See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/345149811

A Case Study of Image Classification Based on Deep Learning Using

Tensorflow

Article in International Journal of Innovative Research in Computer and Communication Engineering · April 2018

CITATIONS READS

11 3,635

1 author:

Jubin Dipakkumar Kothari

Campbellsville University
5 PUBLICATIONS 52 CITATIONS

SEE PROFILE

All content following this page was uploaded by Jubin Dipakkumar Kothari on 01 November 2020.

The user has requested enhancement of the downloaded file.

 ISSN(Online): 2320-9801
ISSN (Print) : 2320-9798

International Journal of Innovative Research in Computer

and Communication Engineering
(A High Impact Factor, Monthly, Peer Reviewed Journal)

Website: www.ijircce.com
Vol. 6, Issue 4, April 2018

A Case Study of Image Classification Based on

Deep Learning Using Tensorflow
Jubin Dipakkumar Kothari
Department of Information Technology, Campbellsville University, Campbellsville, Kentucky

ABSTRACT: This examination learns about image classification by utilizing the deep neural network (DNN) or
otherwise called Deep Learning by utilizing system Tensor Flow. Python is utilized as a programming language since it
meets up with the Tensor Flow system. The information chiefly centres in blossoms class, which there are five (5) sorts
of blossoms that have been utilized in this paper. Deep neural network (DNN) has been picking as the ideal alternative
for the preparation cycle since it delivered a high level of exactness. Results are talked about as far as the exactness of
the image classification in rate. Roses get 90.585%, and same goes to another sort of blossoms where the normal of the
outcome is up to 90% or more.

KEYWORDS: Image classification, Deep Learning, Tensor flow

I. INTRODUCTION

As of late, image classification is developing and turning into a pattern among technology designers particularly with
the development of information in various pieces of industry, for example, online business, car, medical care, and
gaming. The most evident case of this technology is applied to Facebook. Facebook presently can distinguish up to
98% precision to recognize your face with just a couple of labelled images and arranged it into your Facebook's
collection. The technology itself nearly beats the capacity of human in image classification or recognition (What is the
Working of Image Recognition and How it is Used, 2017).One of the prevailing methodologies for this technology is
deep learning. Deep Learning falls under the class of Artificial Intelligence, where it can act or adopt the thought
process of a human[1]. Typically, the framework itself will be set with hundreds or perhaps a considerable number of
info information to make the 'training'session to be more productive and quick.

Machine learning is likewise the continuous frameworks that have been applied toward image classification.
Notwithstanding, there are still parts that can be improved inside machine learning. Along these lines, image
classification will be busy with a deep learning system.Machine Vision has its setting when it accompanies Image
Classification. This technology can perceive individuals, objects, places, activity and writing in images.The
fundamental undertaking of image classification is to ensure its particular areas or groups order all the images.
Classification is simple for people, yet it has ended up being severe issues for machines[2]. It comprises of unidentified
examples contrasted with identifying an item as it to be ordered to the correct classes. The different applications, for
example, vehicle route, robot route and far off detecting by utilizing image classification technology. It is as yet going
through testing work, and restricted assets are expected to improve it.In this paper, a deep neural network, in light of
Tensor Flow is utilized with Python as the programming language for image classification. A large number of images
are utilized as the info information in this undertaking.

II. CHALLENGES IN IMAGE-BASED TAXONOMY IDENTIFICATION

Massive Number of taxonomy to be Biased

In this world, there is an enormous number of species accessible like plants, creatures, bugs thus numerous others. It is
tough to bias them into the taxonomical structure[3]. Regardless, while restricting the focus to the verdure of an area, a

Copyright to IJIRCCE DOI: 10.15680/IJIRCCE.2018.0604109 3888

 ISSN(Online): 2320-9801
ISSN (Print) : 2320-9798

International Journal of Innovative Research in Computer

and Communication Engineering
(A High Impact Factor, Monthly, Peer Reviewed Journal)

Website: www.ijircce.com
Vol. 6, Issue 4, April 2018

great many vegetation classifications ought to be upheld. Let us take a case of German's Flora shows a large number of
blossoming species, and in every single class, there will be numerous different species.

Variations on Huge Intraspecific

A few animal groups plants have some unique agriculture attributes like area, dampness, nourishment, life cycle and
climatic conditions and so forth These adjustments in cultivation qualities can happen on their estimating units,
blossoms, natural products, leaves and at times even entire plants[4]. For instance, Knautia genus commonly known
types of knautia sort. This reaches from colossal entire, or dentate lanceolate ground leaves over significantly lobed and
essentially pinnate stem leaves to pretty much nothing and again lanceolate and entire upper stem leaves.

Accession process for variation

Plants leaves are in 3-d images yet when we catch then it becomes in 2-d, so sine of the highlights change like
appearance. By this, enormous contrasts make between unique images and caught images like shape and appearance.
Outside states of image catching additionally restricted like zoom, focus, sensor,etc.

III. METHODOLOGY

Given Figure 1, it is the system of image classification where deep neural networks are additionally applied. There are
four (4) stages all through this cycle, and every one of the stages will be talked about. Every one of the stages is
remembered for TensorFlow as the open-source programming and Python as its programming language[5]. At that
point, the cycle is kept on gathering a portion of the images (contributions), by applying DNN and ultimately, all
images will be arranged into their gatherings.

Fig 1:The block diagram of Image Classification

Training Images
Information for this paper primarily utilizes a large number of images. These images are taken from ImageNet.
ImageNet additionally was known as Large Scale Visual Recognition Challenge where it is an opposition about
distinguishing and arranged a vast number of the article into its classes[6]. This is a yearly rivalry from 2010 until
today. This is the beginning of unrest on 'big data'.In this paper, a great many pictures of blossoms were acquired
through this ImageNet site, as appeared in Figure 2, and it is free.

Copyright to IJIRCCE DOI: 10.15680/IJIRCCE.2018.0604109 3889

 ISSN(Online): 2320-9801
ISSN (Print) : 2320-9798

International Journal of Innovative Research in Computer

and Communication Engineering
(A High Impact Factor, Monthly, Peer Reviewed Journal)

Website: www.ijircce.com
Vol. 6, Issue 4, April 2018

Fig 2:DNN process towards flower images

It intended to be utilized by the analyst or engineers. This exploration paper exclusively centres around order blossoms
into every one of its classes. There are a great many bloom pictures, and it has five sorts of blossoms here. Each sort of
blossoms contains many pictures with different side and colours[7]. The all outnumber for the entirety of this bloom is
2171 pictures as appeared in Table 2.

Table 1: Number of images according to the type of flowers

Sl.no Type of Flowers No of Images

1 Daisy 631

2 Dandelion 897

3 Roses 643

Implementation Deep Neural Network (DNN)

As proven in Figure 2, it consists of 5 (3) information inputs (three kinds of unique flowers) and undergoes coaching
with more than one hidden layers[8]. The inputs are additionally set with fixed-size of the 224x224 RGB image. The
convolution system is configured with MobileNet as it produces an environment-friendly convolution neural networks.

Flowchart of the classification systems

the flowchart of photo classification that will be applied to the usage of TensorFlow. The programming language that
will be used in the software in Python. The flowchart indicates that the systems will be beginning utilizing amassing
photos of the flowers. After that, DNN is utilized to teach the model. Running for validation or trying out and if it is no
longer the picture of a specific flower that supposedly acts as an output, then it wants to begin over once more from
DNN[9]. The manner ends after the output is categorized into the proper kind of flowers.The flowchart begins with
inserting units of flower photos as an entry in this research. It has 3(3) sorts of the flower which is Roses, Daisy and
Dandelion. After that, all of these enter snapshots endure 'training' with the deep neural network(DNN).

Copyright to IJIRCCE DOI: 10.15680/IJIRCCE.2018.0604109 3890

 ISSN(Online): 2320-9801
ISSN (Print) : 2320-9798

International Journal of Innovative Research in Computer

and Communication Engineering
(A High Impact Factor, Monthly, Peer Reviewed Journal)

Website: www.ijircce.com
Vol. 6, Issue 4, April 2018

Fig 3:The flowchart of an image classification system

IV. CONCLUSION

Classification by way of the use of deep Learning through framework TensorFlow. It has three targets that have
accomplished at some point of this research. The goals are linked without delay with conclusions due to the fact it can
decide whether or not all goals are correctly accomplished or not. It can be concluded that all outcomes that have been
obtained, confirmed pretty mind-blowing outcomes. The deep neural network (DNN) turns into the main agenda for
this research, mainly in picture classification technology.
DNN method used to be studied in more significant important points beginning from assembling, coaching mannequin
and to classify photos. The roles of epochs in DNN was once capable of managing the accuracy and additionally
stopping any issues such as overfitting. Implementation of in-depth gaining knowledge of via the usage of framework
Tensor Flow additionally gave desirable effects as it is in a position to stimulate, instruct and categorized with up to
90% percentage of accuracy in the direction of three kinds of plants that have to grow to be a skilled model.

REFERENCES

[1] Rastegari, M., Ordonez, V., Redmon, J., &Farhadi, A. (2016). XNOR-net: Imagenet classification using binary
convolutional neural networks. Lecture Notes in Computer Science (Including Subseries Lecture Notes in
Artificial Intelligence and Lecture Notes in Bioinformatics), 9908 LNCS, 525–542. https://doi.org/10.1007/978-3-
319-46493-0_32
[2] Korytkowski, M., Rutkowski, L., & Scherer, R. (2016). Fast image classification by boosting fuzzy classifiers.
Information Sciences, 327, 175–182. https://doi.org/10.1016/j.ins.2015.08.030
[3] A. Angelova, A. Krizhevsky, and V. Vanhoucke. Pedestrian detection with a largefield- of-view deep network. In
Proceed- ings of ICRA, pages 704–711. IEEE,2015.www.vision.caltech.edu/anelia/publications/
Angelova15LFOV.pdf.
[4] Ankit Narendrakumar Soni (2018). Feature Extraction Methods for Time Series Functions using Machine
Learning. International Journal of Innovative Research in Science, Engineering and Technology, 7(8), 8661-8665.
doi:10.15680/IJIRSET.2018.0708062

Copyright to IJIRCCE DOI: 10.15680/IJIRCCE.2018.0604109 3891

 ISSN(Online): 2320-9801
ISSN (Print) : 2320-9798

International Journal of Innovative Research in Computer

and Communication Engineering
(A High Impact Factor, Monthly, Peer Reviewed Journal)

Website: www.ijircce.com
Vol. 6, Issue 4, April 2018

[5] Trishen Munisami, Mahess Ramsurn, ―Plant leaf recognition using shape features and colour histogram with k-
nearest neighbor classifiers‖. Second International Symposium on Computer vision and the Internet(VisionNet‖15).
[6] Karunakar pothuganti (2013) ‘An Efficient Architecture for Lifting Based 3D-Discrete Wavelet Transform’
International Journal of Engineering Research & Technology (IJERT), Vol. 2 Issue 12, December – 2013 ISSN:
2278-018
[7] Kody G. Dangtongdee, Dr. Franz Kurfess, ―Plant Identifica-tion Using Tensorflow‖. Plant Phenomics Volume
2019, Article ID 9237136, 14 pages https://doi.org/10.34133/2019/9237136CPE 462, Winter 2018
[8] Jonathan Y.Clark, David P.A. Corney and H.LIlianTang,‖Automated Plant Identification Using Artificial Neural
Networks‖. presented at CIBCB2012 @IEEE12.
[9] Ankit Narendrakumar Soni (2018). Application and Analysis of Transfer Learning-Survey. International Journal of
Scientific Research and Engineering Development, 1(2), 272-278.

Copyright to IJIRCCE DOI: 10.15680/IJIRCCE.2018.0604109 3892

View publication stats