Hello, everyone, my name is Xin, I’m a second-year student in Master of

Agricultural sciences, majoring in animal science. My topic is Non-invasive

visual examination of physiological and behavioural parameters in pig and
cattle: A review, Supervised by Professor Fuentes. It will be presented through
introduction, methods, results, discussion and conclusion.

So, The research aims to review various computer vision technologies to

support the farmer by monitoring health status and early detection of
abnormal behaviours in large scale enterprises. The performance of developed
imaging systems is reviewed in terms of sensitivity, specificity, accuracy, error
rate and precision.

A literature review was conducted to create a list of models that help farms
with livestock production in Australia. Five criteria for model selection were
defined. First, models must include at least one specific livestock species from
either pig or cattle, which excluded models with a generic description of
livestock. Second, the model had to be focused on the farm level, the herd or
flock level, or the animal level. Models at the farm level were selected because
the farm is the main management unit of an agricultural system. Third, only
quantitative models that describe interactions among components within the
boundary of a system were included in the list of models. Fourth, given the
large number of models that has been developed, this study focuses on models
applicable in Australian livestock industry. Fifth, I selected documents
describing an original model, a new combination of existing models, or a
major extension of an existing model.

Following these methods, results are obtained as below. Image acquisition,

which is the first step of any machine vision system, mainly involves three
types of cameras - Complementary Metal-Oxide Semiconductor or CMOS for
short, infrared and depth sensor cameras. CMOS cameras need an additional
source of light to make the image visible. Examples of using this type of
camera in livestock behaviour detection are numerous (Porto et al., 2015;
Nasirahmadi et al., 2016b). The captured images are potentially suitable for
image processing algorithms to observe things like body weight, body
condition score and abnormal behaviours

Infrared or thermal cameras work similarly to optical CMOS cameras, By

receiving and measuring infrared radiation from the surface of an object, the
camera captures information on the heat that the object is emitting and then
converts this to a radiant temperature reading. Thermography can indicate
inflammatory, vascular or neurological disorders and physiological change in
response to environmental conditions.
The use of depth sensor cameras in agricultural products has been recently
described by . The 3D depth sensing makes it possible to overcome common
issues causing problems with 2D imaging systems, such as background
removal, segmentation, feature extraction and sensitivity to lighting variance.
They can give a better spectral resolution rather than a special resolution.
Thus, these cameras are widely used in measuring body shape and condition
of production animals (Nir, Parmet, Werner, Adin, & Halachmi, 2018).

So what kind of parameters can be measured using computer vision? Based on

length and width dimensions of pigs, some researchers have used top-down
view CMOS cameras to obtain estimates of individual pig live weight.
Similarly, image processing has been used to measure cattle live weight due to
the importance of live weight monitoring for milk and meat production. Also,
Body shape and condition of a live pig/cow is an important indicator of its
health, reproductive potential and value, whether for breeding or for carcass
quality Examples of using image processing for pig body condition have used
3D cameras for shape detection and thermal cameras for shape and body
contour detection. top view CMOS cameras, and thermal camera
measurement has been used to assess the thickness of fat and muscle layers
and provide a body condition score. Moreover, The body temperature is one of
the main indicators to evaluate health status of the livestock. Infrared cameras
are useful in this case to detect potential inflammation or diseases. Also,
tracking animals by machine vision has many possible advantages in livestock
monitoring. A top-down view camera has been applied for tracking piglets,
based on blob edge and an ellipse fitting technique. In addition, Early
uncovering of symptoms of disease is critical to successfully address the
animal welfare and disease problems on farm. This monitoring system can
improve the sustainability of dairy farming as well as the health and welfare of
the animals and the farmers themselves. The best opportunities for improving
computer vision solutions lie with hyperspectral imaging which can provide
additional information on the meat composition and structure.

Here is an example of the 3D reconstruction procedure of pigs. In this study,

Images can be collected by either a single sensor moved around or a set of
synchronised sensors installed on a reference frame. The live wright is then
calculated according to dimensions such as length from scapula to snout,
length from tail to scapula, shoulder width, breadth at middle and breadth at
back. Similar image processing has been used on cattle as well. And also for
body condition score estimation. Again, assessment of live animal body
condition by eye or hand is time and labour intensive and highly dependent on
the subjective opinion of the operator. Top-view 3D cameras has been used for
shape and body contour detection.
Livestock behaviours also contain important information that can enable
better management of animals and detection of problems. For example,
changes in feeding and drinking behaviours in pigs can reflect pig health. In
order to register the presence of pigs in a feeding area and detect feeding
behaviour, a multi-camera video system for obtaining top-down view images
has been applied, In dairy cattle, the lying behaviour affects the milk
production, and deprivation of adequate lying time reduces welfare. Again, a
top-down view CMOS camera system was developed. In addition, A reliable
heat detection is essential, as the time of insemination determines its success.
CMOS cameras are used to help precise insemination. It detects behaviours
such as sniffing and chin resting. Finally, abnormal behaviours are monitored
as well to improve animal welfare.

So, It is important to combine multiple factors and consider the integrity of

detection. for example, gathering data of the feeding time of cows with the ear
temperature gives good indication of cow’s activity level and health status.
And No single health parameter would be adequate for health monitoring.
Also, Data itself won’t improve performance. application of such techniques is
important to Measure and interpret the data to make good management
decisions, then implement the management changes to improve performance.

The computer vision technology has been widely considered in many

agricultural and industrial processes. However, such equipment is expensive
and requires skilled personnel. The main research gap is How to record and
present data for farmers to use easily. To convert academic statistics into
straightforward learning material for farm producers is what we should
further investigate. Moreover, The technology to provide producers with real-
time capability is still under development. Further elaboration of real-time
monitoring could be an important step towards continuously and
automatically 1. track individual animals. 2. Control inventory and comply
with traceback reporting. 3. Identify sick animals and poor performers. 4.
Enhance animal well-being and mitigate risks. 5. Identify animals in need of
treatment or determine when individual animals are ready for market. 6.
Measure and predict individual animal gain and market value.

conclusion: Livestock production data is traditionally obtained manually and

in multiple places, which is laborious and stressful to both the animal and the
workers. Assessment of live animal condition by eye or hand is also highly
dependent on the subjective opinion of the stockman. The computer vision
technology provide the ability to associate data to animal performance easily.
Application of modern digital technologies in 3D imaging systems offer
further possibilities for improvement. Although many machine vision
techniques have been recently applied by researchers for livestock behaviour
detection, further elaboration of image processing techniques could be an
important step towards the development of an automated system that can
detect behaviours of animals and decide the best solution or alarm in unusual
situations.

Nir, O., Parmet, Y., Werner, D., Adin, G., & Halachmi, I. (2018). 3D
Computer-vision system for automatically estimating heifer
height and body mass. Biosystems Engineering, 173, 4-10.
doi:https://doi.org/10.1016/j.biosystemseng.2017.11.014
Pezzuolo, A., Guarino, M., Sartori, L., González, L. A., & Marinello,
F. (2018). On-barn pig weight estimation based on body
measurements by a Kinect v1 depth camera. Computers and
Electronics in Agriculture, 148, 29-36.
doi:https://doi.org/10.1016/j.compag.2018.03.003