Non Invasive Estimating of Cattle Live Weight Using Thermal Imaging

The body mass is one of highly important indicators of the development of the young cattle, therefore it is essential to follow it up. Only the animals with sufficient body frame and with well muscled  top – quality body parts can be successfully fattened to high body mass. To obtain the highest precision in measuring of cattle weight gain, weighing should take place repeatedly in the period prior to slaughter.Methods  for  assessing  growth  by  the  weighing  of  animals  are  an  important  factor  in monitoring  production  capacity  of  cattle  according  to  their  inheritance  capacity,  food conversion, average daily gain and carcass yield all of which subsequently affect the output of the herd. The most reliable information about the body mass of cattle can be obtained by weighing the animals. Accurate data on the body mass are obtained by weighing with the use  of  various  weighing  devices,  such  as  the  spring  weighing  devices  or  more  modern electronic  weighing  devices.  Since  in  most  cases,  weighing  is  done  manually  the  process often  needs  at  least  two  stockmen,  and  takes  3  to  5  minutes  per  bull.  In  addition,  the procedure   is   stressful   for   the   whole   herd,   and   from   an   ergonomic   point   of   view, unsatisfactory (Brandl & Jørgensen, 1996).However,  the  raisers  did  not  have  available  weighing  devices  frequently  and  the  animals cannot  be  often  weighed,  thus  the  individual  body  parts  are  measured  alternatively. Usually, measuring of the whither size suffices, because the body frame, size and length are in positive relation to intensity and capacity of meat production. However, the information acquired is less accurate than the one obtained by the weighing. The differences are affected by the breed type, satiety of the animal, manner of feeding and even tightness of the tape during measuring (Kogel & Utz, 1988). It is also known, as much as the calves deviate from the optimal body development, the establishing of the body mass deviates. Thus, in order to calculate the body mass more accurately, it is necessary to consider also the trunk length in addition to the whither size.In  the  following  chapters  the  comparison  between  the  manual  and  automated  electronic weighing systems is going to be presented firstly, which clearly shows the most important advantages of the modern systems. Then, the non invasive, indirect weight determination from  the  dimensions  of  the  animal,  measured  by  an  image-processing  model  will  be presented  and  explained  that  can  be  used  for  automatic  and  precise  estimation  of  body dimensions from images of live animals and for predicting the body weight of individual cattle.

Manual weighing

Manual weighing represents the oldest and simplest method for assessing the body mass of live  animal,  which  served  for  centuries  precise  enough  for  commercial  use.  Nowadays weighing  on  balance  is  still  important  method  for  measuring  the  growth  of  animals  and estimating  the   fodder   conversions,   because   the   device  is   cheap   and   easy   to   handle. However, whenever using manual balance in progeny test of bulls sires a farmer must first insure a fattening barn for each manual weighing, where a particular number of animals are placed together into the group boxes. The groups are usually formed according to the body mass  so  the  animals  remained  together  until  the  end  of  the  growing.  Normally  a  plan  of several  weighing  is  scheduled,  starting  from  housing  till  slaughtering  in  60  days  interval. For each weighing the animal must be moved from the group box on the weighing device, which is very stressful for the cattle as well as the stablemen. The simplest manual weighing device is shown in Fig. 1, which is calibrated manually once the cattle stand calmly on it. After the balance is set correctly, the stablemen can read the weight.

The next generation of balances represents the electronic weighing device, which contrary to manual devices is calibrated automatically and can read the body weight autonomously as well. A sample of automatic electronic device EC 2000 is shown in the Fig. 2. As seen, during the weighing the load bars must be installed under the platform on a firm, level surface and prevent  platform  movement.  Then  the  load  bar  cables  are  connected  to  the  indicator  to ensure that they would not be walked on, chewed by animals or squashed in any way.Once  the  cattle  is  standing  on  the  weighing  scale  additional  measurements  of  the  trunk, length, the chest size, the withers height and the croup height can be measured as well. The height of the withers, the croup height, the body length, the croup length, the chest width, the haunch width and the chest depth are usually done by means of Lydtin’s rod (Fig. 3). The beginning and the end of certain body dimension are taken by two foldable arms, which can be moved vertically on the rod. However, for measuring of animals the stableman must apply additional devices as  the 117 cm long hollow Lydtin’s rod, measuring tape – meter and the compasses (Fig. 3 left). The measuring tape is a 2.00 cm long scale made of linen or metal  and  serve  to  measure  the  chest  size.  Unfortunately,  there  are  certain  differences between  the  body  masses  stated  by  individual  authors.  The  compasses  are  metallic  and ensure measuring of up to 60 cm length (Kräußlich, 1994; Cepon et al., 2006).

Automatic weighing

Automatic weighting systems have been intensively researched for pigs, sheep and poultry, however  in  cattle  production  only  weighting  of  dairy  cows  is  reported  by  Cveticanin  & Wendl  (2004)  and  Pastell  et  al.,  (2006).  Different  automatic  weighing  systems  have  been intensively researched in cattle as well as dairy production.The sophisticated radio frequency identification system was found to be a highly effective tool for documenting bunk attendance, feeding behaviour and weighing of animals on the scale   in   cattle   production,   whenever   animals   were   equipped   with   two   separate   RF transponders (Schwartzkopf-Genswein et al., 2002). There is a worldwide movement towards automation in cattle husbandry, with the objective of fully automating every process from feeding to milking. Automatic milking has become a common  practice  in  dairy  production  and,  at  the  end  of  the  year  2003,  about  2200  farms worldwide  used  a  milking  robot  (Rousing  et  al.,  2004).  Milking  robots  offer  a  unique possibility for the dynamic measurement of body weight. For example, it is highly probable that hoof problems can easily be noted by separately measuring the load on each leg. Dairy farming  is  developing into  full  automated  production  system  in  which  scientists  want  to automate every single process from feeding to milking in order to reduce production costs and maximise milk yield. Dynamic weighing of cattle represents an important segment of this system. Weighing in motion without stopping the animal’s movement after it leaves the milking parlour is of interest due to continuity of the milking process and smaller variations of body weight when measured at the same time of the day. The forces to the ground during the cow’s passage are recorded on walk-through weighers (Rousing et al., 2004). Cveticanin & Wendl (2004) designed the fuzzy-logic dynamic weighing system for dairy cows based on a mathematical model for simulating a cow’s walk, which calculates the body weight with an average error of less than 2 %. The simplified two- legged system measures the force to the ground produced by a cow’s feet when the animal is crossing the scale. Depending on walking speed and force, a body weight is calculated and compared with the database.

Pastell et al. (2006) described another automatic weighing system made from four strain gauge balance devices installed in a milking robot. The computer program was able to measure the average weight, the weight variation of each leg and the total weight with 90 % accuracy. From many different models, two might be suggested for monitoring of young sires.

The GrowSafeTM  system

The GrowSafeTM  (Fig. 4) system consisted of two antennas, each embedded in a rubberized mat lining the outer wall of the 30 m feedbunks, passive transponders (Texas Instruments Inc., Dallas, TX) encased in plastic ear tags (Allflex, Dallas, TX), a data-logging reader panel connected  to  the  antennae,  and   a  computer  to  which  data  were uploaded  and  analysed (Gibb & McAllister, 1999; Schwartzkopf-Genswein et al., 1999). The antennae emitted a 130 kHz electromagnetic field, and detected the transponders borne by the cattle whenever they came within 50 cm of the feedbunk.This system can identify and record bunk attendance at the bunk of an unlimited number of animals simultaneously. The reader panel logs the presence of each transponder every 6 s for as long as the transponders are within the read range of the antenna. These data can be use to derive the bunk visit frequency and  duration. For the purpose of  summarizing the transponder data, different meal criterion can be selected. According to Gibb & McAllister (1999)  a  300  s  meal  criterion  is  most  suitable,  since  it  is  based  on  visual  observation  and validation work carried out by (Schwartzkopf-Genswein et al., 2002) and is the same as that reported by Sowell et al. (1998) for beef cattle. Passive radio transponder data is possible to be collected on all the animals for 24 h per day throughout  the  growing  period.    The  computer  on  which  the  data  are  storied  must  be checked every week to ensure that all cells are read when the animals’ transponders are held next to the cell. Every  time  an  animal  visits  the  water  trough  GrowSafe  BeefTM   measures  weight  and drinking behaviour, so the graph can be plotted illustrating the number of visits hourly at each GrowSafe BeefTM drinking position over a three day period

At  least  two  other  system  similar  to  the  GrowSafe  BeefTM   system  are  known  from  the literature.  First,  Cveticanin  &  Wendl  (2004)  designed  the  fuzzy-logic  dynamic  weighting system  of  dairy  cow  based  on  the  mathematical  model  for  simulating  cow  walk,  which calculates the body weight with the absolute average error less than 2 %. The simplified two legged system measures the force to the ground when the animal is crossing the scale and generate  a  curve,  which  is  compared  with  the  database.  Pastell et  al.  (2006)  described another  automatic  weighing  system  made  from  four  strain  gauge  balances  equipment installed  into  a  milking  robot.  The  computer  program  was  able  to  measure  the  average weight, the weight variation of each leg and total weight with 90 % accuracy.

Weight determination using body measurements

Since weighting of older cattle represents a dangerous job for the stableman, the significant correlation between live weight and dimensions of the cattle has led many authors to study the possibility of estimating body weight from the dimensions of the cattle. Heinrichs et al. (1992) indicated that the linear regression of body weight on heart girth had the highest R2, followed  by  hip  width,  body  length  and  wither  height.  Although  all  measurements  are highly correlated, addition of the second body measure contributes a little predictive benefit in  the  estimation  of  Holstein  heifers’  body  weight.  Also  Wilson  et  al.  (1997)  detected  the addition of heart girth as a second measurement to the wither height as the most important contribution for estimating the body weight of Holstein veal calves.

Enevoldsen & Kristensen (1997) evaluated the use of wither height, hip height and width to predict the body weight. Seven regression models were developed based on indicators, which are relatively simple to obtain precisely because the anatomical locations are easy to identify. Also Willeke & Dürsch (2002) detected high significant correlation in hip height, heart girth and weight also for the Simmental heifers suggested the third order polynomial equation as the most fitting. The chest size as a basis for the determination of the body mass of living cattle is increased close behind the shoulder-bones. During this measuring the animal must stand with the legs placed  parallel  and  the  head  should  be  kept  normally  (Fig.  1).  Each  time  after  the measurement is  taken, the body mass is read according to the table. However, it must be kept into account that the determination of the body mass on the basis of the chest size is only approximate.

Determining dimensions by image analysis

Computer-controlled systems  for  the  remote  monitoring  of  livestock  have  the  potential  to increase   production   efficiency   and   improve   animal   health   and   welfare.   Examples   of potential applications for image analysis based systems in pig husbandry were suggested by Schofeld   (2003)   and   include   recording   of   animal   weight,   growth   rate,   quality   and conformation,  control  of  diet,  monitoring  behavioural  vices  and  providing  management decision  support.  Schofeld  (1993)  describes  how  remote  monitoring  systems  have  the greatest value where continuous observation of intermittent events is essential, and where data  collection  is  tedious  and  labour  intensive,  e.g.  monitoring  weight  gain,  feeding behaviour and breeding processes including heat detection and farrowing.

A visual image analysis (VIA) system can provide continuous automatic collection of size and shape data in pigs. Thus, it serves as an accurate means of for reflecting pig live mass and  for  tracking  changes  in  the  pig  size  over  time  periods  that  are  sufficiently  short  for commercial use (White et al., 2004).

In order to apply image analysis to the weighing of bulls, the determination of animal body dimensions from images must be possible. Prior to using the image processing, a prediction function  was  established  using  the  relationship  between  body  dimensions  from  acquired images and the live weight of the specific cattle breed. Since the image is only a 2-D plain projection of the animal, the loss of one dimension limits the application of such a system to measuring vertical and horizontal dimensions. Therefore, the prediction functions should be precise  enough  to  obtain  valid  information  and  can  be  similar  to  those  described  in  the previous chapter only when the hip height and wither height are included in the polynomial equations.According to Schaefer & Tong (1998), the thermal expression of warm-blooded animals is highly correlated with various tissue composition characteristics of specific animals, which involve the relative proportions and total quantities of different types of tissue in the animal. Therefore  infrared  thermal  images  taken  of  liveanimals  are  suggested  for  detecting  and inspecting the body composition noninvasively. Kmet et al. (2000) studied the application of image processing for slaughter value analysis on  the  basis  of  three  images  (above,  left  and  rear)  captured  from  15-month-old,  live  Simmental  bulls.  Live  weight  was  found  to  be  highly  correlated  with  a  stepwise  linear regression  model  based  on  the  animal  shoulder  width,  lumbar  protuberance  in  the  body width, upper body area and rear thigh area.In the next chapter we are going to present the results of original approach for determining the body dimensions and estimating the body mass of cattle via thermal camera and image analysis developed by Stajnko et al. (2008), which was based on the data collected from May 2006 to June 2008 at the Faculty’s experimental farm in Rogoza.

Image analysis procedure

As  opposed  to  common  RGB  visual  techniques,  thermography  is  based  on  sensing  an object’s own heat radiation. This enables evaluation of different characteristics of observed objects with the use of visual cameras. The authors have found few references concerning application  of thermal  imaging  for  determining  plant  parts,  counting  fruit,  vegetable, seedlings or selecting weeds from plants and background. In our experiment the examined cattle were captured from the side by the AGEMA 570 (Flir SystemsTM) thermal camera with an image resolution of 320×240 pixels. The emissivity of the object was set to 0.98 and the temperature resolution was better than 0.5°C, which enables precise detection of body heat in  any  environmental  condition.  The  measurements  were  performed using  cold  concrete wall  surroundings,  which  enabled  a  temperature  difference  and  therefore  a  sharp  edge between the animal and the background on the capture images.The  uncertainty  of  body  edge  measurements  depends  on  the  image  resolution  and  the position of animal in relation to the camera. Therefore, for holding the distance between the camera  and  the  object  constant  for  all  measurements,  the  animals  were  guided  through  a narrow  corridor,  which  also  prevented  the  animal  from  moving  perpendicular  to  the camera. When the animal was walking through the corridor, we were waiting until it stood in  an  appropriate  position  on  the  image  frame.  Such  procedure  was  possible  because  the animals were not aware of being captured by the thermal camera.

The algorithm for determination of animal’s measures will be illustrated with one sample of a  cattle  image  (Fig.  5),  chosen  to  be  representative  for  demonstrating  the  results  of  the procedure. As seen from the BMP image, the thermal camera was able to separate the cattle from  the  surroundings  accurately,  so  any  additional  image  separation  was  not  required. Prior to every set of measurements a Lydtin’s stick was captured on the image together with the animal. It was applied for the pixel/unit calibration as it ensured enough temperature gradient  to  the  cattle  body,  which  was  required  for  the  automatic  image  processing. Because the thermal image  was relatively small for precise measurements, all calculations were conducted on the subpixel level, which increased the accuracy significantly.

Calculating weight

On the basis of the eight-year data set, three models for calculating the cattle weight were developed, which include main body characteristics of each animal of the given Simmental herd. Since, in the period of progeny test all cattle were in fact sons of eight cattle sire lines, the  herd  was  genetically  relative  unique.  Thus,  the  applied  data  served  excellently  for evaluating our group of experimental animals.As   already   reported   by   Enevoldsen   &   Kristensen   (1997)   and   Heinrichs   et   al.   (1992) measurement  of  heart-girth  (chest  circumference)  was  commonly  used  to  estimate  dairy heifer   body   weight   from   previously   derived   equations   or   tables.   However,   many experiments  showed  that  variability  of  heart  girth  measurements  depended  significantly from the weighing classes (42–590 kg). Thus measured standard deviations can varied from2.19  cm  to  2.74  cm  within  one  observer.  However,  repeatability  between  two  heart-girth measurements by an individual observer on the same animal using a blind heart girth tape was  >0.99.  additionally  correlation  coefficients  between  two  measurements  by  different observers  using  blind  measuring  tapes  on  the  same  animal  also  were  >0.99,  with  99%  of total differences due to observer and heifer, indicating very little random variation.Based on previous experiences of Enevoldsen & Kristensen (1997) and Heinrichs et al. (1992) the wither height and the hip height was determined as most significant body dimensions for estimating the cattle body weight also in our experiment with thermal camera and image analysis   system.   From   this   reason   we   choose   wither   and   height   measures   as   most convenient for estimating the cattle live weight. The following three regression models were chosen to be investigated:

Related Posts

Comments are closed.

© 2024 Mechanical Engineering - Theme by WPEnjoy · Powered by WordPress