Grading Gedong Gincu Mango Using Image Processing Method

Enrico Syaefullah, Maulida Hayuningtyas, Dondy ASB

Indonesian Center for Agricultural Post-harvest Research and Development
Jl. Tentara Pelajar 12, Bogor 16114, Indonesia
Email : [email protected]

Keywords : gedong gincu mango, image processing, grading

Abstract
The domestic and export market of gedong gincu mango require a prime and
consistent quality. However, manual grading system based on eye human vision
resulting in quality inconsistentcy. The objective of this research was to develop a
grading system for the gedong gincu mango using image processing method. The
gedong gincu mango were classified into different quality standars i.e. A, B and C
based on weight (grams) and areas (pixel). A camera was used to capture the image in
16 megapixsel resolution. The results indicated that image processing method can
grade gedong gincu mango nondestructively. There were a strong relationship between
the image processed dimensions and the actual dimensions of gedong gincu mango at
weight (grams) and areas (pixel) with r = 0.993. The validation of algorithm for the
grading system of gedong gincu mango performed a high accuracy of 100% for A, 95%
for B, and 100% for C.

INTRODUCTION
Indonesia has hundreds of mango varieties and also served as fifth biggest mango
producer in the world. Since 2011, there were a significant raise in Indonesia’s mango
productions: from 1,287,287 tons in 2010 and become 2,131,139 tons at 2011. BPS (2011)
also stated that 1.065.259 ton were imported at year 2010. Mango has high of Vitamin C and
can be consumed in fresh or processed forms. Every 100 grams of mango contains 44.00
cal, 0.7 g protein, 0.2 g fat, 11.2 g carbs, dan 41 g vitamin C (Anonymous, 2009).
Indonesian National Standard (Standar Nasional Indonesia) has listed the standard for
gedong mango in SNI 01-3164-1992 based on the weight: big fruits >400 g; medium fruits
200-250 g; small fruits 150-199 g and very small fruits 100-149 g (SNI. 1992).
One of Indonesia most famous mango variety is gedong gincu. Gedong gincu mango
has unique features in its size, shape, color, taste and smell. It has a roundish shape, medium
in size, the reddish to yellowish skin color, bright yellow fruit carp and its delicious smell of
fruits has enchant many people.
The main problems facing on Indonesian mango in international trade are poor
quality due to unproper post-harvest handling techniques, specifically in grading. The fruit
grading is mostly done manually hance inaccurate and subjective, resulted in variety of
grade (almost reach 30%). This kind of subjectivity can be reduced by developing a fine
grading method. One of the methods is using a non-destructive image processing technique.
This method is a visual system build using a CCD (charge coupled device) camera as a
sensor. This device helps grading much easier; through the screen, fruits can be grade
visually by its appearances i.e. size, color, shape and spots on fruit rind. This method is

537
making grading process more objective and consistent. Furthermore, the grading result will
be more equal with acceptable level of diversity.

METHODOLOGY
Research subject is gedong gincu mango wich were obtained from Cirebon, West
Java as the main producer of this mango. Sampling was conducted by using purposive
random sampling method; a random sampling limited by weight criteria (quality standard A,
B and C). Amount of fruit sampling is 40 for each quality standard. Those fruits sample
were packed in cardboard with papers as extraprotection then transported to laboratory by
car and
A set of image processing devices consist of a digital camera, four fluorescent lamp,
a box for image acquisition, luxmeter, a computer set, a digital scale, tripod, boards and
black and white fabric as a fruit pad. Softwares were constructed and operated on MS -
Windows. Delphi-based algorithm is use to image processing. Hardwares used were
computer with Intel Pentium core duo processor.
This research was conducted in Juli – November 2012 at ICAPRD laboratory in
Bogor, Indonesia.

Procedure

Image Acquisitions
A number of image from 120 mango samples of A, B and C were taken using image
processing devices, from the side angle of the fruit. This method of acquisition was used to
measure the parameter area. Fruit reflectance intensity then captured by camera sensor.
Those images were recorded and saved in JPG format.

Image Processing
In order to analize the saved images, researchers have made an image processing
program operated in Windows. The program consists of some modul in image processing
i.e. file opening, filter modul, color processing modul, and graphic-modelling modul. This
program can be used to analyze the characteristics of fruit weight-based quality. First step
are open and process the sample fruit image, on a corresponding weight, to obtain the
average area as a basis of fruit grading. The quality parameter observed is the area of fruit
projection.
Measurement on weight and area was done in order to tell each of the fruit quality.
This measurement was conducted by using Euclidian methods, which is defined as the
length of the farthest distance. Distance is obtained by multiplying the number of pixels with
a pixel size. The formula used to measure the length is:
d i1, j1, i 2 , j 2   i 2  i1 2   j 2  2
j1 

Data Processing
Data that obtained through image processing are the length, width and area.
Afterwards, those data will be processed to find the correlation with the actual fruit length,
width and area. Correlation analysis was used to determine the relationship between the
results of grading by manual and image processing. Furthermore,based on the analysis
correlation coefficient for each quality parameter isthen sought.

Program Validation

538
 A computer program was made to validate the data images, and the using 20 different
fruits for quality A, B and C to look at the program's ability to classify the fruit grade. A
baseline value obtained from the data prosessing, this value then act as a benchmark of the
minimum and maximum parameters of the fruit (length, width, and area). Given the
parameters, the program is expected to conduct an objective and precise fruit grading in
order to meet The Indonesian National Standards (SNI).

RESULT AND DISCUSSION

Mango Image Processing

Images records are made using camera then stored in JPEG files with a resolution of
16 megapixels. The results are stored into a bmp file in the hard disk memory for further
image analysis using Photoshop program. Delphi were used as an image processing
program, the program display is shown in Figure 4.
Program Running Process:
1. Start the image processing program
Process of data files generation and storage, by using the start, next and finish button.
2. Image acquisition and image storage into database
By click the open button, the dialog box will show up. The next step is to find the folder
location of fruit images and opened it, then use the save button to storage the image
processing data using Microsoft Acces 2007 database system.
3. Separating images with background color
Clicking the thresholding button will show the fruit images, that already separate with
the background color. The thresholding button continuosly use to separate the object and
background.
4. Calculating the length, width and area of fruit
By using long, diameter and areas button, the object will be moved to a white
background. The length of the fruit obtained by using Euclidian distance method, in
which the principles of the program was to calculates the horizontal diameter of the loop
coordinates X and Y. Furthermore, the black pixel were obtained with minimum X value
(X1) at Y1 and X of maximum value (X2) in order to get the value of (X1, Y1) and
value of (X2, Y2). Fruit diameter was calculated by averaging the vertical and horizontal
diameter. The vertical and horizontal diameter obtained by the principles of Euclidian
distance.
5. Validation data and grading process
Grading process was done when the entire fruit grades sample A, B and C was included
in the database. Quality parameters with the smallest value of all data Grade-A fruit (40
samples). The basic principle in grading process is when one of the quality parameter
has a greater value than the analysed fruit parameter, then the fruit has a grade B.
However, if all the quality parameter values remain smaller than the analysed fruit
parameter then the fruit has a grade A. The grading steps are: by using open and save
button to open and save; the thresholding button to separate fruit images from
background; length, diameter and area button for geometric operations; and ends with
grade result.

539
Gedong Mango Image Characteristics
Correlation between Dimensional Images Processing with Actual Dimensions
Correlation between the values of mango area with mango weights was shown in
Figure 5.
The Graph (Gig.5) demonstrate a high value of R2 = 0.965. A high value indicates
that the linear model developed explaining that the relationship between the two variables
were fairly reliable. Therefore, there is a correlation between the Gedong mango weight
with the projected image area (pixel). The result demonstrate that the image processing
techniques can be adopted in grading system on post-harvest handling process. The mango
weight measurement simultaneously plotted with image processing, assuming a correlation
between manual and digital measurement. Such assumption can be drawn as the relationship
measurement reliability by using gedong mango images.

Gedong Mango Area Characteristics

There are of grade A mango, is between 8261 pixels to 10505 pixels with an average
value of 9286 pixels, For grade B, it is between 6821 pixels to 8165 pixels with an average
value of 7622 pixels. Whereas for grade C, the area value is between pixels to 6941 pixels
5884 with an average value of 6437 pixels. Comparison between the areas of each grade can
be seen in Figure 6.

Grade Estimation
The grade estimation are developed using image processing techniques based on the
linear relationship model that has been built. It divided into three groups, namely grade A,
grade B and grade C mangoes. In order to estimate the grade samples there were 40 grade A
mangoes, 40 grade B mangoes, and 40 grade C mangoes had been used. The grade
estimation performed by using the mango area parameters. The spesification of gedong
mango grade based on image processing presented in Table 1.
Grading process was done to classify fruits into several grades, based on size, color,
ripening level, defects etc. in order to fulfill consumer needs. Satuhu (1997) explain that
fruit grading process have some benefit, namely: prevent the price decline due to mixed
grade, obtain a similar grade between sellers and buyers, avoid excess supply by excluding
the low grade quality, avoid contact between the good fruit with the damage/disease/rotten
fruit, increase revenue and sales volume as well as simplicity in marketing.

Program Validation
Gedong mangoes with grade A had a range of areas larger than 8261 pixels, grade B
mango area ranged between 8620 pixels to 6821 pixels, whereas grade C mango had a range
areas less than 6820 pixels.
Table 2 ilustrate the result of gedong mango grading validation using image
processing. The grading process showed that from 20 mangoes with estimated grade A, no
defects were found and all of them confirmed as a grade A. The same process applied to 20
mangoes with estimated grade B, 19 mangoes (95%) confirmed as a grade B and yet 1
mango (5%) classified into grade C. No defects were found on 20 mangoes with estimated
grade C, and all of them confirmed as grade C.

CONCLUSION
1. Gedong mango grading (size and area) can be done in non-destructive way, by using
image processing software.

540
2. Direct correlation between mango fruit weight (gram) with area (pixel) shown by value
of r = 0.993.
3. Program validation on image processing algorithm for gedong mango grading showed
accuracy 100% on grade A, 95% on grade B and 100% on grade C.

Literature Cited
Ahmad, U. 2005. Pengolahan Citra Digital dan Teknik Pemrogramannya. Yogyakarta:
Graha Ilmu. (In Bahasa Indonesia).
Ahmad, U., I.D.M. Subrata., S. Gunayanti. 2004. Pemutuan Buah Mangga Berdasarkan
Penampakannya menggunakan Pengolahan Citra.Jurnal Keteknikan Pertanian. Vol.18
No.1 April. (In Bahasa Indonesia).
Ahmad. U. 2010. Aplikasi Teknik Pengolahan Citra dalam Analisis Non-Destruktif Produk
Pangan. Rubrik Teknologi Vol.19 No.1 Maret. (In Bahasa Indonesia).
Anonymous, 2008. Workshop Kajian Pengembangan Mangga dan Manggis.
http://www.hortikultura.deptan.go.id/index.php?option=com_content&task=view&id=1
21&Itemid=1. diakses 20 Mei 2011. (In Bahasa Indonesia).
Anonymous. 2009. Standar prosedur operasional pengolahan mangga direktorat pengolahan
hasil pertanian Direktorat Jenderal Pengolahan dan Pemasaran Hasil Pertanian
Departemen Pertanian. Jakarta. (In Bahasa Indonesia).
Broto, W. 2003. Mangga, Budi Daya, Pascapanen dan tataniaganya. Agromedia Pustaka.
Jakarta. (In Bahasa Indonesia).
Gao, X. and J. Tan. 1996. Analysis of Expanded-Food Texture by Image Processing Part I:
Geometric Properties. Journal of Food Process Engineering 19, 425-444.
Haralick, R.M., K. Shanmugam and I. Dinstein. 1973. Textural feature for image
clasification. IEEE Trans. On System, Man and Cybernetics. 3(6): 610-621.
Jain, R., R. Kasturi, and B.G. Schunck. 1995. Machine Vision. McGraw-Hill Book, Inc.
New York, USA.
Satuhu, S. 1997. Penanganan Mangga Segar untuk Ekspor. PT Penebar Swadaya. Jakarta.
(In Bahasa Indonesia).
Standar Nasional Indonesia. 1992. SNI 01-3164-1992 UDC. (In Bahasa Indonesia).

541
Tables

Table 1. The spesification of gedong mango grade based on image processing

Grade parameters
Grade
Area (A) (Pixel)
Grade A A > 8261
Grade B 6821 < A < 8260
Grade C A < 6820

Table 2. Result of mango grading validation using image processing

Program/Manual Grade A Grade B Grade C Total

Grade A 20 (100%) 0 (0%) 0 (0%) 20
Grade B 0 (0%) 19 (95%) 1 (5%) 20
Grade C 0 (0%) 0 (0%) 20 (100%) 20
Total 20 19 21 60

542
Figures

Fig. 1. Gedong gincu mango

Manual measurement Fruit weight (gram)

Correlation graph between image

processing and manual
measurement
Image processing Area (pixel)

Distribution chart of fruit

weight (gram) and area
(pixel)

Fig. 2. Fruit weight and area measurement diagram

Graphs distribution area Fruit with different quality

(pixel) based on the grade

Image records and storage

Area determination
(pixel) for each grade
Image reading

Integration of quality parameters and

limits for each grade Image analysis for grade characterization:
weights, fruit projection area

Fruit grading program

Mango grading program

Meet quality standards

Fig. 3. Program Validation and Mango grading diagram

543
 Fig. 4. Display of Image Processing Program

korelasi area vs bobot ( r = 0,993; P = 0,000 )

11000

10000
area (piksel)

9000

8000

7000

6000

150 200 250 300 350

bobot (gram)

Fig. 5. Correlation between area and weight values

Fig. 6. Gedong mango area distribution for grade A, B and C

544