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

net/publication/325116934

Image classification using Deep learning

Article  in  International Journal of Engineering & Technology · March 2018

DOI: 10.14419/ijet.v7i2.7.10892

CITATIONS READS

33 35,458

4 authors, including:

Harshali Mane Venu Gopala Rao Matcha

K L University K L University
11 PUBLICATIONS   51 CITATIONS    23 PUBLICATIONS   102 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Venu_Sat View project

All content following this page was uploaded by Venu Gopala Rao Matcha on 20 August 2018.

The user has requested enhancement of the downloaded file.

 International Journal of Engineering & Technology, 7 (2.7) (2018) 614-617

International Journal of Engineering & Technology

Website: www.sciencepubco.com/index.php/IJET

Research Paper

Image classification using Deep learning

M Manoj krishna1*, M Neelima2, M Harshali3, M Venu Gopala Rao4
1,2B
tech Student, Dept. of ECE, K L E F, India
3SRF-CSIR,Dept. of ECE, K L E F, India
4Professor, Dept. of ECE, K L E F, India

*Corresponding author E-mail: manoj.krishna4301@gmail.com

Abstract

The image classification is a classical problem of image processing, computer vision and machine learning fields. In this paper we study
the image classification using deep learning. We use AlexNet architecture with convolutional neural networks for this purpose. Four test
images are selected from the ImageNet database for the classification purpose. We cropped the images for various portion areas and
conducted experiments. The results show the effectiveness of deep learning based image classification using AlexNet.

Keywords: AlexNet; Convolutional Neural Networks; Deep Learning; Image Classification; ImageNet; Machine Learning.

The basic artificial neural network is outlined in Section-II. Section-

1. Introduction III describes about AlexNet. The implementation and results are
discussed in Section-IV. We conclude in section-V and finally the
Classification is a systematic arrangement in groups and categories references are given at the end.
based on its features. Image classification came into existence for
decreasing the gap between the computer vision and human vision
by training the computer with the data. The image classification is
2. Artificial Neural Networks
achieved by differentiating the image into the prescribed category
based on the content of the vision. Motivation by [1], in this paper, A neural network is a combination of hardware bonded or separated
we explore the study of image classification using deep learning. by the software system which operates on the small part in the hu-
The conventional methods used for image classifying is part and man brain called as neuron. A multi layered neural network can be
piece of the field of artificial intelligence (AI) formally called as proposed as an alternative of the above case. The training image
machine learning. The machine learning consists of feature extrac- samples should be more than nine times the number of parameters
tion module that extracts the important features such as edges, tex- essential for tuning the classical classification under very good res-
tures etc and a classification module that classify based on the fea- olution. The multi-layered neural network is so complicated task
tures extracted. The main limitation of machine learning is, while with respect to its architecture in the real world implementa-
separating, it can only extract certain set of features on images and tions[14-17]. The multi-layered neural network is at present ex-
unable to extract differentiating features from the training set of pressed as the Deep Learning.
data. This disadvantage is rectified by using the deep learning [2]. In deep neural networks every node decides its basic inputs by it-
Deep learning (DL) is a sub field to the machine learning, capable self and sends it to the next tier on behalf of the previous tier.
of learning through its own method of computing. A deep learning
model is introduced to persistently break down information with a
homogeneous structure like how a human would make determina-
tions. To accomplish this, deep learning utilizes a layered structure
of several algorithms expressed as an artificial neural system
(ANN). The architecture of an ANN is simulated with the help of
the biological neural network of the human brain. This makes the
deep learning most capable than the standard machine learning
models [3, 4].
In deep learning, we consider the neural networks that identify the
image based on its features. This is accomplished for the building
of a complete feature extraction model which is capable of solving
Fig. 1: Basic Deep Neural Network
the difficulties faced due to the conventional methods. The extrac-
tor of the integrated model should be able to learn extracting the
differentiating features from the training set of images accurately. We train the data in the networks by giving an input image and con-
Many methods like GIST, histogram of gradient oriented and Local veying the network about its output. Neural networks are expressed
Binary Patterns, SIFT are used to classify the feature descriptors in terms of number of layers involved for producing the inputs and
from the image. outputs and the depth of the neural network. Neural networks are
involved in many principles like fuzzy logic, genetic algorithms and

Copyright © 2018 Authors. This is an open access article distributed under the Creative Commons Attribution License, which permits
unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
 International Journal of Engineering & Technology 615

Bavesian methods. These layers are generally referred to as hidden

layers. They are expressed in terms of number of hidden nodes and
number of inputs and outputs every node consists. The Convolu-
tional Neural Network (ConvNet) is most popular algorithm used
for implementing the deep learning technique. The ConvNet con-
sists of Feature detection layers and classification. A ConvNet is
composed of several layers, and they are convolutional layers, max-
pooling or average-pooling layers, and fully-connected layers.

Fig. 3: Test images (a) Sea Anemone of size 375x500x3(b) Barometer

of size 500x375x3 (c) stethoscope of size 375x500x3 (d) Radio inter-
ferometer of size 375x500x3

In the first layer, there are 96 11x11 filters are used at stride 4. The
output volume size is 55x55x96. The AlexNet is trained on the GPU
named GTX580 which is having a small amount of 3GB of
memory. So, the CONV1 output will be halved and sent to two
GPU’s i.e. 55x55x48 is sent to each GPU. The 2nd, 4th, and 5th
convolutional layers bits are related just to the part maps in the pre-
Fig. 2: Architecture of CNN vious layer which dwell on the same GPU said in the figure. The
kernels of the 3rd convolutional layer are associated with all kernel
maps in the 2nd layer. The neurons in the fully connected layers are
3. Alexnet associated with all neurons in the past layer.

The ConvNet is categorized into two types named LeNet and The 3rd, 4th, and 5th convolutional layers are associated with each
AlexNet. The LeNet is expressed as the Shallow Convolutional other with no interceding pooling or standardization layers. The 3rd
Neural Networks which is designed to classify the hand-written dig- convolutional layer has 384 parts of size 3 × 3 × 256 associated with
its. The LeNet comprises of 2 convolutional layers, 2 subsampling the (standardized, pooled) yields of the 2nd convolutional layer.
layers, 2 hidden layers and 1 output layer [5]. The AlexNet is ex- The fourth convolutional layer has 384 kernels of size 3 × 3 × 192
pressed as the deep convolutional neural networks which are used and the fifth convolutional layer has 256 kernels of size 3 × 3 × 192.
for classifying the input image to one of the thousand classes. The first two fully connected layers have 4096 neurons each.

AlexNet is used to solve many problems like indoor sense classifi- We use the local response normalization in the normalization layer.
cation which is highly seen in artificial neural intelligence. It is a There are two normalization layers present in the AlexNet architec-
powerful method of knowing the features of the image with more ture. The Deep Neural Network with ReLU Nonlinearity can train
differential vision in the computer field for the recognition of pat- very fast than with the identical of the function tanh units. The
terns. This paper discuss about the classification of a particular size ReLU considers quicker and more compelling training by mapping
of image of required choice. It can very effectively classify the the negative esteems to zero and keeping up positive esteems. Sig-
training sample of images present in the AlexNet for better vision. nifying by the movement of a neuron figured by applying kernel i
at position (x, y) and after that applying the ReLU nonlinearity, the
The AlexNet comprises of 5 convolutional layers, 3 sub sampling response-normalized movement is expressed as
layers and 3 fully connected layers. The main difference between β
 2
d 
the LeNet and AlexNet are the type of Feature Extractor. We use min(N-1,i+n/2)
i =di 
c(x,y) (x,y) /  k + α 
i
the non-linearity in the Feature Extractor module in AlexNet ( x, y) (1)
whereas Log sinusoid is used in LeNet. AlexNet uses dropout  j=max(0,i-n/2) 
which is not observed in any other data sets of networking.
This kind of response standardization actualizes a type of parallel
4. Implementation, Results and Discussions hindrance roused by the sort found in genuine neurons, making ri-
valry for huge exercises among neuron yields registered utilizing
different kernels. The test images are cropped to various portion ar-
We selected four images Sea Anemone, Barometer, Stethoscope
eas and applied for classification. The results are shown in Fig. 5,
and Radio Interferometer from the ImageNet database for experi-
Fig. 6, Fig. 7 and Fig. 8. From the results, it is observed that in all
mentation purpose (See Fig. 3) [6]. The block diagram of the archi-
cases of the cropped data, the classification is successful.
tecture shown in Fig.4 and the corresponding implementation is il-
lustrated below [7].

Fig. 4: AlexNet Architecture

 616 International Journal of Engineering & Technology

Fig. 5: Sea Anemone cropped to various areas

Fig. 6: Barometer cropped to various areas

Fig. 7: Stethoscope cropped to various areas

Fig. 8: Radio Interferometer cropped to various areas

5. Conclusion
Four test images sea anemone, barometer, stethoscope and radio in-
terferometer are chosen from the AlexNet database for testing and
validation of image classification using deep learning. The convo-
lutional neural network is used in AlexNet architecture for classifi-
cation purpose. From the experiments, it is observed that the im-
ages are classified correctly even for the portion of the test images
and shows the effectiveness of deep learning algorithm.
 International Journal of Engineering & Technology 617

Acknowledgement
The authors would like to express their deep gratitude towards the
Department of ECE and the management of K L E F for their sup-
port and encouragement during this work.

References
[1] https://in.mathworks.com/matlabcentral/fileexchange/59133-
neural-network-toolbox-tm--model-for-alexnet-network
[2] H. Lee, R. Grosse, R. Ranganath, and A.Y. Ng. Convolutional
deep belief networks for scalable unsupervised learning of hierar-
chical representations. In Proceedings of the 26th Annual Interna-
tional Conference on Machine Learning, pages 609–616. ACM,
2009
[3] Deep Learning with MATLAB – matlab expo2018
[4] Introducing Deep Learning with the MATLAB – Deep Learning
E-Book provided by the mathworks.
[5] https://www.completegate.com/2017022864/blog/deep-machine-
learning-images-lenet-alexnet-cnn/all-pages
[6] Berg, J. Deng, and L. Fei-Fei. Large scale visual recognition chal-
lenge 2010. www.imagenet.org/challenges. 2010.
[7] Fei-Fei Li, Justin Johnson and Serena Yueng, “Lecture 9: CNN
Architectures” May 2017.
[8] L. Fei-Fei, R. Fergus, and P. Perona. Learning generative visual
models from few training examples: An incremental bayesian ap-
proach tested on 101 object categories. Computer Vision and Im-
age Understanding, 106(1):59–70, 2007.
[9] J. Sánchez and F. Perronnin. High-dimensional signature com-
pression for large-scale image classification. In Computer Vision
and Pattern Recognition (CVPR), 2011 IEEE Conference on,
pages 1665–1672. IEEE, 2011.
[10] https://in.mathworks.com/help/vision/examples/image-category-
classification-using-deep-learning.html
[11] Alex Krizhevsky, Ilya Sutskever and Geoffrey E. Hinton,
“ImageNet Classification with Deep Convolutional Neural Net-
works” May 2015.
[12] A. Krizhevsky. Learning multiple layers of features from tiny im-
ages. Master’s thesis, Department of Computer Science, Univer-
sity of Toronto, 2009.
[13] https://in.mathworks.com/help/nnet/deep-learning-imageclassifi-
cation.html
[14] KISHORE, P.V.V., KISHORE, S.R.C. and PRASAD, M.V.D.,
2013. Conglomeration of hand shapes and texture information for
recognizing gestures of indian sign language using feed forward
neural networks. International Journal of Engineering and Tech-
nology, 5(5), pp. 3742-3756.
[15] RAMKIRAN, D.S., MADHAV, B.T.P., PRASANTH, A.M.,
HARSHA, N.S., VARDHAN, V., AVINASH, K., CHAITANYA,
M.N. and NAGASAI, U.S., 2015. Novel compact asymmetrical
fractal aperture Notch band antenna. Leonardo Electronic Journal
of Practices and Technologies, 14(27), pp. 1-12.
[16] KARTHIK, G.V.S., FATHIMA, S.Y., RAHMAN, M.Z.U.,
AHAMED, S.R. and LAY-EKUAKILLE, A., 2013. Efficient sig-
nal conditioning techniques for brain activity in remote health
monitoring network. IEEE Sensors Journal, 13(9), pp. 3273-3283.
[17] KISHORE, P.V.V., PRASAD, M.V.D., PRASAD, C.R. and RA-
HUL, R., 2015. 4-Camera model for sign language recognition
using elliptical fourier descriptors and ANN, International Con-
ference on Signal Processing and Communication Engineering
Systems - Proceedings of SPACES 2015, in Association with
IEEE 2015, pp. 34-38.

View publication stats