(IJACSA) International Journal of Advanced Computer Science and Applications,

Vol. 11, No. 3, 2020

Image-based Individual Cow Recognition using

Body Patterns
Rotimi-Williams Bello1, Abdullah Zawawi Talib2, Ahmad Sufril Azlan Mohamed3
Daniel A. Olubummo4, Firstman Noah Otobo5
School of Computer Sciences, Universiti Sains Malaysia, 11800, Pulau, Pinang, Malaysia1, 2, 3
Department of Computer & Information Systems, Robert Morris University, Moon-Township, Pennsylvania, USA4
Department of Mathematical Sciences, University of Africa, Toru-Orua, Bayelsa State, Nigeria1, 5

Abstract—The existence of illumination variation, non-rigid acceptance rate of the methods among the cow breeders. The
object, occlusion, non-linear motion, and real-time classical methods of animal identification are not reliable; they
implementation requirement has made tracking in computer are prone to fraudulent activities such as swapping,
vision a challenging task. In order to recognize individual cow duplication and forgery of the so called unique identification
and to mitigate all the challenging tasks, an image processing numbers tagged on the animal’s body [7], [8], and therefore
system is proposed using the body pattern images of the cow. cannot meet the required level expected from them for the
This system accepts an input image, performs processing monitoring and identification of animal [9].
operation on the image, and output results in form of
classification under certain categories. Technically, convolutional Many automatic systems have been proposed recently for
neural network is modeled for the training and testing of each the monitoring and identification of cow however, most of
pattern image of 1000 acquired images of 10 species of cow which these devices are sensor based and sometimes become burden
will pass it through a series of convolution layers with filters, and injurious when worn on the body of the animal [10].
pooling, fully connected layers and softmax function for the There is need for automatic cow monitoring system in
pattern images classification with probabilistic values between 0 livestock farm to be developed as there is uprising in the
and 1. The performance evaluation of the proposed system for number of cow year in year out in almost every part of the
both training and testing data was carried out for each cow’s world and there is great task involved in monitoring cow
identification and 92.59% and 89.95% accuracies were achieved
manually. Lu et al. [11] proposed cow traceability system that
respectively.
was based on the iris analysis for the enhancement of cow
Keywords—Cow; body patterns; convolutional neural network; management. The image quality assessment of the captured
image; recognition iris sequences was firstly made before the clear iris was
selected. By using segmentation that was based on edge
I. INTRODUCTION detection, the inner and outer boundaries of the iris of the cow
were fitted as ellipse form. The iris image was normalized
Cows in the past were classically monitored with the sole
using geometric method and both the local and the global
aim of aiding tracking, health information, performance
features of the iris of the cow were extracted using 2D
recording, prevention against manipulation and swapping, and
complex wavelet transform. However, in an unconstrained
verification of false insurance claims. There are basically two
environment where there is greater possibility of getting poor
recognition techniques employed for the identification of the
quality image of cow’s iris, this method may not be
animal. One recognition technique leaves a permanent mark
appropriate for a reliable traceability.
on the animal for identification while the other recognition
technique leaves a temporary mark. Examples of the By using video data, there is every possibility that the
recognition technique that leaves a permanent mark are found problems attributed to the classical methods can be mitigated
in [1], [2], [3], [4], [5] with their drawbacks. The tattooing of using the visual based automatic cow recognizing system. The
ears, tagging of ears, microchips implant and branding are recognition of individual cow in the automatic cow monitoring
popular invasive identification techniques that leave a implementation process enables behavior monitoring of
permanent mark on the animal’s body with so many individual cow at long run for body condition score which
challenges such as animal infections, mild sepsis, and plays important role in the health condition of individual cow.
hemorrhaging [2], [3]. The system proposed in this paper is image-based individual
cow recognition using body patterns. The rest of the paper is
Examples of the recognition technique that leaves a
organized as follow. Presented in Section 2 are the literature
temporary mark on the body of the animal for identification
review, followed by the material and method in Section 3, the
purposes are found in the work of Barron et al. [6] with their
results and discussion are in Section 4. The conclusion is in
drawbacks. Among the classical methods of animal
Section 5.
identification are drawing, tagging, tattooing, branding,
notching, and Radio Frequency Identification (RFID). II. LITERATURE REVIEW
However, classical methods of animal identification have
notable adoption problems which have contributed to the low The conventional constructs of identifying animal can be
categorized into: (1) permanent recognition construct (PRC);

92 | P a g e
www.ijacsa.thesai.org
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 11, No. 3, 2020

(2) semi-permanent recognition construct (SRC); and binary pattern (LBP) based model for recognition of cattle
(3) temporary recognition construct (TRC); [12], [13]. The using the texture features of cattle facial representation.
tattooing of ear and body, tagging of ear, microchip implants Proposed in [28] is an approach for cattle recognition based on
and branding are referred to as PRC recognition methods [14] Speeded Up Robust Feature (SURF) descriptor. The approach
but with several limitations [15] such as: (1) lack of large was an enhancement of Petersen’s method for cattle
scale production of various metal clips and plastic tags that identification. The results of experiment was reported based
can be enough for the identification of large-scale animal; on the image datasets of 4 cattle breeds used which were
(2) easy lost of the available ear tags due to ear tearing; captured on A-5 paper with blue inked for the purpose of
(3) infections of animals such as cattle and other ruminant cattle recognition. Proposed in [20] is a matching refinement
animals due to notches [16], [17], [18], moreover, more than technique in scale invariant feature transform (SIFT)
half percentage of the animals are infected from the injury descriptor for cattle recognition using database of 160 muzzle
sustained on their ear due to the implanted plastic ear-tags, print images. By the application of matching refinement
reason being that, the ear-tags cause various health challenges technique in SIFT approach, the matching scores of the key-
such as local inflammation, thickening of the flesh, presence points of muzzle print images were computed. Nevertheless,
of pus-forming bacteria, and loss of blood through the notch the performance of the matching refinement approach and the
[17], [13]. Cattle recognition using methods such as pattern original SIFT approach were compared, and the value of EER
sketching and collar is SRC method. Furthermore, the use of equal to 0.0167 was achieved.
dye or paint and radio frequency identification (RFID) based
recognition are referred to as TRC for the recognition of Proposed in Awad et al. [29] is a framework for
animal [12], [19]. recognizing cattle using SIFT descriptor approach. The
approach is used for localizing and detecting the beads and
According to [20], the sketching pattern is applied for the ridges’ key points in the images of muzzle print for the cattle
recognition of cattle such as Holsteins and Guernsey with identification. The RANdom SAmple Consensus (RANSAC)
broken color. High drawing skills of an individual for technique incorporated in the SIFT algorithm is used for the
sketching is needed which should be comparable to standard palliation of the outliers in muzzle image for an improved,
image quality and positively affect the cattle identification robust, and reliable cattle identification. Database of 90
process. However, this method cannot be used for the muzzle images was used for the experiment where 15 muzzle
identification of solid collared breeds such as Red Poll and images were captured from each cattle of 6 in number.
Brown Swiss breed as some artificial marking methods such Proposed in Tharwat et al. [23] is an approach of cattle
as ear tagging and tattooing that are discrimination based are recognition that was based on muzzle image using the
needed. However, the method of ear tagging damages the technique of local texture descriptor. The technique works in
cattle’s ear at the long run. As iterated in Petersen’s work [21], such a way that the texture extraction algorithm that was based
muzzle print-based cattle recognition method using blue ink on local binary pattern used the local texture features
and A-5 paper [22] was the first attempt to get permanent extraction from the images of muzzle point. The involvement
recognition method for cattle. In the method, skills are of more processing time in the cattle recognition process is a
required to acquire the muzzle pattern’s print image, by major limitation of the technique.
holding firm the cattle.
Object recognition method that is based on CNN was
Lately, the research community has shifted attention to proposed in [30]. The proposed architecture which combines
advancing cattle recognition using image of muzzle print as a RGB image and its corresponding depth image for object
new paradigm for cattle identification [22], [20]. According to recognition is made up of two unconnected CNN processing
[23], print image of muzzle pattern is made up of beads and streams, which are sequentially integrated with a late fusion
ridges patterns. Muzzle dermatoglyphics such as granola, network. ImageNet [31] is employed for the training of the
ridges, and vibrissae from various breeds are not the same CNNs in which the depth image is encoded as a rendered
[16]. Similarly, proposed in Mishra et al. [24] is method of RGB image, making the information that is contained in the
cattle breeds recognition using the beads and ridges features of depth data to go round over all the three RGB channels, and
muzzle print images. Similar to the work of Mishra et al. [14] subsequently, a standard and pre-trained CNN is employed for
is Minagawa et al. [22], they proposed a cattle identification the recognition. Due to limited availability of large scale depth
method using muzzle print, the performance evaluation was datasets that are labeled, CNNs that are pre-trained on
made using filtering techniques for muzzle image analysis and ImageNet [32] are employed. Proposed in [33], is another
morphological approaches. Equal Error Rate (EER) of 0.419 object recognition method, which employs deep CNN. The
was reported by them. proposed method also uses CNN, which is pre-trained for
image classification and provides a robust, semantically
Contrary to Minagawa et al. [22] is a framework proposed meaningful feature set. The depth information is integrated by
by Barry et al. [25]. The framework is a cattle recognition
rendering objects from a canonical perspective and getting the
using muzzle print images. They reported the 241 false non- depth channel colorized according to distance from the object
match rates (FNMR) over 560 genuine acceptance rate (GAR) center.
and 5197 false matches over 12,160 impostors matching
closely with the same value of EER of 0.429, respectively. In Jingqui et al. [34] proposed the method of object
their cattle identification effort, Kim et al. [26] proposed a recognition based on image entropy; this was aimed at
method that could recognize the Japanese black cattle using identifying the behavior of cow object that is on the motion
the cattle face’s pixel intensity [26]. Proposed in [27] is a local against a complicated background. They used the minimum

93 | P a g e
www.ijacsa.thesai.org
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 11, No. 3, 2020

bounding box and contour mapping for the real-time capture

of behavioral and characteristic features displayed by the cow.
Although the approach used has time-saving advantage for
cow breeders and yields a high recognition rate of estrous and
hoof-disease not less than 80%, the time correlation of cow
behaviors was not integrated.
Andrew et al. [35] demonstrated the suitability of
computer vision pipelines that utilize deep neural architectures
to carry out automated Holstein Friesian cattle detection in
addition to individual identification in a farm set up. They
showed that it is possible to perform robustly Friesian cattle
detection and localization with an accuracy of 99.3% on the
available dataset. Although they showed the capability of their Fig. 1. Dimensional Sketch of the Individual Cattle Recognizing System.
method in the scenarios presented, they did not consider
complicated setups such as faster moving, larger herds and The cow recognition and identification system can run on
tight animal gatherings. any Windows-based personal computer. A faster computer
system is recommended for the processing of the images that
In the process of extracting features from an image, Kumar involves calculations and processing on the go. The personal
et al. [36] posited that pre-processing is important for object computer specifications for development of the cow
tracking accuracy but feature extraction and representation recognition and identification system are Intel core i5
algorithms that are based on appearance are unable to perform Processor, 8 Gigabyte of RAM, Graphics card, 2 terabytes of
the recognition of object as a result of image blurriness due to hard disk space, a CCD digital camera, and a computer
noise, low illumination and the unconstrained environment monitor for digitizing, displaying, and processing multiple
under which the images were captured. Therefore a method images. The specification for the execution of the image-
based on feature descriptor techniques is utilized for the processing and computer vision elements is OpenCV and its
unique identification of individual object. Based on the pre- library.
processing process, reliable results were obtained from the
tracking process of the object. Pre-processing which majorly B. Processing of Images
involves particle filtering and segmentation of muzzle point The filtration technique used for this work is Gaussian
images is necessary in the features extraction process. The filtering technique while multi-layer deep learning neural
primary aim of undergoing pre-processing of the muzzle network was used as a classifier for the cow identification and
images using enhancement algorithms before the feature contrast limited adaptive histogram equalization (CLAHE)
extraction and matching process take place is to ensure that was used for enhancement of the contrast between the cow’s
the muzzle images are enhanced before the analysis of the body patterns. The difference of the Gaussian filter was got by
extracted texture features and for better representation in the finding the difference between two Gaussian functions [37].
feature space.
Fig. 2 shows some image samples of cow’s body patterns
III. MATERIAL AND METHOD from the database. Fig. 3 shows the database containing
blurred image patterns of the cow’s body affected by the
A. Equipment for Experiment unconstrained environment and postures of the cow leading to
Ten (10) species of cow were examined in recognizing the poor quality of the images. Using Norouzzadeh et al. [38], we
characteristic of individual cow, each having 100 images filtered the images to get rid of the blurriness, background
making 1000 images in total. The patterns of the black and patches and low illumination.
white body of the 10 species of cow were used for the
calculation of the input parameters values for training. 400
images of body patterns (40 cows (subject) × 10 images of
each subject) were used for the training of the proposed deep
learning approach in the training phase. 600 pairs of testing
(60 cows (subject) × 10 images of each subject) of the body
patterns images in each fold were used for testing the probe
images in the testing phase. By middle of September 2018, a
test was performed in order to get the image data and the
image data was analyzed accordingly by image processing. A
charged coupled device (CCD) camera was employed for the
side image capture of each cow. In order to obtain images of
required width (235-270cm), the CCD camera was placed on a
high pole away from the experimental system centerline. The
image processing system was strategically placed in a location
through which the cow passed everyday with minimized
illumination variation for the production of noiseless and clear
images as shown in Fig. 1. Fig. 2. Images of Cow’s Body Patterns from the Database.

94 | P a g e
www.ijacsa.thesai.org
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 11, No. 3, 2020

Fig. 4. Neural Network with Convolutional Layers for Cow Recognition.

Generally, the more the convolutional steps become, the

more the complex features possibility of being recognized
becomes using the proposed network. Until the system can
dependably recognize objects, the whole process is repeated in
successive layers. Each layer’s neurons of the CNN as seen in
Fig. 4 are in 3D arrangement, making a transformation of a 3D
input to a 3D output. For instance, for an input image, the first
Fig. 3. Blurred and Poor Illumination Images of Cow’s Body Patterns. layer which is the input layer takes the images as 3D inputs,
with height, width and color channels as the dimensions of the
In order to enhance the identification process and remove image. The first convolutional layer’s neurons connect to the
the patches and the noises from the captured images that were input images’ regions and change them into a 3D output. Each
collected, various image pre-processing techniques were layer hidden units learn nonlinear combinations of the original
applied. Low illumination and poor image quality are the most inputs which becomes the inputs for the layer that follows. By
two fundamental challenges confronting image acquisition this, at the end of the network, the learned features become the
especially images of cow’s body patterns. The images inputs to the classifier.
captured in an unconstrained environment were converted to
grayscale images in order to reduce the patches and the noises The intensity values of the gray scale of the background
captured with them. The converted images were improved images are more than 100 but less than 150 in respect to the
upon by contrast limited adaptive histogram equalization colors of the cows’ body surface. 128 was fixed as the pixel’s
based image processing technique. threshold value for the whole image. While 1 is assigned as
the binary values for the intensities that are greater than the
The pre-processing technique accepts the images in their threshold value of 128, 0 is assigned as the binary values for
color form and converts them to grayscale before fetching the intensities that are less than the threshold value of 128.
them into the filter for removal of the patches and the noises Because the threshold value could be changed with
contained in the captured images. The feature extraction illumination and noise, it becomes very important. Individual
involves the convolution and pooling operations on the images cow’s image is captured for the identification of their
until the images get to the classifier for classification analysis individual characteristics. Individual cow identification using
for the generation of the desired output (Fig. 4). The removal unique body patterns is made possible because of the invariant
of the noises was carried out using an auto-encoding of the body patterns to growth. This uniqueness enables the
technique. Stacked denoising auto-encoder (SDAE) technique patterns to be used as the input layer values in the neural
initializes deep network and it is applicable for encoding and network algorithm.
decoding the texture features of the image patterns that were
extracted and encoding the extracted sets of features for IV. RESULTS AND DISCUSSION
optimum representation of the feature [39]. Having tried out the effectuality of the proposed approach
Technically, convolutional neural network (CNN) is using images of cow’s body patterns for the recognition and
modeled for the training and testing of each input image which identification of the cow, the comparison with other
will pass it through a series of convolution layers with filters, recognition algorithms is attained in order to evaluate the
pooling, fully connected layers and softmax function for the accuracy of the identification in proliferation settings.
image classification with probabilistic values between 0 and 1. Evaluating the performance of the experimental results, the
As shown in Fig. 4, the first layer to extract features from the database of the cow’s body images is segmented as follows:
input image is convolution. Convolution primarily conserves (1) the training phase; and (2) the testing phase. 400 body
the relationship between pixels by learning the image features images of different subjects (40 cows (subject) × 10 images of
using squares of input data. It involves a mathematical each subject) were used for the training of the proposed
operation with two inputs such as image matrix and a filter. approach in the training phase. 600 pairs of testing (60 cows
When there are too large images, pooling layers primarily (subject) × 10 images of each subject) of the body patterns
reduce the number of parameters (dimensionality size). In the images in each fold were used for testing the probe images in
proposed CNN as seen in Fig. 4, the operation of the pooling the testing phase.
is applied individually to each feature map.

95 | P a g e
www.ijacsa.thesai.org
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 11, No. 3, 2020

The second part is “pre-training.” For a given training set

of image data = {𝑥1 , 𝑥2 , … , 𝑥𝑁 } , the learning system obtains a
model through learning (or training) to describe the mapping
relationship between input and output variables. This work
assumed that RBM model has this descriptive ability,
therefore it consists of several layers, through which the input
is the image expression data vector while the output is the
abstracted higher-level feature vector. Each layer of RBM
networks undergoes individually unsupervised training to
ensure that feature information is preserved to the uttermost as
Fig. 5. RBM-based DBN Model.
feature vectors are mapped to different feature spaces. To
construct the joint distribution model of visible layer and the
For the training of the proposed deep learning framework
hidden layer through energy function, the joint probability
using deep belief network (DBN) as shown in Fig. 5, there is a
maximum likelihood of training sample under model
need for a monolithic database amount. Although the number
of cow’s body images in the database is encouraging, it is not parameter 𝜃̂ is calculated by
satisfactory enough to train the stacked denoising auto- 1
𝑃(𝑣|𝜃̂ ) = ∑ℎ 𝑒 −𝐸(𝑣,ℎ|𝜃) (3)
encoder with a database of 1000 worth of cow’s body patterns 𝑍(𝜃)
images. Therefore a transfer learning approach is needed to The third part is “fine tuning.” Fine-tuning is a common
fine-tune the weight between the input and the hidden layer strategy in deep learning to carry out supervised learning
and determine the pre-training of the proposed deep learning through tagged sample training set 𝑇=
approach. {(𝑥1′ , 𝑦1 ), (𝑥2′ , 𝑦2 ), … , (𝑥𝑁′ , 𝑦𝑁 ) } . After that, the top feature
The basic mathematical steps that are involved in using the vectors corresponding to sample output by the multi RBM
deep belief network for this work are as follows: network are formed based on the training set of statistical
classification structure. This part is a BP network; it takes a
Problems setting: Given a training set of pre-processed specific dimension feature vector to a softmax function. In
body pattern image data order to get the best connection weights, this work considered
𝑇 = {(𝑥1 𝑦1 ), (𝑥2 𝑦2 ), … , (𝑥𝑁 𝑦𝑁 ) } (1) solving the following optimization problem using particle
swarm optimization (PSO), so that the loss of function in the
of which (𝑥1 𝑦1 ), 𝑖 = 1, 2, … , 𝑁 denotes the sample point, training set is minimized.
𝑥𝑖 ∈ 𝑋 ⊆ 𝑅𝑛 is the sample image data while 𝑦𝑖 ∈ 𝑌 is the 1
corresponding tag of the label; the recognition procedure of 𝜃̂ = argmax 𝑙(𝜃, 𝑥𝑛 , 𝑡𝑛 ) (4)
𝜃 𝑁
proposed system is to input data set 𝑇 to the network, find the
mapping between input 𝑋 and output 𝑌 to form a generative The last part is the “class identification.” Tested sample
joint probability distribution model formula 𝑃(𝑋, 𝑌), generate 𝑥𝑁+1 as network input is subjected to feature learning and
the output 𝑦𝑁+1 by abstraction through a network model training to produce a
corresponding output 𝑦𝑁+1 by
𝑦𝑁+1 = arg 𝑚𝑎𝑥 𝑃̂ (𝑦𝑁+1 | 𝑥𝑁+1 ) (2)
𝑦𝑁+1 𝑦𝑁+1 = 𝑎𝑟𝑔𝑚𝑎𝑥𝑃̂ (𝑦𝑁+1 |𝑥𝑁+1 ) (5)
for a given prediction sample 𝑥𝑁+1 , and judge the image and thus achieve classification.
classification of 𝑥𝑁+1 according to 𝑦𝑁+1 . The system contains
the following parts as shown in Fig. 4: For the evaluation of performance, the local feature
descriptor technique was used to extract and encode texture
The proposed cow’s body patterns image identification features of the cow’s body patterns. As earlier mentioned, the
using deep belief network and a back propagation (BP) normalization and the descriptor process help in mitigating the
network layer, wherein the multi-layer RBM is used to input external factors such as low illumination, poor image quality,
data feature learning to achieve abstraction and dimensionality and background patches affecting the captured images. In
reduction of data through the hierarchical feature learning is as performing the tasks involved in this process, cells are
shown in Fig. 5; BP network layer is a categorical network, converted to blocks. During this process, blocks are
and it is to categorize the abstracted higher-level features overlapped and cells shared among the blocks and normalized
through softmax function. The softmax function, also known separately. Scale-invariant Feature Transform (SIFT) and
as softargmax or normalized exponential function, is a Rectangular-Histogram of Oriented Gradients (R-HOG) are
function that takes as input a vector of K real numbers, and similar though, they don’t align to their dominant orientation
normalizes it into a probability distribution consisting of K (Fig. 8(b)). SDAE produced the best experimental results
probabilities proportional to the exponentials of the input (Fig. 6 and Fig. 7) when compared to other approaches used in
numbers. this work making it fit the most for the denoising. 400 body
The first part of the processes as shown in Fig. 4 is images equivalent to (40 cows (subject) × 10 images of each
“preprocessed cow’s body patterns images” which are subject) were chosen randomly for system training and 600
introduced as inputs to the proposed networks for features body images equivalent to (60 cows (subject) × 10 images of
extraction and classification. each subject) were used for the testing. The experimental
results are reported and analyzed as found in Table I.

96 | P a g e
www.ijacsa.thesai.org
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 11, No. 3, 2020

The significant reason for binary patterns (Fig. 8(a)) is to

sum up the local structure in a block through comparison of
each pixel with its neighborhood [40]. Each pixel coded with a
sequence of bits is colligated with the connection between the
pixel and one of its neighbors. The center pixel’s intensity is
denoted with 1 if it is greater than or equal to its neighbor, and
denoted with 0 if otherwise with a binary number at the end
created for each pixel.
V. CONCLUSION
Image-based individual cow recognition using body
patterns was the main work carried out in this research. Cows
usually are identified to prevent them from being stolen or
protect them from danger, and in many agricultural settings,
Fig. 6. Illustration of 17% Corrupted Images of Cow’s Body Patterns using their behaviors are usually studied using imaging technology
SDAE. to enable timely monitoring and identification of health
challenges. CNN and some other popular image recognizing
techniques such as DBN, SDAE, CLAHE, Gaussian filter,
binary pattern, were employed in this work for the cow
recognition. The various techniques were discussed in details
as they are applicable to the cow recognition process. Datasets
of 1000 images of cow’s body patterns from 10 species of cow
were created for this work where 400 images were employed
for the training and 600 images were used for the testing. The
advantage of using this datasets is the various species of cow
whose images are contained in the database used for the
recognition. Gaussian filtering technique was used as the
filtration technique; this was supported by SDAE for
denoising while multi-layer convolutional neural network was
used as a classifier in comparison to deep belief network
which needs a monolithic database amount for the cow
Fig. 7. Illustration of 5.7% Corrupted Images of Cow’s Body Patterns. identification, and contrast limited adaptive histogram
equalization (CLAHE) was used for enhancement of the
contrast between the cow’s body patterns. The performance
evaluation of the proposed system for both training and testing
data was carried out for each cow’s identification and 92.59%
and 89.95% accuracies were achieved respectively. Although
this work has been able to apply modern image-based
identification method for the recognition of cow using body
patterns, recognition of occluded and non-linear moving
object such as cow in real-time using the object’s multi-
features is a work that we consider worthy of investigating in
the future.
(a) (b)
Fig. 8. (a) Binary Pattern of Cow’s Body; (b) Histogram. ACKNOWLEDGMENT
This work was supported in part by the Institute of
As it is shown in Table I, the evaluation of the system Postgraduate Studies, Universiti Sains Malaysia, 11800 USM,
performance was carried out on the cropping, the training Penang, MALAYSIA.
data, and the testing data for the overall achievement of the
research objective. The average cropping accuracy of the REFERENCES
captured video data is 79.45%, and the identification accuracy [1] Z. Miao, K. M. Gaynor, J. Wang, Z. Liu, O. Muellerklein, M. S.
Norouzzadeh., and W. M. Getz, “Insights and approaches using deep
of the training data is 92.59% with the testing data having the learning to classify wildlife”, Scientific reports, vol. 9, no. 1, pp. 1-9,
identification accuracy of 89.95%. 2019.
[2] S. Kumar, and S. K. Singh, “Cattle Recognition: A New Frontier in
TABLE. I. RECOGNITION ACCURACY (%) Visual Animal Biometrics Research”, Proceedings of the National
Academy of Sciences, India Section A: Physical Sciences 1-20, 2019.
Cropping 79.45%
[3] T. T. Zin, C. N. Phyo, P. Tin, H. Hama, and I. Kobayashi, “Image
Training data 92.59% technology based cow identification system using deep learning”,
Proceedings of the International MultiConference of Engineers and
Testing data 89.95% Computer Scientists 1, 2018.

97 | P a g e
www.ijacsa.thesai.org
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 11, No. 3, 2020

[4] R. W. Bello, and O. M Moradeyo, “Monitoring Cattle Grazing Behavior of Advanced Machine Learning Technologies and Applications, pp.
and Intrusion Using Global Positioning System and Virtual Fencing,”, 236–247, 2014.
Asian Journal of Mathematical Sciences, vol. 3, issue 4, pp. 4-14, 2019. [24] S. Mishra, O. S. Tomer, and E. Kalm, “Muzzle dermatoglyphics: a new
[5] R. W. Bello, “An overview of animal behavioral adaptive frightening method to identify bovines”, Asian Livestock, pp. 91–96, 1995.
system,” International Journal of Mathematics and Physical Sciences [25] B. Barry, U. Gonzales-Barron, K. McDonnell, F. Butler, and S. Ward,
Research, vol. 6, issue 1, pp. 126-133, 2018. “Using muzzle pattern recognition as a biometric approach for cattle
[6] U. G. Barron, F. Butler, K. McDonnell, and S. Ward, “The end of the identification, Trans ASABE, vol. 50, no. 3,pp. 1073–1080, 2007.
identity crisis? Advances in biometric markers for animal [26] H. T. Kim, Y. Ikeda, and H. L. Choi, “The identification of Japanese
identification”, Irish Veterinary J., vol. 62, no. 3, pp. 204–208, 2009. black cattle by their faces”, Asian Australasian Journal of Animal
[7] M. Shen, L. Liu, L. Yan, M. Lu, W. Yao, and X. Yang, “Review of Sciences, vol. 18, no. 6, pp. 868–872, 2005.
monitoring technology for animal individual in animal husbandry”, [27] C. Cai, and J. Li, “Cattle face recognition using local binary pattern
Nongye Jixie Xuebao = Transactions of the Chinese Society for descriptor”, Proceedings of 2013 Asia-Pacific Signal and Information
Agricultural Machinery, vol. 45, no. 10, pp. 245-251, 2014. Processing Association Annual Summit and Conference (APSIPA), 1–4,
[8] R. W. Bello, and S. Abubakar, “Development of a Software Package for 2013.
Cattle Identification in Nigeria,” Journal of Applied Sciences and [28] A. Noviyanto, and A. M. Arymurthy, “Automatic cattle identification
Environmental Management, vol. 23, no. 10, pp. 1825-1828, 2019. DOI: based on muzzle photo using speed-up robust features approach”,
https://dx.doi.org/10.4314/jasem.v23i10.9. Proceedings of the 3rd European Conference of Computer Science,
[9] R. W. Bello, A. Z. H. Talib, and A. S. A. B. Mohamed, “A Framework ECCS, vol. 110, pp. 114, 2012.
for Real-time Cattle Monitoring using Multimedia Networks”, [29] A. I. Awad, H. M. Zawbaa, H. A. Mahmoud, E. H. H. A. Nabi, R. H.
International Journal of Recent Technology and Engineering, vol. 8, Fayed, and A. E. Hassanien, “A robust cattle identification scheme using
issue 5, 2020. muzzle print images”, Proceedings of 2013 Federated Conference on
[10] D. Grooms, “Radio Frequency Identification (RFID) Technology for Computer Science and Information Systems (FedCSIS), pp. 529–534,
Cattle”, Extension Bulletin E-2970, Michigan State University, Jan. 2013.
2007. [30] A. Eitel, J. T. Springenberg, L. Spinello, M. Riedmiller, and W.
[11] Y. Lu, X. He, Y. Wen, and P. S. Wang, “A new cow identification Burgard, “Multimodal deep learning for robust RGB-D object
system based on iris analysis and recognition”, In International Journal recognition”, In 2015 IEEE/RSJ International Conference on Intelligent
of Biometrics, vol.6, no.1, pp.18-32, 2014. Robots and Systems (IROS), pp. 681-687, 2015.
[12] S. Kumar, S. Tiwari, and S. K. Singh, “Face recognition of cattle: can it [31] O. Russakovsky, J. Deng, H. Su, J. Krause, S. Satheesh, S. Ma, ..., and
be done?”, Proceedings of the National Academy of Sciences, India A. C. Berg, “Imagenet large scale visual recognition challenge”,
Section A: Physical Sciences, vol. 86, no. 2, pp. 137–148, 2016. International journal of computer vision, vol. 115, no. 3, pp. 211-252,
[13] S. Kumar, S. Tiwari, and S. K. Singh, “Face recognition for cattle. Third 2015.
IEEE International Conference on Image Information Processing [32] A. Krizhevsky, I. Sutskever, and G. E. Hinton, “Imagenet classification
(ICIIP), pp. 65–72, 2015. with deep convolutional neural networks”, In Advances in neural
[14] J. Marchant, “Secure animal identification and source verification. JM information processing systems, pp. 1097-1105, 2012.
Communications, UK, pp. 1-28, 2002. [33] M. Schwarz, H. Schulz, and S. Behnke, “RGB-D object recognition and
[15] A. Allen, B. Golden, M. Taylor, D. Patterson, D. Henriksen, and R. pose estimation based on pre-trained convolutional neural network
Skuce, “Evaluation of retinal imaging technology for the biometric features”, In 2015 IEEE international conference on robotics and
identification of bovine animals in northern Ireland”, Livest Sci., vol. automation (ICRA), pp. 1329-1335, 2015.
116, no. 1, pp. 42–52, 2008. [34] G. Jingqiu, W. Zhihai, G. Ronghua, and W. Huarui, “Cow behavior
[16] A. Baranov, R. Graml, F. Pirchner, and D. Schmid, “Breed differences recognition based on image analysis and activities”, International
and intra-breed genetic variability of dermatoglyphic pattern of cattle”, J Journal of Agricultural and Biological Engineering, vol. 10, no. 3, pp.
Anim Breed Genet., vol. 110, no. 1–6, pp. 385–392, 1993. 165-174, 2017.
[17] A. Johnston, and D. Edwards, “Welfare implications of identification of [35] W. Andrew, C. Greatwood, and T. Burghardt, “Visual localisation and
cattle by ear tags”, The Veterinary Record, vol. 138, no. 25, pp. 612– individual identification of Holstein friesian cattle via deep learning”,
614,1996. Proceedings of the IEEE International Conference on Computer Vision,
pp. 2850-2859, 2017.
[18] D. D. Wardrope, “Problems with the use of ear tags in cattle”, The
Veterinary record, vol. 137, no. 26, pp. 675-675, 1995. [36] S. Kumar, A. Pandey, K. S. R. Satwik, S. Kumar, S. K. Singh, A. K.
Singh, and A. Mohan, “Deep learning framework for recognition of
[19] Z. Wang, Z. Fu, W. Chen, and J. Hu, “A RFID-based traceability system
cattle using muzzle point image pattern”, Measurement, vol. 116, pp. 1-
for cattle breeding in china”, Proceedings of 2010 International 17, 2018.
Conference on Computer Application and System Modeling (ICCASM
2010), vol. 2, pp. V2–567, 2010. [37] K. P. Risha, K. A. Chempak, and C. S. Sindhu, “Difference of Gaussian
on Frame Differenced Image,” International Journal of Innovative
[20] A. Noviyanto, and A. M. Arymurthy, “Beef cattle identification based Research in Electrical, Electronics, Instrumentation and Control
on muzzle pattern using a matching refinement technique in the sift Engineering, vol. 3, special issue 1, pp. 92-95, 2016.
method”, Comput Electron Agric., vo. 99, pp. 77–84, 2013.
[38] M. S. Norouzzadeh, A. Nguyen, M. Kosmala, A. Swanson, M. S. Palmer,
[21] W. E. Petersen, “The identification of the bovine by means of nose- C. Packer, and J. Clune, “Automatically identifying, counting, and
prints”, Journal of dairy science, vol. 5, no. 3, pp. 249-258, 1922.
describing wild animals in camera-trap images with deep learning,”, In
[22] H. Minagawa, T. Fujimura, M. Ichiyanagi, K. Tanaka, and M. Proceedings of the National Academy of Sciences, vol. 115, no. 25, pp.
Fangquan, “Identification of beef cattle by analyzing images of their E5716-E5725, 2018.
muzzle patterns lifted on paper”, Proceedings of the 3rd Asian [39] P. Vincent, H. Larochelle, I. Lajoie, Y. Bengio, and P. A. Manzagol,
Conference for Information Technology in Asian agricultural “Stacked denoising auto-encoders: learning useful representations in a
information technology & management. Publications of the Japanese deep network with a local denoising criterion”, Journal of Machine
Society of Agricultural Informatics, vol. 8, pp. 596–600, 2002. Learning Research, vol. 11, pp. 3371–3408, 2010.
[23] A. Tharwat, T. Gaber, and A. E. Hassanien, “Cattle identification based [40] P. Kamencay, T. Trnovszky, M. Benco, R. Hudec, P. Sykora, and A.
on muzzle images using gabor features and svm classifier”, Proceedings Satnik, “Accurate wild animal recognition using PCA, LDA and
LBPH”, In 2016 ELEKTRO, pp. 62-67, 2016.

98 | P a g e
www.ijacsa.thesai.org