コンピュータービジョンによるブロッコリー外観品質の客観 …kawasaki-shi,...

コンピュータービジョンによるブロッコリー外観品質の客観的評価

誌名誌名日本食品工学会誌 = Japan journal of food engineering

ISSNISSN 13457942

著者著者

牧野, 義雄若月, 碧衣網野, 元貴大下, 誠一佐藤, 朱里塚田, 正人

巻/号巻/号 17巻4号

掲載ページ掲載ページ p. 107-115

発行年月発行年月 2016年12月

農林水産省農林水産技術会議事務局筑波産学連携支援センターTsukuba Business-Academia Cooperation Support Center, Agriculture, Forestry and Fisheries Research CouncilSecretariat

Japan Journal of Food Engineering, Vol. 17, No. 4, pp. 107 - 113, Dec. 2016 DOI: 10.11301/jsfe.17.107

000 Original Paper 000

Objective Evaluation of External Quality of Broccoli Heads Using a Computer Vision System

Yoshio MAKIN01' t, Aoi W AKATSUKI1, Genki AMIN01, Seiichi OSHITA1, Akari SAT02, Masato TSUKADA2

1Graduate School of Agricultural and Life Sciences, The University a/Tokyo, 1-1-1, Yayoi, Bunkyo-ku, Tokyo 113-8657, japan

2Data Science Laboratories, NEC Corporation, 1753, Shimonumabe, Nakahara-ku, Kawasaki-shi, Kanagawa 211-8666, japan

A method to objectively evaluate the external quality of broccoli heads using a computer vision system (CVS), a type of digital camera, was proposed. The CVS was effective in calculating the spatial distribution of color space values of broccoli heads. Value of -a*/b* (chromaticity of Commission Internationale de l'Eclairage) was effective in evaluating the concentration of chlorophyll a. An -a*/b* value greater than 0.94 indicated fresh buds that remained green. Conversely, the value of a* was effective in evaluating the existence of anthocyanins that damage the external quality of the broccoli head. Buds with low concentrations of anthocyanin had an a* value less than -12. These two thresholds were used for visualizing high-quality buds with high concentrations of chlorophyll a and low concentrations of anthocyanin. The proportion of high-quality buds of typical heads that included low or high concentrations of anthocyanin reduced from 39% or 22% to 15% or 5.4% during 5 d storage. The proportion of high-quality buds of typical heads with low anthocyanin concentrations was usually higher than those with high anthocyanin concentrations. These results suggest that the proposed method using the CVS is suitable for objectively evaluating the external quality of broccoli heads and selecting high-quality heads. Keywords: anthocyanin, Brassica oleracea var. italica, chlorophyll, digital camera, nondestructive

analysis

1. Introduction

Broccoli (Brassica oleracea var. italica) is a popular

vegetable globally, with 22 Mt of the vegetable produced

in 2013, which has linearly increased by approximately

0.5 Mt/y since 1993, according to FAOSTAT (recorded

as the sum of broccoli and cauliflower [Brassica oleracea

var. botrytis ][i]. Since the early 1980s, consumption of

broccoli has been recommended to help prevent cancer

[1] because it contains sulforaphane, which is effective in

inhibiting Helicobacter pylori (Marshall et al. 1985)

Goodwin et al. 1989 [2,3], the main cause of gastric can-

cer [4]. However, degreening is the major limiting factor

that reduces the shelflife of stored broccoli [5] and is

caused by chlorophyll degradation [ 6]. Degreening may

reduce the health-enhancing effects of broccoli because

this pigment is effective in decreasing serum cholesterol

in mammals [7,8]. Lipton and Harris [9] optically evalu-

ated the color of broccoli heads using scales with nine

(Received 11 Jul. 2016: accepted 30 Sep. 2016)

t Fax: 03-5841-8174, E-mail: [email protected]

levels, while Jacobsson et al. [10] used a visual method

to evaluate broccoli color and set a threshold value for

degreening where 30% of the florets had turned yellow.

Objective evaluation of the surface color of broccoli

heads has also been attempted. Shewfelt et al. [5] nonde-

structively measured the color of six varieties of broccoli

using two types of colorimeters. Ren et al. [11] modeled

changes in the color of broccoli heads using values mea-

sured with a colorimeter, while Kasim et al. [12] mea-

sured the hue angle with a colorimeter to evaluate the

appearance of broccoli heads sealed in different types of

pouches after 1-methylcyclopropene treatment.

Conversely, Lipton and Harris [9] reported that degreen-

ing was localized and occurred at random. Their results

suggested that it may be difficult to use a colorimeter as

a point-based scanning instrument to detect degreening

at local sites.

Hyperspectral camera systems, which measure spec-

tral reflectance with spatial information, have been used

in research to evaluate pigment content in forests in

remote sensing areas [13]. This method has also been

© 2016 Japan Society for Food Engineering

108 Yoshio MAKINO, Aoi WAKATSUKI, Genki AMINO, Seiichi OSHITA, Akari SATO, Masato TSUKADA

applied as an analytical tool for food quality and safety

control [14]. Hyperspectral reflectance imaging was used

by Qin et al. [15] to detect citrus cankers and by Ariana

and Lu [16,17] to evaluate internal defects and surface

color of whole pickles and cucumbers (Cucumis sativus

L.). According to these reports, hyperspectral imaging

might be effective in detecting degreened sites on broc-

coli florets because it provides both spatial and spectral

information. For example, the chlorophyll concentration

at leaf and canopy levels in forests was estimated using

hyperspectral camera systems in a remote sensing area

[18,19]. However, this device is likely too expensive to

use to grade fruits and vegetables at farmlands or consol-

idating stations.

Digital cameras have been used for objectively observ-

ing the two-dimensional external quality of substances.

Although information from an image taken by a digital

camera is much less than that taken by a hyperspectral

camera, it may be adequate for evaluating the external

quality on the basis of color images. Ishikawa et al. [20]

objectively observed the green color of broccoli heads on

the basis of RGB data measured using a digital camera.

Schouten et al. [21] reported that 1000/R is sensitive

enough to evaluate broccoli color change. However, RGB

values are device dependent because each camera has its

own different color sensor characteristics [22].

Anthocyanin, a pigment, is known to be present in

broccoli buds [23]. This pigment reduce the quality of

broccoli heads [24]. Makino et al. [25] evaluated antho-

cyanin concentration using a digital camera to calculate

the proportion of red areas on the skin of mango

(Mangifera indica L.) fruits. This method might be useful

for objective grading of fruits.

In the present study, we attempted to objectively evalu-

ate the quality of broccoli heads on the basis of both

chlorophyll a and anthocyanin concentrations using a

computer vision system (CVS), a type of digital camera.

Color space values associated with chlorophyll a and

anthocyanin concentrations as indices of freshness and

appearance were considered. Thresholds to calculate the

proportion of high-quality buds in a head were deter-

mined. This study proposes a new method to visualize

the proportion of high-quality buds in broccoli.

2. Materials and methods

2.1 Samples

Eighteen normal variety (anthocyanin level: high;

hereinafter called H) and nine anthocyanin-less variety

(anthocyanin level: low; hereinafter called L) broccoli

heads were harvested on December 14, 2014, at a farm-

land in Tsu (Mie prefecture, Japan; 34° 42'20.6"N, 136°

24'24.7"E). After harvesting, the samples were trans-

ported to a laboratory within a day at ambient tempera-

tures. Table 1 lists the cultivars used in the current

study.

Immediately upon arrival at the laboratory six H and

three L heads were analyzed as 0 d storage samples

(hereinafter called 0). Another 18 heads were stored in a

unit at 25°C, 90% relative humidity (RH). Six Hand three

L heads each after 3 d and 5 d storage (hereinafter called

3 and 5, respectively) were used for the following experi-

ments. This storage was conducted to alter chlorophyll

concentrations of samples. Thus, samples with two antho-

cyanin levels and three chlorophyll levels were prepared

by changing varieties and storage periods, respectively.

Table 1 Storage period and anthocyanin levels of cultivars.

Storage period (d) Anthocyanin level Cultivar Number of samples

MONACO 3 High (H) SPL156 1

0 SPL172 2

Low (L) 13SP17 2 13SP18 1

AC43 2

High (H) AC49 1 HA39 1

3 MONACO 2

Low (L) 13SP17 2 13SP18 1

HA39 1

High (H) MONACO 1

SPL156 2 5 SPL157 2

Low (L) 13SP17 3


Objective estimate of broccoli 109

2.2 Observing samples using a CVS and

calculation of color space values of region

of interest

The CVS used in this study (USB camera, FMVU-

13S2C-CS, Serial #: 13088045, Point Grey Research Inc.,

Richmond, British Columbia, Canada; lens, 13FM06IR,

Tamron Co., Ltd., Saitama, Japan; color viewer, PIAS

IS-500, Sugiura Laboratory Inc., Tokyo, Japan) was the

same as that used in a previous study [25]. The following

experiment was conducted to determine the thresholds

for selecting high-quality buds (with low concentrations

of anthocyanin and green color retention). The sRGB

images of a head were acquired before and after region

of interest (ROI) sampling. The images were acquired in

bitmap format by FlyCapture2 ver. 2.4.3.11 (Point Grey

Research Inc.). The shutter was set at 30 ms and the

white balance (red and blue) was 530. By comparing

images shot before and after ROI sampling, the mean

sRGB value of the ROI (a square piece of peel measuring

10 mm x 10 mm) was calculated using MATLAB ver.

8.3.0.532 (MathWorks Inc., Natick, MA, USA). The con-

version from sRGB to device-independent Commission

Internationale de l'Eclairage (CIE) 1976 L*a*b* was done

using a method described in a previous study [26]. This

value was used for subsequent mathematical analyses.

Buds equivalent to the ROI sampled were then used for

measuring chlorophyll a and anthocyanin concentra-

tions.

2.3 Calculation of the proportion of high-

quality buds on the head

The proportion of high-quality buds on the head was

calculated using the Image Processing Toolbox comple-

mented in MATLAB. First, background was removed

from the image of a head. The number of pixels (VII) for a

head was then counted. The number of pixels for the

buds that had low concentrations of anthocyanin (a-less)

and retained their green color (gr) were counted from

the image without the background using the thresholds.

The thresholds to define high-quality buds were deter-

mined by plotting color space values of the ROI on a* vs

-a* lb* coordinates. A threshold using a* might be useful

for selecting buds that include low concentrations of

anthocyanin because this value increases with redness

[27]. Conversely, a threshold using -a* lb* might be use-

ful for selecting buds that retain their green color

because this value is reduced with green color loss

according to Makhlouf et al. [28]. High-quality buds (hi)

were designated as those satisfying both anthocyanin

and green color thresholds. The proportion of high-qual-

ity buds on a head was calculated using the following

equations:

Sa-less = 100 X Na-less (1)

w N

sgr = 100 x ___£_ (2) w S. = lOOx Nh;

hz W (3)

2.4 Analysis of chlorophyll a and

anthocyanin concentrations for the ROI

Chlorophyll a and anthocyanin concentrations in the

sampled buds were measured according to the methods

given in Porra et al. [29] and Nagata et al. [30], respec-

tively. These values were expressed on a fresh weight

basis. The relationships between chlorophyll a or antho-

cyanin concentrations and color space values were inves-

tigated.

2.5 Statistical analysis

Significance of differences between measured values

was confirmed by Tukey's honestly significant difference

test using JMP® Pro ver.12.2.0 (SAS Institute, Cary, NC,

USA)

3. Results and discussion

3.1 Relationship between color space

values and pigment concentrations of

broccoli buds in ROI

Color space values of broccoli buds in ROI are plotted

on a* vs -a* lb* coordinates (Fig. 1). The values of -a* lb*

for 0 d storage samples (OH and OL) were significantly

higher than those for 3H and 5 d storage samples (5H

and 5L). Pigment concentrations of broccoli buds are

shown in Fig. 2. Chlorophyll, the origin of the green

color, may be associated with -a* lb* and is well known to

be degraded over time after harvest [9]. In the present

study, only the data for chlorophyll a was shown because

its levels were -3.5 times higher than those of chloro-

phyll b, and chlorophyll a content declines more mark-

edly, leading to a decrease in the chlorophyll alb ratio in

higher plants [19]. Chlorophyll a concentrations for 0 d

storage samples (OH and OL) were significantly higher

than those for 3H and 5 d storage samples (5H and 5L)

the same as the -a* lb* values. Therefore, the value of -

a* I b* might be useful for evaluating green color or fresh-



1.2 I A I

~c I AB I 1-!------l b 1.0 I 0.8

~---------1-i-b~J

B . f2 0.6 c 'P 1-Il-ic c

+b 0.4

0.2 D f--cI---l a

0 -18 -16 -14 -12 -10 -8 -6 -4 -2

a*

Fig. 1 Color space values of regions of interest on broccoli

heads.

Cl -"' Cl .s "' :;, .r::; 0. e 0

:;:: ()

Dashed lines denote thresholds to select high-quality buds

with low anthocyanin concentrations (a' = -12) that retain

their green color (-a'/b' = 0.94). Closed and open symbols

denote buds of the normal variety (anthocyanin level:

high; hereinafter called H) and anthocyanin-less variety

(anthocyanin level: low; hereinafter called L), respectively.

Circles, triangles, and squares denote buds stored 0, 3,

and 5 d, respectively. Mean ± SE of three (L) or six (H)

observations has been plotted. Symbols followed by the same

capital and lower case letters are not significantly different (P

< 0.05; Tukey's honestly significant difference test) in -a'/b* and a' values, respectively.

400

A A 350 f--+--Hc f c 300 A

250 +be

200 B

B

+a +c B

150 ! ab I I 100

40 60 80 100 120 140 160

Anthocyanin (mg kg-1)

Fig. 2 Pigment concentrations of regions of interest on broccoli

heads.

Closed and open symbols denote buds of the normal

variety (anthocyanin level: high; hereinafter called H) and

anthocyanin-less variety (anthocyanin level: low; hereinafter

called L), respectively. Circles, triangles, and squares

denote buds stored 0, 3, and 5 d, respectively. Mean ±

SE of three (L) or six (H) observations has been plotted.

Symbols followed by the same capital and lower case letters

are not significantly different (P < 0.05; Tukey's honestly significant difference test) in chlorophyll a and anthocyanin

concentrations, respectively.


ness of broccoli heads. This value correlated with chloro-

phyll a concentration at a correlation coefficient accu-

racy of 0.89, with a square error of calibration 44 mg kg-1

(y=238x+71). In the present study, the threshold of-a*/

b* = 0.94, almost midway between the data for OH and the

data for 3L, was determined to select green buds as

shown in Fig. 1. Makhlouf et al. [28] reported that -a*/b*

values of fresh broccoli were ~o.90. This is close to the

threshold determined in the present study. Shewfelt et al.

[5] and Kasim et al. [12] reported that the hue angle of

fresh broccoli heads was ~120° and ~113°, respectively.

The threshold proposed in the present study can be con-

verted to a hue angle of 133°. This value may be high

enough to select green buds because the hue angle

increases with the degree of green color retention. The

chlorophyll a concentration of 3L was significantly

higher than that of 3H. Yamauchi et al. [31] reported that

apigenin, a flavonoid, caused accelerated chlorophyll

decomposition. Anthocyanin concentration as a kind of

flavonoid may concern with the difference of chlorophyll

a concentration between the samples including high and

low concentration of anthocyanin.

The values of a* for OL and 5L were significantly lower

than those for OH and 5H, respectively. Therefore, this

value may be useful for selecting buds that have low con-

centrations of anthocyanin. This value correlated with

anthocyanin concentration at a correlation coefficient

accuracy of 0.60, with a square error of calibration 27 mg

kg-1 (y=3.9x+137). In the present study, the threshold of

a*= -12, almost midway between the data for 3H and 3L,

was determined as shown in Fig. 1.

3.2 Objective selection of high-quality

broccoli buds using a CVS

Examples of objectively selected pixels using the

thresholds proposed in Fig. 1 are shown in Fig. 3.

The proportion of buds with low concentrations of

anthocyanin (a-less) in OL samples was higher by 31 per-

centage points than that in OH samples. This result indi-

cates that the method proposed in the present study is

effective in selecting buds with low anthocyanin concen-

trations. This method may also be effective in removing

heads with high concentrations of anthocyanin at farm-

lands or grading lines.

The proportion of buds that retained their green color

(gr) was calculated as 48% in OL samples. This value

reduced over time due to decomposition of chlorophyll

after harvest. This suggests that the threshold may be

effective in evaluating the freshness of buds. The propor-


tion of high-quality buds (hi) was calculated using the

same head as the above example for green color (gr).

The proportion of high-quality buds (hi) was naturally

lower than that of buds that remained green (gr) because

the buds over a* = - 12 were removed as shown in Fig. 3.

Furthermore, the proportion of high-quality buds (hi)

was reduced over time because of the reductions in

green color retention. The same tendency was observed

in the experiments on OH, 3H, and 5H samples. The pro-

portion of high- quality buds (hi) with low concentrations

of anthocyanin was naturally higher than those with high

concentrations of anthocyanin.

Mean values with standard errors for the proportion of

buds satisfying both thresholds were calculated using all

the heads tested in the present study (Fig. 4). The pro-

Fig. 3 Typical proportion of high- quality broccoli buds.

Pixels equivalent to low concentrations of anthocyanin (a' < -12; a-less) or green (- a' /b' > 0.94; gr) buds are shown as white pixels. Pixels that satisfied both thresholds are also

shown as white pixels (hi). 0, 3, and 5 denote heads stored for

0, 3, and 5 d, respectively. H and L denote the normal variety

(anthocyanin level: high) and anthocyanin-less variety

(anthocyanin level: low), respectively.

portion of green color (Fig. 4(b)) was reduced over stor-

age time independent of anthocyanin concentration the

same as shown by examples in Fig. 3. According to Fig.

4(b) , the proportion of green colored buds in 3H samples

was significantly lower than that in 3L samples. This

result agreed with the pigment concentration result in

Fig. 2. This may be caused by the decomposition of chlo-

60

(a) Q-B ----- ;I; ---Q----~):'.

AB AB

40 A

20 c

(b)

60

A

~ 40 c: 0 t 0 c. 2 0..

20

c

(c)

60

A 40

20

D E

4

Storage period ( d)

Fig. 4 Proportion of broccoli buds with low concentrations of

anthocyanin, with retaining green color, and of high quality.

Symbols denote the heads including (closed) high and (open)

low concentrations of anthocyanin. The proportion of buds

(a) including those with low concentrations of anthocyanin

under the threshold at a' = -12, (b) those retaining their

green color over the threshold at-a' lb' = 0.94, and (c) those

satisfying both . Means ± SE of three (open circles) or six

(closed circles) observations have been plotted . Symbols

followed by the same capi tal lette rs are not significantly

different (P < 0.05; Tukey's honestly significant difference test) within the same figure.



rophyll a in buds and a high concentration of anthocy-

anin stimulated by anthocyanin acting as a flavonoid dur-

ing 3 d storage.

The proportions of high-quality buds and, therefore,

heads of L samples were significantly higher than those

of H samples for all storage days (Figs. 4(a)-(c)). This

was naturally caused by differences in anthocyanin con-

centration leading to differences in a* values. This result

indicates that the method proposed in the present study

is effective in objectively selecting high-quality buds.

This method will also be effective in evaluating the exter-

nal quality of broccoli heads in farmlands or on grading

lines. According to Fig. 4(c), the proportion of high-qual-

ity buds in OL samples was ~40%. This low percentage of

high-quality heads may be caused by the stricter thresh-

old (-a*/b* > 0.94) compared with that used in previous studies [5,12,28]. As the method proposed in the present

study is an example, farmers or persons in charge at

grading lines can select any value depending on their

requirements. Overall, the method proposed in the pres-

ent study will support the application of objective and

automatic grading of broccoli heads.

4. Conclusions

A CVS was used for selecting high-quality broccoli

heads. RGB data available from the CVS could be trans-

formed to universal values of CIE 1976 L*a*b* by

MATLAB. Thresholds of a* < -12 and -a* lb* > 0.94 were used for selecting buds with low concentrations of antho-

cyanin and those that retain their green color, respec-

tively. These values were useful for calculating the pro-

portion of high-quality buds in a head. The method pro-

posed in the present study may be useful as an objective

and rapid grading tool.

5. Acknowledgment

All the broccoli heads used in the present study were

provided by Nacos Co. Ltd. (Tsu, Mie prefecture, Japan).

Financial support was provided by the Japan Society for

the Promotion of Science Grant-in-Aid for Scientific

Research (A) 25252045.

NOMENCLATURE

a*: Chromaticity defined by Commission Internationale

de l'Eclairage

b*: Chromaticity defined by Commission Internationale


de l'Eclairage

N: Number of pixels selected using thresholds on a

broccoli head

R: Component of the RGB color model

RGB: An additive color model in which red, green, and

blue light is added together in various ways to

reproduce a broad array of colors

S: Proportion of buds selected using thresholds on a

broccoli head, %

sRGB: An RGB color model standardized by the

International Electrotechnical Commission

W: Number of pixels equivalent to the whole area of a

broccoli head

Subscripts

a-less: low concentrations of anthocyanin

gr: retention of green color

hi: high quality

References

1) M. Nestle; Broccoli sprouts in cancer prevention. Nutr. Rev.,

56, 127-130 (1998).

2) J. W. Fahey, X. Haristoy, P. M. Dolan, T. W. Kensler, I.

Scholtus, K. K. Stephenson, P. Talalay, A. Lozniewski;

Sulforaphane inhibits extracellular, intracellular, and anti-

biotic-resistant strains of Helicobacter pylori and prevents

benzo[a]pyrene-induced stomach tumors. PNAS, 99, 7610-

7615 (2002).

3) J. W. Fahey, K. K. Stephenson, K. L. Wade, P. Talalay; Urease

from Helicobacter pylori is inactivated by sulforaphane and

other isothiocyanates. Biochem. Biophys. Res. Comm., 435,

1-7 (2013).

4) Y. S. Zhang, P. Talalay, C. G. Cho, G. H. Posner; A major

inducer of anticarcinogenic protective enzymes from broc-

coli - isolation and elucidation of structure. PNAS, 89,

2399-2403 (1992).

5) R. L. Shewfelt, E. K. Heaton, K. M. Batal; Nondestructive

color measurement of fresh broccoli. J. Food Sci., 49, 1612-

1613 (1984).

6) S. Hortensteiner; Chlorophyll degradation during senes-

cence. Ann. Rev. Plant Biol., 57, 55-77 (2006).

7) V. R. Altschul, T. A. Smart; EinfluE von chlorophyll und

von pyrrol auf das serumcholesterin von kaninchen.

Arzneimittel Forschung, 11, 762-763 (1961).

8) K. Yamashita, Y. Nohara, M. Kumagai, M. Namiki; Effect of

chlorophyll on plasma lipids in rats. J. Home Econ. Jpn., 42,

589-594 (1991).

9) W. J. Lipton, C. M. Harris; Controlled atmosphere effects on

the market quality of stored broccoli (Brassica oleracea L.,


Italica group). J. Am. Soc. Hort. Sci., 99, 200-205 (1974).

10) A. Jacobsson, T. Nielsen, I. Sjoholm; Effects of type of pack-

aging material on shelf-life of fresh broccoli by means

of changes in weight, colour and texture. Eur. Food Res.

Technol., 218, 157-163 (2004).

11) K. Ren, K. Tu, L. Q. Pan, Y. Y. Chen; Kinetic modelings

of broccoli color changes during chilled storage. J. Food

Process. Preserv., 30, 180-193 (2006).

12) R. Kasim, M. U. Kasim, S. Erkal; The effect of packaging

after 1-MCP treatment on color changes and chlorophyll

degradation of broccoli (Brassica oleracea var. italica cv.

Monopoly). J. Food Agr. Environ., 5, 48-51 (2007).

13) P. M. Treitz, P. J. Howarth; Hyperspectral remote sensing

for estimating biophysical parameters of forest ecosystems.

Progress Phys. Geography, 23, 359-390 (1999).

14) B. Park, L., Renfu; "Hyperspectral Imaging Technology in

Food and Agriculture" , Springer, 2015.

15) J. W. Qin, T. F. Burks, M.A. Ritenour, W. G. Bonn; Detection

of citrus canker using hyperspectral reflectance imaging

with spectral information divergence. J. Food Eng., 93, 183-

191 (2009).

16) D. P. Ariana, R. F. Lu; Evaluation of internal defect and sur-

face color of whole pickles using hyperspectral imaging. J.

Food Eng., 96, 583-590 (2010).

17) D. P. Ariana, R. F. Lu; Hyperspectral waveband selection for

internal defect detection of pickling cucumbers and whole

pickles. Comput. Electron. Agr., 74, 137-144 (2010).

18) G. A. Blackburn; Quantifying chlorophylls and carotenoids

at leaf and canopy scales: An evaluation of some hyper-

spectral approaches. Remote Sens. Environ., 66, 273-285

(1998).

19) A. B. Nicotra, M. Hofmann, K. Siebke, M. C. Ball; Spatial pat-

terning of pigmentation in evergreen leaves in response to

freezing stress. Plant Cell Environ., 26, 1893-1904 (2003).

20) Y. Ishikawa, Y. Makino, H. Sato, T. Hirata; Evaluation of

color changes in broccoli packed with plastic films by a

computerized image analysis. Nippon Shokuhin Kagaku

Kogaku Kaishi, 43, 1170-1175 (1996).

21) R. E. Schouten, X. Zhanga, J. A. Verschoorb, E. C. Otmab, L.

M. M. Tijskensa, 0. van Kooten; Development of colour of

broccoli heads as affected by controlled atmosphere stor-

age and temperature. Postharvest Biol. Technol., 51, 27-35

(2009).

22) F. Mendozaa, P. Dejmekb, J. M. Aguilera; Calibrated color

measurements of agricultural foods using image analysis.

Postharvest Biol. Technol., 41, 285-295 (2006).

23) M. E. Cartea, M. Francisco, P. Soengas, P. Velasco; Phenolic

Compounds in Brassica Vegetables. Molecules. 16, 251-280

(2011).

24) Editorial Office of Agriculture and Horticulture; "Physiology

and Production of Broccoli (Broccoli no seiriseitai to seisan

jirei)" , Seibundo Shinkosha Publishing Co., Ltd., 2010.

25) Y. Makino, K. Goto, S. Oshita, A. Sato, M. Tsukada; A grad-

ing method for mangoes on the basis of peel color measure-

ment using a computer vision system. Agr. Sci., 7, 327-334

(2016).

26) S. Papadakis, S. Abdul-Malek, R. E. Kamdem, K. L. Jam; A

Versatile and inexpensive technique for measuring. Color of

foods. Food Technol., 54, 48-51 (2000).

27) K. McLaren; The development of the CIE 1976 (L*a*b*),

Uniform colour space and colour-difference formula. J. Soc.

Dyers Colourists, 92, 338-341 (1976).

28) J. Makhlouf, F. Castaigne, J. Arul, C. Willemot, A. Gosselin;

Long-term storage of broccoli under controlled-atmo-

sphere. Hortscience, 24, 637-639 (1989).

29) R. J. Porra, W. A. Thompson, P. E. Kriedemann;

Determination of accurate extinction coefficients and

simultaneous equations for assaying chlorophylls a and b

extracted with four different solvents: verification of the

concentration of chlorophyll standards by atomic absorp-

tion spectroscopy. Biochimica Biophysia Acta, 975, 384-394

(1989).

30) M. Nagata, T. Kobayashi, J. Tallada, H. Toyoda, Y. Goto;

Study on anthocyanin pigment distribution estimation for

fresh fruits and vegetables using hyperspectral imaging. J.

Sci. High Technol. Agr., 18, 42-49 (2006).

31) N. Yamauchi, T. Minamide; Chlorophyll degradation by per-

oxidase in parsley leaves. J. Jpn. Soc. Hort. Sci., 54, 265-

271 (1985).

URLs cited

i) faostat3.fao.org/home/E Gul. 3, 2016)


「日本食品工学会誌J,Vol.-17, No. 4, p. 115, Dec. 2016

く〉く〉く＞和文要約く〉く＞く＞

コンビュータービジョンによるブロッコリー外観品質の

客観的評価

牧野義雄 I，若月碧衣 1，網野元貴 I，大下誠一 1，佐藤朱里 2，塚田正人 2

1東京大学大学院農学生命科学研究科， 2日本電気株式会社データサイエンス研究所

ブロッコリーの外観は，クロロフィル，アントシアニ

ンという 2種類の天然色素の影響を受ける．クロロフィ

ルは緑色の元となる物質で，新鮮さの指標［5］である一

方，アントシアニンが多く含まれる花奮は紫がかった外

観となり，品質を損ねる［24］.ブロッコリーの外観品質

は今のところ，農家あるいは選果選別ラインにおいて主

観的に判定されているが，人によって基準が異なるなど

の問題がある．そこで過去には，ブロッコリーの外観品

質を客観的に評価する試みがなされてきた．色彩計で測

定した色空間値から評価する方法［5,12, 28］は，ブロッ

コリーの鮮度評価に効果的であったが，一点測定である

ことから，一株中の任意の箇所の緑色を著しく喪失する

ブロッコリーの外観［9］評価に利用することは困難と考

えられる．一方，デジタルカメラで撮影した二次元RGB

画像を用いた評価もなされているが［20,21］，これは撮

影機器に依存する値であり［22］，普遍的な結果とは言え

ない．さらに，ブロッコリー花昔部のアントシアニンに

ついて非破壊評価を試みた研究例は見当たらない．

そこで本研究では，デジタルカメラの 1種であるコン

ビュータービジョンを利用して，ブロッコリー花昔部の

クロロフィルとアントシアニンの濃度を同時にかつ客

観的に評価する方法について検討した．

三重県内の農場にて通常の品種（アントシアニン濃

度水準：高，以下， Hという．）とアントシアニンレスの

品種（アントシアニン濃度水準：低，以下， Lという．）

を入手した．それぞれの品種を 0,3,5d貯蔵（以下，そ

れぞれ0,3, 5という．）に割り付け，貯蔵期間を変える

ことによりクロロフィル濃度水準を変動させた．すなわ

ち，クロロフィル濃度 3水準×アントシアニン濃度 2水

準の計6種類の試料を用意した（詳細はTable1の通

り）．画像の撮影には， Makinoet al.[25］がマンゴー果皮

のアントシアニン濃度をコンピュータービジョンで客

観的に評価した研究と同一の装置と方法を用いた．

6種類の試料から採取した関心領域（ROI) の色空間

値をゲおよび－a*/b継を軸とする座標にプロットした(Fig. 1）.なお，グは赤昧が増すほど値が高くなることか

ら［27］，アントシアニンの指標として，ーグlb＊はクロロフィルの減少とともに値が低くなることから［28］，クロ

ロフィルの指標として選択した．アントシアニン濃度が

（受付2016年7月11日，受理2016年9月30日）l干 113-8657東京都文京区弥生1-1-12干 211-8666川崎市中原区下沼部1753t F田 03-5841-5361,E-mail: [email protected]』 to匂o.ac.jp

低くクロロフィル濃度が高い個体（OL）が高品質のブ

ロッコリーと言える.Fig. 1のプロット結果から，高品質

の花膏を示す色空間値の範囲を α＊く－12かつ－a*/b*>0.94と決定した．

Fig. 2にはアントシアニンとクロロフィル aを軸とす

る座標に ROIのデータをプロットした．高等植物はクロ

ロフィル bも含有するが，クロロフィルαの方が濃度が

高く，経時的な変化量も大きい.3Hのクロロフィル α濃

度は 3Lに比べて有意に低かった．フラボノイドの 1種

であるアピゲニンがクロロフィルの分解を促進すると

報告されていることから［31］，同じ貯蔵日数でもアント

シアニン濃度が高い個体の方がクロロフィル濃度が低

かったのは，アントシアニンがフラボノイドの 1種であ

ることが関係しているのかもしれない．

Fig. 3には，高品質花蓄の分布をコンビュータービ

ジョンで可視化した例を示した.OLはOHに比べて， 1

株に占めるアントシアニン濃度が低い花曹の割合

（α＊く－12;a-less）が31ポイント高かったさらに， 1株の

OLに占めるクロロフィルα濃度が高い花膏の割合（ーゲ／

b*>0.94; gr）は48%であったが，貯蔵日数が進むにつれ

て低下した．同じ個体で高品質花菅の割合（α叱ー12；かっ

-a* /b*>0.94; hi)を算出したところ，同じ個体のgrより

も低くなったこれは，ゲくー12 (a-less）に相当する花膏

が除外されたためである．また， Lで算出した高品質花

奮の割合（hi）は，全ての貯蔵日数において Hよりも高

かった．以上の結果は，全て事前に想定された当然の結

果であったことから，コンビュータービジョンがブロッ

コリー花蓄の外観品質の客観的な判定に有効であるこ

とを実証するものである

Fig. 4には，本研究で使用した全ての個体から a-less,

gr, hiの割合を算出した結果を示したいずれも Fig.3

の傾向と一致したことから，やはり，コンピュータービ

ジョンがブロッコリー花膏の外観品質の客観的な判定

に有効で、あることが実証された．ただし，最も高品質に

位置づけられる OLのhiが約40%と比較的低い値であっ

たこれは，－a*/b*>0.94という範囲が過去の研究例[5,12,28］に比べて厳しかったことが一因と考えられる．

本研究で提案した色空間値の範囲あるいは闘値は，農家

あるいは集出荷場の都合により変更可能であり，闘値を

変更することで，適切な選別を行えると考える．いずれ

にしても，本研究で提案した方法は，客観的かつ自動的

なブロッコリーの選別に有効であることが明らかに

なった．

。2016Japan Society for Food Engineering

コンピュータービジョンによるブロッコリー外観品質の客観 …kawasaki-shi,...

Documents