Volume 8, Issue 9, September – 2023 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165

Identifying Specific Breed of a Bird using

Deep Learning
U.Chandee Priya1 A.Srinivasan2
Final Year M. Tech., CSE., Associate Professor, CSE,
Sreenivasa Institute of Technology and Management Sreenivasa Institute of Technology and Management
Studies, Chittoor, A.P., India. Studies, Chittoor, A.P., India.

Abstract:- Convolution neural networks are a particular CNN is constructed based on the 2D input image. This task
kind of neural network that deals with the applications is achieved along with several pooling approaches that
involving image processing and visual representation. convert the input data into invariant features with the help of
Feature extraction and classification based on those many local ties and connected weights. CNN has the
characteristics, such as image recognition and object flexibility in processing and development of the system..
detection, are some of CNN's best uses. These days, there
is a serious problem with bird behaviour and population II. REVIEW OF LITERATURE
patterns. Birds' quick responses to ecological changes
enable humans to recognise the various living forms that Using a mobile application called the Internet of Birds
exist on Earth. Convolutional neural networks (CNN) (IoB), D.Arunal et al. (2022) has created a deep learning
technology is utilised in the experimental setting. For platform to help users identify 27 species of birds that are
image recognition, feature extraction is used. To extract peculiar to Taiwan. A convolutional neural network (CNN)
features and classify photos, the method utilised is was trained on bird photos to identify key features. To fine-
adequate. The primary objective of the study is to tune the sizes and hues of the object granularities, we first
identify the specific bird type an species based on the developed and generated a bounded zone of interest, after
image of the bird. which we balanced the distribution of bird species. The
outputs of the prior and subsequent layers were then linearly
Keywords:- Convolution Neural Networks, Deep Learning, combined using the skip connection approach to enhance
Bird Species, Bird Species Identification, Image Recognition feature extraction. The softmax function was then used to
obtain a probability distribution of bird feature probabilities.
I. INTRODUCTION The submitted photographs were identified using the learned
criteria of bird traits.
Today, bird behaviour and population trends are
significant issues. The environment changes are observed by Suleyman A et al. (2021) conducted research on a
the birds and they tend to quickly react to it. The bird’s database that includes 4340 photographs that the paper's
responses can lead to detection of many environmental author had acquired from Jordan. Layers 6 and 7 as well as
changes. The knowledge gathered on birds is backed by statistical operations combining the two layers, such as
human efforts and the manual process is very expensive. average, minimum, maximum, and combine of both layers,
Development of a dependable system that can analyse bird were subjected to the Principal Component Analysis.
data on a wide scale helps to identify the bird species Artificial neural networks, K-Nearest Neighbour, Random
quickly and act as an useful tool for researchers and Forest, Naive Bayes, and Decision Tree were used to classify
governmental organizations. In order to determine the the datasets. While the metrics that each classifier uses are F-
species of bird using a particular image of it belongs to, bird Measure, recall, accuracy, and precision. The investigation's
species identification is crucial. Bird species identification is findings include, but are not limited to, the following: The
determining which category a particular species of bird falls PCA applied on the deep features not only reduces the
under using a photograph. Identification is possible. Deep dimensionality, which greatly shortens training and testing
learning is the number of neurons connected to the layers or time, but also enables raising the identification rate.
a type of machine learning based on artificial neural
networks in which multiple layers of processing are used to In order to create an image identification using tensor
extract progressively higher level information from data. flow, Parhlad Gaveli (2019) employed the deep
convolutional neural network (DCNN) technique on an
Convolution neural networks refers to particular class image that had been transformed to grey scale. The testing
of neural network that excels at tasks requiring advanced dataset is compared to these various nodes, and a score sheet
image processing and visual representation. The finest use is generated as a result. The required bird species can be
of CNN is the classification of a particular aspect based on predicted by utilising the highest score once the score sheet
the characteristics, such as image recognition and object has been analysed. Caltech-UCSD Birds 200 [CUB-200-
CNN is made of dfferent kinds of convolutional layers . 2011] dataset experimental study reveals that algorithm
After these convolutional layers, there are fully connected achieves an accuracy of bird recognition between 80% and
(FC) layers, which resemble the multilayer neural network. 90%. Using the Ubuntu 16.04 operating system and the

IJISRT23SEP580 www.ijisrt.com 128

Volume 8, Issue 9, September – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
Tensor flow library, an experimental investigation was  Technical Feasibility
conducted.  Operational Feasibility
 Economical Feasibility
III. OBJECTIVE

The main objective is to identify the species of the

 Technical Feasibility
This study was created in the convenient, all-around
birds with the bird details by analyzing an image. Some
Python environment. The only platform required for this
professionals, such as ornithologists, struggled to accurately
scenario is Windows. It can be readily developed using
identify a bird's species from an image. Even though subject
python and is technically sound in the Windows
specialists can categorise birds manually, the amount of data
environment. The study is therefore technically possible.
available makes this procedure increasingly laborious and
time-consuming. Therefore, developing a reliable system
that identify the species of bird quickly is the main purpose  Operational Feasibility
of this research. The proposed system is technically possible since it
was created in a way that even a user with no prior
IV. SYSTEM ANALYSIS knowledge can use it with ease. It produces higher accuracy
and makes excellent decisions about patient blood glucose
 Existing System levels. The user may quickly grasp without requiring much
There are a variety of websites and applications that learning because the entire system is user-friendly and was
use various technologies to identify different bird species. created with them in mind.
But the outcomes are incorrect. Input for bird identification
on webpages and Android apps returns several answers  Economical Feasibility
rather than just the name of the species. It displays all bird This study is being done to see what kind of financial
names that share similar traits. Consequently, there is a gap impact the system will have on the company. The
that must be filled by creating software to generate better and corporation has a finite amount of money to invest in the
more accurate outcomes. Convolutional Neural Networks are system's research and development. The costs must be
employed to categorise the many bird species in order to do supported by evidence. As a result, the developed system
this. came in under budget, which was made possible by the fact
that most of the technologies were free to use.
 Proposed System
To classify an image, the suggested approach uses a VI. SYSTEM MODEL
training dataset. Test results plus learned results make up the
trained dataset. To increase recognition accuracy, the dataset Agile Model is used as the base for system
needs to be retrained using a specific file from Google development.
Collab. The training dataset is constructed using 50000
steps, with the understanding that the accuracy increases as Agile is a term that means quick or adaptable. The
the number of steps increases. The training dataset's phrase "Agile process model" describes a method of
accuracy is 93%. The testing dataset has an accuracy of 80% developing software that is iterative in nature. Agile study
and contains around 1000 photos. To improve the system's management techniques divide work into smaller iterations
performance, the dataset is validated with a 75% accuracy. A or pieces without directly including long-term planning. The
user's input file is briefly saved in the database every time study's requirements and scope are established at the start of
they upload it to a website. Once fed, this input file is used the development phase. Plans for the quantity, length, and
by scope of each iteration are spelt out in detail in advance.

V. FEASIBILITY STUDY In the Agile process model, each iteration is viewed as

a brief time "frame" that typically lasts one to four weeks.
An activity focused on management is the feasibility The study risk is reduced and the overall study delivery time
study. It is an assessment and study of the study's potential requirements are lowered thanks to the study's breakdown
that is based on in-depth inquiry and research to fully into smaller components. During each iteration, a team
reassure the decision makers. Finding out whether an works through a
information system study can be implemented and offering
potential alternatives are the two main goals of the feasibility VII. MODULE AND ITS DESCRIPTION
study. It is meant to be a quick evaluation of the facts to
determine whether moving on to the analytical phase is The process of defining the architecture, modules,
warranted. The feasibility analysis is the main tool used by interfaces, and data for a system to meet predetermined
system analysts to determine whether to move the study requirements is known as system design. The disciplines for
forward or stop it altogether. An alternate search is carefully specifying the computer system's hardware and software
considered. There are three parts in feasibility study. architecture, parts, modules, interfaces, and data to meet
predetermined requirements overlap to some extent.

IJISRT23SEP580 www.ijisrt.com 129

Volume 8, Issue 9, September – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
 Modules:  Classification
The results of model is to Identify the Bird Species with
A. System Modules: the details of that bird

 Image Dataset B. User Modules:

 Pre-processing Upload Image
 Training
The user has to upload an image which needs to be
 Classification
classified.
 Image Dataset  View Results
The Caltech-UCSD Birds 200 [CUB-200-2011] dataset The classified image results are viewed by user.
containing images of the 400 Bird Species to be classified
and split into training and testing dataset with the test size of  Input Design
80-20%. The input design establishes a connection between the
user and the information system. In order to transform
 Pre-processing transaction data into a format that can be processed, it
There are a few procedures to be adhered during the includes developing requirements and methods for data
preprocessing: preparation. User can enter the data directly into the system
to accomplish this, or the computer can read the data from a
 Take an input of an image. written or printed document. The input process is created
 Use preprocessing techniques to draw attention to the key with a view towards decreasing the amount of input required,
elements. minimising errors, delays, and additional phases, as well as
 Perform Image scaling and cropping. maintaining a clear workflow. The data is entered in a way
that ensures its security, usability, and privacy. These factors
 Perform data cleansing and black image removal. were taken into account through input design.
 Image rotation and mirroring to restore balance to an
unbalanced dataset. What data it should be given as input?
 Perform Array conversion to NumPy.
 How the data it should be arranged or coded?
Now use for training or testing. Training  The dialog to guide the operating personnel in providing
input.
Use the pre-processed training dataset is used to train
model using Deep learningalgorithm.  Methods for preparing input validations and steps to
follow when error occur.

Fig 1 Input Image

 Output Design and the hard copy output is taken during output design. It is
A quality output is one that meets the needs of the end the most important and direct source of information for the
user and clearly displays the information. The way user. A system's ability to engage with tools that support user
processing results are communicated to users and other decision-making is enhanced by effective and intelligent
systems is through the outputs of each system. How output design.
information will be displaced for both immediate demand

IJISRT23SEP580 www.ijisrt.com 130

Volume 8, Issue 9, September – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
The process of creating computer output should be  Create document, report, or other formats that contain
organised and well-thought out; the proper output must be information produced by the system.
created while ensuring that each output component is The output form of an information system should
structured in a way that will make the system usable by accomplish one or more of thefollowing objectives.
people quickly and efficiently. When analysing computer-
generated output, one should pinpoint the precise output that  Convey information about past activities, current status
is required to satisfy the specifications.Select methods for or studyions of the Future.
presenting information.
 Signal important events, opportunities, problems, or
warnings.
 Trigger an action.
 Confirm an action.

Fig 2 Output Design

VIII. CONCLUSION REFERENCES

 The main idea behind developing the identification [1]. Bengio, Yoshua. 2016. Deep Learning. Adaptive
website is to build awareness regarding bird-watching, Computation And Machine Learning Series. London,
bird and their identification, especially birds found in England: Mit Press.
India. It also meets the requirement to make bird- [2]. Aruna , J. Jyostna , N. Sai Radhika , K.Moulika And
watching simpler by streamlining the identifying N.Aminash (2022), “Bird Species Identification Using
procedure. Deep Learning”, Dogo Rangsang Researchvol-09
Issue-01 No. 01
 Convolutional Neural Networks (CNN) technology is [3]. Pralhad Gavali , Ms. Prachi Abhijeet Mhetre , Ms.
utilised in the experimental setup. It recognises images
Neha Chandrakhant Patil , Ms. Nikita Suresh Bamane
using feature extraction. The technique employed is
And Ms. Harshal Dipak Buva(2019), “Bird Species
adequate for feature extraction and picture classification.
Identification Using Deep Learning”, International
 The main objective of specified system development is Journal Of Engineering Research & Technology ,
to identify the bird species from an image given as input Volume 08, Issue 04
by the user. CNN is used as it is suitable for [4]. Suleyman A. Al-Showarah1 , Sohyb T. Al-Qbailat2
implementing advanced algorithms and gives good (2021), “Birds Identification System Using Deep
numerical precision accuracy. It is also general-purpose Learning”, International Journal Of Advanced
andscientific. The accuracy of 85%-90% is achieved. Computer Science And Applications,Vol. 12, No. 4
 This research extends a great deal of scope in wildlife [5]. http://www.vision.caltech.edu/visipedi a/CUB-200.html
research and monitoring, This system can be [6]. https://www.tensorflow.org/datasets/catal og/overvie
implemented in-camera traps to maintain the record of [7]. https://colab.research.google.com/note books/gpu.ipynb
wildlife movement in specific habitat and behavior of
any species

IJISRT23SEP580 www.ijisrt.com 131