milling quality of rice yadav&jindal

7/31/2019 Milling Quality of Rice Yadav&Jindal

1/15

Computers and Electronics in Agriculture

33 (2001) 19 33 www.elsevier.com/locate/compag

Monitoring milling quality of rice by imageanalysis

B.K. Yadav, V.K. Jindal *

Processing Technology Program, School of En6ironment, Resources and De6elopment,

Asian Institute of Technology, P.O. Box 4, Klong Luang, Pathumthani 12120, Thailand

Received 6 November 2000; received in revised form 18 May 2001; accepted 20 July 2001

Abstract

Rough rice is milled to produce polished edible grain by first subjecting to dehusking or

removal of hulls and then to the removal of brownish outer bran layer known as whitening.

The control of whiteness (degree of milling) and percentage of broken kernels in milled rice

is required to minimize the economic loss to the millers. Digital image analysis was used to

determine the head rice yield (HRY), representing the proportion by weight of milled kernels

with three quarters or more of their original length, and the whiteness of milled rice. Ten

varieties of Thai rice were subjected to varying degrees of milling by adjusting the test

duration from 0.5 to 2.5 min. Three-dimensional features (namely, length, perimeter and

projected area) were extracted from the images of individual kernels in a milled sample and

used to compute a characteristic dimension ratio (CDR) defined as the ratio of the sum of

a particular dimensional feature of all head rice kernels to that of all kernels comprising head

and broken rice in the sample. HRY and CDR were found to be related by power functions

based on the above-mentioned dimensional features, with R2 more than 0.99 in all cases. The

CDR based on the projected area of kernels in their natural rest position provided the best

estimate of the HRY with the lowest root mean square error of 1.1% among all dimensional

features studied. In case of the whiteness of milled samples, the values provided by a

commercial whiteness meter and the mean of gray level distribution determined by image

analysis correlated with an R2 value of 0.99. The results of this study showed that

two-dimensional imaging of milled rice kernels could be used for making quantitative

assessment of HRY and degree of milling for on-line monitoring and better control of the

rice milling operation. 2001 Elsevier Science B.V. All rights reserved.

* Corresponding author. Tel.: +66-2-524-5457; fax: +66-2-524-6200.

E-mail address: [email protected] (V.K. Jindal).

0168-1699/01/$ - see front matter 2001 Elsevier Science B.V. All rights reserved.

P I I : S 0 1 6 8 - 1 6 9 9 ( 0 1 ) 0 0 1 6 9 - 7
mailto:[email protected]:[email protected]


2/15

B.K. Yada6, V.K. Jindal/Computers and Electronics in Agriculture 33 (2001) 193320

Keywords: Head rice yield; Degree of rice milling; Milled rice whiteness; Rice quality determination;

Image analysis; Machine vision

1. Introduction

Rice constitutes the worlds principal source of food, being the basic grain for the

planets largest population. For tropical Asians it is the staple food and is the major

source of dietary energy and protein. In Southeast Asia alone, rice is the staple foodfor 80% of the population (Armienta, 1991).

Milling of rough rice (or paddy) is usually done at about 14% dry basis moisture

content to produce white, polished edible grain, due to consumer preference. From

the economic point of view, the quality of milled rice is of paramount importance

since the grain size and shape, whiteness and cleanliness are strongly correlated with

the transaction price of rice (Conway et al., 1991). All these factors are closely

related to the process of milling, in which rough rice is first subjected to dehusking

or removal of hulls and then to the removal of brownish outer bran layer, known

as whitening. Finally, polishing is carried out to remove the bran particles and

provides surface gloss to the edible white portion. A high percentage of broken

grains in the milled product or low head rice recovery represents a direct economic

loss to the millers. Head rice yield (HRY) represents the weight percent of milled

kernels with three quarters or more of their original length of brown rice relative to

rough rice weight. The degree of milling determines the extent of removal of bran

layer from the surface of milled kernels and thus relates to their whiteness. HRY

reduces with the increased duration of milling. Hence in the milling process, the

pressure in the milling chamber and the duration of milling must be adjusted to get

the maximum output.

The extent of losses during milling depends on many factors, such as variety and

condition of rough rice, degree of milling required, the kind of rice mill used, and

the operators. Modern large capacity commercial rice mills use different machines

for dehusking, whitening and polishing operations. Besides rubber roll dehuskers,

two types of milling machines, namely, abrasive and frictional types, are used forwhitening and polishing of grain, respectively. Among horizontal and vertical

abrasive type milling machines, the vertical type is more popular. The brown rice is

rubbed between the surface of an abrasive cone and sieve fitted with a set of rubber

brakes. In the frictional type machine, brown rice kernels are rubbed against each

other under pressure to get the desired whiteness. The process of bran removal in

commercial milling is through intense pressure and friction in a single or multiple

pass operation over a very short period of time. Proper setting and adjustment of

the clearance between rubber brakes and abrasive cone by the operator are critical

factors in the milling operation. In both types of machines, the degree of milling is

also controlled by adjusting the pressure in the milling chamber by means of a

spring-loaded counterweight at the discharge outlet.In practice most control systems for rice milling equipment are essentially based

on manual operation. Informal contacts with several large-scale commercial mills in


3/15

B.K. Yada6, V.K. Jindal/Computers and Electronics in Agriculture 33 (2001) 1933 21

Pathumthani province in central Thailand revealed that milled rice quality is

regularly monitored manually at approximate time intervals of 12 h d u e t o

unavailability of continuous on-line measurement methods. Actual determinations

of HRY and milled rice whiteness are made by laboratory measurements. The

necessary adjustments made by a trained operator, based on visual inspection and

the results of laboratory measurements, take effect in a few minutes to produce

milled rice with a minimum amount of broken kernels and maximum degree of

kernel whiteness. Usually milled rice samples obtained by test milling in the

laboratory are supplied to the operator and used as reference for each grade of thedegree of milling.

Despite the extensive use of image analysis in manufacturing and medical

industries, its applications are almost non-existent in grain-based industries. The

determination of milled rice quality parameters by image processing techniques will

enable regular monitoring of milling operation in an objective manner, and thus

allow the operator to quickly react within a few minutes to changes in material

properties. Fant et al. (1994) determined the gray scale intensity in the digital

images of rice samples subjected to various degrees of milling and correlated the

mean gray level with lipids concentration on the surface of rice kernels. Liu et al.

(1998) used digital image analysis to estimate the area of the bran layer on the

surface of rice kernels and correlated with the surface lipids concentration deter-mined by chemical analysis. They reported that the degree of milling could be

measured quickly and accurately in terms of the surface lipids concentration in a

milled rice sample. Sometimes the degree of rice milling is characterized relative to

arbitrary whiteness standards depending upon the type of commercial whiteness

meter employed. However, no information is available for the estimation of HRY

based on the monitoring of dimensional parameters of kernels for rapid quality

assessment. Therefore, the main objective of the present study was to

develop techniques that could be used for estimating HRY and degree of milling

based on two-dimensional imaging of milled rice kernels being sampled at regular

intervals.

2. Approach to the problem

Although HRY has been defined as the ratio of the weight of milled head rice

kernels to the total weight of the rough rice or paddy kernels for practical purposes,

it could also be expressed relative to the total weight of milled rice rather than

rough rice. In general, the HRY relative to milled rice weight is about 50% higher

than the HRY values based on the weight of rough rice. In general too, it is

possible to estimate the weight of the objects having regular shapes based on their

dimensional characteristics. Therefore, simple models could be developed for

estimating the HRY and whiteness of the milled rice samples using image analysisfrom the measurements of dimensional features and gray level distribution, respec-

tively, as described in the following sections.


4/15


2.1. Head rice yield (HRY) and characteristic dimension ratio (CDR)

Based on the results of an earlier study for a single variety of rice (Yadav and

Jindal, 1998), it was hypothesized that the weight of individual kernels, whether of

head rice and broken fraction, is proportional to their respective dimensional

features. Accordingly, the weight of a milled rice sample might be expressed as a

power function of a composite characteristic dimensional feature derived from the

whole and broken kernels in that sample. The characteristic dimensional features of

the kernels could be their length (L), perimeter (P) and projected area (A) based onindividual measurements. HRY could then be related to the characteristic dimen-

sion ratio (CDR) by

HRYh(CDR)b

or

HRY=a

%n

i=1

Hi

%n

i=1

Hi+ %m

j=1

Bj

b

(1)

where HRY is the head rice yield defined as the ratio of the weight of head rice to

the combined weight of head and broken rice in a milled rice sample, %; Hi is thedimensional feature of the ith head rice kernel; Bj is the dimensional feature of the

jth broken rice kernel; n is the number of head rice kernels in the milled rice

sample; m is the number of broken rice kernels in the milled rice sample; and a, b

are equation parameters.

The relationship between HRY and CDR hypothesized in the form of Eq. (1)

could be validated by the experimental data on the characteristic dimensional

features of the kernels and respective weight fractions of the head rice and broken

kernels for different rice varieties. HRY of a milled rice sample could also be

expressed on the basis of the initial weight of rough rice, if so desired.

2.2. Whiteness of milled rice

Most commercial whiteness meters operate on the principle of light reflectance

from the surface of the milled rice to measure its whiteness. Although earlier use of

digital image analysis to correlate mean gray level with lipids concentration on the

surface of rice kernels has been cited (Fant et al., 1994), it was hypothesized that

the overall whiteness of milled rice could be estimated simply from the mean value

of the gray level distribution obtained from the digitized image of the bulk sample.

3. Materials and methods

3.1. Rice samples

Rough rice samples of ten varieties, namely, Suphan Buri scented rice (HSPR),


5/15


Suphan Buri-1 (SPR1), Klong Luang scented rice (HKLG), Suphan Buri-90

(SPR90), Royal Rice Department-7 (RD7), Royal Rice Department-23 (RD23),

Suphan Buri-60 (SPR60), Chainat-1 (CNT1), Leuang Pra Tew-123 (LPT123) and

Mali scented rice-105 (KDML105) were obtained from the Rice Experiment Center,

Klong Luang, Pathumthani, with moisture contents ranging from 10 to 13% dry

basis. Five samples weighing individually about 200 g, were taken from each rough

rice variety and kept separately in polyethylene bags. All rough rice samples were

dehusked twice with a testing husker (model THU-35A, Satake Engineering Co.

Ltd., Japan), and the brown rice so obtained was subsequently milled using a testmill (model TM 05, Satake Engineering Co. Ltd., Japan). All brown rice samples of

each variety were then milled for an arbitrarily selected test duration ranging from

a minimum of 0.5 min to a maximum of 2.5 min at intervals of 0.5 min. The

variations in degree of milling of rice samples produced 5 levels of HRY and kernel

whiteness for each variety. Thus a total of 50 samples was used for determining the

milled rice characteristics.

3.2. Imaging of rice kernels

A diagram of the imaging system is shown in Fig. 1. It consisted of a lighting

unit, a color CCD camera and a frame grabber connected to a host Pentium 120MHz computer. The lighting unit comprised a closed cylindrical image chamber

fitted with a circular 32 W fluorescent lamp working at 60 kHz to provide

flicker-free light. The camera (model 2200, Cohu Inc., USA) was equipped with a

manual zoom lens (model S6X11-II, F1.4 and 11.5 69 mm zoom, Cohu Inc., USA)

capable of producing PAL softvideo output with a resolution of 752 (H)582 (V).

An Imascan Chroma-P PCI frame grabber board (IMAGRAPH Corporation,

USA) with 2 MB on board memory was used to receive the video signal from the

camera.

The operation of the system was carefully controlled for extracting reproducible

features from the captured images of milled rice samples through various adjust-

Fig. 1. Schematic diagram of the IA system.


6/15


Table 1

Calibration results for measurements by the imaging setup

Actual measurements Imaging measurements Difference

Diameter Projected area DiameterDiameterProjected area Projected area

(%)(mm) (mm2) (%)(mm2)(mm)

28.1 0.9128.6 1.926.04 5.98

48.5 0.947.93 1.8849.4 7.86

201.1 0.0616.01 0.0016.00 201.1

17.78254.5 253.8 0.11 0.2718.00

457.4 0.5823.87 1.20452.023.99

25.78528.9 533.5 0.42 0.8725.95

ments. All the captured frames were 8-bit (0 255) gray scale images. The images of

rice kernels were obtained in batch mode for HRY determination. Rice kernels

were placed manually without touching each other in a petri dish of about 40

mm35 mm area directly under the CCD camera. This area could be covered in

a 512452 pixels frame with approximately 12 times magnification. The dimen-

sional features of individual kernel images were extracted using the ImageToolprogram developed at the University of Texas Health Science Center at San

Antonio, Texas and available from the Internet (ftp://maxrad6.uthscsa.edu). The

imaging system was calibrated with the help of brass disks having thickness close to

that of milled rice kernels (1.22 2.88 mm) and by determining their diameter and

projected area independently by a micrometer with a least count of 0.01 mm. Table

1 shows the estimated average diameter and projected area of the selected disks.

These estimates show a maximum absolute difference of 0.94 and 1.92%, respec-

tively, relative to the measured values. The linear measurements on kernels dimen-

sions were subsequently converted into actual values based on the calibration

results. The whiteness of milled rice samples was measured with a cylindrical sample

holder, 57 mm in diameter and 13 mm deep, and taking the image of the open

surface. All measurements on the whiteness of milled rice samples were replicated

three times by filling the sample holder each time to attenuate the random error by

averaging. A white plate supplied with the commercial whiteness meter (model

C-300, Kett Electric Laboratory, Japan) was used for the adjustment of gray level

distribution recorded by the camera. The aperture opening of the CCD camera was

set at a position to obtain the mean gray level (MGL) of 240 for a 252252 pixels

image of the white plate.

3.3. Determination of HRY

A laboratory rice grader (model TRG 05A, Satake Engineering Co. Ltd., Japan)

was used to separate the head and broken kernels in milled rice samples throughappropriate adjustment of its settings. The HRY was based on the kernel length

equal to or more than 75% of the average length of whole brown rice kernels, as


7/15


already defined. It was found that the grader required a minimum of 12 g

milled rice for obtaining reproducible values of HRY differing by less than 1%.

Therefore, representative samples of milled rice weighing about 12 g were

obtained with a specially fabricated sample divider and separated into head and

broken fractions by the laboratory grader for determining HRY. The use of small

test samples was necessary to limit the total number of rice kernels for

subsequent image analysis. In view of the imprecise separation of kernels by the

laboratory grader, all head and broken kernels in their respective fractions were

separated manually with the assistance of imaging system and weighed tocompute the actual HRY of the representative sample. Later all kernels in the

head and broken rice fractions were imaged to extract their dimensional

features, namely, length (L), perimeter (P) and projected area (A). The values of

CDR were then computed for different milled rice samples and related with their

actual HRY to check the applicability of Eq. (1). Finally, the HRY obtained from

the laboratory grader was compared with the actual HRY estimated by image

analysis.

3.4. Whiteness of milled rice

The whiteness of milled rice samples was first measured with a commercialwhiteness meter (model C-300, Kett Electric Laboratory, Japan) to serve as the

reference values. The rice samples were then imaged in 252252 pixels size and

analyzed by ImageTool to determine their MGL. The interrelationship between

MGL of 50 milled rice samples along with the white plate and the corresponding

whiteness values determined by the commercial meter was investigated

subsequently.

4. Results and discussion

4.1. Estimation of head rice yield by image analysis

Dimensional features of the rice kernels were extracted from their respective

images by separating them from the background and identifying each image with a

unique number with the help of the ImageTool. Fig. 2 shows typical images of the

numbered rice kernels used for the measurement of characteristic dimensional

features and the computation of CDR for any selected rice variety. Figs. 3 5

present the relationships between HRY and CDR based on kernel length, perime-

ter, and projected area, respectively. Regression analysis showed the existence of a

power-law relationship between CDR and HRY. Table 2 presents the values of

model parameters estimated by nonlinear regression function of SPSS version 9.05

for Windows with 95% confidence interval in each case. These results validated thehypothesis that the HRY of milled rice sample could be estimated with root mean

square error (RMSE) of less than 2% from the dimensional features extracted from


8/15


Fig. 2. Images of rice kernels.

the images of the rice kernels in their natural rest position. The plots of actual and

estimated values of HRY based on length, perimeter and projected area are shown

in Figs. 6 8, respectively. The projected area of kernel images provided the best

estimation of HRY in terms of the RMSE determined from the actual and

estimated values for all ten Thai rice varieties. Also the influence of kernel shape

and size due to differences in rice varieties and the degree of milling could not be

discerned in the development of the composite relationships between HRY and

CDR.

The separation of milled rice samples by the laboratory grader often resulted in

overlapping head and broken rice fractions. Therefore, the HRY obtained from the

laboratory grader was compared with the HRY estimation based on the measure-

ment of projected area of kernels as shown in Fig. 9. These results confirmed the

discrepancies encountered in the determination of HRY by the laboratory grader in

comparison with manual inspection of kernels by image analysis. These differences

in HRY were possibly due to the imprecise separation of the whole and broken

kernels by the laboratory grader than the more accurate vision-based measure-ments. The HRY determination by the laboratory grader was influenced by its

operating settings such as the size of indentation in the rotating cylinder and the


9/15


Fig. 3. Head rice yield as a function of characteristic dimension ratio based on kernel length.

inclination angle of the receiving trough. In this study, the operating conditions of

the grader were identical for all rice varieties. The manual inspection of head andbroken fractions separated by the grader confirmed the presence of rice kernels

belonging to the other group. Also, the proportion of the overlap between the head

and broken fractions depended upon the changes in the dimensional characteristics

of the rice kernels during the milling operation and the differences in rice varieties.

Therefore, the relationships between HRY determined by the grader and image

analysis in Fig. 9 tended to exhibit characteristic differences among rice varieties.

Fig. 4. Head rice yield as a function of characteristic dimension ratio based on kernel perimeter.


10/15


Fig. 5. Head rice yield as a function of characteristic dimension ratio based on kernel projected area.

These results further implied that the HRYs estimated by image analysis and the

laboratory grader were indeed related, and could be used for monitoring the millingoperation of different rice varieties.

4.2. Estimation of milled rice whiteness

A plot of gray level distribution in the image (252252 pixels) of a milled rice

sample is presented in Fig. 10. The values of gray level varied over a range from 70

to 190, depending upon the degree of milling and rice variety. The MGL increased

with the degree of milling of rice samples. However, its range for different rice

varieties when milled for same duration was different. The relationship between the

whiteness of the milled rice samples determined by the commercial meter and MGL

Table 2

Results of regression analysis for estimating HRY from the dimensional characteristics of kernel

images (Eq. (1))

Characteristic dimension RMSEb (% HRY)R2Standard errorRegression

parametersa

a |b|ab

0.01 0.994 1.20.85Length (mm) 0.091.97

0.120.792.71Perimeter (mm) 1.30.9920.01

0.9941.41Projected area (mm

2

) 0.010.06 1.10.93

a Corresponding to confidence interval of 95%.b RMSE (% HRY)=(HRYactHRYest)2/N.


11/15


Fig. 6. Comparison of actual and estimated HRY based on kernel length.

is shown in Fig. 11. A linear relationship existed between the whiteness meter

reading and MGL for 51 samples of ten Thai rice varieties as follows:

WR=0.606(MGL)60.5 (R2=0.991) (2)

where WR is the whiteness reading from the commercial meter in arbitrary units,

and MGL is the mean gray level for gray value ranging from 0 to 255.

A comparison of the two methods used for estimating the whiteness of milled rice

samples is presented in Fig. 12. These results showed that the estimated values of

milled rice whiteness based on Eq. (2) in terms of MGL were close to the

Fig. 7. Comparison of actual and estimated HRY based on kernel perimeter.


12/15


Fig. 8. Comparison of actual and estimated HRY based on kernel projected area.

measurements by whiteness meter as indicated by the slope of regression line equal

to 0.998 and R2 value of 0.973 for 50 samples.

4.3. Discussion

The proposed technique for estimating milled rice whiteness can be implemented

in a straightforward manner. Recently an electronic milling degree control system

has been introduced in the Satake vertical abrasive whitener (model VTA10AB) to

monitor the whiteness of milled rice continuously and automatically adjust the

position of a counterweight to control the discharge and in turn the pressure inside

the milling chamber (http://www.Satake-usa.com/abrasive.htm). However, the esti-

mation of HRY in a milled rice sample from kernel images will require the

Fig. 9. Relationship between HRY determined by the laboratory grader and image analysis.
http://www.satake-usa.com/abrasive.htmhttp://www.satake-usa.com/abrasive.htmhttp://www.satake-usa.com/abrasive.htmhttp://www.satake-usa.com/abrasive.htmhttp://www.satake-usa.com/abrasive.htmhttp://www.satake-usa.com/abrasive.htm


13/15


Fig. 10. The gray level distribution in an image of a milled rice sample.

positioning of the individual kernels without touching each other prior to extracting

their dimensional features. At present, no information is commercially available for

monitoring the amount of head rice kernels in milled rice. However, a SPY grain

grader, recently introduced by Maztech MicroVision Ltd., Canada makes use of the

physical features of grains and surface discoloration (http://www.maztech.com/

spy.htm). Individual grains approximately ranging from 500 to 900 are picked by

SPY picker and placed on a vacuum-assisted sample holder for extracting theirphysical features such as the distribution of kernel sizes by imaging. It seems likely

that similar procedures could be adopted for the inspection of milled rice quality,

based on the techniques described in this paper. The developments of such low-cost

machine vision-based techniques that either enhance or replace currently used

manual methods may pave the way for rapid assessment, and thus better control of

rice milling operations in a conventional setting.

Fig. 11. Relationship between whiteness meter reading and mean gray level of milled rice samples.


14/15


Fig. 12. Comparison of actual and estimated whiteness meter readings for milled rice samples.

5. Conclusions

Results of this study show that HRY and whiteness of milled rice could beestimated from two-dimensional images of the milled rice kernels in their natural

rest position. The HRY showed a distinct power-law relationship with the charac-

teristic dimension ratio defined in terms of length, perimeter and projected area of

head and broken rice kernels. However, the estimation of HRY from CDR based

on kernel projected area yielded slightly better results with the lowest RMSE of

1.1% as compared to CDR based on kernel length or perimeter. The degree of

milling was directly related to the MGL estimated from the gray level distribution

in milled rice kernels images expressed in arbitrary whiteness units as measured by

a commercial whiteness meter. The developed empirical relationships for estimating

HRY and milled rice whiteness could be used for regular monitoring and better

control of rice milling operations.

Acknowledgements

The financial support and laboratory facilities provided by the School of Envi-

ronment, Resources and Development, Asian Institute of Technology, are grate-

fully acknowledged.

References

Armienta, E.S., 1991. Research contribution and perspectives on rice in Mexico. In: Cuevas-Perez, F.

(Ed.), Proceedings of the VIII International Rice Conference for Latin America and the Caribbean,

Villahermosa, Tabasco, Mexico, 10 16 November, pp. 5 8.


15/15


Conway, J.A., Sidik, M., Halid, H., 1991. Quality/value relationships in milled rice stored in conven-

tional warehouses in Indonesia. In: Naewbanij, O.J., Manilay, A.A. (Eds.), Proceedings of the

Fourteenth ASEAN Seminar on Grain Postharvest Technology, Manila, Philippines, 5 8 November,

pp. 55 82.

Fant, E., Casady, W., Goh, D., Siebenmorgen, T., 1994. Grey-scale intensity as a potential for degree

of rice milling. J. Agric. Eng. Res. 58 (2), 89 97.

Liu, W., Tao, Y., Siebenmorgen, T.J., Chen, H., 1998. Digital image analysis method for rapid

measurement of degree of milling of rice. Cereal Chem. 75 (3), 380 385.

Yadav, B.K., Jindal, V.K., 1998. Monitoring milled rice characteristics by image analysis. In: Salokhe,

V.M., Jianxia, Z. (Eds.), Proceedings of the International Agricultural Engineering Conference,

Bangkok, Thailand, 7 10 December, pp. 963 971.

milling quality of rice yadav&jindal

Documents