hunger or thirst, as well as nursing and suckling behavior. Sows’ nursing behavior was monitored Physical Condition using cameras (37). The estimation of stress conditions related The following technologies were identified for monitoring to hunger and thirst was assessed by vocalizations (90–92) physical condition: load cells (force plates, scales), load cells with using microphones and via skin temperature using thermal RFID, cameras, microphones, thermal cameras, and pyrometer cameras (89), applying different stressors to the animals. Feed (Table 4). Measured traits included gait attributes (weight intake was monitored using RFID in an electronic feeding distribution on legs, gait characteristics, axial body movements station (52, 85). Evidence suggests that the performance of RFID trajectory during walking), cough, body temperature, stress (e.g., feeders for monitoring feeding behavior is negatively affected by due to heat or cold, pain, fear), body weight as well as muscle accumulation of debris under the feed trough, the large number score (loss in muscle condition is associated to acute and chronic of pigs per feeder space and pen space allowance (52). Therefore, diseases, and affects strength, immune function, and wound frequent recalibration of the device is needed. Other studies healing). Body weight was the most studied attribute, followed by validated feeding stations for monitoring individual daily feed stress and gait characteristics. Cameras were frequently used to intake (49). RFID systems were also validated for registering assess body weight. One study tested the potential of depth-image feeding and drinking patterns of individual growing-finishing analysis to evaluate axial body movements trajectory during pigs (58, 86, 93). Drinking patterns can be monitored using video walking, as an early indicator of lameness (113). Microphones analysis for evaluating visits to the drinker and contact time were applied for evaluating the features of stress vocalizations, (94, 95), and for distinguishing drinking from drinker-playing applying stressors such as handling, cold, heat, pain, hunger and behavior (96). Cameras for the identification of behavior of sows thirst (eight studies) and for cough detection. Load cells (force were used for identifying feeding and drinking behavior in the plates) were applied for gait characteristics assessment. Thermal farrowing crate (37, 73, 74), as well as in group-housed sows (97). image was used for assessing body-temperature as an alternative of rectal temperature measurement. Also, the usefulness of Other Behavior thermal image to assess piglets’ stress by measuring body- For monitoring other behavior, accelerometers, cameras, and temperature changes when applying stressors (cold, pain, hunger, water flow meters were used (Table 4). Cameras were the most thirst) was tested (89). One study was found using a pyrometer Frontiers in Veterinary Science | www.frontiersin.org 10 May 2021 | Volume 8 | Article 660565 Gómez et al. PLF for Pig Welfare Assessment for continuously measuring body-temperature, showing negative combination of different sensors as part of the same PLF solution validation results (55). Load cells (scales) with and without RFID was identified in several studies of the literature search, but they were validated for assessing body weight. were not found as commercial solutions in our list (except for the combination of RFID and load cells, and accelerometers with Health-Related Traits body temperature sensors attached to ear tags). Seven technologies were identified for assessing health-related Initially, we were searching for externally validated tools. traits: cameras, accelerometers, infrared thermometer, thermal However, only eight tools with external validation records were cameras, microphones, RFID, and water flow meters (Table 4). found. Therefore, the obtained data set has been used to find out The following health-related traits were assessed: lameness, which technologies have potential to contribute to pig welfare claw lesions, detection of signs of disease associated to African monitoring, but are not yet externally validated. Among the Swine fever (decrease in activity), as well as Influenza A market available PLF tools, only 14% were found in validation virus (fever), respiratory disease, diarrhea, and general health studies. However, it needs to be noted that information obtained problems. Respiratory disease was the most frequent studied from the market is an overview of available PFL tools, as only health-related trait (four studies), followed by body-temperature products with websites and commercial information in English to detect fever (three studies), and lameness (three studies). were included. Besides, several solutions may have been left Acceleration in combination with body-temperature data was out of the list, as we excluded technologies not addressing tested for generating early alerts of disease (44). Acceleration animal-based welfare indicators, or without direct involvement was also applied for lameness detection based on sows’ postures with animal welfare, for instance, those measuring reproductive (131). Thermal imaging for assessing health problems was parameters [e.g., (138–141)], or animal identification, such as applied in three studies: one for detecting inflammation related facial recognition (142, 143). to lameness in pregnant sows (132), and two for respiratory Also, the literature search on validation studies may not have disease assessment (measuring skin-temperature at chest level included all relevant PLF technologies for measuring animal- for detecting lung tissue damage) (47, 134). One study tested based welfare indicators. The reason for this was the choice of the use of infrared thermometry for fever detection (133). search criteria. Our search criteria specified the type of sensor Microphones for cough detection to identify sick pigs was applied to title, abstract and keywords. For this reason, some applied in two studies. Moreover, RFID data were used to studies which mention sensor type only in material and methods detect deviations in individual pigs’ feeding patterns to point section were omitted. This was the case for one study on image- diseases or other disturbances, correlating it with the Welfare analysis, one on water meters and one on load-cells (144–146) for Quality R protocol assessment (looking for skin, ear and tail instance. Pen fouling outbreaks, which can cause health problems lesions, soiling, abnormalities in body condition, respiration, due to poor hygiene, can be predicted by analyzing lying behavior locomotion, bursitis, lameness, or diarrhea) (137). Finally, water using machine learning (144) and drinking patterns from water usage data from flow meters have been tested as an early meters (146). The usefulness of load-cells to detect abnormalities indicator of potential presence of diseases at group level, in growth patterns of pigs at group level has been proved, even demonstrating that changes in diurnal drinking patterns of pigs if the animals are not individually identified, by measuring the can predict, for example, a diarrhea outbreak before clinical signs initial body weight, average daily gain and daily fluctuations show up (38). in body weight parameters (145). One of the exclusion criteria was to remove articles not dealing with animal-based welfare DISCUSSION indicators. Hence, all papers that use environmental data for welfare monitoring [e.g., (147), using ambient temperature data] The aim of this review was to explore existing PLF technologies were excluded. that potentially can contribute to measure animal-based welfare The recent development of certain technologies, such as indicators of pigs and investigate their validation status. There is computer vision based technologies (analysis of static images a substantial number of PLF tools (83 in our commercial list) in and video, 2D, 3D and thermal-imaging) begin to appear on the market that can be potentially used to assess animal-based the market (148). In our review, vision-based PLF was the indicators of pig welfare. However, only a limited number of type of technology that could have potentially assessed the technologies have been internally validated, and only four market largest number of animal-based welfare indicators. However, available technologies were externally validated (two thermal most studies using computer vision for monitoring measures cameras, one pyrometer for monitoring body-temperature, and related to animal welfare assessment still report some need of one RFID feeding station for monitoring feed intake and body improvement. For instance, in automatic body-weight detection, weight) (52–55). Through this review, we identified important there is a need for development of algorithms accounting for gaps in terms of validation on commercially available sensors. the effect of gender and genotype (118), and the refinement of PLF tools that can identify stress due to hunger and thirst (90–92) algorithms on automatic detection rate of pig boundaries (116). have been found in the literature search but not in the commercial Similarly, for lameness detection, some reports have suggested search. Similarly, tools that can assess play (64, 96), exploratory the need for algorithms refinement to increase sensitivity and behavior (64), and aggressive behavior (64, 99–104) as well as reliability (113), and the need to incorporate additional elements models trained for recognizing specific diseases, such as African to the system, such as infrared lights (35). None of the reviewed swine fever (31, 44), are not yet commercially available. The systems were externally validated. Frontiers in Veterinary Science | www.frontiersin.org 11 May 2021 | Volume 8 | Article 660565 Gómez et al. PLF for Pig Welfare Assessment Performance of Validated and algorithms for young pigs exclusively. Thus, there is an important Commercially Available PLF Technologies, concern in regard to the usefulness of PLF for monitoring the welfare of young animals. and Its Potential for Pig Welfare As stated by Stygar et al. (28), in the case of the dairy Assessment cow industry, devices used for the official recording of milk According to our results there are no guidelines on the reporting (such as milk samplers) must comply with the requirements of performance information in PLF validation studies. For that of the International Organization for Standardization (ISO) reason, the differences in performance measures reported by to obtain the certification, and must be tested for approval validation studies were not used as exclusion criteria. To be by the International Committee for Animal Recording and considered valid and feasible in commercial conditions, the Analysis (ICAR) (149). Recommendations on proper validation performance of a technology should be tested in multiple procedures for PLF technologies for pig industry are still lacking. practical scenarios, in different types of production systems and There is a constant development of PLF technologies to with different housing environments. In the reviewed studies, offer solutions for animal production including animal welfare. external validation was only performed in 7% of studies. Low Despite the lack of external validation for the majority of number of validation studies can be explained by: (i) insufficient technologies, the link between the feature measured by a sensor reporting (e.g., lack of information on validation place), (ii) and the state of the animal in terms of welfare is not always low scientific interest (e.g., reluctance of scientific journals to clear. For instance, camera-based motion detection is often publish validation studies on tool not applied for research), (iii) mentioned as a tool for welfare assessment. However, few studies high costs and labor intensity of data collection, (iv) reluctance have demonstrated a clear link between features of motion and to publish negative results, and (v) the recent development of specific animal welfare problems, such as lameness (113), or certain technologies. According to our results, validation trials specific diseases (31). The performance of identified types of PLF for commercial purposes were less common than for research for monitoring animal-based welfare indicators and measured purposes, and it could be due to time and resources requirements traits in validation studies will be described below. The types of for validation. Besides, market available PLF technologies for pigs sensors are listed in descending order, according to the number are mostly calibrated by the providers, and its precision and of validation studies compiled for each. Supplementary Table 2 reliability on data management is assumed by them without an shows full information on the validation results of each study. independent validation (Table 3). The fact that PLF companies perform validation trials themselves, and could obtain negative Camera-Based Technologies results without reporting these, has to be considered as an Internal validation of the vision-based technologies in many cases important reason for reluctance of dissemination. reported very promising results with accuracy above 95% [e.g., Concerning the quality of reporting, external validation (30, 33, 58, 73, 74, 93, 95, 97, 107, 145–147)]. However, none of requires specific information on the location of the trials, the outstanding performance results for vision-based monitoring the name of validated device, software provider, and studied have been confirmed by external validation. Image-analysis has population, knowledge about the origin of the animals, if the been used for assessing sows’ postures, such as standing, lying, test procedure was applied in commercial or experimental and sitting (73), evaluating lying patterns in group-housed pigs conditions, and clear information on which golden standard was responding to thermal conditions of the pen (76, 80), detecting used for validation and how it was measured. Information gaps animals’ location (32, 84), distinguishing drinking and drinker- in reporting were found in 22% of studies, for which reason were playing behavior (96), identifying feeding behavior (37, 73, 74, classified as internal validation studies. Few examples are the 97), recognizing aggression events (100–103) and tail biting (50), study of Petry et al. (48), and Guarino et al. (51), which despite estimating body weight (36, 118, 119), and detecting African reporting their results under laboratory and practical conditions, swine fever (31). presented lacks of information in materials and methods Changes in animals’ postures can be used as health indicators (regarding used animals, and study location, respectively). (31). Although the assessment of certain postures (sitting, In addition, internal validation studies with samples smaller kneeling) is not very accurate using vision based technologies than 10 or 20 animals were very frequent (validating some (72), it is possible to distinguish standing active behaviors, cameras, accelerometers, microphones, RFID for tracking, and such as feeding or walking, against resting patterns (30, 69). force plates). It was observed that larger samples (above 20 Lying posture, predicted by image-analysis, can indicate health animals) were mainly used in studies validating feeders and problems. For instance, resting duration and frequency changes drinkers with or without RFID, and sorting scales. According to due to diseases (31), lesions and stressful situations, are at the Royston and Altman (23), an appropriate validation sample is same time associated to damaging behavior outbreaks (50, 107). required to provide a reasonably accurate estimate of a measure Similarly, resting can be used to extrapolate maternal ability of and to avoid the risk of false negatives. Thus, studies with limited sows, as it is associated to nursing behavior (37), and thermal sample sizes could have low validity and are inconclusive (23). comfort in the pen (34). Lying posture can also be used for However, at present, a standardized parameter is not known for assessing diurnal activity patterns of the animals (79). Image what could be considered a reasonable sample size (depending technologies that detect locomotion and axial body-movement on the type of technology to be validated). A remarkable lack are promising tools for assessing lameness, an important welfare was found regarding technologies developed or with adapted issue (113), especially in sows (35). Frontiers in Veterinary Science | www.frontiersin.org 12 May 2021 | Volume 8 | Article 660565 Gómez et al. PLF for Pig Welfare Assessment Image based technologies are also able to accurately assess Load-cell technologies allow to monitor body weight and drinking behavior and water usage, which are acknowledged to growing patterns at group level. When working with RFID, load- be crucial for pig welfare (73, 74). Vision-based technologies cells can monitor feeding and drinking patterns and growing have great potential of assessing animal welfare by continuously performance at individual level, overcoming one of the challenges monitoring behaviors of pigs, which can be used to detect that cannot be achieved by current welfare protocols, which can changes and deviations in normal behavioral patterns related only monitor these aspects at group level. Although a normal to animals’ affective state (150). Some specific features such growth pattern may have little predictive value in terms of as the posture of the tail can provide useful information in good animal welfare, growth deviations or retardations have relation to tail biting outbreaks or can even be related to the been used to identify health issues and other welfare problems affective state of the pig (150). Besides, computer vision can (137). Automatic feeders with RFID are a promising technique provide information on behavioral changes such as interactions to understand animals’ requirements and anticipate welfare between individuals, allowing the detection of aggressive events problems based on feeding patterns deviations, allowing the and affiliative behaviors as nursing and playing (37). The use of implementation of corrective measures and thus improving image-analysis to evaluate the cascade defense has been validated animal health and welfare (46). in just one study, however, it still shows the potential of this tool to assess fear and stress-associated conditions (105). Force Plates Body weight detection, individual recognition, behavior and Lameness is a frequent and important welfare problem, because activity tracking are the most frequent uses of commercial image of the intense pain it causes, the disadvantages that it brings PLF technologies. According to Wurtz et al. (148), one of the in terms of access to food and water (151, 152). Also, in the difficulties of camera-based technologies is to monitor animals normal housing conditions of a pig farm, which mainly use at individual level. Nevertheless, results on studies validating slatted floor (151), may only exacerbate the problem. Due to vision algorithms for individual identification and location, seem stocking density, and subjectivity of observations, the usual visual to be promising (84). Image-based individual recognition is not diagnosis of lameness is challenging. The most affected animals invasive, and can be used in real-time, helping to overcome some often lose feeding times, and consequently body condition of the limitations of RFID systems (stress to the animals when decreases, which gets the attention of farm staff, and that is when attaching an RFID tag, and time requirements to the farmer observation is usually performed. Early diagnosis of lameness in attaching and reading). Current protocols, such as Welfare can prevent the associated high culling and mortality rates, Quality R (2), assess the nutritional state of animals by the especially for sows (152). Force plates are accurate for evaluating body-condition. Image-analysis seems to be a promising tool to gait characteristics and detecting lameness even at an early improve the assessment of the nutritional status continuously, by stage (40). Several validation studies confirm their potential monitoring the body size (117, 122). Compiled results on camera- (35, 40, 109–111, 153). Different features have been extracted based systems in farm conditions for pig weight estimation, show and validated using visual observation as a gold standard (109, potential of these tools for reducing the need of human-animal 110, 153). Weight distribution of legs (percentage of weight, ratio interaction, reducing stress associated to an unfamiliar human between the weights applied by contralateral legs, weight shifting, presence (118, 119, 124). Besides, camera-based PLF allows to amplitude of weight bearing and weight removing) significantly monitor specific situations as farrowing, and the detection of the correlated with the golden standard (CV = 5.22%) (111). Weight number of piglets in the farrowing pen, which has been studied shifting frequency and the ratio between the weights applied by to prevent perinatal asphyxia and piglets’ crushing (83). contralateral legs performed the best in terms of identifying lame individuals (109). Load Cells and Flow Meters Flow meters are discussed in the same section with load-cells, Flow Meters as its application for welfare assessment is strongly related to The use of flow meters to assess drinking patterns and water monitoring of feed-intake. Load-cells also include force plates. usage, have proved useful for prediction of several welfare Scales without individual identification have been used for conditions, such as presence of disease (38), and tail biting body weight measurement. Reported deviations are around 1 kg outbreaks (108). Performance of warning algorithms based on at group (39) and individual level (123). In combination with deviations from expected diurnal pattern in water consumption, RFID, load-cells systems (electronic feeding stations), could showed that the algorithms were capable to predict a diarrhea estimate body weight with a percentage error of 3% (52), showing outbreak 1 day before presentation of clinical signs (38). less accuracy than an ordinary scale. Monitoring the feed intake by measuring the feed weight in an electronic station with RFID Accelerometers was found to reach a 90% accuracy (85). An overestimation Accelerometers have been used to classify postures and activity of 1.1% of feed intake has been found in one study (49). with a performance for detecting and classifying activity ranging Feeding patterns (time and frequency) of individual growing- from 75 to 100% (60–64, 77). By classifying postures and activity finishing pigs can be analyzed by combining RFID and load- nest-building behavior can be monitored to predict farrowing cells, reaching an accuracy of 97% (58, 86). RFID data for time with an accuracy of 86% (98). Acceleration data have also measuring the drinking behavior of individual pigs, showed 93% been used to detect lameness based on sow postures with an of accuracy (93). accuracy of up to 93% (131). Acceleration in combination with Frontiers in Veterinary Science | www.frontiersin.org 13 May 2021 | Volume 8 | Article 660565 Gómez et al. PLF for Pig Welfare Assessment body temperature data was tested for generating early alerts of temperature changes (89). Thermal image for predicting stress disease, reaching 97% of sensitivity and 89% of specificity (44). in piglets, reached accuracies of 50, 86, 91, and 100% when Deviations in activity pattern might point out to health issues stress was related to pain, hunger, thirst, and cold, respectively (44) and lameness (131). Accelerometers can therefore provide (89). Thermal cameras for assessing animals’ space distribution useful information, but the application on pig farms will be (clustering behavior) in function of body temperature and limited because sensors have to be attached to individual pigs, radiated temperature, was validated showing a significant implying handling stress. For instance, accelerometers may be correlation between clustering and temperature response (81). embedded in ear tags, which requires the perforation of an Also, the correlation between thermal image measurements and animal’s ear for placement. Another alternative is the attachment rectal temperature was high (r = 0.80) (126). Inflammation of the accelerometer on the animal’s back or leg, but ensuring related to lameness in pregnant sows was also detected using that the device remains in place can lead to complications. thermal imaging, showing significant correlation between mean Besides, the maintenance of a device attached to the pigs’ bodies upper metatarsal temperature and sows’ parity (132). Therefore, could be difficult under farm conditions, as it can motivate thermal imaging allowed to differentiate between lame and other pigs’ chewing behavior in response to novelty of an object non-lame sows, and to detect temperature differences in the (154). Short battery life of wearable sensors is also a limitation affected leg. Hence, the welfare problem resulting from the pain of its applicability on farm. However, optimization of power caused by the inflammation associated with lameness (151), consumption and battery life are currently being improved (65). can be detected by thermal imaging. Thermal imaging at chest For lameness detection, accelerometers can be mainly relevant to level for the diagnosis of lung tissue alterations associated be used in sows. with Actinobacillus pleuropneumoniae infection, by measuring the body temperature at chest level, reached a specificity Microphones of 100% (134). Microphones accuracy for assessing and classifying vocalizations was >73% (eight out of nine validation trials studies). Photoelectric Sensors One sound-analysis algorithm reached an accuracy of 98% Photoelectric sensors, the only sensor group with external distinguishing stress vocalizations associated to pain, using validation records for activity measurements, showed a precision duration and intensity of vocalization signal as a gold standard lower than 90% (56, 66). The potential of these sensors to detect (91). The detection of vocalizations related to hunger, thirst, position changes in puerperal sows showed 64% of sensitivity cold and heat conditions (ranged from 69 to 71%) (91). Cough and 88% of specificity (59). For monitoring activity levels in pigs, detection for localization of sick pigs at barn level using photoelectric sensors detected movement in <1 s (67). microphones, reached an interval of confidence of 95% (135). Photoelectric sensors can detect movement and therefore It was also found an accuracy from 73 to 93% for correct provide useful information about the activity level and postural identification ratio of sick pigs cough sounds (136). transitions, which contribute to welfare assessment. According Sound analysis has been used for detecting coughing pigs. to Besteiro et al. (56), these sensors work better with recently Coughing is a sign of respiratory problems or at least of poor weaned piglets and assessing play than feeding behavior. As climate conditions (dust, ammonia). Measuring coughing is the body-weight of animals increases, the coverage area of therefore a relevant indicator contributing to animal welfare the photoelectric sensor decreases, resulting in less precise assessment, although it cannot be done at the individual measurements. In contrast, the detection of intense activity is level. Furthermore, if stress and pain related vocalizations more precise than non-intense activity (56). can be reliably identified, it could also be used to further welfare aspects such as stress assessment and fighting events, for instance. Distress vocalizations induced by hunger, or RFID extreme thermal discomfort seem to be more difficult to RFID technology used for individual recognition of multiple pigs classify than vocalizations due to pain (91). Future research at the same RFID reader in the pen, can reach an accuracy is needed in a larger vocal spectrum of vocal signals, not of 92% (41). It has been demonstrated that the use of two only to assess negative welfare aspects but also for assessing RFID tags instead of one, increased the accuracy up to 97% positive welfare. (58, 86). Deviations in feeding patterns as an indicator of disease, monitored by RFID data, have showed an accuracy of 97%, and a Thermal Cameras precision of 71% (137). Thermal cameras are mainly used for remote sensing of body RFID is used for individual identification, and this is essential temperature (17). Body temperature is relevant in relation to if we want to opt for an increasingly individualized welfare animal welfare because over certain thresholds it can evidence evaluation. RFID is very useful in combination with many other hyperthermia or hypothermia and may also reflect fever. devices such as scales and automatic feeders and drinkers. RFID Besides, thermal imaging seems to be a promising tool for allows to track animals’ location. It may offer additional practical monitoring physiological responses as inflammation related to applications such as monitoring social interaction as a possible lameness (132), and animals’ distribution responding to housing transmission path for diseases (155), as the contact intensity thermal conditions (81). Additionally, thermal imaging can be and length between individuals may be an indicator for disease a promising tool for assessing acute stress events by body transmission (156). Frontiers in Veterinary Science | www.frontiersin.org 14 May 2021 | Volume 8 | Article 660565 Gómez et al. PLF for Pig Welfare Assessment Non-contact Body-Temperature Sensors welfare throughout an animal’s lifetime and throughout the Non-contact body-temperature sensors revealed its limited chain may facilitate understanding of factors impairing or usefulness as an alternative to sensing temperature measurement. promoting it. This understanding of animal welfare will further The one study found on pyrometer for continuously measuring be reinforced by accessibility to large data sets, only available with pigs’ body temperature showed that the performance was not the integrated automatic and systematic assessment. reliable (55). Under fever-induced situations, comparing vaginal There is thus a need for an integration of the different aspects thermometer data and pyrometer data in the orbital area of of animal welfare (i.e., health, nutrition, comfort, affective state animals in time periods from 0.25 to 5 h, a positive correlation and natural behavior) into relevant information that could assist was found only in a third of the sample. The longer the measuring stakeholders to make decisions. The combination of sensors period, the fewer animals showed a significant correlation. may provide more relevant information than if taken separately Similar conclusions were obtained from testing the accuracy as animal-based indicators can be related. For instance, the of infrared thermometers for body-temperature measurement, use of one activity sensor may alert when an animal stops compared to rectal temperature as a gold standard (133). moving, which could be a sign of different health problems Several authors conclude that environmental conditions such (e.g., lameness, disease), but if the activity sensor is combined as ambient temperature, sunlight, air movement, barn and pen with a thermal camera informing about body temperature, the configuration, and stocking density, have a significant impact welfare information delivered can be much more precise. In on the reliability of infrared thermometry to assess body- order to cover these needs, block chain technology has been temperature in pigs (133, 157). judged useful to integrate information throughout the entire production chain and monitor welfare at different stages of the Trends and Gaps in PLF Technologies for Pig animal’s lives (158, 159). Welfare Assessment Market availability and validation records of sensor To increase transparency of animal production, there is a need for technologies dedicated to animal-based welfare monitoring reliable data on the welfare of farmed animals. This information in dairy industry has been recently conducted (28). There are is above all important for the animals as if used in a proper clear differences between dairy and pig industry when it comes to way it can improve their lives. It can also assist both consumers market availability, type of sensors used and validation records. and producers to make decisions from an informed perspective. It seems that the pig industry is behind dairy regarding sensor For the sake of the animals and production efficiency, producers availability (and validation), especially when population numbers need to monitor the health and welfare status of animals. are compared [pigs around 677.6 millions of heads (160) against It may be done with reliable and up-to-date information to 270 million oh head in dairy cattle (161)]. Looking on as early-warning systems, before implementing corrective and the nature of production, pigs are mostly kept in groups, timely measures. Consumers are demanding clear information and very often are not individually identified. Since lifespan about farm animal welfare to assist them in identifying and of a productive pig is limited (excluding sows), individual choosing enhanced welfare-friendly products. Recent advances in identification is relatively expensive and more difficult to manage sensor technologies increasingly allow systematic and automated (145, 162). Nevertheless, individual identification allows for a monitoring of several indicators that inform about the welfare more specific picture of any sub-optimal state of well-being, status of farm animals. This data could be transformed to useful which is not captured by group averages (46, 58, 85, 86, 93, 145). information for consumers as labeling. There is a difference in the investment on individual animals’ However, there is a need to identify and select the most identification in function of their productive objectives. Sows appropriate indicators and the relevant PLF technologies to are more commonly identified by RFID tags than fattening pigs, assess them. This review, is the first of its kind, spotting especially in farms using electronic feeding stations, as their relevant technologies that can assist on this task. Nevertheless, we productive lifespan is longer. In fattening pigs, group monitoring identified some challenges and gaps that need to be addressed. is more common as it reduces the costs of assessment. This To date, welfare has been based on focal assessments, and might be a reason why some technologies, which would have as information is mainly applicable to the day in which the great potential for health and welfare monitoring, are so scarcely evaluation is carried out, a limited picture of welfare status represented on the market. of animals is provided. PLF technologies dispense continuous Based on market analyses, it is clear that availability of vision- welfare information using both behavioral (e.g., activity) and based monitoring for pigs are greater than in cattle production. physiological indicators (e.g., body temperature and weight), It could be due to cost concerns (163). For example, in order to which could yield a continuous and systematic assessment at monitor the body weight of fattening pigs, few technologies could different stages of their life, and in the future, may revolutionize be considered. Using a weight sorting system based on load cells the way animal welfare is evaluated. This may allow to investigate and RFID requires substantial investment on farms and might deviations from normality at the individual level, leading to one only be feasible in newly constructed farms (164), while cameras welfare appraisal which is predominantly animal-based, and that can be installed also to already operating systems with potentially is less dependent on environmental-based indicators. Deviations less financial input. Interestingly, neither of those systems are from “normal” patterns at individual level will account for validated externally. individual differences rather than trying to understand “average” In conclusion, existing PLF technologies are potential tools animals. Information on the evolution of animal behavior and for on-farm animal welfare assessment in pig production. A Frontiers in Veterinary Science | www.frontiersin.org 15 May 2021 | Volume 8 | Article 660565 Gómez et al. PLF for Pig Welfare Assessment variety of animal-based welfare indicators can be monitored AUTHOR’S NOTE on an individual scale, continuously and in real time, using PLF. These tools had demonstrated potential for yielding a This study is conducted within ClearFarm project continuous and systematic assessment at different stages of which aim is to co-design, develop, and validate a animals’ lives, overcoming some difficulties and gaps of current software platform powered by the integration of PLF welfare assessment protocols. Thus, in the future, PLF may technologies, to assess animal welfare throughout the revolutionize the way animal welfare is assessed and informed. production chain, and to provide this information to all However, validation studies are lacking for an important parts of it. percentage of market available solutions, and in particular, research and development need to focus on identifying feature AUTHOR CONTRIBUTIONS candidates of the measures (e.g., deviations from diurnal pattern, threshold levels etc.) that are valid signals of either negative YG, AS, and PL conceived the idea of this manuscript. YG or positive animal welfare. An important gap identified are the defined and performed the literature search and the company lack of technologies to assess affective states (both positive and search. YG and AS conducted the literature screening. YG wrote negative states). the paper with input from AS, PL, JN, MP, EB, IB, LP, and In this review, tools were validated against three possible XM. All authors contributed to the article and approved the golden standards: human observer, other tool with well-defined submitted version. performance record, or based on the tool’s ability to detect change in animal behavior or physical condition during planned FUNDING experiment. The need for an established protocol for the validation procedures of PLF technologies can be noticed, as The project has received funding from the European Union’s the measurements presented in the performance reports are Horizon 2020 research and innovation program under grant very heterogeneous. agreement No. 862919. DATA AVAILABILITY STATEMENT SUPPLEMENTARY MATERIAL The original contributions presented in the study are included The Supplementary Material for this article can be found in the article/Supplementary Material, further inquiries can be online at: https://www.frontiersin.org/articles/10.3389/fvets. directed to the corresponding author/s. 2021.660565/full#supplementary-material REFERENCES 10. Xu L, Yang X, Wu L, Chen X, Chen L, Tsai FS. Consumers’ willingness to pay for food with information on animal welfare, lean meat essence 1. Fraser D, Weary DM, Pajor EA, Milligan BN. A scientific conception of detection, and traceability. Int J Environ Res Public Health. (2019) animal welfare that reflects ethical concerns. Anim Welf. (1997) 6:187–205. 16:3616. doi: 10.3390/ijerph16193616 2. Welfare Quality R . Welfare Quality R Assessment Protocol for Pigs. Welfare 11. Cornish AR, Briley D, Wilson BJ, Raubenheimer D, Schlosberg D, McGreevy Quality Consortium, Lelystad (2009). p. 182. Available online at: http://www. PD. The price of good welfare: does informing consumers about what on- welfarequalitynetwork.net/media/1018/pig_protocol.pdf package labels mean for animal welfare influence their purchase intentions? 3. Jääskeläinen T, Kauppinen T, Vesala KM, Valros A. Relationships between Appetite. (2020) 148:104577. doi: 10.1016/j.appet.2019.104577 pig welfare, productivity and farmer disposition. Anim Welf. (2014) 23:435– 12. Sørensen JT, Schrader L. Labelling as a tool for improving animal 43. doi: 10.7120/09627286.23.4.435 welfare-the pig case. Agriculture. (2019) 9:1–13. doi: 10.3390/agriculture90 4. Clark B, Panzone LA, Stewart GB, Kyriazakis I, Niemi JK, Latvala T, et al. 60123 Consumer attitudes towards production diseases in intensive production 13. Czycholl I, Büttner K, Beilage EG, Krieter J. Review of the assessment of systems. PLoS ONE. (2019) 14:e0210432. doi: 10.1371/journal.pone.0210432 animal welfare with special emphasis on the “Welfare Quality R animal 5. Homola JJ, Loftin CS, Survey G, Fish MC, Cammen KM, Sciences M, welfare assessment protocol for growing pigs.” Arch Tierzucht. (2015) et al. Impact of health challenges on pig growth performance, carcass 58:237–49. doi: 10.5194/aab-58-237-2015 characteristics and net returns under commercial conditions. Transl Anim 14. Rutherford KMD, Donald RD, Lawrence AB, Wemelsfelder F. Qualitative Sci. (2019) 2:50–61. doi: 10.1093/tas/txx005 behavioural assessment of emotionality in pigs. Appl Anim Behav Sci. (2012) 6. Sinisalo A, Niemi JK, Heinonen M, Valros A. Tail biting and 139:218–24. doi: 10.1016/j.applanim.2012.04.004 production performance in fattening pigs. Livest Sci. (2012) 15. Berckmans D. Automatic on-line monitoring of animals by 143:220–5. doi: 10.1016/j.livsci.2011.09.019 precision livestock farming. Livest Prod Soc. (2006) 1:287–94. 7. Stygar AH, Chantziaras I, Toppari I, Maes D, Niemi JK. High doi: 10.3920/978-90-8686-567-3 biosecurity and welfare standards in fattening pig farms are 16. Guarino M, Norton T, Berckmans D, Vranken E, Berckmans D. A blueprint associated with reduced antimicrobial use. Animal. (2020) for developing and applying precision livestock farming tools: a key output 14:2178–86. doi: 10.1017/S1751731120000828 of the EU-PLF project. Anim Front. (2017) 7:12–7. doi: 10.2527/af.2017.0103 8. Thorslund CAH, Aaslyng MD, Lassen J. Perceived importance and 17. Benjamin M, Yik S. Precision livestock farming in swinewelfare: a review for responsibility for market-driven pig welfare: literature review. Meat Sci. swine practitioners. Animals. (2019) 9:133. doi: 10.3390/ani9040133 (2017) 125:37–45. doi: 10.1016/j.meatsci.2016.11.008 18. Matthews SG, Miller AL, Clapp J, Plötz T, Kyriazakis I. 9. Denver S, Sandøe P, Christensen T. Consumer preferences for pig welfare – Early detection of health and welfare compromises through can the market accommodate more than one level of welfare pork? Meat Sci. automated detection of behavioural changes in pigs. Vet J. (2016) (2017) 129:140–6. doi: 10.1016/j.meatsci.2017.02.018 217:43–51. doi: 10.1016/j.tvjl.2016.09.005 Frontiers in Veterinary Science | www.frontiersin.org 16 May 2021 | Volume 8 | Article 660565 Gómez et al. PLF for Pig Welfare Assessment 19. Tullo E, Fontana I, Diana A, Norton T, Berckmans D, Guarino 38. Madsen TNTN, Kristensen ARAR. A model for monitoring the condition M. Application note: labelling, a methodology to develop of young pigs by their drinking behaviour. Comput Electron Agric. (2005) reliable algorithm in PLF. Comput Electron Agric. (2017) 48:138–54. doi: 10.1016/j.compag.2005.02.014 142:424–8. doi: 10.1016/j.compag.2017.09.030 39. Schinckel AP, Einstein ME, Miller D. Evaluation of a method to analyze pig 20. Steeneveld W, Hogeveen H, Oude Lansink AGJM. Economic consequences live weight data from animal sorting technologies. Prof Anim Sci. (2005) of investing in sensor systems on dairy farms. Comput Electron Agric. (2015) 21:50–8. doi: 10.15232/S1080-7446(15)31166-9 119:33–9. doi: 10.1016/j.compag.2015.10.006 40. McNeil B, Díaz JC, Bruns C, Stock J, Millman S, Johnson A, et al. 21. Steeneveld W, Vernooij JCM, Hogeveen H. Effect of sensor systems for cow Determining the time required to detect induced sow lameness using an management on milk production, somatic cell count, and reproduction. J embedded microcomputer-based force plate system. Am J Anim Vet Sci. Dairy Sci. (2015) 98:3896–905. doi: 10.3168/jds.2014-9101 (2018) 13:59–65. doi: 10.3844/ajavsp.2018.59.65 22. Wieringa R. Empirical research methods for technology validation: scaling 41. Porto SMC, Arcidiacono C, Cascone G, Anguzza U, Barbari M, Simonini S. up to practice. J Syst Softw. (2014) 95:19–31. doi: 10.1016/j.jss.2013.11.1097 Validation of an active RFID-based system to detect pigs housed in pens. J 23. Royston P, Altman DG. External validation of a Cox prognostic Food, Agric Environ. (2012) 10:468–72. model:principles and methods. BMC Med Res Methodol. (2013) 42. Gertheiss J, Maier V, Hessel EF, Staicu A-M. Marginal functional regression 13:33. doi: 10.1186/1471-2288-13-33 models for analyzing the feeding behavior of pigs. J Agric Biol Environ Stat. 24. Altman DG, Vergouwe Y, Royston P, Moons KGM. Prognosis and (2015) 20:353–70. doi: 10.1007/s13253-015-0212-7 prognostic research: validating a prognostic model. BMJ. (2009) 43. Thompson R, Matheson SMSM, Plötz T, Edwards SA, Kyriazakis I. Porcine 338:b605. doi: 10.1136/bmj.b605 lie detectors: automatic quantification of posture state and transitions 25. Moher D, Liberati A, Tetzlaff J, Altman DG, Altman D, Antes in sows using inertial sensors. Comput Electron Agric. (2016) 127:521– G, et al. Preferred reporting bibitems for systematic reviews 30. doi: 10.1016/j.compag.2016.07.017 and meta-analyses: the PRISMA statement. PLoS Med. (2009) 44. Martínez-Avilés M, Fernández-Carrión E, López García-Baones JM, 6:e1000097. doi: 10.1371/journal.pmed.1000097 Sánchez-Vizcaíno JM. Early detection of infection in pigs through 26. Reitsma JB, Rutjes AWS, Khan KS, Coomarasamy A, Bossuyt PM. an online monitoring system. Transbound Emerg Dis. (2017) 64:364– A review of solutions for diagnostic accuracy studies with an 73. doi: 10.1111/tbed.12372 imperfect or missing reference standard. J Clin Epidemiol. (2009) 45. Van Hirtum A, Berckmans D. Objective recognition of cough 62:797–806. doi: 10.1016/j.jclinepi.2009.02.005 sound as biomarker for aerial pollutants. Indoor Air. (2004) 27. Gold R, Reichman M, Greenberg E, Ivanidze J, Elias E, Tsiouris AJ, et 14:10–5. doi: 10.1046/j.1600-0668.2003.00195.x al. Developing a new reference standard. Acad Radiol. (2010) 17:1079– 46. Cross AJ, Rohrer GA, Brown-Brandl TM, Cassady JP, Keel BN. Feed- 82. doi: 10.1016/j.acra.2010.05.021 forward and generalised regression neural networks in modelling feeding 28. Stygar AH, Gómez Y, Berteselli GV, Dalla Costa E, Canali E, behaviour of pigs in the grow-finish phase. Biosyst Eng. (2018) 173:124– Niemi JK, et al. How precise is precision livestock farming? A 33. doi: 10.1016/j.biosystemseng.2018.02.005 systematic review on commercially available and validated sensor 47. Loughmiller JA, Spire ME, Dritz SS, Fenwick BW, Hosni MH, Hogge SB. technologies for welfare assessment of dairy cattle. Front Vet Sci. (2020) Relationship between mean body surface temperature measured by use of 8:634338. doi: 10.3389/fvets.2021.634338 infrared thermography and ambient temperature in clinically normal pigs 29. Cang Y, He H, Qiao Y. An intelligent pig weights estimate method based and pigs inoculated with Actinobacillus pleuropneumoniae. Am J Vet Res. on deep learning in sow stall environments. IEEE Access. (2019) 7:164867– (2001) 62:676–81. doi: 10.2460/ajvr.2001.62.676 75. doi: 10.1109/ACCESS.2019.2953099 48. Petry A, McGilvray W, Rakhshandeh ARR, Rakhshandeh ARR. Technical 30. Chung Y, Kim H, Lee H, Park D, Jeon T, Chang H-H. A cost-effective pigsty note: assessment of an alternative technique for measuring body temperature monitoring system based on a video sensor. KSII Trans Internet Inf Syst. in pigs. J Anim Sci. (2017) 95:3270–4. doi: 10.2527/jas.2017.1566 (2014) 8:1481–98. doi: 10.3837/tiis.2014.04.018 49. Bruininx EM, Van Der Peet-Schwering CM, Schrama JWW, Den Hartog 31. Fernández-Carrión E, Martínez-Avilés M, Ivorra B, Martínez-López LAA, Everts H, Beynen ACC. The IVOG R feeding station: a tool for B, Ramos ÁMÁM, Sánchez-Vizcaíno JMJM. Motion-based video monitoring the individual feed intake of group-housed weanling pigs. J Anim monitoring for early detection of livestock diseases: the case of African Physiol Anim Nutr. (2001) 85:81–7. doi: 10.1046/j.1439-0396.2001.00305.x swine fever. PLoS ONE. (2017) 12:e0183793. doi: 10.1371/journal.pone. 50. D’Eath RB, Jack M, Futro A, Talbot D, Zhu Q, Barclay D, et al. 0183793 Automatic early warning of tail biting in pigs: 3D cameras can 32. Navarro-Jover JM, Alcañiz-Raya M, Gómez V, Balasch S, Moreno JR, Grau- detect lowered tail posture before an outbreak. PLoS ONE. (2018) Colomer V, et al. An automatic colour-based computer vision algorithm for 13:e0194524. doi: 10.1371/journal.pone.0194524 tracking the position of piglets|Sistema de visión artificial basado en el color, 51. Guarino M, Jans P, Costa A, Aerts J-MM, Berckmans D. Field test of para el seguimiento de la posición de lechones. Spanish J Agric Res. (2009) algorithm for automatic cough detection in pig houses. Comput Electron 7:535–49. doi: 10.5424/sjar/2009073-438 Agric. (2008) 62:22–8. doi: 10.1016/j.compag.2007.08.016 33. Parsons DJ, Green DM, Schofield CP, Whittemore CT. Real-time 52. Faltys GL, Young JM, Odgaard RL, Murphy RB, Lechtenberg KF. Technical control of pig growth through an integrated management system. note: validation of electronic feeding stations as a swine research tool. J Anim Biosyst Eng. (2007) 96:257–66. doi: 10.1016/j.biosystemseng.2006. Sci. (2014) 92:272–6. doi: 10.2527/jas.2013-6808 10.013 53. Farrar KL, Field AE, Norris SL, Jacobsen KO. Comparison of 34. Shao B, Xin H. A real-time computer vision assessment and control of rectal and infrared thermometry temperatures in anesthetized thermal comfort for group-housed pigs. Comput Electron Agric. (2008) swine (Sus scrofa). J Am Assoc Lab Anim Sci. (2020) 59:221– 62:15–21. doi: 10.1016/j.compag.2007.09.006 5. doi: 10.30802/AALAS-JAALAS-19-000119 35. van Riet MMJMMJ, Vangeyte J, Janssens GPJGPJ, Ampe B, Nalon E, Bos EJE- 54. Sykes DJJ, Couvillion JSS, Cromiak A, Bowers S, Schenck E, Crenshaw M, J, et al. On-Farm claw scoring in sows using a novel mobile device. Sensors. et al. The use of digital infrared thermal imaging to detect estrus in gilts. (2019) 19:1473. doi: 10.3390/s19061473 Theriogenology. (2012) 78:147–52. doi: 10.1016/j.theriogenology.2012.01.030 36. White RPP, Schofield CPP, Green DMM, Parsons DJJ, Whittemore CTT. 55. Schmidt M, Ammon C, Christian Schön P, Manteuffel C, Hoffmann G, The effectiveness of a visual image analysis (VIA) system for monitoring Schön PCPC, et al. The suitability of infrared temperature measurements the performance of growing/finishing pigs. Anim Sci. (2004) 78:409– for continuous temperature monitoring in gilts. Arch Anim Breed. (2014) 18. doi: 10.1017/S1357729800058811 57:1–12. doi: 10.7482/0003-9438-57-021 37. Yang A, Huang H, Zheng B, Li S, Gan H, Chen C, et al. 56. Besteiro R, Rodríguez MR, Fernández MD, Ortega JA, Velo An automatic recognition framework for sow daily behaviours R. Agreement between passive infrared detector measurements based on motion and image analyses. Biosyst Eng. (2020) and human observations of animal activity. Livest Sci. (2018) 192:56–71. doi: 10.1016/j.biosystemseng.2020.01.016 214:219–24. doi: 10.1016/j.livsci.2018.06.008 Frontiers in Veterinary Science | www.frontiersin.org 17 May 2021 | Volume 8 | Article 660565 Gómez et al. PLF for Pig Welfare Assessment 57. Schön PC, Puppe B, Manteuffel G. Automated recording of stress 76. Nasirahmadi A, Richter U, Hensel O, Edwards S, Sturm B. Using machine vocalisations as a tool to document impaired welfare in pigs. Anim Welf. vision for investigation of changes in pig group lying patterns. Comput (2004) 13:105–10. Electron Agric. (2015) 119:184–90. doi: 10.1016/j.compag.2015.10.023 58. Maselyne J, Saeys W, De Ketelaere B, Mertens K, Vangeyte J, Hessel EFEF, et 77. Ringgenberg N, Bergeron R, Devillers N. Validation of accelerometers to al. Validation of a high frequency radio frequency identification (HF RFID) automatically record sow postures and stepping behaviour. Appl Anim Behav system for registering feeding patterns of growing-finishing pigs. Comput Sci. (2010) 128:37–44. doi: 10.1016/j.applanim.2010.09.018 Electron Agric. (2014) 102:10–8. doi: 10.1016/j.compag.2013.12.015 78. Thompson RJRJ, Matthews S, Plötz T, Kyriazakis I. Freedom 59. Mainau E, Dalmau A, Ruiz-de-la-Torre JL, Manteca X. Validation of an to lie: how farrowing environment affects sow lying behaviour automatic system to detect position changes in puerperal sows. Appl Anim assessment using inertial sensors. Comput Electron Agric. (2019) Behav Sci. (2009) 121:96–102. doi: 10.1016/j.applanim.2009.09.005 157:549–57. doi: 10.1016/j.compag.2019.01.035 60. Cornou C, Lundbye-Christensen S, Kristensen ARARARAR. Modelling and 79. Kim J, Chung Y, Choi Y, Sa J, Kim H, Chung Y, et al. Depth-based monitoring sows’ activity types in farrowing house using acceleration data. detection of standing-pigs in moving noise environments. Sensors. (2017) Comput Electron Agric. (2011) 76:316–24. doi: 10.1016/j.compag.2011.02.010 17:2757. doi: 10.3390/s17122757 61. Escalante HJ, Rodriguez SV, Cordero J, Kristensen AR, Cornou 80. Nasirahmadi A, Hensel O, Edwards SAA, Sturm B. A new approach for C. Sow-activity classification from acceleration patterns: a categorizing pig lying behaviour based on a Delaunay triangulation method. machine learning approach. Comput Electron Agric. (2013) Animal. (2017) 11:131–9. doi: 10.1017/S1751731116001208 93:17–26. doi: 10.1016/j.compag.2013.01.003 81. Cook NJJ, Bench CJJ, Liu T, Chabot B, Schaefer ALL. The automated 62. Marcon M, Meunier-Salaün MC, Le Mer M, Rousselière Y. Accelerometer analysis of clustering behaviour of piglets from thermal images in technology to perform precision feeding of pregnant sows and follow their response to immune challenge by vaccination. Animal. (2018) 12:122– health status. in Precision Livestock Farming 2017 - Papers Presented at the 8th 33. doi: 10.1017/S1751731117001239 European Conference on Precision Livestock Farming, ECPLF 2017. Nantes, 82. Ahrendt P, Gregersen T, Karstoft H. Development of a real-time computer (2017). p. 666–73. vision system for tracking loose-housed pigs. Comput Electron Agric. (2011) 63. Oczak M, Maschat K, Berckmans D, Vranken E, Baumgartner J. Can an 76:169–74. doi: 10.1016/j.compag.2011.01.011 automated labelling method based on accelerometer data replace a human 83. Oczak M, Maschat K, Berckmans D, Vranken E, Baumgartner labeller? - Postural profile of farrowing sows. Comput Electron Agric. (2016) J. Automatic estimation of number of piglets in a pen during 127:168–75. doi: 10.1016/j.compag.2016.06.013 farrowing, using image analysis. Biosyst Eng. (2016) 151:81– 64. Rodriguez-Baena DS, Gomez-Vela FA, García-Torres M, Divina 9. doi: 10.1016/j.biosystemseng.2016.08.018 F, Barranco CD, Daz-Diaz N, et al. Identifying livestock behavior 84. Nilsson M, Herlin AHH, Ardö H, Guzhva O, Aström K, Bergsten C. patterns based on accelerometer dataset. J Comput Sci. (2020) Development of automatic surveillance of animal behaviour and welfare 41:101076. doi: 10.1016/j.jocs.2020.101076 using image analysis and machine learned segmentation technique. Animal. 65. Liu LS, Ni JQ, Zhao RQ, Shen MX, He CL, Lu MZ. Design and (2015) 9:1859–65. doi: 10.1017/S1751731115001342 test of a low-power acceleration sensor with Bluetooth Low Energy on 85. Jiao S, Tiezzi F, Huang Y, Gray KA, Maltecca C. The use of multiple ear tags for sow behaviour monitoring. Biosyst Eng. (2018) 176:162– imputation for the accurate measurements of individual feed intake by 71. doi: 10.1016/j.biosystemseng.2018.10.011 electronic feeders. J Anim Sci. (2016) 94:824–32. doi: 10.2527/jas.2015-9667 66. Von Jasmund N, Wellnitz A, Krommweh MS, Büscher W. Using passive 86. Maselyne J, Saeys W, Briene P, Mertens K, Vangeyte J, De Ketelaere B, et infrared detectors to record group activity and activity in certain focus areas al. Methods to construct feeding visits from RFID registrations of growing- in fattening pigs. Animals. (2020) 10:792. doi: 10.3390/ani10050792 finishing pigs at the feed trough. Comput Electron Agric. (2016) 128:9– 67. Ni JQJ, Liu S, Radcliffe JS, Vonderohe C. Evaluation and 19. doi: 10.1016/j.compag.2016.08.010 characterisation of Passive Infrared Detectors to monitor pig 87. Reiners K, Hegger A, Hessel EFEF, Böck S, Wendl G, Van den Weghe activities in an environmental research building. Biosyst Eng. (2017) HFAHFA. Application of RFID technology using passive HF transponders 158:86–94. doi: 10.1016/j.biosystemseng.2017.03.014 for the individual identification of weaned piglets at the feed trough. Comput 68. Besteiro R, Arango T, Rodríguez MR, Fernández MD, Velo R. Estimation of Electron Agric. (2009) 68:178–84. doi: 10.1016/j.compag.2009.05.010 patterns in weaned piglets’ activity using spectral analysis. Biosyst Eng. (2018) 88. Adrion F, Kapun A, Eckert F, Holland EME-M, Staiger M, Götz S, et 173:85–92. doi: 10.1016/j.biosystemseng.2017.06.014 al. Monitoring trough visits of growing-finishing pigs with UHF-RFID. 69. Ott S, Moons CPHPH, Kashiha MAA, Bahr C, Tuyttens FAMAM, Comput Electron Agric. (2018) 144:144–53. doi: 10.1016/j.compag.2017. Berckmans D, et al. Automated video analysis of pig activity at pen level 11.036 highly correlates to human observations of behavioural activities. Livest Sci. 89. da Fonseca FN, Abe JM, de Alencar Nääs I, da Silva Cordeiro AF, do (2014) 160:132–7. doi: 10.1016/j.livsci.2013.12.011 Amaral FV, Ungaro HC. Automatic prediction of stress in piglets (Sus 70. Kashiha MA, Bahr C, Ott S, Moons CPH, Niewold TA, Tuyttens F, et al. Scrofa) using infrared skin temperature. Comput Electron Agric. (2020) Automatic monitoring of pig locomotion using image analysis. Livest Sci. 168:105148. doi: 10.1016/j.compag.2019.105148 (2014) 159:141–8. doi: 10.1016/j.livsci.2013.11.007 90. da Silva Cordeiro AF, de Alencar Nääs I, Oliveira SRM, Violaro F, de 71. Matthews SG, Miller AL, Plötz T, Kyriazakis I. Automated tracking to Almeida ACM, Neves DP. Understanding vocalization might help to assess measure behavioural changes in pigs for health and welfare monitoring. Sci stressful conditions in piglets. Animals. (2013) 3:923–34. doi: 10.3390/ani30 Rep. (2017) 7:17582. doi: 10.1038/s41598-017-17451-6 30923 72. Zheng C, Zhu X, Yang X, Wang L, Tu S, Xue Y. Automatic recognition of 91. da Silva JP, de Alencar Nääs I, Abe JM, da Silva Cordeiro AF. lactating sow postures from depth images by deep learning detector. Comput Classification of piglet (Sus Scrofa) stress conditions using vocalization Electron Agric. (2018) 147:51–63. doi: 10.1016/j.compag.2018.01.023 pattern and applying paraconsistent logic Eτ . Comput Electron Agric. (2019) 73. Leonard SMM, Xin H, Brown-Brandl TMM, Ramirez BCC. Development 166:105020. doi: 10.1016/j.compag.2019.105020 and application of an image acquisition system for characterizing 92. Moi M, Nääs IA, Caldara FR, Paz ICLA, Garcia RG, Cordeiro AFS. sow behaviors in farrowing stalls. Comput Electron Agric. (2019) Vocalization data mining for estimating swine stress conditions. Eng Agric. 163:104866. doi: 10.1016/j.compag.2019.104866 (2014) 34:445–50. doi: 10.1590/S0100-69162014000300008 74. Lao F, Brown-Brandl T, Stinn JP, Liu K, Teng G, Xin H. Automatic 93. Maselyne J, Adriaens I, Huybrechts T, De Ketelaere B, Millet S, Vangeyte recognition of lactating sow behaviors through depth image processing. J, et al. Measuring the drinking behaviour of individual pigs housed in Comput Electron Agric. (2016) 125:56–62. doi: 10.1016/j.compag.2016.04.026 group using radio frequency identification (RFID). Animal. (2016) 10:1557– 75. Nasirahmadi A, Edwards SASASA, Matheson SMSM, Sturm B. Using 66. doi: 10.1017/S1751731115000774 automated image analysis in pig behavioural research: assessment of the 94. Kashiha M, Bahr C, Haredasht SA, Ott S, Moons CPH, Niewold TA, et al. The influence of enrichment substrate provision on lying behaviour. Appl Anim automatic monitoring of pigs water use by cameras. Comput Electron Agric. Behav Sci. (2017) 196:30–5. doi: 10.1016/j.applanim.2017.06.015 (2013) 90:164–9. doi: 10.1016/j.compag.2012.09.015 Frontiers in Veterinary Science | www.frontiersin.org 18 May 2021 | Volume 8 | Article 660565 Gómez et al. PLF for Pig Welfare Assessment 95. Zhu W-X, Guo Y-Z, Jiao P-P, Ma C-H, Chen C. Recognition and drinking 114. Chedad A, Moshou D, Aerts JMM, Van Hirtum A, Ramon H, Berckmans D. behaviour analysis of individual pigs based on machine vision. Livest Sci. Recognition system for pig cough based on probabilistic neural networks. J (2017) 205:129–36. doi: 10.1016/j.livsci.2017.09.003 Agric Eng Res. (2001) 79:449–57. doi: 10.1006/jaer.2001.0719 96. Chen C, Zhu W, Steibel J, Siegford J, Han J, Norton T. Classification of 115. Zhao J, Li X, Liu W, Gao Y, Lei M, Tan H, et al. Dnn-hmm based acoustic drinking and drinker-playing in pigs by a video-based deep learning method. model for continuous pig cough sound recognition. Int J Agric Biol Eng. Biosyst Eng. (2020) 196:1–14. doi: 10.1016/j.biosystemseng.2020.05.010 (2020) 13:186–93. doi: 10.25165/j.ijabe.20201303.4530 97. Zhang Y, Cai J, Xiao D, Li Z, Xiong B. Real-time sow behavior 116. Buayai P, Piewthongngam K, Leung CK, Saikaew KR. Semi-automatic pig detection based on deep learning. Comput Electron Agric. (2019) weight estimation using digital image analysis. Appl Eng Agric. (2019) 163:104884. doi: 10.1016/j.compag.2019.104884 35:521–34. doi: 10.13031/aea.13084 98. Oczak M, Maschat K, Berckmans D, Vranken E, Baumgartner J. 117. Fernandes AFA, Dórea JRR, Fitzgerald R, Herring W, Rosa GJM. A novel Classification of nest-building behaviour in non-crated farrowing automated system to acquire biometric and morphological measurements sows on the basis of accelerometer data. Biosyst Eng. (2015) and predict body weight of pigs via 3D computer vision. J Anim Sci. (2019) 140:48–58. doi: 10.1016/j.biosystemseng.2015.09.007 97:496–508. doi: 10.1093/jas/sky418 99. Šustr P, Špinka M, Cloutier S, Newberry RC. Computer-aided method 118. Kashiha M, Bahr C, Ott S, Moons CPH, Niewold TA, Ödberg FO, Berckmans for calculating animal configurations during social interactions from two- D. Automatic weight estimation of individual pigs using image analysis. dimensional coordinates of color-marked body parts. Behav Res Methods Comput Electron Agric. (2014) 107:38–44. doi: 10.1016/j.compag.2014.06.003 Instruments Comput. (2001) 33:364–70. doi: 10.3758/BF03195390 119. Wongsriworaphon A, Arnonkijpanich B, Pathumnakul S. An 100. Chen C, Zhu W, Liu D, Steibel J, Siegford J, Wurtz K, et al. Detection approach based on digital image analysis to estimate the live of aggressive behaviours in pigs using a RealSence depth sensor. Comput weights of pigs in farm environments. Comput Electron Agric. (2015) Electron Agric. (2019) 166:105003. doi: 10.1016/j.compag.2019.105003 115:26–33. doi: 10.1016/j.compag.2015.05.004 101. Chen C, Zhu W, Ma C, Guo Y, Huang W, Ruan C. Image motion feature 120. Condotta ICFSICFS, Brown-Brandl TMTM, Silva-Miranda extraction for recognition of aggressive behaviors among group-housed pigs. KOKO, Stinn JPJP. Evaluation of a depth sensor for mass Comput Electron Agric. (2017) 142:380–7. doi: 10.1016/j.compag.2017.09.013 estimation of growing and finishing pigs. Biosyst Eng. (2018) 102. Oczak M, Viazzi S, Ismayilova G, Sonoda LTLT, Roulston N, Fels M, 173:11–8. doi: 10.1016/j.biosystemseng.2018.03.002 et al. Classification of aggressive behaviour in pigs by activity index 121. Jun K, Kim SJ, Ji HW. Estimating pig weights from images without and multilayer feed forward neural network. Biosyst Eng. (2014) 119:89– constraint on posture and illumination. Comput Electron Agric. (2018) 97. doi: 10.1016/j.biosystemseng.2014.01.005 153:169–76. doi: 10.1016/j.compag.2018.08.006 103. Lee J, Jin L, Park D, Chung Y. Automatic recognition of aggressive 122. Wang K, Guo H, Ma Q, Su W, Chen L, Zhu D. A portable and automatic behavior in pigs using a kinect depth sensor. Sensors. (2016) Xtion-based measurement system for pig body size. Comput Electron Agric. 16:631. doi: 10.3390/s16050631 (2018) 148:291–8. doi: 10.1016/j.compag.2018.03.018 104. Viazzi S, Ismayilova G, Oczak M, Sonoda LTT, Fels M, 123. Schinckel AP, Wastell M, Einstein ME, Hubbs JT, Preckel PV. Guarino M, et al. Image feature extraction for classification of Evaluation of alternative methods to analyze pig body weight aggressive interactions among pigs. Comput Electron Agric. (2014) data from animal sorting technologies. Prof Anim Sci. (2009) 104:57–62. doi: 10.1016/j.compag.2014.03.010 25:654–62. doi: 10.15232/S1080-7446(15)30774-9 105. Statham P, Hannuna S, Jones S, Campbell N, Robert Colborne G, 124. Alsahaf A, Azzopardi G, Ducro B, Hanenberg E, Veerkamp RF, Petkov N. Browne WJ, et al. Quantifying defence cascade responses as indicators of Estimation of muscle scores of live pigs using a kinect camera. IEEE Access. pig affect and welfare using computer vision methods. Sci Rep. (2020) (2019) 7:52238–45. doi: 10.1109/ACCESS.2019.2910986 10:8933. doi: 10.1038/s41598-020-65954-6 125. Lu M, He J, Chen C, Okinda C, Shen M, Liu L, et al. An automatic ear base 106. Nasirahmadi A, Hensel O, Edwards SA, Sturm B. Automatic detection of temperature extraction method for top view piglet thermal image. Comput mounting behaviours among pigs using image analysis. Comput Electron Electron Agric. (2018) 155:339–47. doi: 10.1016/j.compag.2018.10.030 Agric. (2016) 124:295–302. doi: 10.1016/j.compag.2016.04.022 126. Feng Y-Z, Zhao H-T, Jia G-F, Ojukwu C, Tan H-Q. Establishment of 107. Li YZ, Johnston LJLJ, Dawkins MSMS. Utilization of optical flow algorithms validated models for non-invasive prediction of rectal temperature of sows to monitor development of tail biting outbreaks in pigs. Animals. (2020) using infrared thermography and chemometrics. Int J Biometeorol. (2019) 10:323. doi: 10.3390/ani10020323 63:1405–15. doi: 10.1007/s00484-019-01758-2 108. Larsen MLV, Pedersen LJ, Jensen DB. Prediction of tail biting events in 127. Cordeiro AFDS, Nääs IDA, Oliveira SRDM, Violaro F, De Almeida finisher pigs from automatically recorded sensor data. Animals. (2019) ACM. Efficiency of distinct data mining algorithms for classifying stress 9:458. doi: 10.3390/ani9070458 level in piglets from their vocalization. Eng Agric. (2012) 32:208– 109. Conte S, Bergeron R, Gonyou H, Brown J, Rioja-Lang FC, Connor L, et 16. doi: 10.1590/S0100-69162012000200001 al. Measure and characterization of lameness in gestating sows using force 128. Vandermeulen J, Bahr C, Tullo E, Fontana I, Ott S, Kashiha M, et al. plate, kinematic, and accelerometer methods. J Anim Sci. (2014) 92:5693– Discerning pig screams in production environments. PLoS ONE. (2015) 703. doi: 10.2527/jas.2014-7865 10:e0123111. doi: 10.1371/journal.pone.0123111 110. Mohling CMM, Johnson AKK, Coetzee JFF, Karriker LAA, Abell 129. Cordeiro AFDS, Nääs IDA, da Silva Leitão F, de Almeida ACM, de Moura DJ. CEE, Millman STT, et al. Kinematics as objective tools to evaluate Use of vocalisation to identify sex, age, and distress in pig production. Biosyst lameness phases in multiparous sows. Livest Sci. (2014) 165:120– Eng. (2018) 173:57–63. doi: 10.1016/j.biosystemseng.2018.03.007 8. doi: 10.1016/j.livsci.2014.04.031 130. Moura DJ, Silva WT, Naas IA, Tolón YA, Lima KAO, Vale MM. Real time 111. Sun G, Fitzgerald RF, Stalder KJ, Karriker LA, Johnson AK, Hoff SJ. computer stress monitoring of piglets using vocalization analysis. Comput Development of an embedded microcomputer-based force plate system for Electron Agric. (2008) 64:11–8. doi: 10.1016/j.compag.2008.05.008 measuring sow weight distribution and detection of lameness. Appl Eng 131. Scheel C, Traulsen I, Auer W, Müller K, Stamer E, Krieter J. Detecting Agric. (2011) 27:475–82. doi: 10.13031/2013.37063 lameness in sows from ear tag-sampled acceleration data using wavelets. 112. Grégoire J, Bergeron R, D’Allaire S, Meunier-Salaün M-CC, Animal. (2017) 11:2076–83. doi: 10.1017/S1751731117000726 Devillers N. Assessment of lameness in sows using gait, footprints, 132. Amezcua R, Walsh S, Luimes PH, Friendship RM. Infrared thermography postural behaviour and foot lesion analysis. Animal. (2013) to evaluate lameness in pregnant sows. Can Vet J. (2014) 55:268–72. 7:1163–73. doi: 10.1017/S1751731113000098 doi: 10.13140/RG.2.2.11670.80968 113. Stavrakakis S, Li W, Guy JH, Morgan G, Ushaw G, Johnson GR, 133. Bowman ASS, Nolting JMM, Workman JDD, Cooper M, Fisher AEE, et al. Validity of the Microsoft Kinect sensor for assessment of Marsh B, et al. The inability to screen exhibition swine for influenza normal walking patterns in pigs. Comput Electron Agric. (2015) 117:1– A virus using body temperature. Zoonoses Public Health. (2016) 63:34– 7. doi: 10.1016/j.compag.2015.07.003 9. doi: 10.1111/zph.12201 Frontiers in Veterinary Science | www.frontiersin.org 19 May 2021 | Volume 8 | Article 660565 Gómez et al. PLF for Pig Welfare Assessment 134. Menzel A, Beyerbach M, Siewert C, Gundlach M, Hoeltig D, Graage 151. Jensen TB, Kristensen HH, Toft N. Quantifying the impact of lameness on R, et al. Actinobacillus pleuropneumoniae challenge in swine: diagnostic welfare and profitability of finisher pigs using expert opinions. Livest Sci. of lung alterations by infrared thermography. BMC Vet Res. (2014) (2012) 149:209–14. doi: 10.1016/j.livsci.2012.07.013 10:199. doi: 10.1186/s12917-014-0199-2 152. Heinonen M, Peltoniemi O, Valros A. Impact of lameness and claw 135. Silva M, Ferrari S, Costa A, Aerts J-MM, Guarino M, Berckmans D. lesions in sows on welfare, health and production. Livest Sci. (2013) 156:2– Cough localization for the detection of respiratory diseases in pig houses. 9. doi: 10.1016/j.livsci.2013.06.002 Comput Electron Agric. (2008) 64:286–92. doi: 10.1016/j.compag.2008. 153. Pluym LMLM, Maes D, Vangeyte J, Mertens K, Baert J, Van Weyenberg S, et 05.024 al. Development of a system for automatic measurements of force and visual 136. Exadaktylos V, Silva M, Ferrari S, Guarino M, Taylor CJ, Aerts J-M, stance variables for objective lameness detection in sows: SowSIS. Biosyst et al. Time-series analysis for online recognition and localization of Eng. (2013) 116:64–74. doi: 10.1016/j.biosystemseng.2013.06.009 sick pig (Sus scrofa) cough sounds. J Acoust Soc Am. (2009) 124:3803– 154. Day JEL, Kyriazakis I, Lawrence AB. An investigation into the 9. doi: 10.1121/1.2998780 causation of chewing behaviour in growing pigs: the role of 137. Maselyne J, Van Nuffel A, Briene P, Vangeyte J, De Ketelaere exploration and feeding motivation. Appl Anim Behav Sci. (1996) B, Millet S, et al. Online warning systems for individual 48:47–59. doi: 10.1016/0168-1591(95)01022-X fattening pigs based on their feeding pattern. Biosyst Eng. (2018) 155. Will MK, Büttner K, Kaufholz T, Müller-Graf C, Selhorst T, Krieter J. 173:143–56. doi: 10.1016/j.biosystemseng.2017.08.006 Accuracy of a real-time location system in static positions under practical 138. Manteuffel C, Hartung E, Schmidt M, Hoffmann G, Schön PC. Validation of conditions: prospects to track group-housed sows. Comput Electron Agric. individual parturition indicators for sows based on light barriers. in Precision (2017) 142:473–84. doi: 10.1016/j.compag.2017.09.020 Livestock Farming 2015 - Papers Presented at the 7th European Conference on 156. Maselyne J, Saeys W, Ketelaere B De, Briene P. How do fattening pigs Precision Livestock Farming, ECPLF 2015. Milan, (2015). p. 605–13. spend their day? In: Proceedings of the 6th International Conference on the 139. Manteuffel C, Hartung E, Schmidt M, Hoffmann G, Schön PCPC. Towards Assessment of Animal Welfare at Farm and Group Level. Clermont-Ferrand qualitative and quantitative prediction and detection of parturition onset (2014). p. 157. doi: 10.3921/978-90-8686-798-1 in sows using light barriers. Comput Electron Agric. (2015) 116:201– 157. Dewulf J, Koenen F, Laevens H, Kruif A. Infrared thermometry is not 10. doi: 10.1016/j.compag.2015.06.017 suitable for the detection of fever in pigs. Vlaams Diergeneeskd Tijdschr. 140. Manteuffel C. Parturition detection in sows as test case for measuring activity (2003) 72:373–9. behaviour in farm animals by means of radar sensors. Biosyst Eng. (2019) 158. Patelli N, Mandrioli M. Blockchain technology and traceability 184:200–6. doi: 10.1016/j.biosystemseng.2019.06.018 in the agrifood industry. J Food Sci. (2020) 85:3670– 141. Oczak M, Maschat K, Baumgartner J. Dynamics of sows’ activity 8. doi: 10.1111/1750-3841.15477 housed in farrowing pens with possibility of temporary crating might 159. Kamilaris A, Fonts A, Boldv́ FXP. The rise of blockchain technology in indicate the time when sows should be confined in a crate before agriculture and food supply chains. Trends Food Sci Technol. (2019) 91:640– the onset of farrowing. Animals. (2020) 10:6. doi: 10.3390/ani100 52. doi: 10.1016/j.tifs.2019.07.034 10006 160. Shahbandeh M. Number of Pigs Worldwide From 2012 to 2020. Statista 142. Marsot M, Mei J, Shan X, Ye L, Feng P, Yan X, et al. An adaptive (2020). Available online at: https://www.statista.com/statistics/263963/ pig face recognition approach using Convolutional Neural Networks. number-of-pigs-worldwide-since-1990/ (accessed December 4, 2020). Comput Electron Agric. (2020) 173:105625. doi: 10.1016/j.compag.2020. 161. Shahbandeh M. Number of Milk Cows Worldwide in 2019, by Country. 105386 Statista (2020). Available online at: https://www.statista.com/statistics/ 143. Hansen MFMF, Smith ML, Smith LN, Salter MGMG, Baxter 869885/global-number-milk-cows-by-country/ (accessed December 4, EM, Farish M, et al. Towards on-farm pig face recognition 2020). using convolutional neural networks. Comput Ind. (2018) 162. Cost-Effectiveness Assesment of Improving Animal Welfare Standards 98:145–52. doi: 10.1016/j.compind.2018.02.016 in European Agriculture. (2004). p. 4. Available online at: http://www. 144. Jensen DB, Larsen MLV, Pedersen LJ. Predicting pen fouling ganaderia.gob.do/index.php/component/k2/bibitemlist/category/146- in fattening pigs from pig position. Livest Sci. (2020) articulos-tecnicos 231:103852. doi: 10.1016/j.livsci.2019.103852 163. Majewski E, Hamulczuk M, Malak-rawlikowska A, Gȩbska M, Harvey D. 145. Stygar AH, Kristensen AR. Detecting abnormalities in pigs’ Cost-effectiveness assessment of improving animal welfare standards in growth – a dynamic linear model with diurnal growth pattern the European Agriculture. In: Selected Paper prepared for presentation at for identified and unidentified pigs. Comput Electron Agric. (2018) the International Association of Agricultural Economists (IAAE) Triennial 155:180–9. doi: 10.1016/j.compag.2018.10.004 Conference. Foz do Iguaçu (2012). p. 22. doi: 10.22004/ag.econ.126741 146. Dominiak KN, Hindsborg J, Pedersen LJ, Kristensen AR. Spatial modeling 164. Stygar AH, Dolecheck KA, Kristensen AR. Analyses of body weight of pigs’ drinking patterns as an alarm reducing method II. Application patterns in growing pigs: a new view on body weight in pigs for of a multivariate dynamic linear model. Comput Electron Agric. (2019) frequent monitoring. Animal. (2018) 12:295–302. doi: 10.1017/S1751731117 161:92–103. doi: 10.1016/j.compag.2018.10.037 001690 147. Jensen DB, Kristensen AR. Temperature as a predictor of fouling and diarrhea in slaughter pigs. Livest Sci. (2016) Conflict of Interest: The authors declare that the research was conducted in the 183:1–3. doi: 10.1016/j.livsci.2015.11.007 absence of any commercial or financial relationships that could be construed as a 148. Wurtz K, Camerlink I, D’Eath RB, Fernández AP, Norton T, Steibel J, potential conflict of interest. et al. Recording behaviour of indoor-housed farm animals automatically using machine vision technology: a systematic review. PLoS ONE. (2019) Copyright © 2021 Gómez, Stygar, Boumans, Bokkers, Pedersen, Niemi, Pastell, 4:e0226669. doi: 10.1371/journal.pone.0226669 Manteca and Llonch. This is an open-access article distributed under the terms 149. ICAR. The Global Standard for Livestock Data. Section 11 - Guidelines for of the Creative Commons Attribution License (CC BY). The use, distribution or Testing, Approval and Checking of Milk Recording Devices. Available online reproduction in other forums is permitted, provided the original author(s) and the at: https://www.icar.org/Guidelines/11-Milk-Recording-Devices.pdf (2020). copyright owner(s) are credited and that the original publication in this journal 150. Schrøder-Petersen DL, Simonsen HB. Tail biting in pigs. Vet J. (2001) is cited, in accordance with accepted academic practice. No use, distribution or 162:196–210. doi: 10.1053/tvjl.2001.0605 reproduction is permitted which does not comply with these terms. Frontiers in Veterinary Science | www.frontiersin.org 20 May 2021 | Volume 8 | Article 660565