J. Dairy Sci.

100:1–9
https://doi.org/10.3168/jds.2016-12055
© American Dairy Science Association®, 2017.

The use of infrared thermography and accelerometers

for remote monitoring of dairy cow health and welfare
M. Stewart,*1 M. T. Wilson,* A. L. Schaefer,† F. Huddart,‡ and M. A. Sutherland‡
*InterAg, Ruakura Research Centre, 10 Bisley Rd., Hamilton 3214, New Zealand
†Animal Inframetrics, Box 5451, Lacombe, AB, T4L 1X2, Canada
‡AgResearch Ltd., Private Bag 3123, Hamilton 3240, New Zealand

ABSTRACT integrated into existing systems for remote monitoring

of dairy cows’ health and welfare on-farm.
Increasing reliance on automated systems on-farm Key words: dairy cow, respiration rate, accelerometers,
has led to a need for remote monitoring of health and heart rate, infrared thermography
welfare. We aimed to validate 2 methods that could
be integrated into automated systems currently in use:
INTRODUCTION
infrared thermography (IRT) to measure respiration
rate (RR), and accelerometers to measure the flinch, In recent years in the dairy industry, reliance on
step, kick (FSK) response and assessing stress and dis- automated in-shed systems (such as automatic cup re-
comfort. We monitored 22 multiparous, nonlactating, movers and drafting gates) to reduce labor has been in-
Friesian and Friesian × Jersey cows (average 5.1 yr creasing. The wide use of radiofrequency identification
of age) during a baseline period (2 min), a restraint systems has opened up opportunities for the remote
in a crush (2 min), and then a recovery period after monitoring of individual cows. Remote monitoring
exposure to a startle (2 min). We measured RR with systems allow noninvasive, non-contact data collection
continuous IRT imaging of airflow through the nostrils that can be streamed in real time or downloaded to
and by counting flank movements from video and provide information about an animal’s biological state
live recordings. We recorded heart rate (HR) and HR (e.g., activity, behavior, physiology). For example, ru-
variability using HR monitors, and we recorded FSK mination and activity monitors (using accelerometer or
from continuous video analysis of leg movements and pedometer technology) attached to the leg or neck col-
indirectly using accelerometers attached to both hind lar have become common for detecting estrus in dairy
legs. The FSK response was scored between 1 and 4 cows (Roelofs and van Erp-van der Kooij, 2015). Due to
based on the height and direction of each leg move- this shift toward automation and the declining number
ment. We observed no change in RR, HR variability, or of experienced stock people entering the industry, along
FSK in response to the startle; however, HR increased with increasing herd sizes (Lyons et al., 2016; DairyNZ,
briefly by 10 bpm. Bland-Altman plots indicated good 2015), automated remote monitoring of health and
agreement between the different methods of measuring welfare on-farm is needed. For example, the size of
RR, with average differences of −0.01 ± 0.87, 0.83 ± the average dairy herd in New Zealand (419 cows) has
0.57, and 0.37 ± 1.02 breaths/min for video versus live, tripled in the last 30 years, and herds of 2,000 to 3,000
IRT versus live and IRT versus video, respectively. Ac- cows are not uncommon (DairyNZ, 2015). Increasing
celeration was also highly correlated with FSK scores herd size leads to less individual monitoring of cows.
of ≤3 (R2 = 0.96) and ≤2 (R2 = 0.89) and moderately Respiration rate (RR) can be affected by stress and
correlated with FSK scores of 1 (R2 = 0.66) over the heat exposure (Schütz et al., 2014), pain (Stewart et
4-min sampling period. The results show that acceler- al., 2013), and disease (Gershwin et al., 2015); there-
ometers can provide an indirect measure of the FSK fore, changes in RR can be a useful indicator of these
response, and IRT can be used reliably to measure RR. physiological states. Respiration is the physiological act
With further development, both technologies could be of breathing, or more specifically, the act of taking in
oxygen, which is conveyed to the tissues and cells in the
body, followed by the release of carbon dioxide. One
Received September 26, 2016.
method for measuring RR in dairy cows is to record
Accepted December 18, 2016. the time it takes for a cow to take 10 breaths: the
1
Corresponding author: [email protected] number of flank movements as the animal inhales and

1
2 STEWART ET AL.

exhales. Although this method has been used reliably, bas et al., 2011; Lewis et al., 2011). Lewis et al. (2011)
for example to assess heat stress in cows (Schütz et al., developed an algorithm that could accurately extract
2014), it is very labor-intensive, and flank movements RR and relative tidal volume in 25 adult participants.
can be difficult to observe in colder climates or with Using continuous IRT video recording of the nostrils,
shallow breathing (e.g., during disease onset). This changes in temperature during the breathing cycle can
method also requires that observers be near the animal, be detected: inhaling brings in external air and cools
which could cause a stress response in some animals the nostrils, whereas exhaling expels air from the body
and affect the validity of the recordings. Some remote core and warms the nostrils. This technology could be
devices have been developed to measure RR in cattle. developed for non-contact RR monitoring, without in-
Typically, they involve sensors that measure chest-wall terfering with the animal or fitting equipment.
movements (e.g., strain gauges or pressure transduc- It is becoming increasingly common for dairy farms to
ers). For example, Eigenberg et al. (2000) developed a use accelerometers, mainly to assist farmers with estrus
RR monitoring system for cattle that used a thoracic detection. One survey of dairy producers reported that
belt to keep the transducer in place. Although remote 41.3% of respondents used technologies that measured
monitoring is possible, this equipment is cumbersome activity (Borchers and Bewley, 2015). These small de-
and not practical for long-term recording. Similar to vices are attached to the hind leg or neck collar and are
heart rate (HR) monitoring systems, this approach is likely to be less invasive for the animal and less sus-
impractical and susceptible to displacement and dam- ceptible to damage or displacement than other devices,
age. Also, having such bulky equipment fitted to the such as thoracic belts. It would be beneficial for the
animal could cause changes in normal behavioral pat- farmer if other valuable animal information indicative
terns and increase stress, affecting the accuracy of the of pain or discomfort due to lameness, mastitis, or other
results. Pastell et al. (2006) developed a laser-based disease onset could be collected from these devices. One
contactless method for measuring RR in an automatic method may be automated monitoring of the flinch,
milking system. However, a major limitation was that it step, kick (FSK) response. The FSK response during
could not be used on black cows, because the laser was milking has been associated with milk production (Wil-
absorbed by their dark hair. To be practical on-farm, lis, 1983; Breuer et al., 2000; Hedlund and Løvlie, 2015);
a RR monitoring system must be contactless, capable udder health and clinical signs of mastitis (Rousing et
of real-time monitoring, and account for variations in al., 2004; Pastell et al., 2006; Medrano-Galarza et al.,
animal characteristics. 2012; Fogsgaard et al., 2015); and lameness (Pastell et
Infrared thermography (IRT) measures radiated al., 2006). However, the method used to measure FSK
heat, and studies in cattle have shown that this tech- involves manual scoring of kicks and foot movements,
nology can detect thermal changes before clinical signs which is labor-intensive and prone to inaccuracies, due
occur in the onset of bovine viral diarrhea (Schaefer et to the speed of the movements being recorded and low
al., 2004), bovine respiratory disease (Schaefer et al., inter-observer reliability.
2012), and neonatal calf diarrhea (Lowe et al., 2016). The aim of this study was to validate (1) IRT to
Other studies have investigated IRT as a noninvasive measure RR, and (2) accelerometers to measure FSK
tool for detecting mastitis (Polat et al., 2010), lameness in dairy cows, using a startle as a stress model. Both
(Alsaaod and Büscher, 2012), stress (Stewart et al., methods could be developed and integrated into a non-
2008), pain (Stewart et al., 2009), and estrus (Talukder invasive, automated system and replace manual record-
et al., 2014) in cattle. The study by Schaefer et al. ings for the on-farm monitoring of animal health and
(2012) was the first demonstration of an automated welfare.
IRT system, whereby IRT images were captured every
time the animal visited a water trough to drink. The MATERIALS AND METHODS
diseases mentioned above are all associated with lo-
calized inflammatory response (e.g., respiratory tract, Animals and Handling Facilities
intestines), and during their onset, animals use other
mechanisms for heat loss (e.g., radiated heat) to main- The protocol and procedures in this study were
tain a normal core body temperature. By combining approved by the Ruakura Animal Ethics Committee
RR monitoring with IRT, it could be possible to detect under the New Zealand Animal Welfare Act 1999. This
other conditions that may not cause an inflammatory study was conducted in June 2014 (Southern hemi-
response, such as metabolic diseases (e.g., ketosis, ru- sphere winter) in a cattle-handling facility at Ruakura
men acidosis, bloat) or stress and discomfort. Human Research Centre, Hamilton, New Zealand. We used 22
studies have investigated IRT for measuring RR (Ab- pregnant, multiparous, nonlactating Friesian and Frie-

Journal of Dairy Science Vol. 100 No. 5, 2017

 REMOTE MONITORING OF DAIRY COW WELFARE 3

sian × Jersey cows, average age 5.1 ± 1.42 yr (range complete 10 breaths, observed from flank movements
3 to 9 yr). Cows were managed full-time at pasture as during baseline and restraint, and following the startle.
part of normal farm practice and brought into handling These recordings were later converted to breaths per
yards for the study. The yards (approximately 35 m2) min by combining counts from both observers for each
had a concrete floor, steel-pipe fences (155 cm high), 2-min period. We used the same method to count flank
and a corrugated steel roof. We used a cattle restraint movements (during restraint and following the startle
with a head bail (Technipharm Ultimate; Technipharm only) from video recordings using a video camera
International Ltd., Rotorua, NZ) to restrain the cows (HDR-CX220E digital HD; Sony, Shanghai, China)
during procedures. mounted approximately 2 m above the cattle restraint.
We used an IRT camera (ThermaCam S60; FLIR
Treatments and Procedures Systems AB, Danderyd, Sweden) to collect continuous
image recordings (60 frames/s) of the nose during base-
The study was conducted over 2 days, 6 d apart. On line and restraint, and following the startle. All images
d 1 (n = 15), we measured RR, HR, and FSK response; were collected by the same camera operator, standing
on d 2 (n = 7), we measured only FSK. To fit equipment 1.0 m in front of the cow. Ambient temperature and
(HR monitors and accelerometers, described below) to relative humidity were measured every 30 min using a
the cows, the cows were brought into the restraint twice hand logger (Kestrel 3000; Nielsen-Kellerman, Booth-
on each experimental day: first for fitting equipment wyn, PA), and we accounted for ambient temperature,
and then for sampling. After equipment fitting, cows relative humidity, distance (1.0 m), and emissivity
were released into an adjacent yard for a 30-min rest (0.98) by entering these details into the camera set-
period before they were brought in again for sampling. tings. Figure 1 shows IRT images of a cow inhaling
On d 1, recordings were taken during a 2-min base- and exhaling. Inhaling brings in the external air and
line period in a laneway, after the 30-min rest period cools the nostrils, whereas exhaling air from the body
but before the cow entered the restraint (baseline), and core warms the nostrils (as illustrated by the lighter
then for 2 min in a restraint without the head in a head colors in the image). To calculate RR, we used image
bail (restraint). The cow’s head was then restrained in analysis software (ThermaCam Researcher 2.7; FLIR
a head bail and the cow was exposed to an experimen- Systems AB) to view the frames from each animal, and
tally induced startle (startle). This involved a person 1 observer manually recorded the time it took for each
standing approximately 1.5 m from the cow’s head at cow to complete 10 breaths, observed from air move-
a right angle, shouting and rapidly opening and shut- ments in the nostrils during baseline and restraint, and
ting an umbrella for approximately 3 s. The startle was following the startle.
followed by a final 2-min recording period before the
equipment was removed and the cow was released back HR and HR Variability
into the yard.
On d 2, cows fitted only with accelerometers were On d 1, we recorded continuous HR and HR vari-
brought straight into the restraint and held with their ability continuously on 10 cows during the sampling
heads unrestrained for a 2-min recording period. They period using Polar HR monitors (RS800; Polar Electro
were then exposed to the startle, followed by a further Oy, Helsinki, Finland). A Polar HR strap was fastened
2-min recording period before the equipment was re- firmly around the thorax, directly behind the shoulder,
moved and they were released back into the yard. with the monitors attached. At each electrode contact
point, ultrasound transmission gel (Aquasonic 100;
RR Parker Laboratories Inc., Fairfield, NJ) was applied to
the area, from which the hair had been removed. We
On d 1, we measured RR in 15 cows using 3 methods: used Polar software (Polar Precision Performance Soft-
(1) manual counting of flank movements in real time ware, version 4.03; Polar Electro Oy) to extract con-
(live observations); (2) manual counting of flank move- tinuous recordings of HR and inter-beat interval (R–R
ments from video recordings; and (3) manual counting interval, the time interval between consecutive beats),
of air movement from the nostrils using IRT image and we calculated the root mean square of successive
recordings. differences (RMSSD). Because inter-beat interval is
The live observations consisted of 2 experienced RR prone to measurement errors (Marchant-Forde et al.,
observers standing approximately 1 m from the side 2004), we used the correction function in the Polar
of the cow, recording the time it took for a cow to software to correct for artifacts.

Journal of Dairy Science Vol. 100 No. 5, 2017

4 STEWART ET AL.

Figure 1. Infrared thermography images showing the cow (A) inhaling, and (B) exhaling.

Flinch, Step, Kick Response this information with the downloaded data from the
accelerometers.
On d 1 and 2, cows (n = 22) were fitted with 2
Hobo Pendant G accelerometer data loggers (64k; On-
Statistical Analysis
set Computer Corp., Bourne, MA). The accelerometers
were programmed to measure leg orientation via the We tested the differences between pairs of methods
x-, y-, and z-axes using the fast setting of 5 Hz, which for measuring RR using a mixed model with random
records every 0.2 s. The accelerometers were placed animal effects, examined visually using Bland-Altman
in a durable fabric pouch and attached to the lateral plots (differences plotted against averages). We assessed
side of both hind legs above the metatarsophalangeal the change in HR, RR, and FSK after the startle using
joint. The pouch was held in place by a strap around a repeated measures analysis and residual maximum
the cow’s leg. The data were downloaded using Onset likelihood in Genstat 18 (version 18, VSN International
HOBOware Pro software (version 3.4.1; Onset Com- Ltd., Hemel Hempstead, UK). We examined the cor-
puter Corp.). relation between the number of movements seen on
We mounted a Sony HDR-CX220E digital HD video video with the number of high accelerations in the ac-
camera recorder (Sony, Shanghai, China) facing the celerometer data. We were not testing for equality for
right-hand side of the restraint at a height of approxi- this relationship because, for example, an up-and-down
mately 2 m, to continuously record all leg movements. leg movement would be expected to generate 2 high
One person observed the resulting video files and re- accelerations but be assessed as 1 movement. For a cow
corded the number of times each cow lifted her left or standing, the accelerometer should read y = −1.0 for 1
right hind leg. We scored the FSK response using defi- g gravity, and x = z = 0. Therefore, we added 1 to the
nitions modified from Sutherland et al. (2012), which vertical (y) axis and calculated the overall acceleration
give a description of the height and direction of each as the square root of the sum of squares of the 3 axis
leg movement (Table 1). We recorded the start and end values (after 1 had been added to the y-axis values). A
time of each “cluster” of leg movements and compared visual inspection of the graphs of acceleration over time

Table 1. Description of flinch, step, kick (FSK) behaviors

Score Description
1 The hoof slides forward or backward, or leg moves up or down or both, but without the hoof losing contact with the ground
2 The hoof is lifted no higher than 20 cm off the ground
3 The hoof is lifted higher than 20 cm off the ground
4 Backward kick of hind leg

Journal of Dairy Science Vol. 100 No. 5, 2017

 REMOTE MONITORING OF DAIRY COW WELFARE 5
Table 2. Mean respiration rate (RR) measured using 3 methods for cows (n = 15) during the baseline period,
while standing in the restraint, and following a startle

Mean RR (breaths/min)

Method Baseline Restraint Startle rdf1 SED2 P-value

Live 36.2 33.4 33.4 21.1 1.3 0.085
Video — 37.2 35.8 9.0 1.6 0.412
Infrared thermography 44.6 39.0 39.2 16.7 2.8 0.118
1
rdf = residual degrees of freedom.
2
Standard error of the difference.

showed that the values were almost constant (baseline 9.2 ± 1.68, P = 0.503) or in the FSK score (restraint
values), apart from a small percentage that were clearly 1.9 ± 0.09, startle: 2.0 ± 0.09, P = 0.712). Overall,
much larger and associated with leg movements. Base- the behavioral response of cows to the restraint or the
line levels varied a little between animals (range 0.25 startle was low: no cows scored a 4, and only 4 cows
to 0.50 g), with the 95 percentile varying from 0.32 to scored a 3. Acceleration was highly positively correlated
0.73. We found that taking any acceleration over 0.75 with FSK scores of ≤3 (total FSK movements) and ≤2
g as a leg movement ensured that we were above the over the entire 4 min sampling period, and moderately
scatter of baseline levels and gave us a good correlation correlated with FSK scores of 1 (Figure 3) using data
between the number of movements seen in the videos collected from all 3 axes.
and the number of movements picked up by the accel- Correlations were also positively correlated with FSK
erometers. We also analyzed how well the acceleration scores of 1, ≤2, and ≤3 over the entire 4 min sampling
data correlated with the number of movements if data period when acceleration data were used from just 1 or
from only 1 or 2 of the axes were available. 2 of the axes; the x- and y-axes were the best if using
only 2, and x was the best if using only 1.
RESULTS

RR

We observed no change (P ≥ 0.085) in RR in response

to the startle using any of the methods (Table 2). The
Bland-Altman plots indicated good agreement between
the different methods of measuring RR. Overall, we
observed little difference in RR between the pairs of
methods (Figure 2). The average difference in RR was
−0.01 ± 0.87, 0.83 ± 0.57, and 0.37 ± 1.02 breaths/
min for video versus live, IRT versus live, and IRT ver-
sus video, respectively (means ± SE).

HR and HR Variability

Heart rate increased in response to the startle (aver-

age over 2 min: 60.7, 60.4, and 70.4 ± 3.1 beats/min for
baseline, restraint, and startle, respectively, P < 0.001),
and returned to baseline levels within 40 s. We observed
no change in RMSSD in response to the startle (average
89.5, 75.8, and 79.1 ± 16.1 ms for baseline, restraint,
and startle, respectively, P = 0.631).

Flinch, Step, Kick Response

Figure 2. Difference in respiration rate (breaths/min) between
pairs of the 3 methods: video versus live, infrared thermography (IRT)
We observed no change in the number of leg lifts versus live, and IRT versus video. Box plots express median, 25 to 75%
in response to the startle (restraint 7.6 ± 1.68, startle quartiles, minimum, maximum, and outliers.

Journal of Dairy Science Vol. 100 No. 5, 2017

6 STEWART ET AL.

Figure 3. Correlation between flinch, step, kick (FSK) analyzed from video recordings and acceleration measured by accelerometers using
data collected from all 3 axes over the 4-min sampling period while the cow was standing or restrained in a crush. Correlation between accelera-
tion and number of FSK scores: (A) ≤3, (B) ≤2, and 1 (C). Color version available online.

Journal of Dairy Science Vol. 100 No. 5, 2017

 REMOTE MONITORING OF DAIRY COW WELFARE 7

DISCUSSION validated as a reliable method of measuring RR (Schütz

et al., 2014), but it is labor-intensive and can be prone
Remote, noninvasive monitoring of physiological and to inaccuracies (e.g., because of shallow breathing).
behavioral responses to stress, discomfort, and disease The use of IRT has advantages, including the nonin-
onset could provide valuable information about the vasive nature of the data collection, opportunities to
health status of an animal. This study is the first dem- automate recording using appropriate algorithms, and
onstration that IRT can be used reliably to measure the possibility of measuring the depth of the breath as
RR remotely in dairy cows and shows that this tech- well as RR. Lewis et al. (2011) described a method of
nology could be a noninvasive alternative to counting extracting respiratory parameters from IRT recordings
flank movements. In addition, FSK response can be across a range of breathing patterns in human subjects,
measured remotely using accelerometers. including relative tidal volume and RR, using facial
The lack of response to the startle as measured by tracking algorithms. They found strong correlations
RR, RMSSD (a measure of HR variability), and FSK, between IRT and traditional devices that use strain
as well as the relatively brief effect on HR, suggests that gauges to measure RR using thoracic movements.
the model used to simulate a startle or reaction was The ability to use IRT to measure relative tidal
not effective in this study. Shouting has been shown to volume and extract this information automatically in
be aversive to cattle (Pajor et al., 2000) and has been dairy cows could be useful for early disease detection,
used to simulate a fright response in cattle (Stewart when tidal volume can change (e.g., during the onset
et al., 2008). The lack of response in the cows used of metabolic or respiratory disease). This information
in the present study may have been because they had could be combined with thermal changes on other body
been extensively handled and were experienced with sites using an IRT system. For example, Schaefer et al.
human contact. Responses to other types of stress, such (2012) used an automated IRT system installed at a
as anxiety or pain, may be more apparent. Heart rate water trough and found eye temperature responses in
and HR variability have been used in recent studies bovine respiratory disease that occurred several days
investigating the autonomic nervous system response to to a week before clinical signs were apparent. Further-
various husbandry procedures in dairy cattle, and these more, Martello et al. (2016) reported positive correla-
studies (along with some of the benefits and limitations tions between RR and IRT temperatures of different
of measuring cardiac responses) have been thoroughly body areas, including the eye, cheek, and back in beef
reviewed (von Borell et al., 2007; Kovács et al., 2014). cattle. Combining automated IRT thermal and RR
One study in dairy calves reported significant HR in- measures could strengthen a predictive index for early
creases in response to disbudding and ear tagging (by disease detection.
18 and 10 bpm, respectively) and significant increases Environmental factors need to be considered when
in RR (by 9 breaths/min) after disbudding but not after using IRT on-farm, such as sunlight, dust, and the dis-
ear tagging (Stewart et al., 2013). In the present study, tance between the camera and the animal, which can
the changes in HR and the lack of response in RR were affect temperature outputs. Still, IRT lends itself to
similar to calves’ response to ear tagging. However, integration into other systems (e.g., automatic feeder
many factors need to be considered when comparing or milking system, rotary platform, or drinking trough)
studies, including age, experience, and the level and because the camera can be positioned at a set distance
nature of the stressor (e.g., pain, fear, disease). from the animal. As well, because the animals visit
Using IRT, it was easy to observe and count breaths daily, deviations from baseline levels can be determined
by changes in temperature due to air movement through using historic data. Furthermore, most of these systems
the nostrils, and IRT was highly correlated with the have some degree of cover that would reduce solar load-
other methods (video footage and real-time counting ing, or cover could be provided.
of flank movements). Breaths could be automatically In the present study, data collected from accel-
detected by an algorithm and used to measure RR in erometers were positively correlated with manually
an automated system. We observed small differences in recorded FSK scores. In addition, it may be possible
RR estimated from each method; this would be expect- to detect different degrees of FSK responsiveness us-
ed for counts based on physical flank movements vs. ing accelerometers. Several studies have reported a
nostril air flow, given that the flank movements involve possible relationship between cow restlessness in the
more movement of the diaphragm versus lung function- milking parlor (flinch, step, or kick) and milk produc-
ality, and each counting method could be influenced tion (Willis, 1983; Breuer et al., 2000; Hedlund and
more or less by different factors (e.g., sighs, vocaliza- Løvlie, 2015); this possible association may indicate a
tions, rumination). Counting flank movements has been stressful situation. The FSK response in cows during

Journal of Dairy Science Vol. 100 No. 5, 2017

8 STEWART ET AL.

milking has also been associated with udder health and REFERENCES
clinical signs of mastitis (Rousing et al., 2004; Pastell
Abbas, A. K., K. Heimann, K. Jergus, T. Orlikowsky, and S. Leon-
et al., 2006; Medrano-Galarza et al., 2012; Fogsgaard et hardt. 2011. Neonatal non-contact respiratory monitoring based
al., 2015), and with lameness (Pastell et al., 2006). In on real-time infrared thermography. Biomed. Eng. Online 10:93.
these studies, the FSK response was measured manu- Alsaaod, M., and W. Büscher. 2012. Detection of hoof lesions using
digital infrared thermography in dairy cows. J. Dairy Sci. 95:735–
ally using live observations or videos, or using equip- 742.
ment such as strain gauge scales installed under the Borchers, M. R., and J. M. Bewley. 2015. An assessment of producer
floor of an automatic milking machine, which measured precision dairy farming technology use, prepurchase consider-
ations, and usefulness. J. Dairy Sci. 98:4198–4205.
the weight of each leg while cows were standing during Breuer, K., P. H. Hemsworth, J. L. Barnett, L. R. Matthews, and
milking (Pastell et al., 2006). Both of these methods G. J. Coleman. 2000. Behavioural response to humans and the
for assessing FSK response are impractical in a large productivity of commercial dairy cows. Appl. Anim. Behav. Sci.
66:273–288.
rotary or herringbone milking platform. The advantage DairyNZ. 2015. New Zealand Dairy Statistics 2014–2015. Accessed
of accelerometers is that they are already commonly Jan. 10, 2017. https://www.dairynz.co.nz/media/3136117/new-
used for other types of monitoring on-farm, such as zealand-dairy-statistics-2014-15.pdf.
Eigenberg, R. R., G. L. Hahn, J. A. Nienaber, T. M. Brown-Brandl,
estrus detection, and therefore would require no further and D. E. Spiers. 2000. Development of a new respiration rate
investment in equipment or software for many farms. monitor for cattle. Am. Soc. Agric. Eng. 43:723–728.
However, further investigations into cows’ response Fogsgaard, K. K., T. W. Bennedsgaard, and M. S. Herskin. 2015.
Behavioral changes in freestall-housed dairy cows with naturally
to different health issues is still required, as well as occurring clinical mastitis. J. Dairy Sci. 98:1730–1738.
optimizing the data collected and development of ap- Gershwin, L. J., A. L. Van Eenennaam, M. L. Anderson, H. A. McEli-
propriate algorithms (i.e., specific to FSK). got, M. X. Shao, and R. Toaff-Rosenstein. 2015. Single patho-
gen challenge with agents of the bovine respiratory disease com-
plex. PLoS One 10:e0142479. https://doi.org/10.1371/journal.
CONCLUSIONS pone.0142479.
Hedlund, L., and H. Løvlie. 2015. Personality and production: Nervous
cows produce less milk. J. Dairy Sci. 98:5819–5828.
In conclusion, our results show that accelerometers Kovács, L., V. Jurkovich, M. Bakony, O. Szenci, P. Póti, and J. Tőzsér.
can provide an indirect measure of the FSK response, 2014. Welfare implication of measuring heart rate and heart rate
and IRT can be used reliably to measure RR. As a variability in dairy cattle: Literature review and conclusions for
future research. Animal 8:316–330.
next step, we will further develop image-processing Lewis, G. F., R. G. Gatto, and S. W. Porges. 2011. A novel method for
techniques to automate the extraction of RR from extracting respiration rate and relative tidal volume from infrared
IRT images in real time so this information can be thermography. Psychophysiology 48:877–887.
Lowe, G. L., A. L. Schaefer, J. R. Waas, M. T. Wilson, M. A. Suther-
integrated into other automated systems on-farm. This land, and M. Stewart. 2016. Brief communication: The use of in-
information, combined with remote sensing of FSK, frared thermography and feeding behaviour for early disease detec-
integrated into existing systems where other produc- tion in New Zealand dairy calves. Pages 177–179 in Proc. N.Z. Soc.
Anim. Prod., Adelaide, Austrailia.
tion and behavioral information is available (e.g., the Lyons, N. A., C. R. Eastwood, B. T. Dela Rueb, and K. L. Kerrisk.
number of visits, feed intake, milk yield) could provide 2016. Current and future adoption of milking related technologies
noninvasive, real-time assessment of animal health and on Australian and New Zealand dairy farms. Pages 445–450 in
Proc. Int. Precision Dairy Farming Conf., Leeuwarden, the Neth-
welfare. Combined with other thermal data using IRT, erlands.
an automated system may be able to identify animals Marchant-Forde, R. M., D. J. Marlin, and J. N. Marchant-Forde. 2004.
with early onset of pathological or metabolic diseases Validation of a cardiac monitor for measuring heart rate variabil-
ity in adult female pigs: Accuracy, artefacts and editing. Physiol.
and distress or discomfort, allowing early intervention Behav. 80:449–458.
by the farmer and improving animal health, produc- Martello, L. S., S. da Luz E Silva, R. da Costa Gomes, R. R. da Silva
tion, and welfare. Corte, and P. R. Leme. 2016. Infrared thermography as a tool to
evaluate body surface temperature and its relationship with feed
efficiency in Bos indicus cattle in tropical conditions. Int. J. Bio-
ACKNOWLEDGMENTS meteorol. 60:173–181.
Medrano-Galarza, C., J. Gibbons, S. Wagner, A. M. de Passillé, and J.
Rushen. 2012. Behavioral changes in dairy cows with mastitis. J.
The authors thank Gemma Lowe, Melissa Hempstead, Dairy Sci. 95:6994–7002.
Chloé Picchiottino, and Suzanne Dowling (AgResearch Pajor, E. A., J. Rushen, and A. M. B. de Passillé. 2000. Aversion
Ltd., Hamilton, New Zealand) for their technical assis- learning techniques to evaluate dairy cattle handling practices.
Appl. Anim. Behav. Sci. 69:89–102.
tance during this study, and Vanessa Cave (AgResearch Pastell, M., A. M. Aisla, M. Hautala, V. Poikalainen, J. Praks, I.
Ltd.) and Neil Cox (NeilStat Ltd., Hamilton, New Zea- Veermäe, and J. Ahokas. 2006. Contactless measurement of cow
land) for their assistance with the statistical analysis. behavior in a milking robot. Behav. Res. Methods 38:479–486.
Polat, B., A. Colak, M. Cengiz, L. E. Yanmaz, H. Oral, A. Bastan, S.
This project was funded by the New Zealand Ministry Kaya, and A. Hayirli. 2010. Sensitivity and specificity of infrared
of Business, Innovation and Employment, Wellington, thermography in detection of subclinical mastitis in dairy cows. J.
and InterAg, Hamilton, New Zealand. Dairy Sci. 93:3525–3532.

Journal of Dairy Science Vol. 100 No. 5, 2017

 REMOTE MONITORING OF DAIRY COW WELFARE 9
Roelofs, J. B., and E. van Erp-van der Kooij. 2015. Estrus detection periences on responses of dairy calves to routine husbandry proce-
tools and their applicability in cattle: Recent and perspectival situ- dures. Animal 7:828–833.
ation. Anim. Reprod. 12:498–504. Stewart, M., J. M. Stookey, K. J. Stafford, C. B. Tucker, A. R. Rogers,
Rousing, T., M. Bonde, J. H. Badsberg, and J. T. Sørensen. 2004. S. K. Dowling, G. A. Verkerk, A. L. Schaefer, and J. R. Webster.
Stepping and kicking behaviour during milking in relation to re- 2009. Effects of local anesthetic and a non-steroidal anti-inflamma-
sponse in human-animal interaction test and clinical health in tory drug on pain responses of dairy calves to hot-iron dehorning.
loose housed dairy cows. Livest. Prod. Sci. 88:1–8. J. Dairy Sci. 92:1512–1519.
Schaefer, A. L., N. J. Cook, C. Bench, J. B. Chabot, J. Colyn, T. Sutherland, M. A., A. R. Rogers, and G. A. Verkerk. 2012. The ef-
Liu, E. K. Okine, M. Stewart, and J. R. Webster. 2012. The non- fect of temperament and responsiveness towards humans on the
invasive and automated detection of bovine respiratory disease on- behavior, physiology and milk production of multi-parous dairy
set in receiver calves using infrared thermography. Res. Vet. Sci. cows in a familiar and novel milking environment. Physiol. Behav.
93:928–935. 107:329–337.
Schaefer, A. L., N. J. Cook, S. V. Tessaro, D. Deregt, G. Desroches, P. Talukder, S., K. L. Kerrisk, L. Ingenhoff, P. C. Thomson, S. C. Garcia,
L. Dubeski, A. K. W. Tong, and D. L. Godson. 2004. Early detec- and P. Celi. 2014. Infrared technology for estrus detection and as
tion and prediction of infection using infrared thermography. Can. a predictor of time of ovulation in dairy cows in a pasture based
J. Anim. Sci. 84:73–80. system. Theriogenology 81:925–935.
Schütz, K. E., N. R. Cox, and C. B. Tucker. 2014. A field study of the von Borell, E., J. Langbein, G. Despres, S. Hansen, C. Leterrier, J.
behavioral and physiological effects of varying amounts of shade Marchant-Forde, R. Marchant-Forde, M. Minero, E. Mohr, A.
for lactating cows at pasture. J. Dairy Sci. 97:3599–3605. Prunier, D. Valance, and I. Veissier. 2007. Heart rate variability as
Stewart, M., A. L. Schaefer, D. B. Haley, J. Colyn, N. J. Cook, K. a measure of autonomic regulation of cardiac activity for assess-
J. Stafford, and J. R. Webster. 2008. Infrared thermography as a ing stress and welfare in farm animals—A review. Physiol. Behav.
non-invasive method for detecting fear-related responses of cattle 92:293–316.
to handling procedures. Anim. Welf. 17:387–393. Willis, G. L. 1983. A possible relationship between the flinch, step and
Stewart, M., H. M. Shepherd, J. R. Webster, J. R. Waas, L. M. kick response and milk yield in lactating cows. Appl. Anim. Ethol.
McLeay, and K. E. Schütz. 2013. Effect of previous handling ex- 10:287–290.

Journal of Dairy Science Vol. 100 No. 5, 2017