object classification via extreme learning machinethe purpose of image processing by computers is to...
Post on 30-Jul-2021
0 Views
Preview:
TRANSCRIPT
Abstract—Object recognition process is used in many areas for
different purposes, notably for systems performing intelligent data
analysis for automated systems. In this process, first of all, image
processing-based features of the object are extracted. Then, by
considering these features, class labeling is performed according to
the class features determined with controlled learning. The
purpose of this process is to conduct Extreme Learning Machine
(ELM)-based classification of 4 image features of the herb iris
found in UC Irvine Machine Learning Repository database. To
assess results, a comparison was made with other machine
learning methods (SVM, NB) and ELM was detected to have the
highest achievement of 99.33%.
Keywords—Artificial intelligence, Extreme Learning Machine
(ELM), Object attributes and classification.
I. INTRODUCTION
Since digital imaging has become a fundamental part of many
studies, computer-aided assessment by using advanced imaging
analysis is now an inseparable part of many research projects.
As a scientific research process, model recognition examines
and models data for the identification of a system design.
Recognition system was developed in tandem with computer
vision as a result of an extensive accumulation of knowledge.
The purpose of image processing by computers is to enable
multiple perceptions with a wide range of categories under
various conditions and to increase computer control and data
accuracy [1-5].
Object recognition is the identification of figures and objects
that have common characteristics and can be associated with
each other by using the identified features. A group of figures
and objects with common characteristics is called an object
class. Two important goals of object recognition are to:
• Identify an object belonging to a known class.
• Classify objects with unknown features.
Object recognition systems are generally made up of two
parts. In the first part, features required for object recognition or
sizes to be measured are selected. This process is called feature
extraction and the system performing this process is called a
feature extractor. Manuscript received Feb. 26, 2018 Phd. Yasin Sönmez. is now with the
Dicle University, Diyarbakır, Turkey (e-mail: yasin.sonmez@dicle.edu.tr ).
Dr. Türker Tuncer. is now with the Fırat University, Elazığ, Turkey (e-mail:
turkertuncer@firat.edu.tr ).
Prof. Dr. Engin Avcı is now with Fırat University, Elazığ, Turkey. (e-mail:
engin.avci@firat.edu.tr).
A feature is the actual number that represents the
measurement result of each feature used in the recognition of
figures or objects or size to be measured. The primary factor
affecting the performance of classification is that all features
selected must represent the objects thoroughly.
In the second part of object recognition, objects are classified
with the use of obtained feature vectors. The block diagram of
an object recognition system is given in Figure I [2].
FEATURE EXTRACTION
CLASSIFICATION
LEARNING
OBJECT REPOSITORY
IDENTIFICATION
Fig. 1. Block diagram of an object recognition system.
TABLE I: FEATURES OF IRIS
Input Output
Feature Class
Sepal length 1 Iris Setosa
2 Iris Versicolour
3 Iris Virginica
Sepal width
Petal length
Petal width
Table 2 shows the three types of the herb iris, A-setosa,
B-versicolor and C-virginica, according to their sepal and petal
feature data
Fig. 2. The kinds of iris plant A-setosa, B-versicolor ve C-virginica.
II. MATERIAL AND METHOD
Procedural steps for solving the classification problem
presented is as follows:
A. Identification of the problem
This study attempts to solve the problem as to how phishing
analysis data will be classified.
B. Data set
Approximately 150 data containing the 4 features extracted
based on the features of iris image analysis in UC Irvine
Machine Learning Repository database.
Object Classification via Extreme Learning Machine
Yasin Sönmez, Türker Tuncer, and Engin Avcı
International Journal of Computer Science and Electronics Engineering (IJCSEE) Volume 6 Issue 1 (2018) ISSN 2320-4028 (Online)
5
C. Modeling
After the data is ready to be processed, modeling process
for the learning algorithm is initiated. The model is basically the
construction of the need for output identified in accordance with
the task qualifications.
D. Classification
Classification is to determine the class to which each data
sample of the methods belongs, which methods are used when
the outputs of input data are qualitative. The purpose is to divide
the whole problem space into a certain number of classes. A
wide range of classification methods are present. This is due to
the fact that different classification methods have been
constructed for different data as there is no perfect method that
works on every data set. As mentioned in literature studies, the
aim of classification is to assign the new samples to classes by
using the pre-labeled samples. The most commonly used
classification methods are described below.
• Artificial Neural Networks (ANN)
• Support Vector Machine (SVM)
• Naive Bayes (NB)
E. Extreme Learning Machine (ELM)
Extreme Learning Machine (ELM) is a feed-forward
artificial neural network (ANN) model with a single hidden
layer. For the ANN to ensure a high-performing learning,
parameters such as threshold value, weight and activation
function must have the appropriate values for the data system to
be modeled. In gradient-based learning approaches, all of these
parameters are changed iteratively for appropriate values. Thus,
they may be slow and produce low-performing results due to the
likelihood of getting stuck in local minima. In ELM Learning
Processes, differently from ANN that renews its parameters as
gradient-based, input weights are randomly selected while
output weights are analytically calculated. As an analytical
learning process substantially reduces both the solution time
and the likelihood of error value getting stuck in local minima, it
increases the performance ratio. In order to activate the cells in
the hidden layer of ELM, a linear function as well as non-linear
(sigmoid, sinus, Gaussian), non-derivable or discrete activation
functions can be used [8-15].
ELM structure is given in Figure 3.
• Artificial Neural Networks (ANN)
• Support Vector Machine (SVM)
• Naive Bayes (NB)
INPUT LAYER HIDDEN LAYER OUTPUT LAYER
i=1,2, .n j=1,2, .m k=1,2, .p
y(p)
X (1)
X (2)
X(n-1)
X(n)
.
.
b(m-1)
b (m)
b(1)
b (2)
Fig. 3. An artificial neural network model with a single hidden layer
with forward feed.
(1)
In equation 1, xi refers to input vector and yp refers to output
vector (m and n neuron count) , wi,j indicates input layer to
hidden layer weights and βj indicates output layer to hidden
layer weights, bj represents the threshold value of neurons in the
hidden layer and g(.) represents activation function. Input layer
weights (w) and bias (bj) values in the equation are randomly
assigned. Activation function (g(.)), input layer neuron count (n)
and hidden layer neuron count (m) are assigned in the beginning
step [8-15].
F. Model performance evaluation
The topics addressed in this section are the two measures that
affect the performance of the model and the algorithm used, the
first one being the division of data set into training and test data
set and the second one being the definition of expressions
measuring the performance. In the first measure, the data set is
divided into three parts as training, validation and test data by
three-phase division in K-Fold method, and model selection and
performance status are simultaneously performed. In the second
measure, performance assessment of classifier models generally
uses a validation value. Validation value can be measured as the
ratio of data count detected or estimated correctly by the
algorithm into all data in the data set.
(2)
III. EXPERIMENTAL RESULTS
100% performance was achieved in network training via
ELM and 99.33% performance was achieved in the training test
These results were obtained by using MATLAB 2103b software
and a PC with Intel i7-6500 CPU and 8 GB RAM. While
attaining these results, cell count in the hidden layer is 50 and
activation count is sigmoid for ELM.
International Journal of Computer Science and Electronics Engineering (IJCSEE) Volume 6 Issue 1 (2018) ISSN 2320-4028 (Online)
6
Fig. 4. ELM performance chart.
Comparison of the results of different classification methods
Achieved performance of ELM method and achieved
performance of other machine learning methods (Support
Vector Machine (SVM), Naive Bayes (NB)) are presented in
Table 2. As deduced from these data, ELM achieved higher
performance compared to other methods in terms of
performance and speed.
TABLE I. ACCURACY OF MACHINE LEARNING METHODS.
Methods Train
Accuracy Test / True Accuracy
ELM 100% 99.33%
NB 100% 96,00%
SVM 100% 96,00%
IV. CONCLUSION AND DISCUSS
In this study, features in the database created for features
from images are classified by determining the input and output
parameters for the ELM classifier. Results obtained by ELM
show that ELM has higher achievement compared to other
classifier (SVM and NB) methods. This study is considered to
be an applicable design in automated systems with
high-performing classification against the image analyzing.
REFERENCES
[1] Dedeoğlu, Y., Töreyin, B. U., Güdükbay, U., & Çetin, A. E. (2006).
Videoda Nesne Sınıflandırması için Siluet Tabanlı Yöntem. Signal
Processing and Communications Applications, 1-4.
[2] Şirvan, O., Karlık, B., & Tunalı, T. (2004). Yapay Sinir Ağları
Kullanılarak Güvenlik Amaçlı Biyometrik Tanıma.
[3] Hsu, W., Lee, M. L., & Zhang, J. (2002). Image mining: Trends and
developments. Journal of Intelligent Information Systems, 19(1), 7-23.
[4] Basheer, I. A., & Hajmeer, M. (2000). Artificial neural networks:
fundamentals, computing, design, and application. Journal of
microbiological methods, 43(1), 3-31.
[5] Domeniconi, C., Peng, J., & Gunopulos, D. (2001). An adaptive metric
machine for pattern classification. In Advances in Neural Information
Processing Systems (pp. 458-464).
[6] Lichman, M. (2013). UCI Machine Learning Repository
[http://archive.ics.uci.edu/ml]. Irvine, CA: University of California,
School of Information and Computer Science.
[7] Umut Orhan “Makine Öğrenmesi ders notları” , Çukurova Üniversitesi
http://bmb.cu.edu.tr/uorhan/MLearn.htm 02.08.2015
[8] Huang, G.-B., Zhu, Q., Siew, C., Ã, G. H., Zhu, Q 2006. Extreme learning
machine: Theory and applications. Neurocomputing, 70(1–3), 489–501.
[9] Huang, G., Zhu, Q., & Siew, C. 2004. Extreme Learning Machine : A
New Learning Scheme of Feedforward Neural Networks. IEEE
International Joint Conference on Neural Networks, 2, 985–990.
[10] Kaya, Y., Ertuğrul, Ö. F., & Tekin, R. 2014. An expert spam detection
system based on extreme learning machine. Computer Science and
Applications, 1(2), 132-137.
[11] Ertuğrul, Ö. F. 2016. Aşırı Öğrenme Makineleri ile biyolojik sinyallerin
gizli kaynaklarına ayrıştırılması. D.Ü. Mühendislik Dergisi Cilt: 7, 1, 3-9
[12] Tağluk, M. E., Mamiş, M. S., Arkan, M., & Ertuğrul, Ö. F. Aşırı Öğrenme
Makineleri ile Enerji İletim Hatları Arıza Tipi ve Yerinin Tespiti In Signal
Processing and Communications Applications Conference (SIU), 2015
23st (pp. 1-4).
[13] Faruk Ertuğrul, Ö., & Kaya, Y. 2014. A detailed analysis on extreme
learning machine and novel approaches based on ELM. American Journal
of Computer Science and Engineering, 1(5), 43–50.
[14] Ertugrul, Ö. F. 2016. Forecasting electricity load by a novel recurrent
extreme learning machines approach. International Journal of Electrical
Power and Energy Systems, 78, 429–435.
[15] Ertugrul, Ö. F. 2016. Determining the order of risk factors in diagnosing
heart disease by Extreme Learning Machine International Conference on
Natural Science and Engineering (ICNASE’16), pp.389-396
First A. Author a Yasin Sönmez received the master
graduated in computer science from University Fırat,
Turkey in 2012. He is currently Phd student in
software engineering. end working for the computer
technology at the University of Dicle. His research
interests include Computer vision and video analysis.
yasin.sonmez@dicle.edu.tr
Türker TUNCER was born in Elazig, Turkey, in
1986. He received the B.S. degree from the Firat
University, Technical Education Faculty, Department
of Electronics and Computer Education in 2009, M.S.
degree in telecommunication science from the Firat
University in 2011 and Ph.D. degree department of
software engineering at Firat University in 2016. He
works as research assistant Digital Forensic
Engineering, Firat University. His research interests
include data hiding, image authentication, cryptanalysis, cryptography,
image processing.
Engin Avcı received the PhD degree from University of Fırat in 2000. He is
currently a professor of software engineer at Fırat University His research
interests include Computer vision and video analysis. Fuzzy Logic, Artificial
Intelligence Techniques, Biomedical, Signal Processing, Intelligent Systems,
Radar Target Recognition, Pattern Recognition, Image Processing, Data
Mining, Information Security, Cryptography
International Journal of Computer Science and Electronics Engineering (IJCSEE) Volume 6 Issue 1 (2018) ISSN 2320-4028 (Online)
7
top related