International Journal of Electrical and Computer Engineering (IJECE) Vol.
No.
June 2013, pp.
ISSN: 2088-8708
Nondestructive Determination of Beans Water Absorption Capacity using CFA Images Analysis for Hard-to-Cook Evaluation Ousman Boukara,b.
Laurent Bitjokaa,b.
Gamraykryo Djaowya,b Electrical Engineering and Industrial Automation Laboratory.
Modelisation Image Processing and Applications Research Group.
The University of Ngaoundere.
Cameroon Biophysics and Food Biochemistry Laboratory.
National School of Agro-Industrial Sciences.
The University of Ngaoundere.
PO Box 455 Ngaoundere.
Cameroon
Article Info
ABSTRACT
Article history:
Hard to cook (HTC) phenomenon is developed by storing bean grains under the adverse conditions of high temperature (Ou 25 AC ) and high humidity (Ou 65 %).
Bean grains that have undergone this HTC phenomenon are characterized by loss of color lightness, development of browning and darkening, and decrease of Water Absorption Capacity (WAC).
The objective of this study was to develop a CFA (Color Filter Arra.
image processing system to measure Water Absorption Capacity (WAC) of bean grains with high precision in short time intervals .
The relationships between the CFA image features, extracted from raw images captured by CCD .
harge coupled devic.
camera, and the measured WAC were The calibration models using multiple linear regression (MLR) were developed to predict WAC.
The MLR models for prediction samples resulted in correlation coefficient (R.
in the range of 0.
811 to 0.
standard error of prediction (SEP) in the range of 7.
587 to 11.
669, and Fisher variable value (F) in the range of 52.
300 to 221.
Results indicate that computer vision system (CVS) based on CFA image analysis technique can provide an accurate, reliable and nondestructive measurement method of WAC to evaluate the hard to cook defect in bean grains.
Received Jan 10, 2013 Revised Apr 13, 2013 Accepted May 18, 2013 Keyword:
CFA images Bean grains Hard to cook Regression models Nondestructive method Copyright A 2013 Institute of Advanced Engineering and Science.
All rights reserved.
Corresponding Author:
Ousman Boukar.
The University of Ngaoundere.
Box 455 Ngaoundere.
Cameroon Tel: 237 96 19 90 80.
Fax: 237 22 15 81 89 E-mail address: boukarousman@gmail.
INTRODUCTION
Beans (Phaseolus vulgari.
are the third most important legume in the world on the basis of total grain production after soybeans and peanuts.
they are a staple food in some tropical and subtropical countries .
Bean grains are a good source of carbohydrates and proteins.
They provide important quantities of protein, starch, dietary fiber, protective phytochemicals, oil, vitamins and mineral elements .
Bean grains quality is determined by soaking characteristics, cooking and nutritive value .
Acceptability characteristics have not received enough attention in breeding programs.
These traits include grain size, shape, color, appearance, stability under storage conditions, cooking properties, quality of the product obtained and flavor .
Despite the advantages that beans are a good food source, bean grains have some undesirable characteristics that limit their acceptability or nutritional value, such as: hard to cook (HTC) phenomenon, antinutriments or antinutritional factors or limitation in some amino acids of high biological value .
, .
, .
Hard to cook phenomenon is developed by storing bean grains under the adverse conditions of high temperature (Ou 25 AC ) and high humidity (Ou 65 %).
This phenomenon is characterized by extended cooking Journal homepage: http://iaesjournal.
com/online/index.
php/IJECE A ISSN:2088-8708 times for cotyledon softening .
Bean grains that have undergone this HTC phenomenon require increased energy .
cost for preparation.
are less acceptable to the consumer due to changes in flavor, color, and texture.
and have decreased nutritive quality .
Several hypotheses have been proposed to explain the cause of bean grains hardening: lipid oxidation and/or polymerization, formation of insoluble pectates, lignification of middle lamella, and multiple mechanisms.
Several methods have been proposed to study the hard to cook of bean, namely the determination of the cooking time .
, the determination of the water absorption capacity .
and recently the determination of the cooking timeconstant .
These methods are laborious, time consuming and invasive since they destroy the analyzed Therefore, color and Water Absorption Capacity (WAC) might be related to each other.
This has been proved by previous studies .
, in which color varied as WAC changed in bean grains during the storage at high temperature and high relative humidity.
However, no clear relationship was established in these studies and any prediction models were not found to measure WAC as a function of color attributes.
Therefore, the objective of this study was to develop an image processing system to measure WAC of beans with high precision in short time intervals .
Images features of beans were obtained by computer vision while WAC was obtained by physico-chemical analysis.
The sensor is usually the most expensive component of the digital camera.
To reduce cost and complexity, digital camera manufacturers often use a single CCD .
harge coupled devic.
or CMOS .
omplementary metal-oxide semiconducto.
sensor covered by a color filter array (CFA).
The acquired image is a gray-scale image and thus, digital image processing solutions should be used to generate a camera output comparable to the one obtained using a three-sensor device.
The acquired image is called CFA image.
In order to analyze images in higher precision, rapidly with a low cost method.
CFA images were used in this work.
This paper is divided in the following principal sections: section 2 describes the materials And methods, section 3 shows experimental results and discussion, and section 4 presents the the authors Conclusions.
MATERIALS AND METHODS
Beans samples and hardening procedure Bean (Phaseolus vulgari.
grain samples of five freshly harvested varieties, namely.
DOR-701.
MERINGUE.
SENEGALAIS.
ECAPAN-021 and MAC-55 were obtained from the Regional Center for Research and Innovation West.
Bafoussam.
Cameroon.
The different varieties of bean grains samples used in this study are given in Table1.
Hardened beans were produced using the accelerated storage procedure .
Everyday, 40 bean grains were sampled and previously analyzed through algorithms developed using the computer vision system (CVS) before using them to measure the WAC.
Table 1.
Bean samples used in this study.
Red bean
DOR-701
Red bean
MERINGUE
Speckled bean
ECAPAN-021
Speckled bean
MAC-55
Red bean IJECE Vol.
No.
June 2013: 317Ae328
SENEGALAIS
Captured images IJECE
ISSN: 2088-8708
Method of water absorption capacity (WAC) determination WAC was determined using the Paredes-Lypez et al.
procedure with minor variations.
Before analysis, samples were adjusted to the same moisture content .
%).
40 beans were selected.
Whole seeds were soaked in two volumes of distilled water at 25 AC.
After 24h of soaking samples were removed, drained, blotted and weighed.
The increment in weight was reported as WAC per 100 g of dry weight and was calculated using the following formula:
WAC A% A = 100
M 1 Aweightofso akedbeans A A M 0 Aweightofdr ybeans A M 0 Aweightofdr ybeans A .
CFA Images acquisition CFA images were captured using an image acquisition system of a digital color camera similar to that developed by Bitjokaet al.
, and Mery and Pedreschi .
(Figure .
This image acquisition system is a box type enclosure.
Samples were illuminated by using four parallel fluorescent lamps with a color temperature of 6500 K (Philips.
Natural Daylight, 40 W) and a color rendering index (R.
near to 95%.
The four lamps were situated 40 cm above the samples and at angle of 45A of the food sample plane.
This illumination system gave a uniform light intensity over the food plane.
A CCD digital color camera (FUJIFILM Finepix HS.
, was located vertically at a distance of 25cm from the samples through a hole on the top surface of the box.
The angle between the camera lens axis and the lighting sources was around 45A.
Sample illuminators were inside a wood box whose internal walls were painted white, firstly to increase contrast and eliminate areas with grey levels close to white .
and secondly to assure a uniform illumination inside the room .
Images were captured with the mentioned CCD digital color camera with its maximum resolution .
4 X 2742 pixel.
, and stored as a raw image without compression.
The raw
CFA
The white balance of the CCD digital color camera image stored was a CFA image and was denoted I was set using a simple white reference.
Color standards were photographed and analyzed periodically to ensure that the lighting system and the CCD digital color camera were working properly.
The CFA images of beans were acquired everyday over a period of project to show different levels of hardening.
Figure 1.
Schematic representation for bean grains image acquisition system CFA Image pre-processing Images captured by CCD digital color camera are subject to various types of noises.
These noises may degrade the quality of an image and subsequently it cannot provide correct information for subsequent image processing.
In order to improve the quality of an image, some operations need to be performed on it to remove or decrease degradations suffered by the image during its acquisition.
The purpose of pre-processing is an improvement of the image data, which suppresses unwilling distortions or enhances some image features that are important for further processing and creates a more suitable image than the original for a specific application.
Pre-processing method used in this study is similar to the one reported by Du and Sun .
Our local pre-processing method focus on acquiring CFA image of bean grains with a white scene The white scene content in the CFA image, denoted L , had equal responses in all color components (RCFA,= GCFA=BCFA ), with the midtones being rendered near the middle of the tone scale, regardless of the illuminant or content of the scene.
Correction image adjustment was accomplished by multiplying pixels in each color component (RCFA.
GCFA and BCFA ) by any different gain factor .
, b, .
which compensates for a Nondestructive Determination Of Beans Water Absorption Capacity Using CFA Images.
(Ousman Bouka.
A ISSN:2088-8708 non-neutral CCD digital color camera response and illuminant imbalance.
The corrected CFA image was CFA The corrected CFA image or the corrected pixels values were obtained as follows:
E R CFA E Ea 0 0E E R CFA E
E CFA E E
E E CFA E
EG E = E0 b 0E EG E
E CFA E E0 0 c E E B CFA E
EE B EE E
AeA
a = 0L
AiA
n c= 0
AiA
A eARL = I CFA A p A Eu
CFA
Card AR EN L A pEaRCFA EN L
GL=
I CFA A p A Eu
CFA
EN L A pEaG CFA EN L
Card AG
AiA
n BL = I CFA A p A Eu
CFA
Card AB EN L A pEaBCFA EN L
AiA
Where I CFA
= AuR
CFA
G CFA B CFA Ay .
I CFA = AuR CFA G CFA B CFA Ay .
I CFA A p A is the level value of the pixel p.
K 0 isthe white reference value, and R .
pixels of R
CFA
correspond to the mean value of the EN L .
G CFA EN L and B CFA EN L , respectively.
CFA image segmentation .
Figure 2.
Segmentation procedure: .
original CFA image and .
segmented CFA image IJECE Vol.
No.
June 2013: 317Ae328
IJECE
ISSN: 2088-8708
Since the acquired images contained both the bean sample and background, a technique capable of removing the background from the images is a prerequisite for the color analysis procedures that follow.
algorithm for segmenting bean sample from the background was developed using ImageJ code.
The developed segmentation algorithm was based on the Otsu thresholding method .
to automatically calculate the optimum threshold value.
the computing method of optimum threshold value was described in details by Chen and Qin .
Segmentation was performed using the following three steps.
separating bean sample from the background, .
removing of noise from the binary image, and .
filling of holes in the segmented binary image to obtain an actual binary image.
The bean sample .
r beans pixel.
were denoted H and the background H .
One image of the sample and the segmented result are illustrated in Figure 2.
CFA images features extraction Features of the bean samples were extracted from R .
G .
and B .
components of the CFA images.
For each of the three color components, their mean and standard deviation values were calculated and, consequently, 6 CFA images features including three means values ( AA .
AA .
AA ) and three standards deviations ( E .
E .
E ) were obtained .
The means show the average color properties of bean grains and the standard deviations represent a measure of color un-uniformity over a bean They were calculated by the following equations:
AAR =
AAG =
AAB =
ER =
EG =
EB =
Card H EN R CFA
A EuI
Card H EN G CFA
A EuI
Card H EN B CFA
A EuI
Card H EN R CFA
A Eu AuI
Ap AA AA R Ay
Eu AuI
Ap AA AA G Ay
CFA
CFA
AH EN G
CFA
Card H EN B CFA
CFA
Where I .
ApA .
ApA .
pEaH EN G CFA
Card
ApA pEa H EN R CFA
CFA
pEaH EN B CFA
CFA
p Ea H EN R CFA
CFA
p Ea H EN G CFA
A Eu AuI
CFA
pEa H EN B CFA Ap AA AA B Ay A p A is the level value of the corrected pixel p.
Data analysis To develop a prediction model of WAC of the bean grains using their extracted CFA image features, multiple linear regression (MLR) analysis was applied to build the model of prediction .
MLR
analysis between WAC and CFA image features extracted was conducted using Statgraphics Plus for Windows software.
Version 5.
1 at 95% of confidence level.
The aim of MLR analysis is to find a mathematical relationship between a set of independent variables.
X matrix .
storage days observation y 6 CFA image feature.
, and the dependent variable.
Y matrix .
storage days observation y 1 WAC).
The values of the attribute WAC of the calibration set were used to represent the dependent variable Y.
Meanwhile, the 6 CFA image features of the bean samples represented the independents variables or the predictors (X).
The linear equation between the 6 CFA image features ( AA .
AA .
AA .
E .
E .
E ) and the Water Absorption Capacity (WAC), was defined as follows:
WAI C = b0 b1 AA R b2 AA G b3 AA B b4 E R b5 E G b6 E B Au Nondestructive Determination Of Beans Water Absorption Capacity Using CFA Images.
(Ousman Bouka.
A ISSN:2088-8708 I C , predicted value of the attribute.
b0, b1, b2, b3, b4, b5, b6 , regression coefficients of the equation Where WA to be estimated.
and Au , the standard error between the predicted and measured values.
Once the linear regression model was determined, the equations were used to predict the attributes of samples in the calibration and validation sets.
The quality of the calibration model was evaluated by the standard error of calibration (SEC), standard error of prediction (SEP), the multiple correlation coefficient (R.
between the predicted and measured value of the attribute .
, the Fisher variable value (F).
A good model should have a low SEC, a low SEP, a high R, a small difference between SEC and SEP .
and a high F.
The predictive ability of the models were also quantified by these criteria.
The results of future predictions with a 95% confidence interval can be expressed as the predicted value WACi A 1.
96ySEC .
These criteria are defined as follows:
W AI C i A WAC i Eu n c i=1 SEC = A A .
A 1 p I SEP = Eu WACi A WACi A bias
n p i=1 Eu W AI C i A WAC i n p i= 1
E SEC E
R2 = E
E SEO E
SEO =
Eu AW A C A WAC i A n c i= 1
WA C =
Eu WAC i n p i= 1
n A k A 1 E SEC E
F= c
E SER E
SER 2 = SEO 2 A SEC 2
I C represents the predicted value of the i-th observation.
WAC represents the measured value Where.
WA of the i-th observation.
nc is the number of observations in calibration set.
n p is the number of observations in prediction set.
k is thenumber of CFA image features.
RESULTS AND DISCUSSION
Water Absorption Capacity (WAC) determination Figure 3 shows the effect of storage on WAC of the five bean grains varieties.
In the fresh state.
MERINGUE variety had higher WAC value than the other varieties.
Accelerated storage caused a significant .
< 0.
decrease in the WAC of the whole bean grain samples, being more pronounced in the MERINGUE SENEGALAIS variety had lower decrease of WAC than the other varieties.
The decrease in the WAC of the bean grains was reported by other authors .
, .
, .
these results showed that beans stored under tropical conditions absorb less water during soaking, which itself may contribute to a harder bean texture.
The results obtained in the present study were in agreement with the results presented by Reyes-Moreno et al.
, whichsuggested that changes of a biochemical and/or physico-chemical nature in both the cotyledon and seed coat result in a lower water uptake capacity.
IJECE Vol.
No.
June 2013: 317Ae328
IJECE
ISSN: 2088-8708
The curves presented in Figure 3 show that a WAC constant was found for each bean variety.
These WAC constants could characterize the age and the hardening of each bean grain variety and could be used to predict the maximum storage time of each bean variety under the tropical conditions.
high temperature and high relative humidity.
The WAC constants obtained were in the range of -0.
198 to -0.
This study suggests that higher is the absolute value of WAC constant, rapidly could be the hardening process of the bean grain.
Furthermore, lower is the absolute value of WACconstant.
slowest could be the hardening process of the bean grain.
Water AbsorptionCapacity(%) DOR-701 Exponentielle (DOR-.
MERINGUE Exponentielle (MERINGUE) SENEGALAIS Exponentielle (SENEGALAIS) ECAPAN-021 Exponentielle (ECAPAN-.
MAC-55 Exponentielle (MAC-.
Storage time .
Exponentielle (DOR-.
Exponentielle (MERINGUE) Exponentielle (SENEGALAIS) f.
= 204.
832 exp( -0.
188 x ) RA = 0.
= 217.
369 exp( -0.
134 x ) RA = 0.
= 148.
820 exp( -0.
113 x ) RA = 0.
Exponentielle (ECAPAN-.
= 169.
975 exp( -0.
130 x ) RA = 0.
Exponentielle (MAC-.
= 178.
615 exp( -0.
198 x ) RA = 0.
Figure 3.
Relation between water Absorption capacity and storage time in the five varieties of common bean (Phaseolus vulgari.
CFA image features extraction Six features characterizing color of the bean grains were extracted from the CFA images of bean The effect of storage on CFA image features of the five varieties is outlined in Figures 4a-e.
general, at all times and for the whole bean grain varieties, the AA value was ( p < 0.
higher than both the AA G and AA B values.
except the MERINGUE variety where AA R value was lower than both the AA G and AA B values after ninth storage day.
Furthermore, the E R .
E G .
E B values were ( p < 0.
similar for all times and for the whole bean grain varieties.
Storage produced a significant ( p < 0.
decrease of the AA .
AA G and AA B values of the five bean varieties, however a significant ( p < 0.
decrease of the E R .
E G .
E B values was obtained in the ECAPAN-012 and MAC-55 varieties.
However, no significant ( p < 0.
decrease of the E .
E .
E values was obtained in the DOR-701.
MERINGUE and SENEGALAIS varieties and could not be used also as a CFA image features indicators to establish WAC calibration models.
The AA .
AA .
AA .
E .
E and E curves of the DOR-701.
MERINGUE.
SENEGALAIS, and ECAPAN021 varieties presented an exponential effect, while the same features presented a logarithmic effect for the MAC-55 variety.
Theses curves effects showed that the five studied varieties of beans were browned and darkened during the storage at high temperature and high humidity.
Other researchers .
reported a similar behavior in common beans during storage at high temperature and high The results of this study suggest that a significant browning and darkening were developed at the bean grain surface during the storage at high temperature and high humidity, being more pronounced in red bean than speckled bean.
Nondestructive Determination Of Beans Water Absorption Capacity Using CFA Images.
(Ousman Bouka.
A ISSN:2088-8708 MERINGUE CFA image feature CFA image feature DOR-701 Storage time .
Storage time .
= 21.
806 exp( -0.
022 x ) RA = 0.
= 60.
710 exp( -0.
034 x ) RA = 0.
= 60.
627 exp( -0.
040 x ) RA = 0.
= 21.
570 exp( -0.
017 x ) RA = 0.
= 55.
299 exp( -0.
035 x ) RA = 0.
= 22.
188 exp( -0.
014 x ) RA = 0.
= 76.
679 exp( -0.
060 x ) RA = 0.
= 59.
869 exp( -0.
036 x ) RA = 0.
= 62.
061 exp( -0.
030 x ) RA = 0.
Figure 4a.
Relation between 6 CFA image features and storage time in the DOR-701 variety Figure 4b.
Relation between 6 CFA image features and storage time in the MERINGUE variety SENEGALAIS CFA image feature CFA image feature f.
= 20.
816 exp( -0.
017 x ) RA = 0.
= 21.
595 exp( -0.
017 x ) RA = 0.
= 23.
223 exp( -0.
019 x ) RA = 0.
ECAPAN-021
= 104.
121 exp( -0.
043 x ) RA = 0.
= 76.
319 exp( -0.
040 x ) RA = 0.
= 69.
538 exp( -0.
029 x ) RA = 0.
= 19.
198 exp( -0.
019 x ) RA = 0.
= 19.
227 exp( -0.
019 x ) RA = 0.
= 86.
754 exp( -0.
074 x ) RA = 0.
= 75.
798 exp( -0.
062 x ) RA = 0.
= 65.
394 exp( -0.
056 x ) RA = 0.
= 20.
316 exp( -0.
019 x ) RA = 0.
Figure 4c.
Relation between 6 CFA image features and storage time in the SENEGALAIS variety f.
=32.
726 exp( -0.
038 x ) RA = 0.
= 32.
847 exp( -0.
043 x ) RA = 0.
=30.
305 exp( -0.
039 x ) RA = 0.
Figure 4d.
Relation between 6 CFA image features and storage time in the ECAPAN-021 variety MAC -55 CFA image feature S torage time .
= -21.
515 ln.
RA = 0.
= -15.
109 ln.
RA = 0.
= -9.
685 ln.
RA = 0.
= -7.
192 ln.
RA = 0.
= -4.
819 ln.
RA = 0.
= -2.
128 ln.
RA = 0.
Figure 4e.
Relation between 6 CFA image features and storage time in the MAC-55 variety IJECE Vol.
No.
June 2013: 317Ae328 Storage time .
Storage time .
IJECE
ISSN: 2088-8708
Multiple linear regression calibration models analysis For each bean varieties, only the significant CFA image features were used to build a multiple linear regression (MLR) model between the significant features among AA .
AA .
AA .
E .
E .
E .
s Xvariable.
and the measured values of the WAC .
s Y-variable.
The calibration set was used to develop the MLR models for predicting the WAC.
The MLR results for calibration are shown in Table 2.
The obtained results show that the behavior of the selected features depends on model of each variety.
The feature AA had a positive contribution in all models for WAC.
being more pronounced in MAC-55 variety.
This feature had the similar contribution in the models of DOR-701 and MERINGUE, this mean that the color of DOR701 and MERINGUE is dominated by redness appearance rather than both greenness and blueness color.
Furthermore.
AA and AA had an opposite behavior in same model for WAC.
The E feature had the largest contribution in the models for WAC of ECAPAN-021 and MAC-55 varieties than the both of the E and E contributions, this mean that ECAPAN-021 and MERINGUE had both the more dispersion in redness color.
In order to assess the accuracy of the calibration model and to avoid over fitting, calibration parameters were obtained (Table .
As shown in Table 3, the values of SEC.
R2and F were in the range of 65131 to 15.
8547, 0.
923 to 0.
993 and 52.
68 to 299.
92, respectively.
These results show that the values of R2and F obtained were in the range for a good calibration model, while the SEC values were higher than those recommended by Elmasriet al.
The calibration equations presented here is consistent with the finding of Jhaet al.
using MLR on variables a, b and the product ab .
btained from HunterLab colorimete.
to calibrate for predicting the maturity of mango.
However, to verify the generalization ability of calibration models, the present MLR models should be adopted to further investigate the predictive performance for prediction samples.
Table 2.
Form of WAC models calibration as a function of CFA image features Models DOR-701
WAC
= Oe 286.
918AA Oe 0.
698AA 3.
MERINGUE
WAC
= Oe 33.
758AA 5.
338AA Oe 7.
SENEGALAIS
WAC
= Oe 81.
241AA Oe 1.
692AA 3.
ECAPAN-021
WAC
= Oe 75.
415AA 1.
352AA Oe 3.
Oe 0.
766E R 7.
089EG 1.
MAC-55
WAC
= Oe 175.
31AA Oe 23.
656AA 6.
745E R 11.
795E G 0.
787E B
Table 3.
Statistical results of MLR models calibration.
Varieties models SEC
DOR-701
MERINGUE
SENEGALAIS
ECAPAN-021
MAC-55
Prediction results for WAC In order to assess the accuracy of the calibration models and to avoid over fitting, validation values were obtained (Table .
a calibration model without validation is nonsense.
Based on the above results.
MLR models were used for measuring WAC in prediction samples.
Measured values of WAC from physicochemical test .
and its predicted values resulting from MLR models .
are shown in Figures 5aAee.
Tables 4 and 3 show that the model was accurate for predicting WAC with R2in the range of 0.
923 to 993 and 0.
811 to 0.
947 for calibration and validation sets, respectively.
The SEC and SEP were in the range Nondestructive Determination Of Beans Water Absorption Capacity Using CFA Images.
(Ousman Bouka.
A ISSN:2088-8708 651 to 15.
855 and 7.
587 to 11.
669 for calibration and validation sets, respectively.
The accuracy of the models in the validation set for predicting WAC was with F in the range of 52.
300 to 221.
It is obvious for the attribute under study (WAC) that the validation tests gave close results as the calibration set indicating good performance of the models for predicting WAC nondestructively.
Table 4.
Performance of MLR models for predicting WAC.
Varieties models SEP
DOR-701
MERINGUE
SENEGALAIS 10.
ECAPAN-021
MAC-55
Figure 5a.
The prediction results for WAC of DOR701 variety by using MLR model, resulting in SEP = 669.
R2 = 0.
878 and F = 87.
Figure 5b.
The prediction results for WAC of MERINGUE variety by using MLR model, resulting in SEP = 11.
R2 = 0.
811 and F = 52.
Figure 5c.
The prediction results for WAC of SENEGALAIS variety by using MLR model, resulting in SEP = 10.
R2 = 0.
879 and F = Figure 5d.
The prediction results for WAC of ECAPAN-021 variety by using MLR model, resulting in SEP = 7.
R2 = 0.
947 and F = Figure 5e.
The prediction results for WAC of MAC-55 variety by using MLR model, resulting in SEP = 816.
R2 = 0.
912 and F = 197.
IJECE Vol.
No.
June 2013: 317Ae328
IJECE
ISSN: 2088-8708
CONCLUSION
Storage of these five bean varieties under adverse conditions of high temperature and relative humidity rendered them susceptible to the hardening phenomenon.
Several changes in grain quality characteristics were affected by storage under these adverse conditions: loss of color lightness and development of browning and darkening .
and decrease in WAC.
For practical and low cost application.
WAC of beans was predicted from CFA image features.
By multiple linear regression (MLR), precise calibration equations were obtained.
The calibration models were validated successfully.
The results show that the nondestructive determination of WAC of common beans using CFA image features is reliably It is also known that the established models is low cost and high precision, but less robust due to the higher values of SEC and SEP.
In the future, more work will be done to optimize and implement this technique by using other algorithms, such as neural networks, wavelet transform and genetic algorithm, for many other bean grain varieties that differ by season or growing region.
ACKNOWLEDGEMENTS
The authors gratefully acknowledge Prof NJINTANG Nicolas and Dr BEKA Robert Germain of Biophysics and Food Biochemistry Laboratory for the technical assistance of physico-chemical analysis.
would like to express our gratitude to Ludovic MACAIRE and Olivier LOSSON of LAGIS Laboratory who contributed to analyze the CFA images in this study.
We are also indebted to the Regional Center for Research and Innovation West for providing bean grains varieties.
The work presented in this paper was supported by AuService de Coopyration et dAoActionCulturelle.
Ambassade de France au CamerounAy and the Cameroon Ministry of High Education.
REFERENCES