See discussions, stats, and author profiles for this publication at: [Link]
net/publication/335813356
Animal Detection Using Deep Learning Algorithm
Conference Paper · September 2019
CITATIONS READS
6 11,353
1 author:
Banupriya Prakash
Sri Ramakrishna Engineering College
2 PUBLICATIONS 7 CITATIONS
SEE PROFILE
Some of the authors of this publication are also working on these related projects:
Animal Detection Using Deep Learning Algorithm View project
A Cloud Based Risk Prediction of Coronary Heart Disease View project
All content following this page was uploaded by Banupriya Prakash on 14 September 2019.
The user has requested enhancement of the downloaded file.
First International Conference on Intelligent Digital Transformation ICIDT - 2019 (11-13 July 2019, Volume - I)
Animal Detection Using Deep Learning Algorithm
Rashmi Jayakumar, Student, ECE, Sri Ramakrishna Engineering College, Coimbatore, India. E-mail: rashmi.1602202@[Link]
Rashmi Swaminathan, Student, ECE, Sri Ramakrishna Engineering College, Coimbatore, India. E-mail: rashmi.1602203@[Link]
Sanchithaa Harikumar, Student, ECE, Sri Ramakrishna Engineering College, Coimbatore, India.
E-mail: sanchithaa.1602217@[Link]
N. Banupriya, Assistant Professor, ECE, Sri Ramakrishna Engineering College, Coimbatore, India. E-mail: banupriya.n@[Link]
S. Saranya, Assistant Professor, ECE, Sri Ramakrishna Engineering College, Coimbatore, India. E-mail: saranya.s@[Link]
Abstract: Efficient and reliable monitoring of wild animals in their natural habitat is essential. This project develops
an algorithm to detect the animals in wild life. Since there are large number of different animals manually identifying
them can be a difficult task. This algorithm classifies animals based on their images so we can monitor them more
efficiently. Animal detection and classification can help to prevent animal-vehicle accidents, trace animals and
prevent theft. This can be achieved by applying effective deep learning algorithms.
Keywords: Animal Detection and Classification, Deep Learning Algorithms.
1. Introduction Monitoring wild animals, therefore, is essential as it
provides researchers evidences to inform conservation
Machine learning is an application of artificial and management decisions to maintain diverse,
intelligence (AI) that provides systems the ability to balanced and sustainable ecosystems in the face of those
automatically learn and improve from experience changes.
without being explicitly programmed. Machine learning
focuses on the development of computer programs that 2. Literature Survey
can access data and use it learn for [Link] The purpose of animal detection is to prevent or
process of learning begins with observations or data, reduce the number of animal-vehicle collisions. These
such as examples, direct experience, or instruction, in systems are specifically aimed at the wild animals that
order to look for patterns in data and make better can cause human death, injury and property damage.
decisions in the future based on the examples that we This system detects the wild animals before they enter
provide. The primary aim is to allow the computers learn the road.
automatically without human intervention or assistance
and adjust actions accordingly. Historically animal-vehicle collisions have been
addressed by putting up signs that warn peoples of
Deep learning is a subset of machine learning. Deep potential animal crossings. In other cases, wildlife
artificial neural networks are a set of algorithms that warning reflectors or wildlife fences have been installed
have set new records in accuracy for many important to keep animals away from the road. In some selected
problems, such as image recognition, sound recognition, areas wildlife fencing has been combined with a series of
etc., In deep learning, a convolutional neural network wildlife crossing structures. In most cases however, such
(CNN) is a class of deep neural networks, most crossing structures are limited in number and width,
commonly applied to analyzing visual [Link] use mostly because of their relatively high costs.
relatively little pre-processing compared to other image
classification algorithms. This means that the network A. Animal Detection Using Template Matching
learns the filters that in traditional algorithms were Algorithm
hand-engineered. This independence from prior Animal detection is useful in prevention of animal-
knowledge and human effort in feature design is a major vehicle accidents and will increase human and wildlife
advantage. They have applications in image and video safety, it will detect large animals before they enter the
recognition, recommender systems, image classification, road and warn the driver through audio and visual
medical image analysis, and natural language processing. signals. This also helps in saving crops in farm from
One of the applications of the deep learning animals. In this project there is survey of different object
technique called Convolutional Neural Network is animal detection techniques and for object identification as
detection. Observing wild animals in their natural animal techniques such as object matching, edge based
environments is a central task in ecology. The fast matching, skeleton extraction. After survey the most
growth of human population and the endless pursuit of appropriate method is selected for animal detection and
economic development are making over-exploitation of efficiency is measured. Proposed system has low false
natural resources, causing rapid, novel and substantial positive rate and false negative rate.
changes to Earth’s ecosystems. Template Matching
An increasing area of land surface has been Template matching is a technique in digital image
transformed by human action, altering wildlife processing for finding small parts of an image which
population, habitat and behaviour. More seriously, many match a template image. To perform template matching
wild species on Earth have been driven to extinction, and the concept of normalized cross co relation can be used.
many species are introduced into new areas where they In signal processing, cross-correlation is a measure of
can disrupt both natural and human systems.
11
First International Conference on Intelligent Digital Transformation ICIDT - 2019 (11-13 July 2019, Volume - I)
similarity of two waveforms as a function of a time-lag layers. Some of these layers are convolutional, using a
applied to one of them. mathematical model to pass on results to successive
This is also known as a sliding dot product or sliding layers.
inner-product. It is commonly used for searching a long- • Input will hold the raw pixel values of the image
duration signal for shorter, known feature. For image- and with three colour channels R, G, B.
processing applications in which the brightness of the • CONV layer will compute the output of neurons
image and template can vary due to lighting and that are connected to local regions in the input,
exposure conditions, the images can be first normalized. each computing a dot product between their
This is typically done at every step by subtracting the weights and a small region they are connected
mean and dividing by the standard deviation. Here we to in the input volume.
have used feature based template matching mechanism
using NCC • RELU layer will apply an element wise
activation function. This leaves the size of the
B. Automatically Identifying, Counting, and volume unchanged.
Describing Wild Animals in Camera-Trap Images
• POOL layer will perform a down sampling
with Deep Learning
operation along the spatial dimensions (width,
Having accurate, detailed, and up-to-date information height), resulting in volume such as
about the location and behaviour of animals in the wild [16x16x12].
would revolutionize our ability to study and conserve FC (i.e. fully-connected) layer will compute the class
ecosystems. This paper investigates the ability to scores, resulting in volume of size. As with ordinary
automatically, accurately, and inexpensively collect such Neural Networks and as the name implies, each neuron
data, which could transform many fields of biology, in this layer will be connected to all the numbers in the
ecology, and zoology into “big data” [Link] previous volume.
sensor “camera traps” enable collecting wildlife pictures
inexpensively, unobtrusively, and frequently. However, B. Convolutional Layer
extracting information from these pictures remains an Fig. 1 shows the convolution which is the first layer to
expensive, time-consuming, manual task. We extract features from an input image. Convolution
demonstrate that such information can be automatically preserves the relationship between pixels by learning
extracted by deep learning, a cutting edge type of image features using small squares of input data. It is a
artificial intelligence. mathematical operation that takes two inputs such as
We train deep convolutional neural networks to image matrix and a filter or kernel. Convolution of an
identify, count, and describe the behaviours of 48 image with different filters can perform operations such
species in the 3.2-million-image Snapshot Serengeti as edge detection, blur and sharpen by applying filters.
dataset. Our deep neural networks automatically identify
animals with over 93.8% accuracy, and we expect that
number to improve rapidly in years to come. More
importantly, if our system classifies only images it is
confident about, our system can automate animal
identification for 99.3% of the data while still
performing at the same 96.6% accuracy as that of crowd
sourced teams of human volunteers, saving more than
8.4 years (at 40 hours per week) of human labelling
effort (i.e. over 17,000 hours) on this 3.2-million-image
dataset.
Those efficiency gains immediately highlight the
importance of using deep neural networks to automate Fig. 1: Convolutional Layer
data extraction from camera-trap images. Our results
C. Pooling
suggest that this technology could enable the
inexpensive, unobtrusive, high-volume, and even real- Pooling layers section would reduce the number of
time collection of a wealth of information about vast parameters when the images are too large. Max pooling
numbers of animals in the wild. take the largest element from the rectified feature map.
The objective is to down-sample an input representation
3. Block Diagram (image, hidden-layer output matrix, etc.), reducing its
A. Convolutional Neural Network dimensionality. This is shown in Fig. 2.
A convolutional neural network (CNN) is a specific
type of artificial neural network that uses perceptron’s, a
machine learning unit algorithm, for supervised
learning, to analyze data. CNNs apply to image
processing, natural language processing and other kinds
of cognitive tasks. A convolutional neural network has
an input layer, an output layer and various hidden
Fig. 2: Pooling Layer
12
First International Conference on Intelligent Digital Transformation ICIDT - 2019 (11-13 July 2019, Volume - I)
D. Flattening 4. Flow Diagram
Flattening is the process of converting all the
resultant 2 dimensional arrays into a single long
continuous linear vector. It gets the output of the
convolutional layers, flattens all its structure to create a
single long feature vector to be used by the dense layer
for the final classification. This is shown in Fig. 3
Fig. 3: Flattening Layer
E. Fully connected
Fig.4 shows the hidden layers inside a Convolutional
Neural Network w are called Fully Connected Layers.
These are a specific type of hidden layer which must be
used within the CNN. This is used to combine the
features into more attributes that predict the outputs
more accurately.
Fig. 6: Flow Diagram
Fig. 6 shows the flow diagram of animal detection.
5. Dataset
The dataset used here is a collection of image data
Fig. 4: Fully Connected Layer
that contains various images of animals. The Dataset is
splitted into train and test in the ratio of 75:25
respectively.
A. Train Dataset
Elephant Train Dataset
Fig. 5: Block Diagram
Fig. 5 shows the block diagram of the animal
detection using deep learning algorithm. Fig. 7: Elephant Train Dataset
13
First International Conference on Intelligent Digital Transformation ICIDT - 2019 (11-13 July 2019, Volume - I)
Cheetah Train Dataset
Fig. 12: Detected Elephant
B. Detection of Cheetah
Fig. 8: Cheetah Train Dataset
B. Test dataset
Elephant Test Dataset
Fig. 9: Elephant Test Dataset
Cheetah Test Dataset
Fig. 13: Accuracy of Detected Cheetah
Fig. 10: Cheetah Test Dataset
6. Result
A. Detection of Elephant
Fig. 14: Detected Cheetah
7. Conclusion
Thus this project uses Convolutional Neural Network
(CNN) algorithm to detect wild animals. The algorithm
classifies animals efficiently with a good number of
accuracy and also the image of the detected animal is
displayed for a better result so that it can be used for
other purposes such as detecting wild animals entering
into human habitat and to prevent wildlife poaching and
Fig. 11: Accuracy of Detected Elephant even human animal conflict.
14
First International Conference on Intelligent Digital Transformation ICIDT - 2019 (11-13 July 2019, Volume - I)
8. Future Scope
This work can be further extended by sending an
alert in the form of a message when the animal is
detected to the nearby forest office. Furthermore it can
be used to reduce human wildlife conflict and also
animal accidents.
References
[1] Xie, Z., A. Singh, J. Uang, K.S. Narayan and [Link].
Multimodal blending for high-accuracy Instance cognition.
In: 2013 IEEE/RSJ International Conference on Intelligent
Robots and Systems. Tokyo: IEEE 2013, pp. 2214-2221.
ISBN 978-1-4673-6356-3. DOI: 10.1109/IROS.2013.
[2] Tiber Trnovszky, Patrik Kamencay, Richard Orjesek,
Miroslav Benco, Peter Sykora. Animal recognition system
based on convolutional neural network.
[3] Ahonen, T., Hadid, A., Pietikainen, and M.: Face description
with local binary patterns:Application to face recognition.
IEEE TPAMI 28(12), 2037-2041 (2006).
[4] Burghardt, T., Calic, J.: Real-time face detection and
tracking of animals. In:Neural Network Applications in
Electrical Engineering. pp. 27{32. IEEE (2006).
[5] Felzenszwalb, P.F., Girshick, R.B., McAllester, D., Ramanan,
D.: Object detection with discriminatively trained part-
based models. IEEE TPAMI 32(9), 1627-1645(2010).
[6] He, K., Zhang, X., Ren, S., Sun, J.: Deep residual learning for
image recognition. In: CVPR. pp. 770-778 (2016).
[7] Kamencay, P., T. Trnovszky, M. Benco, R. Hudec, P. Sykora
and A. Satnik. Accurate wild animal recognition using PCA,
LDAand LBPH, In: 2016 ELEKTRO. Strbske Pleso: IEEE,
2016, pp. 62–67. ISBN [Link]:
10.1109/ELEKTRO.2016.7512036.
[8] WU, J. L. and W. Y. MA. A Deep Learning Framework for
Coreference Resolution Based on Convolutional Neural
Network. In: 2017IEEE 11th International Conference on
Semantic Computing (ICSC). San Diego: IEEE,2017, pp. 61–
64. ISBN [Link]: 10.1109/ICSC.2017.57.
[9] P. M. Vitousek, H. A. Mooney, J. Lubchenco, J. Melillo,
"Human domination of Earth's ecosystems", Science, vol.
277, no. 5325, pp. 494-499, 1997.
[10] G. C. White, R. A. Garrott, Analysis of wildlife radio-
tracking data, Elsevier, 2012.
[11] B. J. Godley, J. Blumenthal, A. Broderick, M. Coyne, M.
Godfrey, L. Hawkes, M. Witt, "Satellite tracking of sea
turtles: Where have we been and where do we go next?",
Endangered Species Research, vol. 4, no. 1–2, pp. 3-22,
2008.
[12] A. Gomez, A. Salazar, F. Vargas, towards automatic wild
animal monitoring: Identification of animal species in
camera-trap images using very deep convolutional neural
networks, 2016.
15
View publication stats