PREDICTION OF THERMAL STRESS IN INDIGENOUS AND CROSSBRED CATTLE USING INFRA-RED THERMOGRAPHY AND EMERGING MACHINE LEARNING TECHNIQUES
Loading...
Date
2023
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
ICAR-NDRI, KARNAL
Abstract
The present study was conducted at the Livestock Research Centre (LRC) of ICAR-NDRI, Karnal to study the
thermogram and physiological responses of Karan Fries and Tharparkar cattle during different seasons and to
develop and validate machine learning models for intelligent prediction of thermal stress (rectal temperature) in
the Karan Fries and Tharparkar cattle. Twenty each of Tharparkar and Karan Fries cattle were selected and
managed under standard managemental practices followed at LRC and fed as per the ICAR standard (2013).
The research work was carried out during the extremes of winter (December to January), spring (February to
March) and summer (May to June) season. The physiological parameters and infrared thermography of
experimental animals was recorded twice weekly and the blood samples were collected fortnightly from six
animals of each breed for the haematological and biochemical parameters. The environmental parameters viz
dry and wet bulb temperature, black globe temperature, wind velocity, relative humidity and temperature
humidity index and black globe humidity index was calculated from the recorded environmental parameters.
Infrared thermograms were analyzed for the skin temperature at different anatomical regions i.e. muzzle,
forehead, eye, ear base, ear tip, bridge of nose, neck, brisket, hump, dorsal, ventral, fetlock, knee, elbow,
shoulder and lateral body. The environmental, physiological and infrared thermography readings were used for
the development of the predictive models by different machine learning algorithms such as Bayesian
Regularized Neural Network (BRNN), K-Nearest Neighbors (KNN), Support Vector Machine (SVM) and
Multiple Linear Regression (MLR).There was significant (P>0.05) difference in mean values of physiological
parameters viz RT and RR being higher during summer followed by spring and winter season within the breeds.
RT was significantly (P<0.05) higher in Karan Fries than Tharparkar cattle during different seasons. Whereas,
RR was significantly (P<0.05) higher in Karan Fries than Tharparkar cattle during spring and summer season.
Within the breed a significant (P<0.05) difference was observed in mean values of thermal imaging temperature
at different anatomical sites viz. muzzle, forehead, eye, base of ear, tip of ear, bridge of nose, neck, brisket,
hump, dorsal, ventral, fetlock, knee, elbow, shoulder and lateral body region during different season. These
values of skin temperature were highest during summer followed by spring and winter season. Among the both
breeds of cattle, thermal imaging temperature at different anatomical sites viz. brisket, ventral and shoulder
regions was significantly (P<0.05) higher in KF than TP during winter season. During the spring season a
significant (P<0.05) difference was observed in the mean values of the muzzle and neck region temperature
among the breeds. Significantly (P<0.05) lower mean values of red blood cell (million/mm3) and haemoglobin
content (gm %) was found in both breeds during summer season compared to the other seasons. Within breed
and among the breeds, a significant (P<0.05) difference was observed in mean values of plasma cortisol levels
(ng/ml) during different seasons. Within the breed of cattle, a significant (P<0.05) difference was found in mean
values of catalase activity (μmole of H2O2 consumed/min/mg Hb) during different seasons and values also
differed significantly (P<0.05) among Karan Fries and Tharparkar cattle during spring and summer season.
There was a (P<0.05) significant difference in mean values of superoxide dismutase activity (u/mg/Hb) during
different seasons within the breeds. The levels of both the antioxidant enzymes were higher in KF than TP
during all the seasons. The mean values of malondialdehyde (MDA) (nmol/mg) differed significant (P<0.05)
within the breeds during different seasons and also differed significantly (P<0.05) among the both breeds of
cattle during winter seasons. The prediction models developed by using Bayesian Regularized Neural Network
(BRNN), K-Nearest Neighbors (KNN), Support Vector Machine (SVM), and Multiple Linear Regression
(MLR) algorithms. The animal parameters like breed, infrared temperature of eyes, the respiration rate and the
environmental stress indicator temperature humidity index (THI) used as input variables for the prediction of
the rectal temperature/ heat stress. The four input parameters were selected based on the correlation among the
parameters to the rectal temperature. The final predictive model by KNN algorithm exhibited the mean absolute
error of 0.1627, root mean square error of 0.2059 and the R2 value of 0.7788 (R value as 0.8825), which is
comparatively higher than the other models developed in this study. The BRNN model got the R2 value of 0.75
and the SVM and MLR models got R2 of 0.74. both the classical and machine learning models works well, but
the machine learning model (KNN , BRNN, SVM) works slightly better than the classical Multiple Linear
Regression models. Based on the results obtained, it can be concluded that the physiological parameters
(thermal imaging temperature of eye and respiration rate) and THI are the most accurate parameters for the
prediction of stress levels (Rectal temperature) of cattle.