The Imaging Methods for Classifying Crispiness of Freeze-Dried Durian using Fuzzy Logic

World Academy of Science, Engineering and Technology 25 2007

The Imaging Methods for Classifying
Crispiness of Freeze-Dried Durian using Fuzzy
Logic
Sitthichon Kanitthakun, Pinit Kumhom, and Kosin Chamnongthai



II. BASIC CONCEPTS
Abstract—In quality control of freeze-dried durian, crispiness is
Our proposed method is based on the analysis of freeze-
a key quality index of the product. Generally, crispy testing has to be
durian images taken from a Scanning Electron Microscope
done by a destructive method. A nondestructive testing of the (SEM). In the analysis, we start by trying to find features of
crispiness is required because the samples can be reused for other the images that will represent the crispiness of the durians.
kinds of testing. This paper proposed a crispiness classification
method of freeze-dried durians using fuzzy logic for decision According to the study of the freeze dry process, the texture of
making. The physical changes of a freeze-dried durian include the durian has changed because the water molecular flow through
pores appearing in the images. Three physical features including (1) and damage the texture of durian [11], [12]. Due to the effect
the diameters of pores, (2) the ratio of the pore area and the pores as illustrated in Fig. 1 are created. According to the
remaining area, and (3) the distribution of the pores are considered to
study in [1], the characteristic of these pores indicates the
contribute to the crispiness. The fuzzy logic is applied for making the
crispiness of the freeze-dried durian. In our previous work,
decision. The experimental results comparing with food expert we propose to use 3 features of the pores as the input to fuzzy
opinion showed that the accuracy of the proposed classification interference system for the classification.
method is 83.33 percent.


Keywords—Durian, crispiness, freeze drying, pore, fuzzy logic.
I. INTRODUCTION
LONG with Mangosteen, Durian is a famous fruit of
A Thailand, which is the first county that exports durians.
Thailand exports both fresh and processed durians for more
than 10,000 million baht per year [1]. One of the popular
processed durians is the freeze-dried durian. It had been
shown by a research result [1], and by its popularity that

freeze-dried durian is preferable in the market comparing to
Fig. 1 Freeze-dried Durian Image
other dried durian products. As a result, in order to keep high
quality freeze-dried durian, it is very important to control the
The three features include (1) the average diameter of the
quality of the product.
pores, (2) the ratio between the areas with and without pores,
On the quality control issue, recent research works in [2]-
and (3) the distribution of the pores. However, in our previous
[6] investigated the quality of the freeze dried products. These work, we assume that all pores have the same size. As a result,
researches studied the effects of the freeze-drying process to the ratio of the areas and the distribution of the pores are
the products. Moreover, image analysis is a tool to support the identical. Although it gave a good result, we propose to
above investigations [7]-[9]. However, these previous works improve these features as follows.
did not address a quality of freeze-dried durian in term of
For the first feature, if the diameter of the pore is large, it
crispiness, which is a key quality parameter for the processed means the freeze-dried durian is crispier. This first case is
durians.
shown in Fig. 2 (a) and (b). Therefore, we propose to use the
We had introduced a classification of freeze-dried durian average diameter of the pores as the first feature. However,
crispiness based on image processing and fuzzy logic in [10]. observing only the average diameter of pores is not enough to
However, our previous method had some drawbacks on the consider the crispiness since freeze-dried durians with too few
feature extraction and the fuzzy rules. Therefore, in this paper, large-size pores mean less crispiness than those with more
we propose its improvement. The remainder of the paper is small-size pores as shown in Fig. 2 (c) and (d). This feature
organized as follows. The basic concept of our proposal is can be represented by the ratio between the total pore area and
described in section 2. Then, the detail method is explained in the remaining area. For the third feature, although the average
section 3. Section 4 describes the experiments and results. size and the ratio of the areas are large, the freeze-dried
Section 5 and 6 give the discussion and conclusion durians might not be crispy if all the pores are not distributed
respectively.
well as shown in Fig. 2 (e) and (f). Therefore, we measure the

62

---

World Academy of Science, Engineering and Technology 25 2007

mean distance between a pore to all of it neighbors. Then, we
use the average of these mean values as the representation of
the distribution.
Since these three features of the pores that indicate the
crispiness are against each others, we propose to use the fuzzy
logic principle to decide the level of crispiness.

Find centers and diameters
of all pores
Calculating the features


Fig. 3 Flow chat of proposed method
A. Freeze-Dried Durian Image
The images of the freeze-dried durians under consideration

are taken from an SEM. All images are taken under the same

conditions such as same image size.
Fig. 2 Crispiness based on feature of the pores
B. Feature Extraction
This step is for feature detection. First, an edge detection
III. PROPOSED METHOD
method is applied to find all edges in the images. Then, the
Based on the basic concept, the proposed classification of Hough transform is used to find the centers and corresponding
freeze-dried durian crispiness is concluded in Fig. 3. The diameters of all circles, which are considered to be the images
following subsections describe the steps in the proposed of the pores. Finally, all 3 features are calculated as follows.
method.
The first feature is the average of the diameters. Then, the

ratio between the areas of pores and space is computed by

computing the sum of all circle’s areas first. Finally, the third

feature is computed by (1) computing the distances between a

pore to its neighbors, (2) computing the mean of the distance

and (3) computing the average of the mean distances.


C. Fuzzy Logic System

Fuzzy Logic (FL) is a step of decision process. The three

features computed in previous step are the inputs of if-then

rules of the FL. The AND operator is used for making the

decision. The detail of the fuzzy logic system is as follows.






63

---

World Academy of Science, Engineering and Technology 25 2007

1. Membership functions
2. IF-THEN rules
Each feature is considered to be a member of the FL,
The fuzzy knowledge base contains IF-THEN rules
wherein each member must have a membership function for describing the system behaviours. The AND operation is used
making the decision. Fig. 4 shows each input membership to construct the set of rules. Since there are 3 inputs and each
function of the 3 features. In these functions, the value of each have 3 possible values, there are totally 27 rules.
feature is divided into 3 levels, Low, Medium, and High. The
inputs and output can be classified in linguistic variables and
D. Classification
fuzzy sets as follows. The appropriate values for defining the
The output of FL is a probability value ranging from 0.0 to
membership functions are come from the study in [1].
1.0. Therefore, the values are converted to linguistic variables.
1. Feature 1 denoted by d: d ?{L ,
ow Medi
,
um Hi
}
gh In our case, crispy-A, crispy-B, crispy-C, and not-crispy are
classified by using the equal interval of output value between
2. Feature 2 denoted by R: R ?{Lo ,
w Me
,
dium Hig }
h
0.0 and 1.0; i.e. the output intervals for the not-crispy, crispy-
3. Feature 3 denoted by ?:? ?{Lo ,
w Me
,
dium Hig }
h
A, crispy-B, crispy-C are 0.0-0.25, 0.25-0.5, 0.5-0.75, and
0.75-1.0, respectively.

IV. EXPERIMENTAL RESULTS
We tested the proposed method with 12 freeze-dried durian
samples, which were taken the images using an SEM. The
Matlab 7.0 running on a PC with Pentium IV processor 3.0
GHz, 256 RAM MB was used for all image processing steps.
The results of fuzzy logic decision are shown as Table I.

TABLE I
FUZZY LOGIC CLASSIFICATION
No. Image
output
Class
1
0.153
not-crispy
2
0.387
crispy-C
3
0.250
not-crispy
4
0.618
crispy-B
5
0.836
crispy-A
6
0.550
crispy-B
7
0.174
not-crispy
8
0.471
crispy-C
9
0.651
crispy-B

10 0.797
crispy-A

Fig. 4 Inputs membership function
11 0.752
crispy-A

12
0.370
crispy-C
To set output membership function, we consider 4 classes
of crispiness including crispy with grade A, B, C and not-
crispy. Fig. 5 shows the output membership function.
We compare the results from our proposed method with

results from the subjective opinions of food experts using
same images. The results are shown in Table II.











Fig. 5 Output membership function









64

---

World Academy of Science, Engineering and Technology 25 2007

TABLE II
[4] D. C. Iwaniw, and G. S. Mittal, “Process optimization of freeze-dried
COMPARISON BETWEEN PROPOSED AND EXPERT DECISION
strawberries,” J. Agricultural, vol. 32, pp. 133-154, 1990.
[5] A. A. Khan, and Vincent, “Mechanical damage induced by controlled
No. Image Proposed
Expert
Result
freezing in apple and potato,” J. Texture Studies., vol. 27, pp. 143-157,
1996.
1 not-crispy
not-crispy
Correct [6] M. K. Krokida, V. T. Karathanos, and Z. B. Maroulis, “Effect of freeze-
drying conditions on shrinkage and porosity of dehydrated agricultural
2 crispy-C
crispy-C
Correct
products,” J. Food Enginrring., vol. 35, pp. 369-380, 1998.
3 not-crispy
not-crispy
Correct [7] E. R. Davies, “Image Processing for the food industry,” 2000.
4 crispy-B
crispy-B
Correct [8] R. C. Gonzalez, and R. E. Woods, “Digital Image Processing,” 1992.
[9] A. Von, “Fuzzy Logic & NeuroFuzzy Applications Explained,” 2000.
5 crispy-A
crispy-A
Correct [10] S. Kanitthakun, P. Kumhom, and K. Chamnongthai, “Classifying
6 crispy-B
crispy-B
Correct
Crispiness of Freeze-dried Durian using Fuzzy Logic,” Int. Computer
Symposium 2004.
7 not-crispy
not-crispy
Correct [11] S. A. Goldblith, L. Rey, and W. W. Rothmayr, “Freeze Dring and
8 crispy-C
crispy-C
Correct
Advanced Food Technology,” 1975.
[12] Olivera, “Processing Food: quality optimization,” 1999.
9 crispy-B
crispy-B
Correct
10 crispy-A
crispy-B
Incorrect
11 crispy-A
crispy-A
Correct
12 crispy-C
not-crispy
Incorrect

V. DISCUSSIONS
The experiment result shows that the proposed method can
classify crispiness of freeze-dried durian. However, some
images are wrongly classified comparing to the expert
opinions. Fig. 6 shows an example of the incorrect images. In
this image, some pores have the twist shape or non circle, so
these pores are not detected. As a result it is wrongly
classified.


Fig. 6 Image with non circle pores

VI. CONCLUSION
A method for no-expert classification of freeze-dried
durians based on its crispiness is proposed. The method is
based on the analysis of SEM images of freeze-dried durians.
Three features of circular pores are used as the inputs to the
fuzzy system. The experiments on samples of freeze-dried
durians showed that the method can classify crispiness with
83.33 % accuracy comparing with the expert options.
REFERENCES
[1] W. Khuaying, “Effect of Processing Conditions on Quality of Freeze-
dried Durian,” 2003.
[2] C. Hammami and F. Rene, “Determination of Freeze-dried process
variables for strawberries,” J Food Engineering, vol.32, pp.133-154,
1997
[3] C. Hammami, and F. Rene, “Process-quality optimization of the vacuum
freeze-drying of apple slices by the response surface method,” J. science
and technology., vol. 34, pp. 145-160, 1999.

65

---