IDENTIFICATION OF BIRD SPECIES USING

CONVOLUTION NUERAL NETWORKS

A PROJECT REPORT
Submitted by

B V SATYA SAI 316126510128

S APUROOP 316126510115
S MOUNIKA 316126510110
D SAI KUMAR 316126510137
In partial fulfilment for the award of the

degree Of

BACHELOR OF TECHNOLOGY
IN

COMPUTER SCIENCE ENGINEERING

Under the guidance of

Dr. R. Sivaranjani
Head of the Department
Dept. of CSE

DEPARTMENT OF COMPUTER SCIENCE AND ENGINEERING

ANIL NEERUKONDA INSTITUTE OF TECHNOLOGY &
SCIENCES (AUTONOMOUS)
(Permanently Affiliated to Andhra University)
SANGIVALASA: VISAKHAPATNAM –
531162
2016-2020
 ANIL NEERUKONDA INSTITUTE OF
TECHNOLOGY AND SCIENCES(A)
(Affiliated to Andhra University)
SANGIVALASA, VISAKHAPATNAM-
531162

BONAFIDE CERTIFICATE

This is to certify that project report “IDENTIFICATION OF BIRD

SPECIES USING CONVOLUTIONAL NEURAL NETWORKS” is
the bonafide work of “B.V.SATYA SAI(316126510128),S.APUROOP
(316126510115),S.MOUNIKA(316126510110),D.SAIKUMAR(31612651
0137)” who carried out the project work under my supervision.

Dr. R. Sivaranjani Dr. R. Sivaranjani

HEAD OF THE DEPARTMENT PROJECT GUIDE
Department of Computer Science Department of Computer
and Engineering Science and Engineering
ANITS ANITS
 DECLARATION

This is to certify that the project work entitled “IDENTIFICATION OF

BIRD SPECIES USING CONVOLUTIONAL NEURAL
NETWORKS” is a bonafide work carried out by B.V.SATYA
SAI,S.APUROOP,S.MOUNIKA,D.SAI KUMAR as a part of B.TECH
final year 2nd semester of computer science &Engineering of Andhra
University, Visakhapatnam during the year 2016-2020.

We,B.V.SATYA SAI, S.APUROOP, S.MOUNIKA, D.SAI KUMAR ,

of final semester B.Tech., in the department of Computer Science
Engineering from ANITS, Visakhapatnam, hereby declare that the project
work entitled IDENTIFICATION OF BIRD SPECIES USING
CONVOLUTIONAL NEURAL NETWORKS is carried out by us and
submitted in partial fulfillment of the requirements for the award of
Bachelor of Technology in Computer Science and Engineering , under
Anil Neerukonda Institute of Technology & Sciences during the academic
year 2016-2020 and has not been submitted to any other university for the
award of any kind of degree.

B V SATYA SAI 316126510128

S APUROOP 316126510115
S MOUNIKA 316126510110
D SAI KUMAR 316126510137
 ACKNOWLEDGEMENT

The satisfaction and euphoria that accompany the successful

completion of task would be incomplete without the mention of the people
who made it possible, whose constant guidance and encouragement always
boosted the morale. We take a great pleasure in presenting a project, which
is the result of a studied blend of both research and knowledge.

We first take the privilege to thank the Head of our Department,

Dr.R.SIVARANJANI, for permitting us in laying the first stone of
success and providing the lab facilities, we would also like to thank the
other staff in our department and lab assistants who directly or indirectly
helped us in successful completion of the project.

We feel great to thank Dr.R.SIVARANJANI , who are our project

guides and who shared their valuable knowledge with us and made us
understand the real essence of the topic and created interest in us to work
day and night for the project; we also thank our B.Tech coordinator
Dr.K.S.DEEPTHI, for her support and encouragement.

We also thank all the staff members of CSE department who

extended their part of support in the successful completion of the project.

B V SATYA SAI 316126510128

S APUROOP 316126510115
S MOUNIKA 316126510110
D SAI KUMAR 316126510137
 ABSTRACT
In our world, there are above 9000 bird species. Some bird species are being found rarely
and if found also prediction becomes very difficult. In order to overcome this problem, we
have an effective and simple way to recognize these bird species based on their features.
Also, the human ability to recognize the birds through the images is more understandable
than audio recognition. So, we have used Convolutional Neural Networks (CNN). CNN’s
are the strong assemblage of machine learning which have proven efficient in image
processing. In this paper, a CNN system classifying bird species is presented and uses the
Caltech-UCSD Birds 200 [CUB-200-2011] dataset for training as well as testing purpose.
By establishing this dataset and using the algorithm of similarity comparison, this system
is proved to achieve good results in practice. By using this method, everyone can easily
identify the name of the particular bird which they want to know.

Keywords: Bird species, Machine Learning, Convolutional Neural Networks.

 CONTENTS

TITLE Page No.

1. Introduction ……………………………………………. 1
1.1 Introduction …………. …………………………………. 1
1.1.1 Machine Learning …………………………………………… 2
1.1.2 Supervised Learning ………………………………………… 3
1.1.3 Unsupervised Learning ……………………………………… 4
1.1.4 Reinforcement Learning ……………………………………... 5
1.1.5 Neural Networks ………………………………………….…. 6
1.1.6 Deep Learning ………………………………………………. 9

1.2 Problem Statement ………………………………………. 11

2. Literature Survey ……………………………………… 12
2.1 Introduction ……………………………………………… 12
2.2 Survey of different classification methods……………… 12
2.3 Existing system …………………………………………. 14
3. Methodology …………………………………………… 15
3.1 Proposed System ………………………………............... 15
3.2 Architecture ……………………………………………... 17
3.3 Data Set Description ……………………………………. 21
3.4 Algorithm ………………………………………………. 27
3.4.1 ReLu Activation Function …………………………………… 31
3.4.2 Softmax Activation Function ………………………………… 32

3.5 Training …………………………………………………… 33

3.6 Flow of System …………………………………………… 36
4. Experimental analysis and results ……………………. 38
4.1 System Configuration ………………………………........ 38
4.1.1 Software Requirements …………………………… 38
 4.1.2 Hardware Requirements …………………………. 40
4.2 Sample Code …………………………………………… 41
4.3 Screenshots………………………………………………. 52
4.4 Experimental Analysis/Testing …………………………. . 56
5.Conclusion and Future Work………………………… 58
5.1 Conclusion ………………………………......................... 58
5.2 Future Work ……………………………………………... 59
References 60
Basepaper 63
Published Paper 66
 LIST OF FIGURES
Fig No. Topic Name Page No.

1 Machine Learning vs Traditional Programming 2

2 Neuron Model 7

3 Basic Neural Network 8

4 A CNN Sequence 17

5 Covolution operation with kernel 18

6 Performing Pooling operation 19

7 Sample Images Dataset 21

8 Sample Image of Painted_Bunting 22

9 Sample Image of Painted_Bunting 23

10 Sample Image of Brewer_Blackbird 24

11 Sample Image of Brewer_Blackbird 24

12 Sample Image of Eared_Grebe 25

13 Sample Image of Eared_Grebe 25

14 Three layers of Neural Network 27

15 Convolution Neural Network Layers 28

16 Understanding of CNN layers 29

17 Working of an Activation Function 31

18 ReLU Activation Function 31

19 Visualization of 5x5 filter 32

20 3-D graph of Neural Net 34

21 Flow of System 36
22 Inserting user input 52
 LIST OF TABLES

Table No. Table Name Page No.

1 Dataset Details 26

2 Efficiency Table 57
 1.INTRODUCTION
1.1. Introduction:
Bird behaviour and populace patterns have become a significant issue now a days. Birds
help us to recognize different life forms on the earth effectively as they react rapidly to
ecological changes. Be that as it may, assembling and gathering data about bird species
requires immense human exertion just as it turns into an extremely costly technique. In
such a case, a solid framework that will give enormous scale preparation of data about
birds and will fill in as a significant apparatus for scientists, legislative offices, and so
forth is required. In this way, bird species distinguishing proof assumes a significant job
in recognizing that a specific picture of birds has a place with which categories. Bird
species identification means predicting the bird species belongs to which category by
using an image.

The recognition of bird species can be possible through a picture, audio or video. An
audio processing method makes it conceivable to recognize by catching the sound sign
of different birds. Be that as it may, because of the blended sounds in condition, for
example, creepy crawlies, objects from the real world, and so forth handling of such data
turns out to be progressively convoluted. Normally, people discover images more
effectively than sounds or recordings. So, an approach to classify birds using an image
over audio or video is preferred. Bird species identification is a challenging task to
humans as well as to computational procedures that carry out such a task in an automated
fashion.

As image-based classification systems are improving the task of classifying, objects are
moving into datasets with far more categories such as Caltech-UCSD. Recent work has
seen much success in this area. Caltech UCSD Birds 200(CUB-200-2011) is a well-
known dataset for bird images with photos of 200 categories. The dataset contains birds
that are mostly found in Northern America. Caltech-UCSD Birds 200 consists of 11,788
images and annotations like 15 Part Locations, 312 Binary Attributes, 1 Bounding Box.

In this project, rather than recognizing an oversized number of disparate categories, the
matter of recognizing an oversized number of classes within one category is investigated
– that of birds. Classifying birds pose an additional challenge over categories, as a result
1
of the massive similarity between classes. additionally, birds are non-rigid objects which
will deform in many ways and consequently there's also an oversized variation within
classes. Previous work on bird classification has taken care of a little number of classes,
or through voice.

1.1.1. Machine Learning

Machine Learning is the most popular technique of predicting or classifying
information to help people in making necessary decisions. Machine Learning algorithms
are trained over instances or examples through which they learn from past experiences
and analyse the historical data.

Simply building models is not enough. You must also optimize and tune the
model appropriately so that it provides you with accurate results. Optimization
techniques involve tuning the hyperparameters to reach an optimum result.

As it trains over the examples, again and again, it is able to identify patterns in
order to make decisions more accurately. Whenever any new input is introduced to the
ML model, it applies its learned patterns over the new data to make future predictions.
Based on the final accuracy, one can optimize their models using various standardized
approaches. In this way, Machine Learning model learns to adapt to new examples and
produce better results.

Fig 1. Machine Learning vs Traditional Programming

2
Types of Learnings

Machine Learning Algorithms can be classified into 3 types as follows:

1. Supervised learning
2. Unsupervised Learning
3. Reinforcement Learning

1.1.2. Supervised Learning

Supervised learning is the most popular paradigm for machine learning. It is the
easiest to understand and the simplest to implement. It is the task of learning a function
that maps an input to an output based on example input-output pairs. It infers a function
from labelled training data consisting of a set of training examples. In supervised
learning, each example is a pair consisting of an input object (typically a vector) and a
desired output value (also called the supervisory signal). A supervised learning algorithm
analyses the training data and produces an inferred function, which can be used for
mapping new examples. Supervised Learning is very similar to teaching a child with the
given data and that data is in the form of examples with labels, we can feed a learning
algorithm with these example-label pairs one by one, allowing the algorithm to predict
the right answer or not. Over time, the algorithm will learn to approximate the exact
nature of the relationship between examples and their labels. When fully trained, the
supervised learning algorithm will be able to observe a new, never-before-seen example
and predict a good label for it.

Most of the practical machine learning uses supervised learning. Supervised

learning is where you have input variable (x) and an output variable (Y) and you use an
algorithm to learn the mapping function from the input to the output.

Y = f(x)

The goal is to approximate the mapping function so well that when you have new
input data (x) that you can predict the output variables (Y) for the data. It is called
supervised learning because the process of an algorithm learning from the training
dataset can be thought of as a teacher supervising the learning process. Supervised
learning is often described as task oriented. It is highly focused on a singular task, feeding
more and more examples to the algorithm until it can accurately perform on that task.
3
This is the learning type that you will most likely encounter, as it is exhibited in many of
the common applications like Advertisement Popularity, Spam Classification, face
recognition.

Two types of Supervised Learning are:

1. Regression:
Regression models a target prediction value based on independent
variables. It is mostly used for finding out the relationship between
variables and forecasting. Regression can be used to estimate/
predict continuous values (Real valued output). For example, given a picture of
a person then we have to predict the age on the basis of the given picture.

2. Classification:
Classification means to group the output into a class. If the data
is discrete or categorical then it is a classification problem. For example, given
data about the sizes of houses in the real estate market, making our output about
whether the house “sells for more or less than the asking price” i.e. Classifying
houses into two discrete categories.

1.1.3. Unsupervised Learning

Unsupervised Learning is a machine learning technique, where you do not need

to supervise the model. Instead, you need to allow the model to work on its own to
discover information. It mainly deals with the unlabelled data and looks for previously
undetected patterns in a data set with no pre-existing labels and with a minimum of
human supervision. In contrast to supervised learning that usually makes use of human-
labelled data, unsupervised learning, also known as self-organization, allows for
modelling of probability densities over inputs.

Unsupervised machine learning algorithms infer patterns from a dataset without

reference to known, or labelled outcomes. It is the training of machine using information
that is neither classified nor labelled and allowing the algorithm to act on that
information without guidance. Here the task of machine is to group unsorted information
according to similarities, patterns, and differences without any prior training of data.
Unlike supervised learning, no teacher is provided that means no training will be given
4
to the machine. Therefore, machine is restricted to find the hidden structure in unlabelled
data by our-self. For example, if we provide some pictures of dogs and cats to the
machine to categorized, then initially the machine has no idea about the features of dogs
and cats so it categorize them according to their similarities, patterns and differences.
The Unsupervised Learning algorithms allows you to perform more complex processing
tasks compared to supervised learning. Although, unsupervised learning can be more
unpredictable compared with other natural learning methods.

Unsupervised learning problems are classified into two categories of algorithms:

 Clustering: A clustering problem is where you want to discover the inherent

groupings in the data, such as grouping customers by purchasing behaviour.
 Association: An association rule learning problem is where you want to discover
rules that describe large portions of your data, such as people that buy X also
tend to buy Y.

1.1.4. Reinforcement Learning

Reinforcement Learning (RL) is a type of machine learning technique that enables

an agent to learn in an interactive environment by trial and error using feedback from its
own actions and experiences. Machine mainly learns from past experiences and tries to
perform best possible solution to a certain problem. It is the training of machine learning
models to make a sequence of decisions. Though both supervised and reinforcement
learning use mapping between input and output, unlike supervised learning where the
feedback provided to the agent is correct set of actions for performing a task,
reinforcement learning uses rewards and punishments as signals for positive and
negative behaviour. Reinforcement learning is currently the most effective way to hint
machine’s creativity.

5
1.1.5. Neural Networks

Neural Network (or Artificial Neural Network) has the ability to learn by
examples. ANN is an information processing model inspired by the biological neuron
system. ANN biologically inspired simulations that are performed on the computer to
do a certain specific set of tasks like clustering, classification, pattern recognition etc. It
is composed of a large number of highly interconnected processing elements known as
the neuron to solve problems. It follows the non-linear path and process information in
parallel throughout the nodes. A neural network is a complex adaptive system. Adaptive
means it has the ability to change its internal structure by adjusting weights of inputs.

Artificial Neural Networks can be best viewed as weighted directed graphs,

where the nodes are formed by the artificial neurons and the connection between the
neuron outputs and neuron inputs can be represented by the directed edges with weights.
The ANN receives the input signal from the external world in the form of a pattern and
image in the form of a vector. These inputs are then mathematically designated by the
notations x(n) for every n number of inputs. Each of the input is then multiplied by its
corresponding weights (these weights are the details used by the artificial neural
networks to solve a certain problem). These weights typically represent the strength of
the interconnection amongst neurons inside the artificial neural network. All the
weighted inputs are summed up inside the computing unit (yet another artificial neuron).

If the weighted sum equates to zero, a bias is added to make the output non-zero
or else to scale up to the system’s response. Bias has the weight and the input to it is
always equal to 1. Here the sum of weighted inputs can be in the range of 0 to positive
infinity. To keep the response in the limits of the desired values, a certain threshold value
is benchmarked. And then the sum of weighted inputs is passed through the activation
function. The activation function is the set of transfer functions used to get the desired
output of it. There are various flavours of the activation function, but mainly either linear
or non-linear set of functions. Some of the most commonly used set of activation
functions are the Binary, Sigmoid (linear) and Tan hyperbolic sigmoidal (non-linear)
activation functions.

6
 Fig 2. Neuron Model

The Artificial Neural Network contains three layers

1. Input Layer: The input layers contain those artificial neurons (termed as units)
which are to receive input from the outside world. This is where the actual
learning on the network happens or corresponding happens else it will process.
2. Hidden Layer: The hidden layers are mentioned hidden in between input and the
output layers. The only job of a hidden layer is to transform the input into
something meaningful that the output layer/unit can use in some way. Most of the
artificial neural networks are all interconnected, which means that each of the
hidden layers is individually connected to the neurons in its input layer and also
to its output layer leaving nothing to hang in the air. This makes it possible for a
complete learning process and also learning occurs to the maximum when the
weights inside the artificial neural network get updated after each iteration.
3. Output Layer: The output layers contain units that respond to the information
that is fed into the system and also whether it learned any task or not.

7
 Fig 3. Basic Neural Network

Learning process of a neural network:

1. Start with values (often random) for the network parameters (wij weights
and bj biases).

2. Take a set of examples of input data and pass them through the network to obtain
their prediction.

3. Compare these predictions obtained with the values of expected labels and calculate
the loss with them.

4. Perform the backpropagation in order to propagate this loss to each and every one
of the parameters that make up the model of the neural network.

5. Use this propagated information to update the parameters of the neural network with
the gradient descent in a way that the total loss is reduced, and a better model is
obtained.

6. Continue iterating in the previous steps until we consider that we have a good model.

8
1.1.6 Deep Learning

Deep learning is a branch of machine learning which is completely based

on artificial neural networks. Deep learning is an artificial intelligence function that
imitates the workings of the human brain in processing data and creating patterns for use
in decision making. Deep learning is a subset of machine learning in artificial
intelligence (AI) that has networks capable of learning unsupervised from data that is
unstructured or unlabelled. It has a greater number of hidden layers and known as deep
neural learning or deep neural network. Deep learning has evolved hand-in-hand with
the digital era, which has brought about an explosion of data in all forms and from every
region of the world. This data, known simply as big data, is drawn from sources like
social media, internet search engines, e-commerce platforms, and online cinemas,
among others. This enormous amount of data is readily accessible and can be shared
through fintech applications like cloud computing. However, the data, which normally
is unstructured, is so vast that it could take decades for humans to comprehend it and
extract relevant information. Companies realize the incredible potential that can result
from unravelling this wealth of information and are increasingly adapting to AI systems
for automated support. Deep learning learns from vast amounts of unstructured data that
would normally take humans decades to understand and process. Deep learning and
utilizes a hierarchical level of artificial neural networks to carry out the process of
machine learning. The artificial neural networks are built like the human brain, with
neuron nodes connected like a web. While traditional programs build analysis with data
in a linear way, the hierarchical function of deep learning systems enables machines to
process data with a nonlinear approach.

Deep Neural Network

It is a neural network with a certain level of complexity (having multiple hidden

layers in between input and output layers). They are capable of modelling and processing
non-linear relationships.

9
Disadvantages of previous bird classification through voice:

 Background noise- especially while using data recorded in a city.

 Multi-label classification problem-when there are many species singing at the
same time.
 Different types of bird songs.
 Inter-species variance-there might be difference in bird song between the same
species living in different regions or countries.
 Data set issues-the data can be highly imbalanced due to bigger popularity of one
species over another, there is a large number of different species and recordings
can have different length, quality of recordings.

Objective of the project:

 The main objective of the project is to identify the species of the birds by
analysing an image. Some of the experts like ornithologists couldn’t identify
species of the bird correctly by looking at an image.
 Although bird classification can be done manually by domain experts, with
growing amounts of data, this rapidly becomes a tedious and time-consuming
process. So, by this model we can identify the species of the birds accurately and
in less time.

10
1.2. Problem Statement:

Identifying a bird can be a challenge, even for experienced birders. And

if you're new to using field guides, it can be daunting to figure out where to even begin
searching in the hundreds of pages of species. By some features like size, shape and
colour birds can be classified. By using CNN, we can classify the species of the birds.

11
 2. LITERATURE SURVEY

2.1. Introduction:

In particular, it was discovered that ecologists monitor them to determine the factors
causing population fluctuation and to help in conserving and managing threatened and
endangered species. The various surveys used in counting bird species including data
collection techniques were succinctly reviewed. It was established that a small but
growing number of researchers have studied the use of computer vision for monitoring
species of birds.

This chapter evaluates reports of studies found in literature that are related to
monitoring and classification of species. In particular, it focuses on reviewing bird
techniques used for these species are often similar, and motion features which this
research seeks to investigate for classification of birds. First techniques which perform
classification using single images were explored. This was done by reviewing them
separately as those that are used for classification of bird species and those for other
species specifically bats.

2.2. Survey of different classification methods:

John Martinsson et al (2017) [1], presented the CNN algorithm and deep residual neural
networks to detect an image in two ways i.e., based on feature extraction and signal
classification. They did an experimental analysis for datasets consisting of different
images. But their work didn’t consider the background species. In Order to identify the
background species larger volumes of training data are required, which may not be
available.

Juha Niemi, Juha T Tanttu et al (2018) [2], proposed a Convolutional neural network
trained with deep learning algorithms for image classification. It also proposed a data
augmentation method in which images are converted and rotated in accordance with the
desired color. The final identification is based on a fusion of parameters provided by the
radar and predictions of the image classifier.

12
Li Jian, Zhang Lei et al (2014)[3], proposed an effective automatic bird species
identification based on the analysis of image features. Used the database of standard
images and the algorithm of similarity comparisons.

Madhuri A. Tayal, Atharva Magrulkar et al (2018)[4], developed a software application

that is used to simplify the bird identification process. This bird identification software
takes an image as an input and gives the identity of the bird as an output. The technology
used is transfer learning and MATLAB for the identification process.

Andreia Marini, Jacques Facon et al (2013) [5], proposed a novel approach based on
color features extracted from unconstrained images, applying a color segmentation
algorithm in an attempt to eliminate background elements and to delimit candidate
regions where the bird may be present within the image. Aggregation processing was
employed to reduce the number of intervals of the histograms to a fixed number of bins.
In this paper, the authors experimented with the CUB-200 dataset and results show that
this technique is more accurate.

Marcelo T. Lopes, Lucas L. Gioppo et al (2011) [6], focused on the automatic

identification of bird species from their audio recorded song. Here the authors dealt with
the bird species identification problem using signal processing and machine learning
techniques with the MARSYAS feature set. Presented a series of experiments conducted
in a database composed of bird songs from 75 species out of which problem obtained in
performance with 12 species.

Peter Jancovic and Munevver Kokuer et al (2012) [7], investigated acoustic modelling
for recognition of bird species from audio field recordings. Developed a hybrid deep
neural network hidden Markov model (DNN-HMM). The developed models were
employed for bird species identification, detection of specific species and recognition of
multiple bird species vocalizing in a given recording. In this paper, the authors achieved
an identification accuracy of 98.7% and recognition accuracy of 97.3%.

Mario Lasseck et al (2013) [8], presented deep convolutional neural networks and data
augmentation techniques for audio-based bird species identification. In this paper, the
author used the Xeno-Canto set of audio recordings of bird species.

13
2.3. Existing System:

To identify the bird species there are many websites produces the results using different
technologies. But the results are not accurate. For suppose if we will give an input in
those websites and android applications it gives us multiple results instead of single bird
name. It shows us the all bird names which are having similar characteristics. So, we
aimed to develop a project to produce better and accurate results. In order to achieve this,
we have used Convolutional Neural Networks to classify the bird species.

14
 3. METHODOLOGY

3.1. Proposed System:

Convolution neural network algorithm is a multilayer perceptron that is the special

design for the identification of two-dimensional image information. It has four layers: an
input layer, a convolution layer, a sample layer, and an output layer. In a deep network
architecture, the convolution layer and sample layer may have multiple. CNN is not as
restricted as the Boltzmann machine, it needs to be before and after the layer of neurons
in the adjacent layer for all connections, convolution neural network algorithms, each
neuron doesn’t need to experience the global image, just feel the local region of the
image. In addition, each neuron parameter is set to the same, namely, the sharing of
weights, namely each neuron with the same convolution kernels to the deconvolution
image.

The key era of CNN is the local receptive field, sharing of weights, subsampling by using
time or space, with a purpose to extract features and reduce the size of the training
parameters. The advantage of CNN algorithm is to avoid the explicit feature extraction,
and implicitly to learn from the training data. The same neuron weights on the surface of
the feature mapping, thus the network can learn parallel, and reduce the complexity of
the network Adopting sub-sampling structure by time robustness, scale, and deformation
displacement. Input information and network topology can be a very good match. It has
unique advantages in image processing. The Convolution Neural Network involves these
steps

Convolution Layer:

The convolutional layer is the core constructing block of a CNN. The convolution
layer comprises a set of independent feature detectors. Each Feature map is
independently convolved with the images.

15
Pooling Layer:

The pooling layer feature is to progressively reduce the spatial size of the
illustration to reduce the wide variety of parameters and computation in the network. The
pooling layer operates on each function map independently. The approaches used in
pooling are:

· Max Pooling
· Mean Pooling

· Sum Pooling

Why to use pooling layers:

 Pooling layers are used to reduce the dimensions of the feature maps. Thus, it
reduces the number of parameters to learn and the amount of computation
performed in the network.
 The pooling layer summarises the features present in a region of the feature map
generated by a convolution layer. So, further operations are performed on
summarised features instead of precisely positioned features generated by the
convolution layer. This makes the model more robust to variations in the position
of the features in the input image.

Fully Connected Layer:

Neurons in the fully connected layer have full connections to all activations inside
the preceding layer. In this, the output obtained from max pooling is converted to a one-
dimensional array and that should be the input layer and the process continues the same
as the ANN model.

16
3.2. Architecture:

Fig4. A CNN Sequence

A Convolutional Neural Network (ConvNet/CNN) is a Deep Learning algorithm which

can take in an input image, assign importance (learnable weights and biases) to various
aspects/objects in the image and be able to differentiate one from the other. The pre-
processing required in a ConvNet is much lower as compared to other classification
algorithms. While in primitive methods filters are hand-engineered, with enough
training, ConvNets have the ability to learn these filters/characteristics.

A ConvNet is able to successfully capture the Spatial and Temporal dependencies in an

image through the application of relevant filters. The architecture performs a better
fitting to the image dataset due to the reduction in the number of parameters involved
and reusability of weights. In other words, the network can be trained to understand the
sophistication of the image better.

The role of the ConvNet is to reduce the images into a form which is easier to process,
without losing features which are critical for getting a good prediction.

17
Convolution Layer
Filters (Convolution Kernels)
A filter (or kernel) is an integral component of the layered architecture.

Generally, it refers to an operator applied to the entirety of the image such that it
transforms the information encoded in the pixels. In practice, however, a kernel is a
smaller-sized matrix in comparison to the input dimensions of the image, that consists of
real valued entries.

The real values of the kernel matrix change with each learning iteration over the training
set, indicating that the network is learning to identify which regions are of significance
for extracting features from the data.

Fig 5. Convolution operation with kernel

In Fig 5 we a convoluting a 5x5x1 image with a 3x3x1 kernel (which change each
iteration to extract a significant features) to get a 3x3x1 convolved feature. The filter
moves to the right with a certain stride value till it parses the complete width. In case of
images with multiple channels (e.g. RGB) the kernel has a same depth as that of input
image. Matrix multiplication is performed between Kn and In stack ([K1, I1]; [K2, I2];

18
[K3, I3]) and all results are summed with bias to give us a squashed 1 depth channel
convoluted feature output.

The objective of the Convolution Operation is to extract the high-level features such as
edges, from the input image, the first Convolutional Layer is responsible for capturing
the Low-Level features such as edges, color, gradient orientation, etc.

Kernel Size - 3x3

Pooling Matrix Size - 2x2

Image will be Resized to - (Height=250, Width=250)

Pooling Layer:

(a)Convoluted Output (b) Pooling Output

Fig 6. Performing Pooling operation

The Pooling layer is responsible for reducing the spatial size of the Convolved Feature.
This is to decrease the computational power required to process the data through
dimensionality reduction. Furthermore, it is useful for extracting dominant features which
are rotational and positional invariant, thus maintaining the process of effectively training
of the model.

There are three types of Pooling: Max Pooling, Average Pooling and Global Pooling.

19
Max Pooling: Max pooling is a pooling operation that selects the maximum element
from the region of the feature map covered by the filter. Thus, the output after max-
pooling layer would be a feature map containing the most prominent features of the
previous feature map. In Fig. 3 we perform maximum pooling operation by considering
convoluted feature output obtained from convolution layer.

Average Pooling: Average pooling computes the average of the elements present in the
region of feature map covered by the filter. Thus, while max pooling gives the most
prominent feature in a particular patch of the feature map, average pooling gives the
average of features present in a patch.

Global Pooling: Global pooling reduces each channel in the feature map to a single value.
Thus, an nh x nw x nc feature map is reduced to 1 x 1 x nc feature map. This is equivalent
to using a filter of dimensions nh x nw i.e. the dimensions of the feature map.
Further, it can be either global max pooling or global average pooling.

20
3.3. Data Set Description:

For the project, we will be using the dataset that includes 200 categories of bird species,
11,788 total number of images, and other information such as labelled visible bird parts,
binary attributes, and bounding boxes surrounding the birds. The authors also include a
recommended test and training set split of the data. From these 200 categories we are
taking 3 basic categories and performs the test.

Fig 7. Sample images dataset

Input data for the identification system consist of digital images and parameters
from the radar. All images for training the CNN are of wild birds in flight and they have
been taken manually at the test location. There are also constraints concerning the area
where the images have to be taken. Here, the area refers to the air space in the vicinity
of the pilot wind turbine. We have used the wind turbine swept area as a suitable altitude
level constraint for taking the images, because birds flying below or above the swept area
are not in danger.

The bird species used in this dataset are:

 Painted_Bunting
 Breweer_Blackbird
21
  Eared_Grebe

1.Painted_Bunting:

The painted bunting (Passerina ciris) is a species of bird in the cardinal

family, Cardinalidae, that is native to North America. The bright plumage of the male
only comes in the second year of life; in the first year they can only be distinguished
from the female by close inspection.

Sometimes called the "Nonpareil," meaning "unrivalled," a fair way to describe the
unbelievable colors of the male Painted Bunting. This species is locally common in the
Southeast, around brushy areas and woodland edges. It is often secretive, staying low in
dense cover. However, males sing their bright warbling songs from higher in the trees,
partly hidden among foliage or sometimes out in the sun on an exposed perch. Some
lucky Floridians have Painted Buntings coming to their bird feeders in winter.

Here are few pictures of Painted_Bunting:

Fig 8 . Sample Image of Painted_Bunting

22
 Fig 9. Sample image of Painted_Bunting

2.Brewer_Blackbird:
The Brewer's blackbird (Euphagus cyanocephalus) is a medium-sized New World
blackbird. It is named after the ornithologist Thomas Mayo Brewer.

This is the common blackbird of open country in the West, often seen walking on the
ground with short forward jerks of its head. It adapts well to habitats altered by humans,
and in places it may walk about on suburban sidewalks or scavenge for crumbs around
beachfront restaurants. In winter, Brewer's Blackbirds gather in large flocks, often with
other blackbirds, and may be seen foraging in farmland all across the western and
southern states.

Adult males have black plumage with an iridescent purple head and neck and glossy
bluish-green highlights on the rest of the body. The feet and legs are black and the eye is
bright yellow. The female is brownish-grey with slight hints of the male's iridescence.
The female's eye is dark brown, while the male's is bright yellow. Overall, they resemble
the eastern member of the same genus, the rusty blackbird; however, the Brewer's
blackbird has a shorter bill and the male's head is iridescent purple. This bird is often
mistaken for the common grackle but has a shorter tail. The call is a sharp check which
is also distinguishable. This bird is in a different family from the Eurasian blackbird.

Here are few pictures of Brewer_Blackbird:

23
 Fig 10. Sample image of Brewer_Blackbird

Fig 11. Sample Image of Brewer_Blackbird

3.Eared_Grebe:

The black-necked grebe (Podiceps nigricollis), known in North America as the eared
grebe, is a member of the grebe family of water birds. It was described in 1831
by Christian Ludwig Brehm. There are currently three accepted subspecies, including
the nominate subspecies. Its breeding plumage features a distinctive ochre-coloured
plumage which extends behind its eye and over its ear coverts. The rest of the upper
24
parts, including the head, neck, and breast, are coloured black to blackish brown. The
flanks are tawny rufous to maroon-chestnut, and the abdomen is white. When in its non-
breeding plumage, this bird has greyish-black upper parts, including the top of the head
and a vertical stripe on the back of the neck. The flanks are also greyish-black. The rest
of the body is a white or whitish colour. The juvenile has more brown in its darker areas.
The subspecies californicus can be distinguished from the nominate by the former's
usually longer bill.

Here are some pictures of Eared_Grebe:

Fig 12. Sample Image of Eared_Grebe

Fig 13. Sample Image of Eared_Grebe

25
Dataset Details:

Sno. Bird Name Samples Dimensions of the

count image

1. Painted_Bunting 250 Any Size

2. Brewer_Blackbird 175 Any Size

3. Eared_Grebe 200 Any Size

Table 1. Dataset Details

Here we can give any size of the image it will be resized to the dimensions 250x250.

The Features considered in this classification are : Beak ,Shape ,Neck, Color, Tail etc.,

26
3.4. Algorithm:
In this experiment, unsupervised learning algorithm has been used for developing
the system, because the inputted image defined is not known. Also, the data which is
given to unsupervised learning algorithm are not labeled, i.e. only the input variables(X)
are given with no corresponding output variables. In unsupervised learning, algorithms
discover interesting structures in the data themselves. In detail, clustering is used for
dividing the data into several groups.
In depth, deep learning models used to find vast number of neurons. Deep learning
algorithms learn more about the image as it goes through each neural network layer.
For classifying Neural Network is used. Figure 5 represents layers of neural networks
for feature extraction. The neural network is a framework for many machine learning
algorithms. Neural networks consist of vector of weights(W) and the bias (B).

Fig 14. Three layers of Neural Network

In deep learning, convolutional neural network (CNN) is a class of deep neural network
mostly used for analyzing visual images. It consists of an input layer and output layer
as well as multiple hidden layers. Every layer is made up of group of neurons and each
layer is fully connected to all neurons of its previous layer. The output layer is
responsible for prediction of output. The convolutional layer takes an image as input,
and produces a set of feature maps as output. The input image can contain multiple
channels such as color, wings, eyes, beak of birds which means that the convolutional
layer perform a mapping from 3D volume to another 3D volume. 3D volumes
considered are width, height, depth. The CNN have two components:

27
1) Feature extraction part: features are detected when network performs a series of
convolutional and pooling operation.
2) Classification part: extracted features are given to fully connected layer which acts
as classifier.

Fig 15. Convolutional Neural Network Layers

CNN consists of four layers: convolutional layer, activation layer, pooling layer and
fully connected. Convolutional layer allows extracting visual features from an image in
small amounts. Pooling is used to reduce the number of neurons from previous
convolutional layer but maintaining the important information. Activation layer passes
a value through a function which compresses values into range. Fully connected layer
connects a neuron from one layer to every neuron in another layer. As CNN classifies
each neuron in depth, so it provides more accuracy.
Image classification: image classification in machine learning is commonly done in two
ways:
1) Gray scale
2) Using RGB values
Normally all the data is mostly converted into gray scale. In gray scale algorithm,
28
 computer will assign values to each pixel based on how the value of the pixel is it. All
the pixel values are put into an array and the computer will perform operation on that
array to classify the data.

Fig 16. Understanding of CNN layers

Image classification is the task of taking an input image and outputting a class (a cat,
dog, etc) or a probability of classes that best describes the image. For humans, this task
of recognition is one of the first skills we learn from the moment we are born and is one
that comes naturally and effortlessly as adults.

When a computer sees an image (takes an image as input), it will see an array of pixel
values. Depending on the resolution and size of the image, it will see a 32 x 32 x 3 array
of numbers (The 3 refers to RGB values). Just to drive home the point, let's say we have
a color image in JPG form and its size is 480 x 480. The representative array will be 480
x 480 x 3. Each of these numbers is given a value from 0 to 255 which describes the
pixel intensity at that point. These numbers, while meaningless to us when we perform
image classification, are the only inputs available to the computer. The idea is that you
give the computer this array of numbers and it will output numbers that describe the
probability of the image being a certain class (.80 for cat, .15 for dog, .05 for bird, etc).

Now that we know the problem as well as the inputs and outputs, let’s think about how
to approach this. What we want the computer to do is to be able to differentiate between
all the images it’s given and figure out the unique features that make a dog a dog or that
make a cat a cat. This is the process that goes on in our minds subconsciously as well.
When we look at a picture of a dog, we can classify it as such if the picture has
identifiable features such as paws or 4 legs. In a similar way, the computer is able
29
perform image classification by looking for low level features such as edges and curves,
and then building up to more abstract concepts through a series of convolutional layers.
This is a general overview of what a CNN does.

The first layer in a CNN is always a Convolutional Layer. First thing to make sure you
remember is what the input to this conv (I’ll be using that abbreviation a lot) layer is.
Like we mentioned before, the input is a 32 x 32 x 3 array of pixel values. Now, the best
way to explain a conv layer is to imagine a flashlight that is shining over the top left of
the image. Let’s say that the light this flashlight shines covers a 5 x 5 area. And now,
let’s imagine this flashlight sliding across all the areas of the input image. In machine
learning terms, this flashlight is called a filter(or sometimes referred to as a neuron or
a kernel) and the region that it is shining over is called the receptive field. Now this
filter is also an array of numbers (the numbers are called weights or parameters). A very
important note is that the depth of this filter has to be the same as the depth of the input
(this makes sure that the math works out), so the dimensions of this filter is 5 x 5 x 3.
Now, let’s take the first position the filter is in for example. It would be the top left
corner. As the filter is sliding, or convolving, around the input image, it is multiplying
the values in the filter with the original pixel values of the image (aka computing element
wise multiplications). These multiplications are all summed up (mathematically
speaking, this would be 75 multiplications in total). So now you have a single number.
Remember, this number is just representative of when the filter is at the top left of the
image. Now, we repeat this process for every location on the input volume. (Next step
would be moving the filter to the right by 1 unit, then right again by 1, and so on). Every
unique location on the input volume produces a number. After sliding the filter over all
the locations, you will find out that what you’re left with is a 28 x 28 x 1 array of numbers,
which we call an activation map or feature map. The reason you get a 28 x 28 array is
that there are 784 different locations that a 5 x 5 filter can fit on a 32 x 32 input image.
These 784 numbers are mapped to a 28 x 28 array.

Activation Functions:
An activation function is a very important feature of an artificial neural network,
they basically decide whether the neuron should be activated or not. They introduce non-
linear properties to our Network whose main purpose is to convert an input signal of a

30
node in an A-NN to an output signal. That output signal now can be used as an input in
the next layer in the stack.

Fig 17. Working of an activation function

In the above figure, (x1, x2 … xn) is the input signal vector that gets multiplied
with the weights (w1, w2, …wn). This is followed by accumulation (i.e. summation +
addition of bias b). Finally, an activation function f is applied to this sum.

3.4.1 ReLU (Rectified Linear Unit) Activation Function

The ReLU is the most used activation function in the world right now

Fig 18. ReLU Activation Function

Equation: f(x) = max (0, x)
Range: (0 to infinity)

Pros:
1. The function and its derivative both are monotonic.
2. Due to its functionality it does not activate all the neuron at the same time
31
 3. It is efficient and easy for computation.

Cons:
1. The outputs are not zero centred similar to the sigmoid activation function
2. When the gradient hits zero for the negative values, it does not converge towards
the minima which will result in a dead neuron while back propagation.

3.4.2 Softmax Activation Function

The softmax function is also a type of sigmoid function but it is very useful to
handle classification problems having multiple classes.

𝓏
𝑒 𝑗
𝜎(𝓏)𝑗 = ∑𝕂 𝓏𝑘
𝑘=1 𝑒

The softmax function is shown above, where z is a vector of the inputs to the
output layer (if you have 10 output units, then there are 10 elements in z). And again, j
indexes the output units, so j = 1, 2, …, K.

The softmax function is ideally used in the output layer of the classifier where we

are trying to attain the probabilities to define the class of each input .

Fig 19. Visualization of 5x5 filter

32
3.5. Training
The way the computer is able to adjust its filter values (or weights) is through a training
process called backpropagation.

Before we get into backpropagation, we must first take a step back and talk about what
a neural network needs in order to work. At the moment we all were born, our minds
were fresh. We didn’t know what a cat or dog or bird was. In a similar sort of way,
before the CNN starts, the weights or filter values are randomized. The filters don’t know
to look for edges and curves. The filters in the higher layers don’t know to look for paws
and beaks. As we grew older however, our parents and teachers showed us different
pictures and images and gave us a corresponding label. This idea of being given an image
and a label is the training process that CNNs go through. Before getting too into it, let’s
just say that we have a training set that has thousands of images of dogs, cats, and birds
and each of the images has a label of what animal that picture is. Back to backprop.

So backpropagation can be separated into 4 distinct sections, the forward pass, the loss
function, the backward pass, and the weight update. During the forward pass, you take
a training image which as we remember is a 32 x 32 x 3 array of numbers and pass it
through the whole network. On our first training example, since all of the weights or
filter values were randomly initialized, the output will probably be something like [.1
.1] particular. The network, with its current weights, isn’t able to look for those low level
features or thus isn’t able to make any reasonable conclusion about what the
classification might be. This goes to the loss function part of backpropagation.
Remember that what we are using right now is training data. This data has both an image
and a label. Let’s say for example that the first training image inputted was a 3. The label
for the image would be [0 0 0 1 0 0 0 0 0 0]. A loss function can be defined in many
different ways but a common one is MSE (mean squared error), which is ½ times (actual
- predicted) squared.

Etotal = ∑ 1/2(target – output)2

Let’s say the variable L is equal to that value. As you can imagine, the loss will be
extremely high for the first couple of training images. Now, let’s just think about this

33
intuitively. We want to get to a point where the predicted label (output of the ConvNet)
is the same as the training label (This means that our network got its prediction right).In
order to get there, we want to minimize the amount of loss we have. Visualizing this as
just an optimization problem in calculus, we want to find out which inputs (weights in
our case) most directly contributed to the loss (or error) of the network.

Fig 20. 3-D Graph of Neural Net

This is the mathematical equivalent of a dL/dW where W are the weights at a particular
layer. Now, what we want to do is perform a backward pass through the network,
which is determining which weights contributed most to the loss and finding ways to
adjust them so that the loss decreases. Once we compute this derivative, we then go to
the last step which is the weight update. This is where we take all the weights of the
filters and update them so that they change in the opposite direction of the gradient.

W = Wi - ƞ (dL/dW)

where W = Weight

Wi = Initial Weight

Ƞ = Learning Rate
The learning rate is a parameter that is chosen by the programmer. A high learning rate
means that bigger steps are taken in the weight updates and thus, it may take less time
for the model to converge on an optimal set of weights. However, a learning rate that is
too high could result in jumps that are too large and not precise enough to reach the
optimal point.

34
The process of forward pass, loss function, backward pass, and parameter update is one
training iteration. The program will repeat this process for a fixed number of iterations
for each set of training images (commonly called a batch). Once you finish the parameter
update on the last training example, hopefully the network should be trained well enough
so that the weights of the layers are tuned correctly.

Testing:
Finally, to see whether or not our CNN works, we have a different set of images and
labels (can’t double dip between training and test!) and pass the images through the CNN.
We compare the outputs to the ground truth and see if our network works!

35
3.6. Flow of System:

Fig 21. Flow of System

The above figure represents the actual flow of the proposed system. To develop such a
system a trained dataset is required to classify an image. The trained dataset consists of
train results and test results. Whenever a user will upload an input file, the image is
temporarily stored in the database. This input file is then fed to the system and given to
CNN where CNN is coupled with a trained dataset. CNN consists of various
convolutional layers. Various features such as head, color, body, shape, and the entire
image of the bird are considered for classification to yield maximum accuracy. Each
alignment is given through a deep convolutional network to extract features out from
multiple layers of the network. Then an unsupervised algorithm called deep learning
using CNN is used to classify that image.

Further, a grey scale method is used to classify the image pixel by pixel. These features
are then aggregated and forwarded to classifier. Here, the input will be compared against
36
the trained dataset to generate possible results. During classification, an autograph is
generated which consist of nodes that ultimately forms a network. On basis of this
network, a score sheet is generated and with the help of score sheet output will be
produced.

37
 4.EXPERIMENTAL ANALYSIS AND RESULTS
4.1. System Configuration
4.1.1 Software Requirements:

Programming Language : Python

Operating System : Windows or Linux

Tools : Anaconda Navigator,Tensorflow,Keras

TENSORFLOW:
The standard name for Machine Learning in the Data Science industry is TensorFlow. It
facilitates building of both statistical Machine Learning solutions as well as deep learning
through its extensive interface of CUDA GPUs. The most basic data type of TensorFlow
is a tensor which is a multi-dimensional array.

It is an open-source toolkit that can be used for build machine learning pipelines so that
you can build scalable systems to process data. It provides support and functions for
various applications of ML such as Computer Vision, NLP and Reinforcement Learning.
TensorFlow is one of the must-know tools of Machine Learning for beginners.

KERAS:

Keras is an open-source neural network library that provides support for Python. It is
popular for its modularity, speed, and ease of use. Therefore, it can be used for fast
experimentation as well as rapid prototyping. It provides support for the implementation
of convolutional neural networks, Recurrent Neural Networks as well as both. It is
capable of running seamlessly on the CPU and GPU. Compared to more widely popular
libraries like TensorFlow and Pytorch, Keras provides user-friendliness that allows the
users to readily implement neural networks without dwelling over the technical jargon.

How to install Keras :

Before we go ahead with installing Keras, let us look at the installation of
TensorFlow.

TensorFlow installation:

38
Mainly, there are two stable TensorFlow versions:

TensorFlow: for running on CPU.

TensorFlow-GPU: for running on GPU

Although with the GPU we can make very heavy computations without
interrupting the work of the CPU, we are going to install the version for CPU for
being simpler and is the ideal one to take the first steps with TensorFlow.
There are also the same versions with the postfix nightly, which are unstable versions
in development but include the latest updates and developments.

So, to install the stable release of TensorFlow we have to write like this in our
terminal:

$ pip install TensorFlow

Installing Keras on Python:

Installing Keras is no different from installing any other library in Python:

$ pip install keras

By default, while installing Keras, Keras will use TensorFlow as its tensor
manipulation library

39
4.1.2 Hardware Requirements:

Processor : Intel Multicore Processor (i3 or i5 or i7)

RAM : 4GB or Above

Hard Disk : 100GB or Above

40
 4.2. Sample Code

1.System Training :
import numpy as np

import pandas as pd

from keras.preprocessing.image import ImageDataGenerator, load_img

from keras.utils import to_categorical

import matplotlib.pyplot as plt

import random

FAST_RUN = False

IMAGE_WIDTH=128

IMAGE_HEIGHT=128

IMAGE_SIZE=(IMAGE_WIDTH, IMAGE_HEIGHT)

IMAGE_CHANNELS=3 # RGB color

from keras.models import Sequential

from keras.layers import Conv2D, MaxPooling2D, Dropout, Flatten,

Dense, Activation, BatchNormalization

model = Sequential()

model.add(Conv2D(32, (3, 3), activation='relu',

input_shape=(IMAGE_WIDTH, IMAGE_HEIGHT,
IMAGE_CHANNELS)))

model.add(BatchNormalization())

model.add(MaxPooling2D(pool_size=(2, 2)))

41
model.add(Dropout(0.25))

model.add(Conv2D(64, (3, 3), activation='relu'))

model.add(BatchNormalization())

model.add(MaxPooling2D(pool_size=(2, 2)))

model.add(Dropout(0.25))

model.add(Conv2D(128, (3, 3), activation='relu'))

model.add(BatchNormalization())

model.add(MaxPooling2D(pool_size=(2, 2)))

model.add(Dropout(0.25))

model.add(Flatten())

model.add(Dense(512, activation='relu'))

model.add(BatchNormalization())

model.add(Dropout(0.5))

model.add(Dense(200, activation='softmax'))

model.compile(loss='categorical_crossentropy', optimizer='rmsprop',
metrics=['accuracy'])

from keras.callbacks import EarlyStopping, ReduceLROnPlateau

earlystop = EarlyStopping(patience=10)

learning_rate_reduction = ReduceLROnPlateau(monitor='val_acc',

patience=2,

verbose=1,

42
 factor=0.5,

min_lr=0.00001)

callbacks = [earlystop, learning_rate_reduction]

train_datagen = ImageDataGenerator(

rotation_range=15,

rescale=1./255,

shear_range=0.1,

zoom_range=0.2,

horizontal_flip=True,

width_shift_range=0.1,

height_shift_range=0.1

test_datagen = ImageDataGenerator(rescale=1./255)

x_train=train_datagen.flow_from_directory(r'C:\Users\user\Desktop\datas
et\traindata',

target_size = (128, 128),

batch_size = 32,

class_mode = 'categorical')

x_test=test_datagen.flow_from_directory(r'C:\Users\user\Desktop\dataset\t
estdata',

target_size = (128, 128),

batch_size = 32,
43
 class_mode = 'categorical')

print(x_train.class_indices)

model.fit_generator(x_train,

steps_per_epoch =130 ,

epochs = 15,

validation_data = x_test,

validation_steps = 20,callbacks=callbacks)

model.save("mybird.h5")

from keras.models import load_model

Matplotlib.pyplot:

Matplotlib.pyplot is a collection of command style functions that make matplotlib work

like MATLAB. Each pyplot function makes some change to a figure: e.g., creates a
figure, creates a plotting area in a figure, plots some lines in a plotting area, decorates
the plot with labels, etc. In matplotlib.pylot various states are preserved across function
calls, so that it keeps track of things like the current figure and plotting area, and the
plotting functions are directed to the current axes (please note that "axes" here and in
most places in the documentation refers to the axes part of a figure and not the strict
mathematical term for more than one axis).

Model = Sequential ():

The sequential model is a linear stack of layers. You can create a sequential model by
passing a list of layer instances to the constructor:

from keras.models import Sequential

from keras.layers import Dense, Activation
model = Sequential([

44
 Dense(32, input_shape=(784,)),
Activation('relu'),
Dense(10),
Activation('softmax'),
])

You can also simply add layers via the .add() method:

model = Sequential()
model.add(Dense(32, input_dim=784))
model.add(Activation('relu'))

You can save a model built with the Functional API into a single file. You can later
recreate the same model from this file, even if you no longer have access to the code
that created the model.

This file includes:

 The model's architecture

 The model's weight values (which were learned during training)

 The model's training config (what you passed to compile), if any

 The optimizer and its state, if any (this enables you to restart training where you
left)

Batch Normalization ():

One possible reason for this difficulty is the distribution of the inputs to layers deep in
the network may change after each mini-batch when the weights are updated. This can
cause the learning algorithm to forever chase a moving target. This change in the
distribution of inputs to layers in the network is referred to the technical name “internal
covariate shift.”
Batch normalization is a technique for training very deep neural networks that
standardizes the inputs to a layer for each mini-batch. This has the effect of stabilizing
the learning process and dramatically reducing the number of training epochs required
to train deep networks.
45
 Conv2D ():

This layer creates a convolution kernel that is convolved with the layer input to produce
a tensor of outputs. If use_bias is True, a bias vector is created and added to the outputs.
Finally, if activation is not None, it is applied to the outputs as well.

ReduceLROnPlateau ():

Reduce learning rate when a metric has stopped improving. Models often benefit from
reducing the learning rate by a factor of 2-10 once learning stagnates. This call back
monitors a quantity and if no improvement is seen for a 'patience' number of epochs, the
learning rate is reduced.

Dropout ():

Dropout is a technique where randomly selected neurons are ignored during training.
They are “dropped-out” randomly. This means that their contribution to the activation of
downstream neurons is temporally removed on the forward pass and any weight updates
are not applied to the neuron on the backward pass.

ImageDataGenerator ():

Data augmentation is what Kera’s’ “ImageDataGenerator” class implements. Using this

type of data augmentation, we want to ensure that our network, when trained, sees new
variations of our data at each and every epoch.

1. An input batch of images is presented to the ImageDataGenerator.

2. The ImageDataGenerator transforms each image in the batch by a series of random

translations, rotations, etc.

3. The randomly transformed batch is then returned to the calling function.

46
2.Extraction of Images:

from keras.models import load_model

import numpy as np

import cv2

model= load_model('mybird.h5')

model.compile(optimizer='adam',loss='categorical_crossentropy',metrics=[
'accuracy'])

from skimage.transform import resize

def detect(frame):

try:

img = resize(frame,(128,128))

img=np.expand_dims(img,axis=0)

if(np.max(img)>1):

img= img/255.0

prediction = model.predict(img)

print(prediction)

prediction_class= model.predict_classes(img)

print(prediction_class)

except AttributeError:

print("Shape not found")

from matplotlib.pyplot import *

47
frame=cv2.imread(r'C:\Users\user\Desktop\dataset\testdata\002.Painted_B
unting\Painted_Bunting_0091_15198.jpg')#path of image

imshow(frame)

data= detect(frame)

frame=cv2.imread(r'C:\Users\user\Desktop\dataset\testdata\009.Brewer_Bl
ackbird\Brewer_Blackbird_0112_2340.jpg')#path of image

imshow(frame)

data= detect(frame)

frame=cv2.imread(r'C:\Users\user\Desktop\dataset\testdata\050.Eared_Gre
be\Eared_Grebe_0088_34067.jpg')#path of image

imshow(frame)

data= detect(frame)

CV2:

OpenCV, which is an image and video processing library with bindings in C++, C,
Python, and Java. OpenCV is used for all sorts of image and video analysis, like facial
recognition and detection, license plate reading, photo editing, advanced robotic vision,
optical character recognition, and a whole lot more.

imRead ():

cv2.imread() method loads an image from the specified file. If the image cannot be read
(because of missing file, improper permissions, unsupported or invalid format) then this
method returns an empty matrix.

imShow ():

cv2.imshow() method is used to display an image in a window. The window

automatically fits to the image size.

48
3.gui.py:

import numpy as np

from keras.preprocessing import image

from tkinter import *

from PIL import ImageTk, Image

from tkinter import filedialog

import os

from keras.models import load_model

classifier = load_model(r'C:\Users\user\Desktop\Birds-identification-
master\mybird.h5')

classifier.compile(optimizer = 'adam', loss = 'categorical_crossentropy',

metrics = ['accuracy'])

root = Tk()

root.geometry("550x300+300+150")

root.resizable(width=True, height=True)

def openfn():

filename = filedialog.askopenfilename(title='open')

return filename

def open_img():

x = openfn()

test_image = image.load_img(x, target_size = (128, 128))

test_image = image.img_to_array(test_image)
49
 test_image = np.expand_dims(test_image, axis = 0)

result = classifier.predict_classes(test_image)

print(result)

index=['002.Painted_Bunting','009.Brewer_Blackbird','050.Eared_Grebe']

label = Label( root, text="Prediction : "+index[result[0]])

label.pack()

img = Image.open(x)

img = img.resize((250, 250), Image.ANTIALIAS)

img = ImageTk.PhotoImage(img)

panel = Label(root, image=img)

panel.image = img

panel.pack()

btn = Button(root, text='open image', command=open_img).pack()

root.mainloop()

tKinter:

The tkinter package (“Tk interface”) is the standard Python interface to the Tk GUI
toolkit. Both Tk and tkinter are available on most Unix platforms, as well as on Windows
systems. (Tk itself is not part of Python; it is maintained at Active State.)

Running python -m tkinter from the command line should open a window demonstrating
a simple Tk interface, letting you know that tkinter is properly installed on your system,
and also showing what version of Tcl/Tk is installed, so you can read the Tcl/Tk
documentation specific to that version.

50
Label ():

The tkinter label widgets can be used to show text or an image to the screen. A label can
only display text in a single font. The text can span multiple lines. We can put any text
in a label and we can have multiple labels in a window (just like any widget can be placed
multiple times in a window).

Button ():

The Button widget is used to add buttons in a Python application. These buttons can
display text or images that convey the purpose of the buttons. You can attach a function
or a method to a button which is called automatically when you click the button.

mainloop ():

There is a method known by the name mainloop () is used when your application is ready
to run. mainloop () is an infinite loop used to run the application, wait for an event to
occur and process the event as long as the window is not closed.

51
4.3. Screenshots

When we run the code, the dataset is divided into training and testing. First, it trains the
system then it tests for accuracy then it asks for user input. It gives the label as a result
with an input image. Following are the steps of execution:

INPUT:

Fig 22. Inserting User Input

52
OUTPUT:

1.Output-Species Name and Input Image : Painted_Bunting

Here first we uploaded the image of Painted_Bunting of different images as an input and
gone through CNN classification to obtain the name of the bird. Finally it gives us the
bird name with image as an output.

53
2.Output-Species Name and Input Image :Brewer_Blackbird
Here we uploaded the image of Brewer_Blackbird of different images as an input and
gone through CNN classification to obtain the name of the bird. Finally it gives us the
bird name with image as an output.

54
3.Output-Species Name and Input Image :Eared_Grebe
Here we uploaded the image of Eared_Grebe of different images as an input and gone
through CNN classification to obtain the name of the bird. Finally it gives us the bird
name with image as an output.

55
4.3. Experimental Analysis/Testing
The evaluation of the proposed approach for bird species classification by considering
color features and parameters such as size, shape, etc. of the bird on the Caltech-UCSD
Birds 200 (CUB-200-2011) dataset. This is an image dataset annotated with 200 bird
species which includes 11,788 annotated images of birds where each image is annotated
with a rough segmentation, a bounding box, and binary attribute annotations. In this the
training of dataset is done by using Google-Collab, which is a platform to train dataset
by uploading the images from your local machine or from the Google drive.
After training labeled dataset is ready for classifiers for image processing. There
are probably average 200 sample images per species are included in dataset of 3 species
which are directly captured in their natural habitat hence also include the environmental
parameters in picture such as grass, trees and other factors. Here bird can identify in
their any type of position as main focus is on the size, shape and color parameter. That
is the image converted into number of pixels by using gray scale method, where value
for each pixel is created and value-based nodes are formed which also referred as
neurons. These neurons relatively defined the structure of matched pixels is simply like
graph of connected nodes.
According to the nodes formed the autograph is generated which understandable
by Tensorflow to classify the image. This autograph is then taken by classifiers and
image is compared with the pre trained dataset images of Caltech UCSD.

56
Tested Results:

Sr. No Bird name (species) Tested Results Accuracy (%)

1. Painted_Bunting 10/10 95%

2. Brewer_Blackbird 10/10 95%

3. Eared_Grebe 7/10 70%

Total 27/30 90%

Table 2. Efficiency Table

57
 5.CONCLUSION AND FUTURE WORK
5.1. Conclusion:
The main idea behind developing the identification website is to build awareness
regarding bird-watching, bird and their identification, especially birds found in India. It
also caters to the need of simplifying the bird identification process and thus making
bird-watching easier. The technology used in the experimental setup is Convolutional
Neural Networks (CNN). It uses feature extraction for image recognition. The method
used is good enough to extract features and classify images.

The main purpose of the project is to identify the bird species from an image given as
input by the user. We used CNN because it is suitable for implementing advanced
algorithms and gives good numerical precision accuracy. It is also general-purpose and
scientific. We achieved an accuracy of 85%-90%. We believe this project extends a great
deal of scope as the purpose meets. In wildlife research and monitoring, this concept can
be implemented in-camera traps to maintain the record of wildlife movement in specific
habitat and behaviour of any species.

58
5.2. Future Work:
1) Create an android/iOS app instead of website which will be more convenient to user.

2) System can be implemented using cloud which can store large amount of data for
comparison and provide high computing power for processing (in case of Neural
Networks).

59
 REFERENCES
[1] Indian Birds [Online] https://play.google.com /store /apps /details ?id =
com.kokanes.birds info & hl = en

[2] Bird Watching Apps: Five Useful Apps to Get Started With Birding [Online]
https://gadgets.ndtv.com/apps/features/bird-watching-apps-five-useful-apps-to-get-
started-with-birding-1640679

[3] Transfer learning using Alex Net [Online] https://in.mathworks.com /help /nnet
/examples / transfer-learning-using- alexnet.html

[4] Feature extraction using AlexNet [Online] https://in.mathworks.com /help /nnet

/examples / feature-extraction-using-alexnet.html # d119e4167

[5] Dipta Das, Sourya & Kumar, Akash” An Exploration of Computer Vision Techniques
for Bird Species Classification”, December 15, 2017.
[6] N. Dalal and B. Triggs, ”Histograms of oriented gradients for human detection,”
IEEE Computer Society Conference on Computer Vision and Pattern Recognition
(CVPR), San Diego, CA, 2005.
[7] C. Wah et al. The Caltech-UCSD Birds-200-2011 Dataset. Tech. rep. CNS-TR-2011-
001. California Institute of Technology, 2011.
[8] Thomas Berg, Jiongxin Liu et. Al ” Large-scaleFine-
grainedVisualCategorizationofBirds”,
[9] Yuning Chai Electrical Engineering Dept. ETH Zurich,Victor Lempitsky Dept. of
Engineering Science University of Oxford, Andrew Zisserman Dept. of Engineering
Science University of Oxford BiCoS:A BiSegmentation Method for Image
Classification.

[10] Tóth, B.P. and Czeba, B., 2016, September. Convolutional Neural Networks for
Large-Scale Bird Song Classification in Noisy Environment. In CLEF (Working Notes)
(pp. 560-568).
[11] Fagerlund, S., 2007. Bird species recognition using support vector machines.
EURASIP Journal on Applied Signal Processing, 2007(1), pp.64-64.

60
[12] U.D.Nadimpalli, R.R.Price, S.G.Hall, and P.Bomma,” A Comparison of image
processing techniques for bird recognition”, Biotechnology Progress, Vol. 22, no. 1,pp.
9-13,2006.

[13] Toth, B.P. and Czeba, B.,2016, September. Convolutional Neural Networks for
Large-Scale Bird Song Classification in Noisy Environment. In CLEF (Working Notes)
(pp. 560-568).

[14] Elias Sprengel, Martin Jaggi, Yannic Kilcher, and Thomas Hofmann. Audio Based
Bird Species Identification using Deep Learning Techniques. 2016.

[15] Jaderick P. Pabico, Anne Muriel V. Gonzales, Mariann Jocel S. Villanueva, and
Arlene a. Mendoza. Automatic identification of animal breeds and species using
bioacoustics and artificial neural networks.arXiv preprint, pages 1-17, 2015.

[10] T. S. Brandes, “Automated sound recording and analysis techniques for bird surveys
and conservation,” Bird Conservation International, Vol. 18, pp.v 163-173, 2008.

[11] Stefan Kahl, Thomas Wilhelm-Stein, Hussein Hussein, Holger Klinck, Danny
Kowerko, Marc Fitter, and Maximilian Eibl Large-Scale Bird Sound Classification using
Convolutional Neural Networks.

[12] Daniel T. Blumstein, Daniel J. Mennill, Patrick Clemins, Lewis Girod, Kung Yao,
Gali Patricelli, Jill L. Deppe, Alan H. Krakauer, Christopher Clark, Kathryn A.
Cortopassi, Sean F. Hanser, Brenda Mccowan, Andreas M. Ali, and ALexander N G
Kirsch. Acoustic monitoring in terrestrial environments using microphone arrays:
applications, technological considerations, and prospectus. Journal of Applied Ecology,
48(3):758-767, 2011.

[13] R. Bardeli, D. Wolff, F. Kurth, M. Koch, K.-H. Tauchert, and K.-H. Frommolt,
“Detecting bird songs in a complex acoustic environment and application to bioacoustic
monitoring,” Patt Recog Letters, vol. Vol.31, pp. Pp. 1524-1534, 2010. 64.

[15] Yann LeCun, Yoshua Bengio, and Hinton Geoffrey. Deep learning. Nature
Methods, 13(1):35-35, 2015.

61
[16] Sergey loffe and Christian Szegedy. Batch Normalization: Accelerating Deep
Network Training by Reducing Internal Covariate Shift. arXiv:1502.03167, pages 1-11,
2015.

62
Published by : International Journal of Engineering Research & Technology (IJERT)
http://www.ijert.org ISSN: 2278-0181
Vol. 8 Issue 04, April-2019

Bird Species Identification using Deep Learning

Prof. Pralhad Gavali, Ms. Prachi Abhijeet Mhetre, Ms. Neha Chandrakhant Patil,
Ms. Nikita Suresh Bamane, Ms. Harshal Dipak Buva
Rajarambapu Institute of Technology,
Rajaramnagar, India

Abstract: Now a day some bird species are being found rarely algorithms that carries out such a task in an automatic
and if found classification of bird species prediction is fashion.
difficult. Naturally, birds present in various scenarios appear Since many decades, ornithologists are facing problems
in different sizes, shapes, colors, and angles from human in bird species identification. Ornithologists require
perspective. Besides, the images present strong variations to
studying all the details of birds such as their existence in
identify the bird species more than audio classification. Also,
human ability to recognize the birds through the images is environment, their biology, their distribution, their
more understandable. So this method uses the Caltech-UCSD ecological impact, etc. Bird identification is usually done
Birds 200 [CUB-200-2011] dataset for training as well as by ornithology experts based on classification proposed by
testing purpose. By using deep convolutional neural network Linnaeus: Kingdom, Phylum, Class, Order, Family, and
(DCNN) algorithm an image converted into grey scale format Species [1].
to generate autograph by using tensor flow, where the As image based classification systems are
multiple nodes of comparison are generated. These different improving the task of classifying, objects is moving into
nodes are compared with the testing dataset and score sheet is
datasets with far more categories such as Caltech-UCSD.
obtained from it. After analyzing the score sheet it can
predicate the required bird species by using highest score.
Recent work has seen much success in this area. Caltech-
Experimental analysis on dataset (i.e. Caltech-UCSD Birds UCSD Birds 200(CUB-200-2011) is a well-known dataset
200 [CUB-200-2011]) shows that algorithm achieves an for bird images with photos of 200 categories [4]. The
accuracy of bird identification between 80% and 90%.The dataset contains birds that are mostly found in Northern
experimental study is done with the Ubuntu 16.04 operating America. Caltech-UCSD Birds 200 consists of 11,788
system using a Tensor flow library. images and annotations like 15 Part Locations, 312 Binary
Attributes, 1 Bounding Box.
Index Terms: Autograph; Caltech-UCSD; grey scale pixels; In this paper, instead of recognizing a large
Tensorflow
number of disparate categories, the problem of recognizing
a large number of classes within one category is
I. INTRODUCTION1
investigated – that of birds. Classifying birds pose an extra
BIRD behavior and population trends have become an challenge over categories, because of the large similarity
important issue now a days. Birds help us to detect other between classes. In addition, birds are non-rigid objects that
organisms in the environment (e.g. insects they feed on) can deform in many ways, and consequently there is also a
easily as they respond quickly to the environmental changes large variation within classes. Previous work on bird
[2]. But, gathering and collecting information about birds classification has deal with a small number of classes, or
requires huge human effort as well as becomes a very through voice.
costlier method. In such case, a reliable system that will
provide large scale processing of information about birds
and will serve as a valuable tool for researchers,
governmental agencies, etc. is required. So, bird species
identification plays an important role in identifying that a
particular image of bird belongs to which species. Bird
species identification means predicting the bird species
belongs to which category by using an image.
The identification can be done through image, audio or
video. An audio processing technique makes it possible to
identify by capturing the audio signal of birds. But, due to
the mixed sounds in environment such as insects, objects
from real world, etc. processing of such information
becomes more complicated. Usually, human beings find
images more effective than audios or videos. So, an
approach to classify bird using an image over audio [8] or
video is preferred. Bird species identification is a
challenging task to humans as well as to computational Figure No.1: Process of classification

IJERTV8IS040112 www.ijert.org 68
(This work is licensed under a Creative Commons Attribution 4.0 International License.)
Published by : International Journal of Engineering Research & Technology (IJERT)
http://www.ijert.org ISSN: 2278-0181
Vol. 8 Issue 04, April-2019

The figure 1 represents the process of detecting the bird Birds 200), Deep Convolutional Neural Network,
from image. The image is getting upload first then from that Unsupervised learning algorithm, etc.
image the various alignments will be considered such as
head, body, color, beak and entire image. Further, each Algorithm:
alignment is given through deep convocational network to In this experiment, unsupervised learning algorithm has
extract features out from multiple layers of network [3]. been used for developing the system, because the inputted
After that representation of the image will get consider. image defined is not known. Also, the data which is given
Then on the basis of it the classifying result will get to unsupervised learning algorithm are not labeled, i.e. only
generated (i.e. features are aggregated to transfer it to the input variables(X) are given with no corresponding
classifier) and the bird species will get found. output variables. In unsupervised learning, algorithms
This paper is assembled in format: Section II covers the discover interesting structures in the data themselves. In
parameters one can consider while identifying a bird detail, clustering is used for dividing the data into several
visually. Section III contains methodologies used for groups [4].
developing the proposed system. Section IV represents In depth, deep learning models used to find vast number
overall flow of the system in detail. of neurons. Deep learning algorithms learn more about the
image as it goes through each neural network layer. For
II. BACKGROUND classifying Neural Network is used. Figure 2 represents
Basically bird identification is done visually or layers of neural networks for feature extraction. The neural
acoustically. The main visual components comprise of network is a framework for many machine learning
bird’s shape, its wings, size, pose, color, etc. However, algorithms. Neural networks consist of vector of weights
while considering the parameters time of year must be (W) and the bias (B).
taken into consideration because bird’s wings changes
according to their growth. The acoustics components
comprise the songs and call that birds make [7]. The marks
that distinguish one bird from another are also useful, such
as breast spots, wing bars which are described as thin lines
along the wings, eye rings, crowns, eyebrows. The shape of
the beak is often an important aspect as a bird can Figure No. 2: Three layers of Neural Network
recognized uniquely. The characteristics of bird such as
shape and posture are the mostly used to identify birds. In deep learning, convolutional neural network (CNN) is
Mostly experts can identify a bird from its silhouette a class of deep neural network mostly used for analyzing
because this characteristic is difficult to change. A bird can visual images. It consists of an input layer and output layer
also be differentiated using its tail. The tail can be as well as multiple hidden layers. Every layer is made up of
recognized in many ways such as notched, long and group of neurons and each layer is fully connected to all
pointed, or rounded. Sometimes legs are also used for neurons of its previous layer. The output layer is
recognizing an image in format long, or short [10]. responsible for prediction of output. The convolutional
By considering a single parameter will not yield an layer takes an image as input, and produces a set of feature
accurate result. So, multiple parameters are to be maps as output [2]. The input image can contain multiple
considered in order to get appropriate output. The size of a channels such as color, wings, eyes, beak of birds which
bird in an image varies depending upon factors such as the means that the convolutional layer perform a mapping from
resolution, distance between the birds and the capturing 3D volume to another 3D volume. 3D volumes considered
device, and the focal distance of the lens. Therefore, based are width, height, depth. The CNN have two components:
on a practical observation for large number of images, 1) Feature extraction part: features are detected when
images are differentiated on the basis of color which network performs a series of convolutional and pooling
consists of various pixel. In depth it is found that greater operation.
the image quality greater is its accuracy. 2) Classification part: extracted features are given to fully
The automatic bird species identification for bird images connected layer which acts as classifier.
project present a series of comparison conducted in a CUB-
200 dataset composed of more than 6,000 images with 200
different category [6]. In this paper, they have considered
two different color spaces, RGB and HSV, and a different
number of species to be classified. If the image consists of
more than 70% of the pixels the accuracy of output was
ranging from 8.82% to 0.43% [1].

III. METHODOLOGY
For developing the system certain methodologies have
been used. They are as follows: Dataset (Caltech-UCSD

IJERTV8IS040112 www.ijert.org 69
(This work is licensed under a Creative Commons Attribution 4.0 International License.)
Published by : International Journal of Engineering Research & Technology (IJERT)
http://www.ijert.org ISSN: 2278-0181
Vol. 8 Issue 04, April-2019

of images: 11,788, Annotations per image: 15 Part

Locations, 312 Binary Attributes, 1 Bounding Box

Figure No. 4: Caltech-UCSD Dataset

Figure No. 3: Convolutional Neural Network Layers

IV. PROPOSED APPROACH
CNN consists of four layers: convolutional layer,
activation layer, pooling layer and fully connected.
Convolutional layer allows extracting visual features from
an image in small amounts. Pooling is used to reduce the
number of neurons from previous convolutional layer but
maintaining the important information. Activation layer
passes a value through a function which compresses values
into range. Fully connected layer connects a neuron from
one layer to every neuron in another layer. As CNN
classifies each neuron in depth, so it provides more
accuracy.
Image classification: image classification in machine
learning is commonly done in two ways:
1) Gray scale
2) Using RGB values
Normally all the data is mostly converted into gray
scale. In gray scale algorithm, computer will assign values
to each pixel based on how the value of the pixel is it. All
the pixel values are put into an array and the computer will
perform operation on that array to classify the data.

Library:
Tensorflow is open source software library which is
created by Google. It gives developers to control each
neuron known as a “node”, so that the parameters can be
adjusted to achieve desired performance. Tensorflow has
many built-in libraries for image classification [3].
Tensorflow is responsible for creating an autograph which
consists series of processing nodes. Each processing node
in the graph represents an operation such as mathematical
operation and connection or edge between nodes. With the Figure No.5: Flow of System
help of python language Tensorflow provides programmer
to perform these operations. The above figure no. 5 represents the actual flow of the
proposed system. To develop such system a trained dataset
Dataset: is required to classify an image. Trained dataset consists of
A dataset is a collection of data. For performing action two parts trained result and test result. The dataset has to be
related to birds a dataset named Caltech-UCSD Birds 200 retrained to achieve higher accuracy in identification using
(CUB-200-2011) is used. It is an extended version of retrain.py in Google Collab. The training dataset is made
the CUB-200 dataset, with roughly double the number of using 50000 steps taking into consideration that higher the
images per class and also has new part location annotations number of steps higher is its accuracy. The accuracy of
for higher accuracy [8]. The detailed information about the training dataset is 93%. The testing dataset consists of
dataset is as follows: Number of categories: 200, Number nearly 1000 images with an accuracy of 80%. Further,

IJERTV8IS040112 www.ijert.org 70
(This work is licensed under a Creative Commons Attribution 4.0 International License.)
Published by : International Journal of Engineering Research & Technology (IJERT)
http://www.ijert.org ISSN: 2278-0181
Vol. 8 Issue 04, April-2019

dataset is validated with an accuracy of 75% to increase the For example, consider below Figure No.6 as input image
performance of system. given to the system for classification of bird which belongs
Whenever a user will upload an input file on website, the to Northern America. Let’s see how it is being evaluated.
image is temporarily stored in database. This input file is
then feed to system and given to CNN where CNN is
coupled with trained dataset. A CNN consists of various
convolutional layers. Various alignments/features such as
head, body, color, beak, shape, entire image of bird are
considered for classification to yield maximum accuracy.
Each alignment is given through deep convocational
network to extract features out from multiple layers of
network. Then an unsupervised algorithm called deep
learning using CNN is used to classify that image.
Further, a grey scale method is used to classify the image
pixel by pixel. These features are then aggregated and
forwarded to classifier. Here, the input will be compared
against the trained dataset to generate possible results.
During classification, an autograph is generated which
consist of nodes that ultimately forms a network. On basis Figure No.6: Input Image
of this network, a score sheet is generated and with the help The system generates following scoresheet after
of score sheet output will be produced. classification which tells us the possibilities that above
selected bird belongs to which species.
V. EXPERIMENTAL ANALYSIS
Table No. 1: Score Sheet
The evaluation of the proposed approach for bird species Sr.No Species Score Obtained
classification by considering color features and parameters 1 Elegant tern 0.00943
such as size, shape, etc. of the bird on the Caltech-UCSD 2 Red faced cormorant 0.00924
Birds 200 (CUB-200-2011) dataset. This is an image 3 Brant cormorant 0.0085
dataset annotated with 200 bird species which includes 4 Pelagic cormorant 0.0082
11,788 annotated images of birds where each image is 5 White pelican 0.00808
annotated with a rough segmentation, a bounding box, and
binary attribute annotations. In this the training of dataset is The table no.1 shows the scoresheet based on the result
done by using Google-Collab, which is a platform to train generated by the system. After analysis of these result it has
dataset by uploading the images from your local machine or observe that,the species those are having the highest score
from the Google drive. has been predicted as a required species. this result can be
After training labeled dataset is ready for classifiers for shown in the follwing graph.
image processing. There are probably average 200 sample
images per species are included in dataset of 5 species
which are directly captured in their natural habitat hence
also include the environmental parameters in picture such
as grass, trees and other factors. Here bird can identify in
their any type of position as main focus is on the size, shape
and color parameter. Firstly these factors are considered for
segmentation where RGB and gray scale methods are used
for histogram. That is the image converted into number of
pixels by using gray scale method, where value for each
pixel is created and value based nodes are formed which
also referred as neurons. These neurons relatively defined
the structure of matched pixels is simply like graph of
connected nodes. Figure No.7: Graph format of Score sheet
According to the nodes formed the autograph is Comparison:
generated which understandable by Tensorflow to classify Sr. No Method Accuracy
the image. This autograph is then taken by classifiers and 1) Pose Norm 80%
2) Part-based R-CNN 76.4%
image is compared with the pre trained dataset images of 3) Multiple granularity CNN 80%
Caltech UCSD and the score sheet is generated. The score 4) Diversified visual attention network 79%
sheet is a result which contains top 5 match results by (DVAN)
which the highest matching value of score sheet is the result 5) The deep LAC localization, alignment, 80.3%
of bird species. Here a trial has made to implement 80% and classification
6) Deep Learning Using CNN 85%
accuracy in result by training the Caltech UCSD.

IJERTV8IS040112 www.ijert.org 71
(This work is licensed under a Creative Commons Attribution 4.0 International License.)
Published by : International Journal of Engineering Research & Technology (IJERT)
http://www.ijert.org ISSN: 2278-0181
Vol. 8 Issue 04, April-2019

REFERENCES

[1] Tóth, B.P. and Czeba, B., 2016, September. Convolutional Neural
Networks for Large-Scale Bird Song Classification in Noisy
Environment. In CLEF (Working Notes) (pp. 560-568).
[2] Fagerlund, S., 2007. Bird species recognition using support
vector machines. EURASIP Journal on Applied Signal
Processing, 2007(1), pp.64-64.
[3] Pradelle, B., Meister, B., Baskaran, M., Springer, J. and Lethin,
R., 2017, November. Polyhedral Optimization of TensorFlow
Computation Graphs. In 6th Workshop on Extreme-scale
Programming Tools (ESPT-2017) at The International
Conference for High Performance Computing, Networking,
Storage and Analysis (SC17).
[4] Cireşan, D., Meier, U. and Schmidhuber, J., 2012. Multi-column
deep neural networks for image classification. arXiv preprint
arXiv:1202.2745.
[5] Andr´eia Marini, Jacques Facon and Alessandro L. Koerich
After analyzing the data, it is observed that if a single Postgraduate Program in Computer Science (PPGIa) Pontifical
parameter is used the accuracy obtained is lesser. But, if a Catholic University of Paran´a (PUCPR) Curitiba PR, Brazil
80215–901 Bird Species Classification Based on Color Features
combined method is used that is by considering parameters
[6] Image Recognition with Deep Learning Techniques ANDREI-
such as pose, wings, color, beak, legs, etc. the accuracy of PETRU BĂRAR, VICTOR-EMIL NEAGOE, NICU SEBE
the project get increase. Faculty of Electronics, Telecommunications & Information
Technology Polytechnic University of Bucharest.
VI. CONCLUSION [7] Xception: Deep Learning with Depthwise Separable
Convolutions François Chollet Google, Inc.
The present study investigated a method to identify the [8] Zagoruyko, S. and Komodakis, N., 2016. Paying more attention
bird species using Deep learning algorithm (Unsupervised to attention: Improving the performance of convolutional neural
networks via attention transfer. arXiv preprint arXiv:1612.03928.
Learning) on the dataset (Caltech-UCSD Birds 200) for [9] Inception-v4,Inception-ResNetand the Impact of Residual
classification of image. It consists of 200 categories or Connectionson Learning Christian Szegedy, Sergey Ioffe,Vincent
11,788 photos. The generated system is connected with a Vanhoucke ,Alexandrr A.Alemi
user-friendly website where user will upload photo for [10] Stefan Kahl, Thomas Wilhelm-Stein, Hussein Hussein, Holger
Klinck, Danny Kowerko, Marc Ritter, and Maximilian Eibl
identification purpose and it gives the desired output. The Large-Scale Bird Sound Classification using Convolutional
proposed system works on the principle based on detection Neural Networks
of a part and extracting CNN features from multiple [11] Thomas Berg, Jiongxin Liu, Seung Woo Lee, Michelle L.
convolutional layers. These features are aggregated and Alexander, David W. Jacobs, and Peter N. Belhumeur Birdsnap:
Large-scaleFine-grainedVisualCategorizationofBirds
then given to the classifier for classification purpose. On [12] Bo Zhao, Jiashi Feng Xiao Wu Shuicheng Yan A Survey on
basis of the results which has been produced, the system Deep Learning-based Fine-grained Object Classification and
has provided the 80% accuracy in prediction of finding bird Semantic Segmentation
species. [13] Yuning Chai Electrical Engineering Dept. ETH Zurich,Victor
Lempitsky Dept. of Engineering Science University of Oxford,
Andrew Zisserman Dept. of Engineering Science University of
VII. FUTURE SCOPE Oxford BiCoS:A BiSegmentation Method for Image
1) Create an android/ios app instead of website which Classification.
will be more convenient to user.
2) System can be implemented using cloud which can
store large amount of data for comparison and
provide high computing power for processing (in
case of Neural Networks).

IJERTV8IS040112 www.ijert.org 72
(This work is licensed under a Creative Commons Attribution 4.0 International License.)
© 2020 IJRAR April 2020, Volume 7, Issue 2 www.ijrar.org (E-ISSN 2348-1269, P- ISSN 2349-5138)

IDENTIFICATION OF BIRD SPECIES USING

CONVOLUTIONAL NEURAL NETWORKS
Ms.Satya Sai,Prof.Sivaranjani,Mr.Apuroop,Ms.Mounika,Mr.Sai Kumar
Anil Neerukonda Institute of Technology And
Sciences,Visakhapatnam,India

ABSTRACT

In our world, there are above 9000 bird species. Some bird species are being found rarely and if found also prediction
becomes very difficult. In order to overcome this problem, we have an effective and simple way to recognize these
bird species based on their features. Also, the human ability to recognize the birds through the images is more
understandable than audio recognition. So, we have used Convolutional Neural Networks (CNN). CNN’s are the strong
assemblage of machine learning which have proven efficient in image processing. In this paper, a CNN system
classifying bird species is presented and uses the Caltech-UCSD Birds 200 [CUB-200-2011] dataset for training as
well as testing purpose. By establishing this dataset and using the algorithm of similarity comparison, this system is
proved to achieve good results in practice. By using this method, everyone can easily identify the name of the particular
bird which they want to know.

Keywords: Bird species, Machine Learning, Convolutional Neural Networks.

INTRODUCTION

Bird behavior and populace patterns have become a significant issue nowadays. Birds help us to recognize different
life forms on the earth effectively as they react rapidly to ecological changes. Be that as it may, assembling and
gathering data about bird species requires immense human exertion just as it turns into an extremely costly technique.
In such a case, a solid framework that will give enormous scale preparation of data about birds and will fill in as a
significant apparatus for scientists, legislative offices, and so forth is required. In this way, bird species distinguishing
proof assumes a significant job in recognizing that a specific picture of birds has a place with which categories. Bird
species identification means predicting the bird species belongs to which category by using an image.

The recognition of bird species can be possible through a picture, audio or video. An audio processing method makes
it conceivable to recognize by catching the sound sign of different birds. Be that as it may, because of the blended
sounds in condition, for example, creepy crawlies, objects from the real world, and so forth handling of such data turns
out to be progressively convoluted. Normally, people discover images more effectively than sounds or recordings. So,
an approach to classify birds using an image over audio or video is preferred. Bird species identification is a challenging
task to humans as well as to computational procedures that carry out such a task in an automated fashion.

As image-based classification systems are improving the task of classifying, objects are moving into datasets with far
more categories such as Caltech-UCSD. Recent work has seen much success in this area. Caltech UCSD Birds
200(CUB-200-2011) is a well-known dataset for bird images with photos of 200 categories. The dataset contains birds
that are mostly found in Northern America. Caltech-UCSD Birds 200 consists of 11,788 images and annotations like
15 Part Locations, 312 Binary Attributes, 1 Bounding Box.

In this paper, rather than recognizing an oversized number of disparate categories, the matter of recognizing an
oversized number of classes within one category is investigated – that of birds. Classifying birds pose an additional
challenge over categories, as a result of the massive similarity between classes. additionally, birds are non-rigid

IJRAR2004172 International Journal of Research and Analytical Reviews (IJRAR) www.ijrar.org 260
© 2020 IJRAR April 2020, Volume 7, Issue 2 www.ijrar.org (E-ISSN 2348-1269, P- ISSN 2349-5138)
objects which will deform in many ways and consequently there's also an oversized variation within classes. Previous
work on bird classification has taken care of a little number of classes, or through voice.

LITERATURE SURVEY

John Martinsson et al (2017)[1], presented the CNN algorithm and deep residual neural networks to detect an image
in two ways ie., based on feature extraction and signal classification. They did an experimental analysis for datasets
consisting of different images. But, their work didn’t consider the background species. In Order to identify the
background species larger volumes of training data are required, which may not be available.

Juha Niemi, Juha T Tanttu et al (2018)[2], proposed a Convolutional neural network trained with deep learning
algorithms for image classification. It also proposed a data augmentation method in which images are converted and
rotated in accordance with the desired color. The final identification is based on a fusion of parameters provided by
the radar and predictions of the image classifier.

Li Jian, Zhang Lei et al (2014)[3], proposed an effective automatic bird species identification based on the analysis of
image features. Used the database of standard images and the algorithm of similarity comparisons.

Madhuri A.Tayal, Atharva Magrulkar et al (2018)[4], developed a software application that is used to simplify the bird
identification process. This bird identification software takes an image as an input and gives the identity of the bird as
an output. The technology used is transfer learning and MATLAB for the identification process.

Andreia Marini, Jacques Facon et al (2013)[5], proposed a novel approach based on color features extracted from
unconstrained images, applying a color segmentation algorithm in an attempt to eliminate background elements and
to delimit candidate regions where the bird may be present within the image. Aggregation processing was employed
to reduce the number of intervals of the histograms to a fixed number of bins. In this paper, the authors experimented
with the CUB-200 dataset and results show that this technique is more accurate.

Marcelo T.Lopes, Lucas L.Gioppo et al (2011)[6], focused on the automatic identification of bird species from their
audio recorded song. Here the authors dealt with the bird species identification problem using signal processing and
machine learning techniques with the MARSYAS feature set. Presented a series of experiments conducted in a
database composed of bird songs from 75 species out of which problem obtained in performance with 12 species.

Peter Jancovic and Munevver Kokuer et al (2012)[7], investigated acoustic modeling for recognition of bird species
from audio field recordings. Developed a hybrid deep neural network hidden Markov model(DNN-HMM). The
developed models were employed for bird species identification, detection of specific species and recognition of
multiple bird species vocalizing in a given recording. In this paper, the authors achieved an identification accuracy of
98.7% and recognition accuracy of 97.3%.

Mario Lasseck et al (2013)[8], presented deep convolutional neural networks and data augmentation techniques for
audio-based bird species identification. In this paper, the author used the Xeno-Canto set of audio recordings of bird
species.

IJRAR2004172 International Journal of Research and Analytical Reviews (IJRAR) www.ijrar.org 261
© 2020 IJRAR April 2020, Volume 7, Issue 2 www.ijrar.org (E-ISSN 2348-1269, P- ISSN 2349-5138)
DATASET DESCRIPTION

For this project, we have used the dataset named Caltech-UCSD Birds 200 that includes 200 categories of bird
species, 11,788 total number of images, and other information such as labeled visible parts, binary attributes and
bounding boxes surrounding the birds. From these 200 categories, we are taking 3 basic categories and performing
the test.

The bird species used in this dataset are:

1.Painted_Bunting

2.Eared_Grebe

3.Brewer_Blackbird

PROPOSED SYSTEM

Convolution neural network algorithm is a multilayer perceptron that is the special design for the identification of two-
dimensional image information. It has four layers: an input layer, a convolution layer, a sample layer, and an output
layer. In a deep network architecture, the convolution layer and sample layer may have multiple. CNN is not as
restricted as the Boltzmann machine, it needs to be before and after the layer of neurons in the adjacent layer for all
connections, convolution neural network algorithms, each neuron doesn’t need to experience the global image, just
feel the local region of the image. In addition, each neuron parameter is set to the same, namely, the sharing of
weights, namely each neuron with the same convolution kernels to the deconvolution image.

The key era of CNN is the local receptive field, sharing of weights, subsampling by using time or space, with a purpose
to extract features and reduce the size of the training parameters. The advantage of CNN algorithm is to avoid the
explicit feature extraction, and implicitly to learn from the training data. The same neuron weights on the surface of
the feature mapping, thus the network can learn parallel, and reduce the complexity of the network Adopting sub-
sampling structure by time robustness, scale, and deformation displacement. Input information and network topology
can be a very good match. It has unique advantages in image processing. The Convolution Neural Network involves
these steps

IJRAR2004172 International Journal of Research and Analytical Reviews (IJRAR) www.ijrar.org 262
© 2020 IJRAR April 2020, Volume 7, Issue 2 www.ijrar.org (E-ISSN 2348-1269, P- ISSN 2349-5138)
Convolution Layer:

The convolutional layer is the core constructing block of a CNN. The convolution layer comprises a set of
independent feature detectors. Each Feature map is independently convolved with the images.

Pooling Layer:

The pooling layer feature is to progressively reduce the spatial size of the illustration to reduce the wide variety
of parameters and computation in the network. The pooling layer operates on each function map independently. The
approaches used in pooling are:

· Max Pooling

· Mean Pooling

· Sum Pooling

Fully Connected Layer:

Neurons in the fully connected layer have full connections to all activations inside the preceding layer. In this,
the output obtained from max pooling is converted to a one-dimensional array and that should be the input layer and
the process continues the same as the ANN model. In this proposed system we considered the following species:

Sr.No Bird Species Names Samples Count

1. Painted_Bunting 210

2. Brewer_Blackbird 185

3. Eared_Grebe 190

IJRAR2004172 International Journal of Research and Analytical Reviews (IJRAR) www.ijrar.org 263
© 2020 IJRAR April 2020, Volume 7, Issue 2 www.ijrar.org (E-ISSN 2348-1269, P- ISSN 2349-5138)
FLOW OF SYSTEM

Input image Upload

user File store
file image file

Feed
Image

Apply
CNN

Feature
extraction

Apply classifier

Show
result

The above figure represents the actual flow of the proposed system. To develop such a system a trained dataset is
required to classify an image. The trained dataset consists of train results and test results. Whenever a user will upload
an input file, the image is temporarily stored in the database. This input file is then fed to the system and given to CNN
where CNN is coupled with a trained dataset. CNN consists of various convolutional layers. Various features such as
head, color, body, shape, and the entire image of the bird are considered for classification to yield maximum accuracy.
Each alignment is given through a deep convolutional network to extract features out from multiple layers of the
network. Then an unsupervised algorithm called deep learning using CNN is used to classify that image.

RESULTS AND DISCUSSION

When we run the code, the dataset is divided into training and testing. First, it trains the system then it tests for
accuracy then it asks for user input. It gives the label as a result with an input image. Following are the steps of
execution:

IJRAR2004172 International Journal of Research and Analytical Reviews (IJRAR) www.ijrar.org 264
© 2020 IJRAR April 2020, Volume 7, Issue 2 www.ijrar.org (E-ISSN 2348-1269, P- ISSN 2349-5138)
Figure 1:Inserting User Input

Figure 2:Output-Species Name and Input Image

Sr.No Bird name (species) Tested Results Accuracy (%)

1. Painted_Bunting 10/10 95%

2. Brewer_Blackbird 10/10 95%

3. Eared_Grebe 7/10 70%

Total 27/30 90%

Table 1: Efficiency Table

CONCLUSION

The main idea behind developing the identification website is to build awareness regarding bird-watching, bird and
their identification, especially birds found in India. It also caters to the need of simplifying the bird identification process
and thus making bird-watching easier. The technology used in the experimental setup is Convolutional Neural
Networks (CNN). It uses feature extraction for image recognition. The method used is good enough to extract features
and classify images.

This paper presents a summary of our project. The main purpose of the project is to identify the bird species from an
image given as input by the user. We used CNN because it is suitable for implementing advanced algorithms and
gives good numerical precision accuracy. It is also general-purpose and scientific. We achieved an accuracy of 85%-
90%. We believe this project extends a great deal of scope as the purpose meets. In wildlife research and monitoring,
this concept can be implemented in-camera traps to maintain the record of wildlife movement in specific habitat and
behavior of any species.

ACKNOWLEDGMENT

The project team members would like to express our thanks to every individual of the Computer Science Department
at Anil Neerukonda Institute of Technologies and Sciences for their suggestions and their help with any problems that
arose during our work. We would like to thank our guide Dr.R.Sivaranjani, for her valuable suggestions and guidance

IJRAR2004172 International Journal of Research and Analytical Reviews (IJRAR) www.ijrar.org 265
© 2020 IJRAR April 2020, Volume 7, Issue 2 www.ijrar.org (E-ISSN 2348-1269, P- ISSN 2349-5138)
for completing our research work. We must admit that she always inspired and motivated us during the completion of
the work. We also thank the other teaching staff in the department for their great efforts and for giving us the advice
that has helped us to achieve our project’s goals.

REFERENCES

[1] Indian Birds [Online] https://play.google.com /store /apps /details ?id = com.kokanes.birds info & hl = en

[2] Bird Watching Apps: Five Useful Apps to Get Started With Birding [Online]
https://gadgets.ndtv.com/apps/features/bird-watching-apps-five-useful-apps-to-get-started-with-birding-1640679

[3] Transfer learning using Alex Net [Online] https://in.mathworks.com /help /nnet /examples / transfer-learning-using-
alexnet.html

[4] Feature extraction using AlexNet [Online] https://in.mathworks.com /help /nnet /examples / feature-extraction-
using-alexnet.html # d119e4167

[5] U.D.Nadimpalli, R.R.Price, S.G.Hall, and P.Bomma,” A Comparison of image processing techniques for bird
recognition”, Biotechnology Progress, Vol. 22, no. 1,pp. 9-13,2006.

[6] Toth, B.P. and Czeba, B.,2016, September. Convolutional Neural Networks for Large-Scale Bird Song
Classification in Noisy Environment. In CLEF (Working Notes) (pp. 560-568).

[7] Elias Sprengel, Martin Jaggi, Yannic Kilcher, and Thomas Hofmann. Audio Based Bird Species Identification using
Deep Learning Techniques. 2016.

[8] Jaderick P. Pabico, Anne Muriel V. Gonzales, Mariann Jocel S. Villanueva, and Arlene a. Mendoza. Automatic
identification of animal breeds and species using bioacoustics and artificial neural networks.arXiv preprint, pages 1-
17, 2015.

[9] Yoonchang Han and Kyogu Lee. Acoustic scene classification using convolutional neural network and multiple-
width frequency-delta data augmentation. 14(80:1-11, 2016.

[10] T. S. Brandes, “Automated sound recording and analysis techniques for bird surveys and conservation,” Bird
Conservation International, Vol. 18, pp.v 163-173, 2008.

[11] Stefan Kahl, Thomas Wilhelm-Stein, Hussein Hussein, Holger Klinck, Danny Kowerko, Marc Fitter, and Maximilian
Eibl Large-Scale Bird Sound Classification using Convolutional Neural Networks.

[12] Daniel T. Blumstein, Daniel J. Mennill, Patrick Clemins, Lewis Girod, Kung Yao, Gali Patricelli, Jill L. Deppe, Alan
H. Krakauer, Christopher Clark, Kathryn A. Cortopassi, Sean F. Hanser, Brenda Mccowan, Andreas M. Ali, and
ALexander N G Kirsch. Acoustic monitoring in terrestrial environments using microphone arrays: applications,
technological considerations, and prospectus. Journal of Applied Ecology, 48(3):758-767, 2011.

[13] R. Bardeli, D. Wolff, F. Kurth, M. Koch, K.-H. Tauchert, and K.-H. Frommolt, “Detecting bird songs in a complex
acoustic environment and application to bioacoustic monitoring,” Patt Recog Letters, vol. Vol.31, pp. Pp. 1524-1534,
2010.

[14] Fagerlund, S., 2007. Bird species recognition using support vector machines. EURASIP Journal on Applied Signal
Processing, 2007(1), pp.64-64.

[15] Yann LeCun, Yoshua Bengio, and Hinton Geoffrey. Deep learning. Nature Methods, 13(1):35-35, 2015.

[16] Sergey loffe and Christian Szegedy. Batch Normalization: Accelerating Deep Network Training by Reducing
Internal Covariate Shift. arXiv:1502.03167, pages 1-11, 2015.

IJRAR2004172 International Journal of Research and Analytical Reviews (IJRAR) www.ijrar.org 266