You are on page 1of 4

2009 International Association of Computer

2009 IACSIT
Science
Spring
andConference
Information Technology - Spring Conference

Classification and Evaluation of Tomato Images Using Several


Classifier
Hassan Asadollahi
Department of computer
Islamic Azad University Firouzkuh Branch
Tehran,IRAN

Morteza Sabery Kamarposhty


Department of computer
Islamic Azad University Firouzkuh Branch
Tehran, IRAN
msaberyk@yahoo.com

Hasan_asadollahi@yahoo.com

Mir Majid Teymoori


Department of Electrical Engineering
Islamic Azad University Firouzkuh Branch
Tehran,IRAN
mteymoor@yahoo.com
Abstract: The goal of this paper is tomato
classification by images processing. Data was
used in this paper in related to 90 images of
tomato. We extract 10 features from 90 images
by machine vision technique. We evaluate
results that accepted in considered images. Also
we have classified these data using different
classification methods and then compare them.

used different classifiers to classify the tomato


data and then compare them.
II. Technical Description
We want to classify tomato by three groups;
high quality, medium quality, low quality. We
extract 10 Characteristics of 90 images based on
machine vision techniques. These characteristics
are:

Key-words: Image Processing, tomato, Neural


Networks, Classification

1.
2.
3.
4.

I. Introduction
Automatic fruit sorting using machine vision
can improve the quality of the product, abolish
inconsistent manual evaluation, and reduce
dependence on available manpower. Machine
vision is known to be a useful tool for external
features measurement (e.g. size, shape, color and
defects).

5.
6.
7.
8.
9.
10.

At first, we selected 40 images randomly. In


these images, some data considered for
learning data and the others used for test
data.

For example, qualification classify of potato,


pepper, cucumber, tomato and etc [1-8]. In this
paper we have focused on tomato classification.
Some research has been done in this subject.
Some of them are based on tomato color which
is classified based on the amount of redness or
greenness of tomatoes [9, 10].

Figure 1 shows the samples of tomato that is


used in this paper.

Some of the other methods are focused on


neural networks and fuzzy logic techniques for
classification of qualification [9, 10]. We have

978-0-7695-3653-8/09 $25.00 2009 IEEE


DOI 10.1109/IACSIT-SC.2009.47

Greenness grade
Redness grade
Yellowness grade
Average of greenness, redness and
yellowness grade
Entropy
Energy
Contrast
Sympathetic
Circularity
Area

459
471

a given percent of all data. This diagram is


based on the correct classification percent
for each method, versus education data of
total data.

Based on the table we can conclude that


firstly the results are not based on certain
regularity. Secondly the correct percent of
result are not interesting and suitable.
To understand of this subject, we
investigated the used characteristic. Result
had showed that some of characteristics do
not any affect on result. That is, if we ignore
these characteristics it will not affect on the
classification results. These unaffected
characteristics are:
1234-

Yellowness, greenness, redness Average


Sympathetic
Circularity
Area

Some weak points of these characteristics

are:

1- Less amount of data


2- Unnecessary characteristic extraction
3- Dont
extraction
of
necessary
characteristic such as the deterioration
area of tomato.

For this purpose we collect the 50 other data


and ignore the above Unnecessary
characteristics, then we consider the amount
and percent of deterioration area of tomato
versus total area of tomato in the
classification.

Fig 1. Images of Tomato used in this paper.


We use the below classifier methods for data
classification:
1.
2.
3.
4.
5.
6.
7.
8.

We can see that the random tree and random


forest method have a best accuracy than the
other method.

NaiveBayes
Multi Layer Perceptron
RBF Network
Neural Binary Tree
Random Tree
Random Forest
Instance Base K-Nearest Neighbor
K-Star(K*)

The obtained results show a good


improvement that is shown in Figure 3.

The Result of this method has showed in the


Figure 2. It is mentioned that the test and
education data has been considered based on

472
460

Correct Classification Percent

Accuracy Chart











%

%

%

%

%

%

%

%

%

NaiveBayes

.

.


. 

.





.

MLP

.


.

.

.











RBF


.

.

 .

.



 .



.

NB Tree

.

. 

 .

 .

.



Random T

 .

.


.

.







Random F


.

.

.





 .







KNN

 .

 .

 .





.

 .

.

K-Star

 .

 .



.



 .

.





Correct Classification Percent

Figure 2- Correct Classification Rate for 40 images

Accuracy Chart













%

%

%

%

%

%

%

%

%

NaiveBayes

.

.

 .


.

.

.

 .

.

MLP


.

.

.

 .


.

.

 .

.

.

RBF

.

.

.

.

.

.

.

.

NB Tree

 .

 .

 .

.

.

.

.

Random T

.

.

.

.

 .

 .

 .

.

.

Random F

.

.

 .

.

.


.


.

.

.

KNN

.

.

.

 .

.

.

.

.

K-Star

 .


.

.

.

 .

 .

.

 .

.

Figure 3- Correct Classification Rate for 90 images


473
461

V. Conclusions
[8] Y.Gejima, H.Zhang, M.Nagata, Judgment on
Level of Maturity for Tomato Quality Using L*a*b*
Color image processing, lntemational Conference on
Advanced Intelligent Mechatronics (AIM 2003),
p1355-1359

We can conclude that the below characteristics


have more effect on the tomato data
classification
1.
2.
3.
4.
5.

Greenness grade
Redness grade
Yellowness grade
the deterioration area of tomato
the percent of deterioration area versus
total area tomato
6. Circularity

[9] Masatem Nagata, Pepito Menguito BATO, asafui


Mitarai, Qixin Cao, Taka0 Kitahara, Study on Sorting
System for Strawbeny Using Machine fision, Journal
of JSAM 62( 1) (2000).
[10]O.castillo, R.Cardona, P.Melin, A Hybrid
Approach for Automated Quality Control Combining
Learning Vector Quantization Neural Networks and
Fuzzy Logic. IEEE ASME 2002. p2081-2085

Based on the different classifiers and the


obtained results, we can conclude that the MLP
and random tree classifier method have better
results with respect to other classifier; also the
total data increasing and amount of test data
increasing cause improvement in classifier
performance.
References
[1] S. Nimesh, M. J. Delwiche, R. S. Johnson, Image
analysis methods for real-time color grading of
stonefruit, Computersand-Electronics-in-Agriculture,
9: 1, pp. 71-84, 1993.
[2] B. K. Miller, M. J. Delwiche, Automatic grading
of fresh peaches using color computer vision, ActaHorticulturae,No.254, pp. 161-166, 1989.
[3] B. K. Miller, M. J. Delwiche, A color vision
system for peach grading, Transactions-of-theASAE, 32: 4, pp. 1484-1490,1989.
[4] Singh-N; Delwiche-MJ; Johnson-RS; ThompsonJ, Peach maturity grading with color computer vision,
Paper-American-Society-of-Agricultural-Engineers,
No. 92-3029, pp. 23, 1992.
[5] Z. Varghese, C. T. Morrow, P. H. Heinemann, H.
J. Sommer III, Y. Tao, R. M. Crassweller, Automated
inspection ofGolden Delicious apples using color
computer
vision,
Paper-American-Society-ofAgricultural-Engineers, No. 91-7002,
16pp, 1991.
[6] Y. C. Hung, K. Morita, R. Shewfelt, A.
Resurreccion, S. Prussia, Color evaluation of apples,
ASAE, No. 93-6541, pp.15, 1993.
[7] Y. Tao, P. H. Heinemann, Z. Varghese, C. T.
Morrow, H. J. Sommer, Machine vision for color
inspection of potatoes and apples, Transactions-ofthe-ASAE, 38: 5, pp. 1555-1561, 1995.

474
462