Dogo Rangsang Research Journal UGC Care Group I Journal

ISSN : 2347-7180 Vol-08 Issue-14 No. 03: March 2021

Animal detection using deep learning

1. A. Vijaykumar,assistant professor, CSE,Sri Indu Institute of Engineering&Technology(SIIET),
Sheriguda, Ibrahimpatnam, Hydarabad
2.J.Pujitha,assistant professor,CSE,SIIET,Sheriguda,Ibrahimpatnam,Hydarabad
3.Potham Rachana,Student,CSE,SIIET,Sheriguda,Ibrahimpatnam,Hydarabad
4.Padigala Shirisha,Student,CSE,SIIET,Sheriguda,Ibrahimpatnam,Hydarabad
5.Suram Spoorthy,Student,CSE,SIIET,Sheriguda,Ibrahimpatnam,Hydarabad
6.P Nikhil Raj,Student,CSE,SIIET,Sheriguda,Ibrahimpatnam,Hydarabad

ABSTRACT:
Departments of transportation all across the world are concerned with ever increasing
number of Animal-Vehicle Collisions as they cause thousands of human and animal causalities along
with billions of economic losses every year. This is one of the few areas of transportation that safety
is not improving. As more roads are being built, the areas that animal inhabit is shrinking and thus
causing more crashes between vehicles and animals. The human fatalities and injuries, animal
fatalities and injuries, and material costs of these crashes emphasize the need for a solution to this
problem. Through this paper, new research directions and combination of technologies which are
suitable for covering the research gaps are presented..
transforming into a interpretable and
INTRODUCTION comprehensible format for the future use. It is
mainly used for data pre-processing, data
Having an updated knowledge about different classification and categorization. Image
animals will impact our study in managing Classification
species in the ecosystem. Identifying animals analyses, identifies and discover several
and their features manually remains a manual properties of an image and organizes the image
and expensive, time-consuming task. Thus, we data into various categories using several
propose that such identification and algorithms. It mainly employs two characteristic
classification can be done with utmost accuracy phases of processing: training and testing. In the
using deep learning neural network techniques. training phase, a unique 1 identity of each
The main purpose of using deep learning neural category is obtained. In the testing phase, these
network techniques is that a unique identities are used to classify the image
neural network framework can automatically data into categories. A neural network is a series
learn from the training images by extracting of algorithms that analyses a set of data and
features from the images and predicts the test recognizes the underlying relationships within
images with efficient accuracies. This intends to the data. They are the workhorses of deep
reduce the manual effort and cost and to learning. Deep Learning is an artificial
maintain and conserve the wildlife ecosystem. intelligence technology that is used for
We processing larger sets of data and mainly used
will demonstrate that such detection of animal in decision making and image classification and
can be done by deep convolution neural pattern creation. It is a subset of Machine
network frameworks with high accuracies. Learning that comprises of networks capable of
Data Mining is the process of identifying and training by unsupervised learning from the
discovering trends and patterns in the large sets unstructured data. The reason why we choose to
of data which is a combination of multiple use deep learning is that it is one of that only
fields that include machine learning, statistics methods that can overcome the challenges of
and feature extraction by learning several different
database systems. It focuses on extracting features itself from the large sets of data
information from the large sets of data and
FIG 1.1 Labeled Images
Page | 841 FIG 1.1 Labeled Copyright
Images @ 2021 Authors
FIG 1.1 Labeled Images
Dogo Rangsang Research Journal UGC Care Group I Journal
ISSN : 2347-7180 Vol-08 Issue-14 No. 03: March 2021
without much effort from the programmer. Principal Component Analysis (PCA), Linear
Generally, in a recognition system, when an Discriminant Analysis (LDA), Local Binary
input image is provided, features are extracted Patterns Histograms (LBPH) and Support
from the image. These are used by the network Vector Machine (SVM) are tested and
to train itself from the training data and compared
organizes the data into with proposed convolutional neural network
classes. The gained knowledge from the training (CNN) for the recognition rate of the input
is used in predicting the test data based on the animal 3 images. Their experiments, the overall
features and classifies them accordingly. A recognition accuracy of PCA, LDA, LBPH and
recognition system can be employed with SVM is demonstrated. Next, the time execution
identification for animal recognition process is evaluated. The
and verification. Identification is where the all experimental results on created animal
given image is compared with all the other database were conducted. This created animal
images and produces a ranked list of matches database consist of 500 different subjects (5
while the Verification is where the given image classes/ 100 images for each class). The
is compared, and the identity of the animal is experimental result shows that the PCA features
confirmed or denied. provide better results as LDA and LBPH for
Literature Survey large training set. On the other hand, LBPH is
This chapter gives the overview of literature better than PCA and LDA for small training
survey. This chapter represents some of the data set. For proposed CNN we have obtained a
relevant work done by the researchers. Many recognition accuracy of 98%. The
existing techniques have been studied by the Object detection with discriminatively trained
researchers on Animal Classification problem, part-based models. P. F. Felzenszwalb, R. B.
few of them are discussed below. Girshick, D. McAllester and D. Ramanan,
Animal Detection Using Deep Learning "Object Detection with Discriminatively
Algorithm Prakash, Banupriya, First Trained Part-Based Models," in IEEE
International Conference on Intelligent Digital Transactions on Pattern Analysis and Machine
Transformation ICIDT – 2019 Intelligence, vol. 32, no. 9, pp. 1627-1645, Sept.
Efficient and reliable monitoring of wild 2010. We describe an object detection system
animals in their natural habitat is essential. This based on mixtures of multiscale deformable part
project develops an algorithm to detect the models. Our system is able to represent highly
animals in wild life. Since there are large variable object classes and achieves state- ofthe-
number art results in the PASCAL object detection
of different animals manually identifying them challenges. While deformable part models
can be a difficult task. This algorithm classifies have become quite popular, their value had not
animals based on their images so we can been demonstrated on difficult benchmarks
monitor them more efficiently. Animal such as the PASCAL data sets. Our system
detection and classification can help to prevent relies on new methods for discriminative
animal-vehicle accidents, trace animals and training
prevent theft. This can be achieved by applying with partially labeled data. We combine a
effective deep learning algorithms . Animal margin-sensitive approach for data-mining hard
recognition system based on convolutional negative examples with a formalism we call
neural network Trnovszky, Tibor & Kamencay, latent SVM. A latent SVM is a reformulation of
Patrik & Orješek, Richard & Benco, Miroslav & MI--SVM in terms of latent variables. A latent
Sykora, Peter. (2017). SVM is semiconvex, and the training problem
Animal Recognition System Based on becomes convex once latent information is
Convolutional Neural Network. Advances in specified for the positive examples. This leads
Electrical and Electronic Engineering. The to an iterative training algorithm that alternates
author studied the performances of well-known between fixing latent values for positive
image recognition methods such as

Page | 842 Copyright @ 2021 Authors

Dogo Rangsang Research Journal UGC Care Group I Journal
ISSN : 2347-7180 Vol-08 Issue-14 No. 03: March 2021
examples and optimizing the latent SVM animals before they enter the road and warn the
objective function. driver
4 Accurate wild animal recognition using PCA through audio and visual signals. This also
P. Kamencay, T. Trnovszky, M. Benco, R. helps in saving crops in farm from animals. In
Hudec, P. Sykora and A. Satnik, "Accurate wild this project there is survey of different object
animal recognition using PCA, LDA and detection techniques and for object
LBPH," 2016 ELEKTRO, Strbske Pleso, 2016, identification as animal techniques such as
pp. 62-67. object matching, edge-based matching, skeleton
The author studied performances of image extraction. After survey the most appropriate
recognition methods such as Principal method is selected for animal detection and
Component efficiency is
Analysis (PCA), Linear Discriminant Analysis measured. Proposed system has low false
(LDA) and Local Binary Patterns Histograms positive rate and false negative rate. Existing
(LBPH) are tested and compared for the image study on background ubtraction was done with
recognition of the input animal images. The the assumption of static background. Many
main idea of this paper is to present an models for background segmentation use a
independent, comparative study and some of the base frame that consists of only background
benefits and drawbacks of these most popular explicitly. The challenge of dynamic
image recognition methods. Two sets of background is handled by considering the
experiments are conducted for relative local segmentation
performance evaluations. In the first part of our sensitivity using feedback loops at pixel level of
experiments, the recognition accuracy of PCA, images in a video.
LDA and LBPH is demonstrated. The overall
time execution for animal recognition process is PROPOSED SYSTEM
evaluated in the second set of our ●The aim of the proposed algorithm is to
experiments. We conduct tests on created implement animal classification by using two
animal database. The all algorithms have been deep learning neural network frameworks –
Architecture CNN and Faster RCNN and highlight the
importance
of deep learning technology in classification
problems. The proposed algorithm is consisting
of five main steps.
•
Image collection
• Data set splitting
• Train CNN and Faster RCNN algorithm
• Classify using CNN and Faster RCNN
algorithm
2.5 ADVANTAGES
The classification done using deep learning
models .
The accuracy of classification is high.
SYSTEM ARCHITECTURE
The system comprises of two stages, training
and testing. In the training stage, a set of images
are provided as visual examples. In the testing
EXISTING SYSTEM stage, a newly captured image i.e. the test image
●Animal detection is useful in prevention of is given as input to the classifier. With the help
animal- vehicle accidents and will increase of the knowledge gained from training, the test
human and wildlife safety, it will detect large

Page | 843 Copyright @ 2021 Authors

Dogo Rangsang Research Journal UGC Care Group I Journal
ISSN : 2347-7180 Vol-08 Issue-14 No. 03: March 2021
image is accordingly classified into the most at a time. It is really important to set this because if you
favorable class. have a lower memory problem in
your computer you cant train the model by throwing all
the data at once.
METHODOLOGY Step 5 — Test the model
Step 1 — Install libraries Step 6 — Predicting on single images
For this project you will need following python Department of Information Technology, MRECW
libraries installed on your machine- Page 10
Step 2 — Prepare Dataset
I downloaded photos each for cat, dog, and
panda categories. You can download photos in
batch using some great chrome extensions. In
my case I used Download All Images extension IMPLEMENTATION
in Chrome.

We go into each directory and

go through all the files inside it. Then we read the images Fig.copy the link and paste it in the browser to
using open the app
cv2.imread. After that using Image object from PIL we
convert the image into array. Since RESULTS:
while training the convolutional neural network it is
required that you have images of same
size we resize the images to width and height of 50px.
Then we convert it to numpy array just by passing the
image array in the function np.array() and we append
the numpy array in our
data array. Also add corresponding label to the image. Eg
for cat label is 0, for dog label is 1
and so on. Labels are required as the training is done in
supervised manner.
Step 3 — Making Kera’s model
It is a very easy process to make your deep learning
model. Thanks to the great Keras library.
All you have to do is think about hyper parameters like Path of the folder
Filter size, number of filters, which
type of padding to use, which activation functions to use
etc.
Step 4 — Train the model
Your all work is over. Now just train the model and wait
patiently.“epochs” is the number of
steps you want to train the model. Batch size is the size
of dataset that is thrown to the model

Page | 844 Copyright @ 2021 Authors

Dogo Rangsang Research Journal UGC Care Group I Journal
ISSN : 2347-7180 Vol-08 Issue-14 No. 03: March 2021

Conclusion:
In this study, we expressed our idea to
automatically classify animal images using the
latest
deep learning neural network frameworks in
order to reduce the human effort and cost. We
tested the ability of the neural networks to
LOSS/ACCURACY GRAPH IN classify images with high accuracies. The
(SmallVGGNet) results
highlight the importance of the deep learning
neural network technology in image
classification. Hence, these network
frameworks can be used to reduce human effort
and cost
and extract wildlife information to conserve and
maintain the wildlife system.
Department of Information Technology,
MRECW
Page 44

REFERENCES
[1] Using Deep Learning. In: 2018 International
Journal of Innovative Research in
Computer andCommunication Engineering
(IJIRCCE). Vol. 6, Issue 3, March (2018).
ISSN(Online): 2320- 9801 ISSN (Print): 2320-
9798.
[2] GULLAL SINGH CHEEMA, SAKET
ANAND. Automatic Detection and Recognition
ofIndividuals in Patterned Species. IIIT –Delhi.
[3] Automatically identifying, counting, and
describing wild animals in camera-trap
images withdeep learning. In November
(2017).arXiv:1703.05830v5.
[4] TIBOR TRNOVSZKY, PATRIK
KAMENCAY, RICHARD ORJESEK,
MIROSLAV
BENCO, PETER SYKORA. Animal
Recognition System Based on Convolution
Neural Network. In September (2017).
DOI:10.15598/aeee.v15i3.2202.
[5] SAMER HIJAZI, RISHI KUMAR, CHRIS
ROWEN. Using Convolution Neural
Networksfor Image Recognition. IP
Group,Cadence.

Page | 845 Copyright @ 2021 Authors