=Paper=
{{Paper
|id=Vol-2030/HAICTA_2017_paper75
|storemode=property
|title=SheepIT - An Electronic Shepherd for the Vineyards
|pdfUrl=https://ceur-ws.org/Vol-2030/HAICTA_2017_paper75.pdf
|volume=Vol-2030
|authors=Luís Nóbrega,Paulo Pedreiras,Pedro Gonçalves
|dblpUrl=https://dblp.org/rec/conf/haicta/NobregaPG17
}}
==SheepIT - An Electronic Shepherd for the Vineyards==
https://ceur-ws.org/Vol-2030/HAICTA_2017_paper75.pdf
SheepIT - An Electronic Shepherd for the Vineyards
Luís Nóbrega1, Paulo Pedreiras1, Pedro Gonçalves2
1
DETI/IT, lnobrega@ua.pt, pbrp@ua.pt
2
ESTGA/IT, pasg@ua.pt
University of Aveiro, Aveiro, Portugal
Abstract. This paper presents a new and innovative Internet of Things based
solution for controlling grazing sheep in vineyards. The SheepIT solution
includes a postural control mechanism that prevents animals from eating
grapes and the lower branches of the vines, but allows them feeding from the
unwanted weeds, thus taking advantage of the animal’s biologic need to feed
themselves to have an ecological vineyard weed control solution. Additionally,
a radio-based virtual fence mechanism is used to contain the flock inside the
desired grazing areas, allowing simultaneously to monitor animal’s activity
and to transfer the gathered data to a cloud application, for logging and
analysis purposes. This paper identifies the main requirements and presents the
system architecture. Moreover, the functional blocks that compose the
developed solution are detailed, with special focus on the collar to be carried
by the sheep. The implementation of the solution is also addressed in the
paper, and some preliminary experimental results, concerning the virtual fence
mechanism, are presented.
Keywords: Autonomous herd management, IoT, posture control, virtual fence,
cloud application
1 Introduction
The constant growth of unwanted and undesirable weeds in vineyards, which
compete for soil nutrients, forces the producers to repeatedly remove them through
the use of mechanical and chemical methods (Monteiro and Moreira, 2004). These
methods include machinery usage as plows and brushcutters to remove the weeds
between plant rows, and herbicides on the line between plant feet, in order to kill or
prevent the growth of weeds. Nonetheless, such methods are considered very
aggressive for vines, as well as harmful for the public health, since chemicals may
remain in the environment and hence contaminate water lines. Moreover, such
processes have to be repeated over the year, which entails a significant economic
impact, representing in the case of vineyards, 20 to 35% of total working time, with
costs per hectare ranging from 80€/ha up to 380€/ha (Carlos, 2014).
The use of animals to weed vineyards (Dastgheib and Frampton, 2000)(Bekkers,
2011), usually ovines, is an ancient practice used around the world. Animals grazing
in vineyards, feed from the unwanted weeds and fertilize the soil, in an inexpensive,
621
ecological and sustainable way. However, as depicted in the Little Prince story , 1
sheep may be dangerous to vines since they tend to feed on grapes and on the lower
branches of the vines, which causes enormous production losses. To overcome that
issue, sheep were traditionally used to weed vineyards only before the beginning of
the growth cycle of grapevines, requiring the use of mechanical and/or chemical
methods during the remainder of the production cycle.
The SheepIT project (SheepIT Project, 2017) aims at developing an autonomous
mechanism to control sheep’s posture and location during vineyard grazing periods.
The project includes an Internet of Things (IoT) based solution to monitor and
control the animals. Data concerning the behavior and physical condition of each one
of the animals are sent, in real-time, to a cloud platform. This cloud platform allows
the human operator to oversee, in an easy and efficient way, the flock, namely
browsing the collected data about animals and equipment. It also permits deploying
algorithms to process the data and detect abnormal situations, such as health
conditions, lost animals or attacks from predators, generating automatically alarms
when one of such events occurs.
The system architecture was designed in order to provide a flexible and adaptable
solution regardless of vineyard’s size and shape. Moreover, the human intervention is
maintained at a low level, being only indispensable for setting up the devices. The
system incorporates: a portable electronic collar, carried by the sheep, responsible for
monitoring and controlling its behavior; fixed devices called beacons, installed in
vineyards, responsible for interconnecting the network devices, define the virtual
fence placement and carrying out collar’s relative localization; a gateway device, that
aggregates data from all the beacons and uploads it to the cloud; and a cloud
application, responsible for gathering all the data collected and make them available
to the user.
This paper presents an overview of the SheepIT project, with special emphasis on the
global system architecture, functional blocks and their interactions. Moreover, and
due to its relevance, a special attention is given to the collars design. This paper
continues in Section 2 with a review of the related work in monitoring and
controlling animals. Section 3 describes the SheepIT architecture and the IoT
network. Section 4 describes the solution implementation and Section 5 presents
some preliminary results concerning the virtual fence, which is a key element of this
solution. Section 6 concludes the paper and presents future work.
2 Related work
In the scope of the SheepIT project, there are mainly two issues of interest that must
be explored, namely i) how to control animal’s posture and ii) how to monitor its
behavior, actions and location. For the latter case, many studies and applications can
be found in the literature (Umstatter, 2011), especially concerning monitoring the
location, pastures and welfare. On this, GPS (Global Positioning System), sometimes
combined with accelerometers, is the most portrayed technology found in the
1
Little prince
622
literature, being evaluated to be used for locating cattle (Augustine and Derner, 2013)
(Kjellqvist, 2008) (Turner et al., 2000), white-tailed deers (Bowman et al., 2000),
griffons (Nathan et al., 2012), crocodiles ((Hunter et al., 2013) and sheep (Rutter,
Beresford and Roberts, 1997). In the last case, a GPS, together with a jaw and
lying/standing sensors, are used to monitor the grazing areas of domestic sheep.
However, this solution weights almost 2kg, needs to be transported in the back of the
sheep and has an estimated autonomy of 7 days. The relatively short autonomy
highlights one of the GPS limitations, which is its high-energy consumption, that
together with its high cost and usual loss of satellites connection (Nadimi et al.,
2008) makes it unsuitable for animal’s localization, particularly small to medium size
ones.
To mitigate GPS device limitations, new alternatives started to arise, namely through
the use of algorithms to estimate the relative position between nodes. Within those,
and due to its simplicity, the Received Signal Strength Indicator (RSSI) is the most
popular (Umstatter, 2011). In fact, as the RSSI is a common parameter available in
most of the wireless interfaces and many wireless technologies are designed to
exhibit low energy consumption, this method becomes very appealing for carrying
out relative localization in energy-constrained scenarios. Therefore, several solutions
employing that technique come up, particularly using Zigbee Wireless Sensor
Networks (Nadimi et al., 2008) (Huircán et al., 2010). A solution using UHF radio
tags, able to communicate with network terminals equipped with GPS and GPRS
devices, was presented in (Thorstensen et al., 2004). The denominated e-shepherd
solution, aimed at developing a system capable of monitoring the location of grazing
sheep in mountains. However, this solution only allowed a rough estimation of the
flock location and presented many communication problems.
The control and conditioning of animal’s location is typically made using physical
ground-based fences. However, its cost and lack of flexibility, prompted the use of
virtual fences, in which electronic systems control the animal’s behavior (Anderson,
2007). These systems emit cues/stimulus to animals when they adopt undesirable
behaviors (e.g. approaching the defined boundary of the fence). These cues are
mostly constituted by a pair of stimuli, namely a warning tone or vibration, followed
by an electrostatic discharge, if the animal persists on the unwanted behavior
(Tiedemann, A.R.; Quigley, T.M.; White, 1999). Using a warning cue, allows
animals to associate it with an electrostatic stimulation and consequently revert its
behavior. The success of this process depends on a training process in which not all
breeds and animals react in the same way. It is important to stress that the use of
electrostatic cues raises ethical issues, but it is proven (Lee et al., 2008) that the
effects of using low energy electrostatic cues on animals are similar to weighing or
treatment processes, and hence can be considered suitable and safe to be used on
animals. Among the solutions described in the literature to constrain animals, the
ones tested in goats ((Fay, McElligott and Havstad, 1989) and sheep (Jouven et al.,
2012) are the most closely related with the SheepIT project. Both use training collars
for dogs with audio and electrostatic cues. Although both studies show that both
animals are able to be trained with those cues, they are not enlightening in what
concerns to the characterization of the stimuli (e.g. duration, intensity).
Table 1 compares the SheepIT requirements with the most relevant solutions found
in the literature and introduced before. The SheepIT system shall implement a
623
virtual fence, confining the sheep grazing inside a predefined area defined by
shepherds/farmers as well as it shall allow to know the localization of each sheep.
Moreover, the system comprises a device, integrated on the animal’s collar, that shall
include a posture control mechanism, in which the neck and head position are
monitored, in order to detect possible undesired behaviors. Additionally, the collar
device shall have small dimensions, similar to the available solutions for training
dogs, in order to be comfortably carried by sheep. The autonomy of collars shall be
at least 4 months, in order to avoid too often battery replacements, thus reducing the
maintenance time and costs. These last two requirements require a highly energy-
efficient system, in which communications, localization, sensing, actuation and
processing activities must be properly synchronized and scheduled, to attain the
desired functionality, while, at the same time, keeping the collars in low-power
consumption modes as long as possible. The data gathered shall then be delivered to
a cloud service, to be analyzed and presented to the user, using a simple and user-
friendly interface. Finally, as the solution is intended to be used in vineyards to
remove weeds, it is expected to have thousands of sheep in few hectares, resulting in
the need of a high scalable network.
Table 1. Comparison between the SheepIT requirements and the similar solutions available
Requirements/Solution SheepIT E- Nadimi Huircan Jouven Rutter
Shepherd et al et al et al et al
Small Dimensions (0,5Kg-1Kg) ✔ ✔ ✔ ✔ ✔ ✘
Localization (relative or absolute) ✔ ✔ ✔ ✔ ✘ ✔
Data Collection ✔ ✔ ✔ ✔ ✘ ✔
Communication Infrastructure ✔ ✔ ✔ ✔ ✘ ✘
High Autonomy (~ 4 months) ✔ ✘ NA ✘ NA ✘
Virtual Fence ✔ ✘ ✘ ✘ ✔ ✘
Posture Control ✔ ✘ ✘ ✘ ✘ ✘
Pasture geographic delimitation ✔ ✘ ✘ ✘ ✘ ✘
User interface ✔ ✘ ✘ ✘ ✘ ✘
Network Scalability ✔ NA NA NA NA NA
Year 2017 2016 2008 2010 2012 1997
3 System architecture
As depicted in Fig. 1, SheepIT follows a typical IoT architecture, with a Wireless
Sensor Network (WSN) layer, a cloud computing layer and an application layer. On
the WSN layer, mobiles nodes, named collars, are carried by sheep and are composed
of a set of sensors and actuators, a microprocessor, a radio link and a battery. Sensors
detect animal’s posture, while actuators apply stimulus when sheep adopt undesirable
behaviours. The radio link reports data sensed by collars and provides support to
relative localization, through the measurement of the link’s RSSI. Carrying out the
localization using the same radio infrastructure used for data collection contributes
significantly to reduce the power consumption, as it can be obtained with minimum
energy expenditure, because it requires only processing the RSSI after regular radio
624
data exchanges. Collars move within the range of one or more fixed beacons, which
are nodes that are installed by shepherds in the areas to be grazed, enabling the
implementation of a virtual fence mechanism and the gathering of data transmitted
by collars. The system also contains a gateway node, which collects the data gathered
by the beacons and sends it to a cloud platform. Finally, a web application processes
the data received and makes them available to the user.
The following subsections present in more detail the system, with special focus on
the wireless sensor network layer, collar device, communication infrastructure and
cloud platform.
Fig. 1. Overall system architecture
3.1 Wireless Sensor Network
Confining animals inside the vineyards without continuous human supervision
requires a fence mechanism. In order to allow an easy and flexible definition of
weeding areas, it is adopted a virtual fence approach, supported by an RSSI-based
relative localization mechanism. This one uses the measurement of the RSSI of the
communication between beacons and collars, which are carried out at regular
intervals, to determine an estimate of the distance between them. Moreover, this
relative localization can be converted in an absolute one, as the beacons, which do
not have particular size and weight limitations, shall incorporate a GPS device. To
enable this mechanism, periodic messages are transmitted between beacons and
collars, allowing regular updates of their location as well as providing
synchronization of the network. Moreover, to enhance the localization process,
beacons are installed close to each other to guarantee overlapping of individual areas
of coverage, which in turn allows merging multiple localization data, thus
contributing to improve the accuracy of the relative localization process.
In the general case, beacons have to relay information of other beacons, since it may
be necessary to cover relatively large areas with arbitrary topologies, therefore it is
not possible to ensure that the gateway device can communicate directly with all
bacons. Therefore, the system integrates a routing mechanism to ensure that the data
sent by collars and collected by beacons, are relayed until reaching the gateway and
consequently the cloud application.
As the collars are mobile and beacon’s coverage areas overlap, a collar can, in a short
period of time, be in the range of different beacons or set of beacons. Hence, the data
from a collar can be received by multiple beacons. Relaying replicated data would
625
consume bandwidth unnecessarily, therefore beacons follow a data centric approach
(Ghaffari, Jafari and Shahraki, 2013), merging the information received locally and
the information relayed by other beacons.
The existence of multiple nodes competing for the transmission medium, allied to the
system energy constraints, requires an efficient Medium Access Control (MAC)
mechanism. The system incorporates different types of traffic, with different
purposes and specifications, namely: periodic sensor data sent from collars to
beacons, periodic localization messages sent beacons to collars, periodic relay traffic
between beacons and sporadic traffic to allow nodes to register dynamically in the
network. As the overall bandwidth utilization can be relatively high (the system shall
allow the presence of several hundreds or even thousands of animals over regions of
a few square kilometers) and it is important minimize the energy expended by
collars, it is adopted a temporal multiplexing approach combined with a cyclic
structure. This cyclic structure is composed of different micro-cycles (uC), on which
different traffic is transmitted, forming a macro-cycle (MC) that is repeated in a
cyclic way (Kopetz, 2011). A uC starts with a message from beacons to collars that is
used to perform localization, to synchronize collars with the remaining nodes and to
identify the type of uC. Depending on the uC type, the remaining time is used for
different purposes, eventually employing different access control methods. Collars
registered on the network send sensor data periodically to the beacons. To minimize
the number of collisions and the amount of time that the radios have to be active, this
kind of uC adopts a Time Division Multiplexing Access (TDMA) scheme, in which
each node has a non-overlapping individual communication slot. Communication
between beacons is also periodic, and uCs dedicated to this kind of communication
adopt a similar TDMA scheme. On the other hand, there are communications that are
not periodic. For example, when a sheep enters in a protected area for the first time
or after being absent for a long time, its collar must register on the system to get a
periodic communication slot. For this purpose, there is one uC type that has an
arbitration scheme based on Carrier Sense Multiple Access, which allows the
transmission of unscheduled communications such as the one related with
registration.
3.2 Posture control
The posture control mechanism is crucial for the SheepIT solution, since it enables
animals to be used inside the vineyard, without threatening the vine grapes and lower
branches. This mechanism is enabled by collars applied on animals and it is based on
the three main blocks shown in Fig. 3: a set of sensors, to monitor the animal posture
and movements; an algorithm executed by a microprocessor that analyzes the data
gathered by sensors, applies sensor fusion and decides about the necessity of
applying stimuli; and a set of actuators that apply the actual stimulation to the
animal, when commanded to do so.
At this stage the animal's posture is monitored by two kinds of sensors: an ultrasound
sensor, to measure the distance between the neck of the animal and the ground; and
an accelerometer to monitor the neck tilt. These sensors are periodically read and
their inputs are fused, in order to detect incorrect animal postures. The need for using
626
more than one kind of sensor and fuse their readings arises from the fact that the
terrain has irregularities (e.g. obstacles and holes in the ground) and animals may
adopt different postures (e.g. laying on the ground or walking), together with the
need for having a reliable detection of undesirable behaviors, as the success of the
animal’s learning process depends on this and it is considered essential for animal’s
comfort. The combination of sound and electrostatic stimuli (Umstatter et al., 2013)
are the most effective, and hence used as the posture control actuators.
3.3 Cloud platform
A cloud platform is proposed with a triple purpose: (i) to store sensor data streams
and perform continuously data mining to extract relevant information about the
location, activity and behavior of animals, as well as about device’s state and
operation; (ii) to provide a user interface through a web interface, allowing shepherds
and/or farmers to use a mobile device to access to the collected data; (iii) to
autonomously generate alarms when problems occur on animals or equipment.
The IoT gateway periodically streams the device monitoring information to the cloud
platform through a broker (Fig. 2). This broker delivers the information within the
cloud platform to entities, on a subscription basis. This allows several entities to
subscribe specific subjects of the stream and carry out data mining of specific
subjects. Moreover, the subscriber entities identify critical values and trigger alarms
(e.g. animal out of bounds, animal’s panic, equipment failure), storing this
information on a database.
Fig. 2 – Cloud platform organization
The alarms are sent by the system to the human operator, notifying him about the
occurrence of critical events, so that he can intervene in the system, correcting the
anomalies as soon as possible, preventing undesirable consequences (e.g. loss of
animals, network failures, damages in the cultures).
Together with the dynamic information generated by the system (e.g. animal activity,
battery state), the system database also contains static information about animals (e.g.
gender, birthdate, vaccines) and equipment (e.g. model, firmware version) that can be
inserted by the system operator. This introduces additional value to the solution,
allowing farmers to correlate information gathered on-line by collars installed on the
sheep with specific information of each animal, as veterinary data.
627
4 Implementation
SheepIT collars are a crucial element due to their requirements and features. Its
implementation is based on Texas CC1110 SoC, which includes a microcontroller
with several IO Ports and timers, allowing diverse power saving modes. Moreover,
this SoC also includes a radio module operating at the 433 MHz ISM band. This SoC
was selected because it has a low-cost and low power consumption and the 433MHz
band radio is suitable to be the used on vineyards, considering the radio environment
constraints (e.g. trees, posts, metallic strings and vine relief). In addition to the
CC1110, the collar, whose architecture is illustrated in Fig 3, contains an
accelerometer, an ultrasound-based distance measurement circuit (using a transceiver
similar to the ones used on cars), a buzzer, and a high voltage stimulator.
The first prototype of the collar, shown in Fig. 4, includes all the signal conditioning
circuitry necessary to integrate sensors and actuators, as well as the firmware to
control the circuit peripherals, allowing the complete parametrization of
measurements and actions.
Fig. 3. Collar modules Fig. 4. Collar prototype
SheepIT beacons are also based on the Texas CC1110 SoC. Regarding
communications, no additional components are required, as the same radio used for
collar-beacon communication is also used for beacon-beacon and beacon-gateway
communication. In the future, a GPS module will be added to the beacon hardware
platform, to enable the deployment of the absolute localization services.
5 Results
The project is still at a very early development stage and, so far, it was only possible
to make preliminary tests of the communications and virtual fence operation. As
such, a scenario test, depicted in Fig. 5, composed of two beacons directly connected
to Linux-based PCs, and a collar, was set up. The collar node moves between the two
beacons, which receive the data from the collar, and computes the RSSI of each link.
The PCs are connected, via a serial link, to the beacons, displaying and storing the
received data. Beacons are placed at a higher position (2m) than collars (50 cm).
628
Measured values were captured during 3 minutes, from both beacons, with 5 meters’
intervals.
Fig. 5. Topology of the test
The initial data capture analysis (see Fig 6) allows us to conclude that the RSSI value
decreases polynomially with the distance between the collar and the beacon. Also, a
huge variation of the RSSI can be observed for the same distance. On this basis, it
was implemented a virtual fence mechanism establishing a minimum RSSI value of -
55 dBm, which in our case corresponds to a distance around 40 m. A communication
received by the collar with this RSSI, means the detection of a fence infraction which
in turn triggers an audible warning signal. If during the subsequent communications
the RSSI value doesn’t return to values greater than -55 dBm, the system responds
with an electrostatic stimuli.
Fig. 6. RSSI measured versus the distance between nodes
In order to enable the computation of an estimation of the distance of a collar
according to the RSSI measured, the variables from the graphic represented in Fig. 6
were inverted and a polynomial regression executed (Fig.7). As the virtual fence
mechanism is especially crucial at the border of the fence, we evaluated the behavior
of the polynomial for values measured close to the border of the fence (40m). Hence,
for all the values of the RSSI measured at distance equal to 40m, we calculated the
respective distance estimation using the polynomial. Fig. 8 shows that the error
associated to the polynomial regression and RSSI measurements result in situations
on which the stimulators are triggered even if the collar is still in the border if the
fence (red line). However, this limitation can be minimized if the warning sound are
triggered before the collar reaches areas close to the border.
629
Fig. 7. RSSI measured versus the distance Fig. 8. Distance estimated using the
between nodes polynomial
6 Conclusions
The SheepIT project aims at taking advantage of sheep grazing behavior, to weed
vineyards in a economical and ecological approach, fertilizing the soil and thus
optimizing the wine production. However, sheep presence within the vineyards
creates several challenges, especially the ones related with preserving the integrity of
vines and grapes, to not jeopardize the wine production, and with keeping sheep herd
within a designated region, all of this without direct human supervision. As such,
posture control and virtual fence mechanism are sought in the scope of the project.
Although in the past, sheep have been used to graze in vineyards, nowadays, thanks
to the specialization that the vines suffered, the flock is a foreign element. In order to
facilitate the animal management process by vineyard staff, the project includes
collars and beacons which, combined with a communication infrastructure, feed a
cloud platform with animal data, following IoT-like design principles.
The project still is in its initial period and its development in a very embryonic stage.
This paper presents the project requirements and goals, as well as the proposed
system architecture and a preliminary assessment of the RSSI-based virtual fence
mechanism. Tests have been carried out on a test scenario containing two beacons
and a collar, with a very rudimentary algorithm to control sheep position.
The obtained results show that the relative localization mechanism, despite basic,
offers a precision of around 2m at a 40m distance. This value is of the order of
magnitude of the required one, and can still be improved, thus indicating that this
localization method is suitable for supporting the virtual fence mechanism. It is,
however, important to improve the precision of the localization, namely by
combining the RSSI values of several beacons, eventually adding also dead
reckoning, and evaluate it in a real scenario to check if the mechanism promotes
animal learning.
Beyond assessing the fence mechanism in a real scenario, tests have to be performed
to gather real data in order to learn how to recognize animal posture from the collar
sensory information, and thus to enhance the definition of a suitable posture control
algorithm. Moreover, further work has to be performed in the communication
630
infrastructure, in order to validate and adapt its operation to the Douro vineyards
orography.
Acknowledgments. This work is supported by the European Structural Investment
Funds (ESIF), through the Operational Competitiveness and Internationalization
Programme (COMPETE 2020) [Project Nr. 017640 (POCI-01-0145-FEDER-
017640)].
References
1. Anderson, D. M. (2007) ‘Virtual fencing past, present and future’, in Rangeland
Journal, pp. 65–78. doi: 10.1071/RJ06036.
2. Augustine, D. J. and Derner, J. D. (2013) ‘Assessing herbivore foraging
behavior with GPS collars in a semiarid grassland’, Sensors (Switzerland),
13(3), pp. 3711–3723. doi: 10.3390/s130303711.
3. Bekkers, T. (2011) ‘Weed control options for commercial organic vineyards’,
Wine and viticulture journal, pp. 62–64. Available at:
http://www.tobybekkers.com/uploads/5/4/3/2/5432540/bekkers-
julyaug11wvj.pdf.
4. Bowman, J. L., Kochany, C. O., Demarais, S. and Leopold, B. D. (2000)
‘Evaluation of a GPS collar for white-tailed deer’, Wildlife Society Bulletin,
28(1), pp. 141–145. doi: Cited By (since 1996) 46\rExport Date 12 June 2012.
5. Carlos, C. (2014) ‘Spraying challenges in the Douro Wine Region of Portugal’.
Lien de la Vigne, France. Available at:
http://www.advid.pt/imagens/comunicacoes/13993677624772.pdf.
6. Dastgheib, F. and Frampton, C. (2000) ‘Weed management practices in apple
orchards and vineyards in the South Island of New Zealand’, New Zealand
Journal of Crop and Horticultural Science, 28(1), pp. 53–58. doi:
10.1080/01140671.2000.9514122.
7. Fay, P. K., McElligott, V. T. and Havstad, K. M. (1989) ‘Containment of free-
ranging goats using pulsed-radio-wave-activated shock collars’, Applied Animal
Behaviour Science, 23(1–2), pp. 165–171. doi: 10.1016/0168-1591(89)90016-6.
8. Ghaffari, Z., Jafari, T. and Shahraki, H. (2013) ‘Comparison and Analysis Data-
Centric Routing protocols in wireless sensor networks’, Communication
Systems and.
9. Huircán, J. I., Muñoz, C., Young, H., Von Dossow, L., Bustos, J., Vivallo, G.
and Toneatti, M. (2010) ‘{ZigBee}-based wireless sensor network localization
for cattle monitoring in grazing fields’, Computers and Electronics in
Agriculture, 74(2), pp. 258–264. doi:
https://doi.org/10.1016/j.compag.2010.08.014.
10. Hunter, J., Brooking, C., Brimblecombe, W., Dwyer, R. G., Campbell, H. a.,
Watts, M. E. and Franklin, C. E. (2013) ‘OzTrack -- E-Infrastructure to Support
the Management, Analysis and Sharing of Animal Tracking Data’, 2013 IEEE
9th International Conference on e-Science, pp. 140–147. doi:
10.1109/eScience.2013.38.
631
11. Jouven, M., Leroy, H., Ickowicz, A. and Lapeyronie, P. (2012) ‘Can virtual
fences be used to control grazing sheep?’, Rangeland Journal, pp. 111–123. doi:
10.1071/RJ11044.
12. Kjellqvist, S. (2008) Determining Cattle Pasture Utilization Using GPS-collars.
Sveriges lantbruksuniversitet (Studentarbete (Sveriges lantbruksuniversitet,
Institutionen för husdjurens miljö och hälsa)). Available at:
https://books.google.pt/books?id=zkEBkAEACAAJ.
13. Kopetz, H. (2011) Real-time systems: design principles for distributed
embedded applications. Springer Science & Business Media.
14. Lee, C., Fisher, A. D., Reed, M. T. and Henshall, J. M. (2008) ‘The effect of low
energy electric shock on cortisol, β-endorphin, heart rate and behaviour of
cattle’, Applied Animal Behaviour Science, 113(1–3), pp. 32–42. doi:
10.1016/j.applanim.2007.10.002.
15. Monteiro, A. and Moreira, I. (2004) ‘Reduced rates of residual and post-
emergence herbicides for weed control in vineyards’, Weed Research, 44(2), pp.
117–128. doi: 10.1111/j.1365-3180.2004.00380.x.
16. Nadimi, E. S., Søgaard, H. T., Bak, T. and Oudshoorn, F. W. (2008) ‘ZigBee-
based wireless sensor networks for monitoring animal presence and pasture time
in a strip of new grass’, Computers and Electronics in Agriculture, 61(2), pp.
79–87. doi: https://doi.org/10.1016/j.compag.2007.09.010.
17. Nathan, R., Spiegel, O., Fortmann-Roe, S., Harel, R., Wikelski, M. and Getz, W.
M. (2012) ‘Using tri-axial acceleration data to identify behavioral modes of free-
ranging animals: general concepts and tools illustrated for griffon vultures’, The
Journal of Experimental Biology, 215(6), pp. 986–996. doi: 10.1242/jeb.058602.
18. Rutter, S. M., Beresford, N. A. and Roberts, G. (1997) ‘Use of GPS to identify
the grazing areas of hill sheep’, Computers and Electronics in Agriculture, 17(2),
pp. 177–188. doi: 10.1016/S0168-1699(96)01303-8.
19. SheepIT Project (2017). Available at: http://www.av.it.pt/sheepit/ (Accessed: 15
April 2017).
20. Thorstensen, B., Syversen, T., Bjørnvold, T.-A. and Walseth, T. (2004)
‘Electronic shepherd-a low-cost, low-bandwidth, wireless network system’, in
Proceedings of the 2nd international conference on Mobile systems,
applications, and services. Boston, MA, USA: ACM, pp. 245–255. doi:
10.1145/990064.990094.
21. Tiedemann, A.R.; Quigley, T.M.; White, L. D. . et al. (1999) ‘Electronic
(fenceless) control of livestock’, Res. Pap. PNW-RP-510. Portland.
22. Turner, L. W., Udal, M. C., Larson, B. T. and Shearer, S. a. (2000) ‘Monitoring
cattle behavior and pasture use with GPS and GIS’, Canadian Journal of Animal
Science, 80(3), pp. 405–413. doi: 10.4141/A99-093.
23. Umstatter, C. (2011) ‘The evolution of virtual fences: A review’, Computers and
Electronics in Agriculture, 75(1), pp. 10–22. doi:
10.1016/j.compag.2010.10.005.
24. Umstatter, C., Brocklehurst, S., Ross, D. W. and Haskell, M. J. (2013) ‘Can
the location of cattle be managed using broadcast audio cues?’, Applied Animal
Behaviour Science, 147(1–2), pp. 34–42. doi: 10.1016/j.applanim.2013.04.019.
632