classification and pixel-based segmentation to evaluate
TRANSCRIPT
Journal of Agricultural Science; Vol. 11, No. 13; 2019 ISSN 1916-9752 E-ISSN 1916-9760
Published by Canadian Center of Science and Education
186
Classification and Pixel-Based Segmentation to Evaluate Soybean Seeds Submitted to Tetrazolium Test
Davi Marcondes Rocha1, Lúcia Helena Pereira Nóbrega2, Maria de Fátima Zorato3, Vitor Alex Alves de Marchi1 & Arlete Teresinha Beuren1
1 Computer Science Department, Federal University of Technology-Paraná, Santa Helena, Paraná, Brazil 2 Tec. and Exact Sciences Center, Western Paraná State University, Cascavel, Paraná, Brazil 3 Biologist, Seed Science & Technology, Londrina, Paraná, Brazil
Correspondence: Davi Marcondes Rocha, Computer Science Department, Federal University of Technology-Paraná, Santa Helena, Paraná, Brazil. E-mail: [email protected]
Received: April 25, 2019 Accepted: June 3, 2019 Online Published: August 15, 2019
doi:10.5539/jas.v11n13p186 URL: https://doi.org/10.5539/jas.v11n13p186
Abstract Production and use of high quality seeds are essential for the soybean crop. Thus, the quality control system in seeds industry must be reliable, precise, and fast. Tetrazolium test evaluates not only seeds viability but also their vigor, as well as provides information concerning agents that cause their quality reduction. Although this test does not use expensive devices and reagents, it requires a well-trained analyst. Its precision depends on knowledge of all techniques and required procedures. Besides, also necessary is the observer’s subjectivity. So, this trial aimed at developing a computational tool that could minimize the implicit subjectivity in carrying out this test. It also contributes to generate a greater credibility of information and to guarantee precise answers. Algorithms of supervised classification were applied based on extraction of digital images characterization of tetrazolium test. This procedure aimed at producing pixel-based segmentation of those images, to produce a digital segmented image of tetrazolium test according to damage classes. This tool allows, based on image of tetrazolium test, to identify damage on soybean embryos, as well as its site and extension on tissues, so that the interpretation is less subjective. The applied method allowed identifying damage on images of tetrazolium tests in a straightforward way, as well as extracting safer information about those damages and carrying out management control of tetrazolium test according to a seed data file.
Keywords: machine learning, patterns recognition, seeds vigor
1. Introduction Soybean crop has much contributed to the Brazilian commercial agriculture. It helped on crop mechanization speed, as well as it modernized transportation, expanded agricultural boundaries, and has taken part in the technological evolution and in another crops production. It also helped the development of Brazilian poultry and swine farming (Dall’Gnol, 2000). Its high protein content (40%) is considered a raw material to feed domestic animals, and although it does not have a high oil content (about 20%), soybean and palm are the greatest oil producers (Roessing, Dall’Agnol, Lazzarotto, Hirakuri, & Oliveira, 2007).
According to Marcos-Filho (2015), seed technology has as main objective the development of efficient mechanisms for the productive chain, using high-quality batches. Quality seeds are those that contain high germination potential and vigor, with high physical and genetic purity, characteristics that can be analyzed by multiple tests, following rigorous norms (Marcos-Filho, 2015; FAO, 2013). The tetrazolium test has stood out among the quality tests adopted by the seed industry, especially for soybeans. This occurs not only because of its accuracy and speed when compared to other tests, but also because of the substantial number of information provided as diagnostic of potential causes for quality reduction (mechanical damage, deterioration due to humidity and bed bug damage) and the possibility of evaluating seed batches’ quality and vigor (França Neto, Krzyzanowski, & Costa, 1998). This test enables to distinguish between vigorous (bright red-carmine colored) and deteriorating (deep red) tissues, and dead or heavily deteriorated tissues (which retain their whitish color) (Deswal & Chand, 1997), and it has been shown that each type of damage is associated with typical patterns of injury (Figure 1).
jas.ccsenet.
(Figure 1
The stainin(high and characterisembryonic
Seeds are humidity iregion (IST
Although tperceptionabnormaliteliminate tresults, als
Techniquecapture, prhave madeattractive aprocessing
Image segassigned toare based performingimportant Zaboleeva
As stated damage, thonly the daccuracy.
In this reseimproving
Corrêa, CíconcludedEvaluated effective in
Pereira, Sshowed thallowing to
Silva, Barrusing comuniformityalternative
org
a) Mechanical1. Damage patt
ng tonality of low vigor) an
stics, enables c tissues (Moor
considered nin regions of aTA, 1985).
the establishedn (tetrazolium tty seedling anthe implied subso in addition t
es to analyze irocessing, and e the image anautomated wayg, besides the e
gmentation is o these region
only on pixg segmentatioalternative to
a-Zatova, 2014
by Cícero anhis promising pdamaged areas
earch field, theg information a
ícero, and Abd the system is
mechanical an detecting the
aito, and Bughat the proposeo automate and
ros, Albuquermputerized imay of seedling e for obtaining
l damage terns found in
f embryos tissund non-viable.the determinare, 1973; Fran
non-viable wheapical and root
d criteria are ftest, fast greennd vigor, resubjectivity in soto reduce the ti
mages are preinterpretation
nalysis systemsy. These technextraction of in
defined as a s (pixel set), w
xel intensity ion. For this ro improve the 4; Arganda-Car
nd Banzatto-Juprocedure is ps’ analysis bu
ere are scientifand/or image e
bud (2017) hava viable altern
and stink bug e damages stud
gatti (2016) sted frameworkd accelerate it
que, Marcos-Fage analysis an
development consistent inf
Journal of A
(b) Deteriorsoybean embr
Krzyzanows
ues submitted . Observing su
ation of presennça Neto, Krzy
en contain met meristem, em
followed (ISTAn) varies from olting in greateome tests woulime spent for c
esented as a prn of images. Ths that perform niques allow grnformation, wh
decompositiowhich share sonformation, breason, recentaccuracy of t
rreras, Kaynig
unior (2003) wossible becaus
ut also the loc
fic papers in wextraction infor
ve evaluated tnative to assess
damage in sodied in soybean
tudied an imagk presented no
the process.
Filho, Gomes nd concluded
and on seedformation on th
Agricultural Sci
187
ration by humidryos submittedski and Costa
to tetrazoliumuch color nuance, location, zanowski, & C
echanical dammbryonic axis,
A, 1985; Françone individualer or less clasld contribute toclassification.
romising technhe agility in prclassification
reater precisiohich is impossi
on process intome visual chabut human betly, trainable the areas labe, Schindelin, C
when using imse each seed iscation and ext
which digital prrmation in seed
the cottonseeds the seed vigooybean seed un seed.
ge analysis frotable contribu
Junior, and Nthat system pr
dling growth he physiologic
ience
dity d to the tetrazol(1998)
m test can be uances, combine
and nature ofCosta, 1998).
mage, sting byalong the cen
ça Neto, Krzyl to another, asssification rigoo the best infor
nology due to ocessing plus and evaluationn in the analysible to be obse
to non-intersearacteristics. Meings, naturall
segmentation eling process Cardona, & Se
mage analysis s analyzed inditent of their d
rocessing and d quality tests.
d vigor througor. Pinto, Cicerusing x-ray an
ramework clasutions to the s
ascimento (20rovided data oindicating thaal potential of
V
(c) Bed bug dlium test. Sour
used to classifyed with the knf the damages
y bed bug, anntral cylinder,
yzanowski, & Cs well as the cror. Thus, toolsrmation credib
the improvemthe reduction on of agricultursis, intrinsic toerved (Carvalh
ecting regions,Most traditionally, use other
methods hav(Alekseev, Orung, 2014).
techniques toividually, suggdamage, with
image analysis.
gh computer iro, França Netnd concluded
ssification of oybean vigor
017) evaluatedon indexes of at automated f seeds.
Vol. 11, No. 13;
damage rce: França Ne
fy seeds into vnowledge of ss that can occ
nd deterioratioand in the vas
Costa, 1998), riteria of norms that minimizbility and guara
ments in systemof acquisition ral products a o the computatho, 2010).
, where a labl targeting metinformation w
ve emerged arlova, Rozalie
o characterize gesting in detai
a good degre
s were used to
mage analysisto, and Forti (2the technique
seed damagesdefinition pro
d squash seed vseedling vigorprocess a rel
2019
eto,
viable seeds ur in
on by cular
color ality/ ze or antee
ms of costs more tional
bel is thods when as an ev, &
seed il not ee of
help
s and 2009) was
s and ocess,
vigor r and liable
jas.ccsenet.org Journal of Agricultural Science Vol. 11, No. 13; 2019
188
Given this context, it is possible to use computational systems based on digital processing and image analysis to help obtaining information from quality tests carried out on the studied seeds.
2. Methodology Tool development and experimental tests were carried out in the Seed and Plant Evaluation Laboratory from UNIOESTE (Western Paraná State University)-Campus Cascavel-PR and in the Teaching Laboratories of Biology and Chemistry from UTFPR (Federal Technological University of Paraná), Campus Santa Helena-PR.
The proposed tool, named SATTz (System of Support for Tetrazolium Test), consists on a base system to classify seeds submitted to the tetrazolium test due to the need for an automation source for this test, to assist the analyst in the decision-making process to classify damages present in soybean embryos samples, and to carry out the test control during the batches’ evaluation.
Thus, specific tools were used for each step and, to reduce development cost, open and/or free source technologies were chosen since an open source software is ideal in scientific development, as it can be freely reviewed, modified, and redistributed (Schindelin, Rueden, Riner & Eliceiri, 2015).
Java object-oriented programming language version 1.8.0_91, in the Integrated Development Environment (Eclipse) version 4.5.2, was applied for this system development. This language was chosen because it is a multiplatform, which allows using and integrating libraries and frameworks for system development and digital image analysis.
The maintenance of data was carried out in the MySQL data file managing system, version 6.3. This system uses a relational database, thus, it was necessary to draw a mapping system of an oriented model to the developed application object for the relational model of MySQL data file. This mapping used Hibernate version 4.3.5, open source framework for object relational mapping, which allows keeping an object-oriented application. Also, that changes on the database do not imply in profound consequences on their application (O’Neil, 2008).
The ImageJ platform is a software distributed freely under GPL (General Public License) and has a dynamic users’ community, made up of researchers from several areas of knowledge. Its use allows several applications, ranging from data visualization to advanced image processing and statistical analysis. Due to its extensibility, it attracts biologists and computer scientists who efficiently implement specific image processing algorithms (Schindelin et al., 2015).
Libraries of the FIJI software was used in the process of image analysis, an ImageJ software distribution that adds several functionalities, which make easier scientific image analysis. This software was proposed as a productive collaboration platform between computer science researchers and the research communities on biology (Schindelin et al., 2012).
To perform the image segmentation process, the TWS (Trainable Weka Segmentation) plug-in was used (Arganda-Carreras et al., 2016), which is an integrating part of FIJI software. This plug-in act as a link between the machine learning fields and the digital image processing, providing the framework needed to use and compare the classifiers that make the image segmentation. This combines a collection of machine learning algorithms with a set of image characteristics to produce pixel-based segmentations, by using a set of methods to extract statistical properties from an image based on pixel samples (Arganda-Carreras et al., 2011; Arganda-Carreras et al., 2014). The WEKA (Waikato Environment for Knowledge Analysis) is an open source software consisting of a range of machine learning algorithms destined for data mining, which includes tools for data preprocessing, classification, regression, clustering, rules of association and visualization (Hall et al., 2009). All WEKA classification, regression, and clustering algorithms can be used by TWS (Arganda-Carreras et al., 2014).
In this trial, the classification algorithm based on Random Forest decision trees was used, a classification and regression technique created by Breiman (2001) and composed of a set of decision trees, in which the class prediction for new values is based on a voting system. After the generation of many trees (forest), the class is chosen, based on most tree votes (Breiman, 2001). According to the same author, a Random Forest is a classifier consisting of a set of tree-based classifiers, formally described as h (x, Θk), where h is the decision tree, x is the input to be sorted and Θk is the k-th identically distributed random vector. Thus, each tree votes in the most popular class for x input to be classified.
Random Forests are obtained by means of a multi-variant version of a predictor, known as bagging (bootstrapping aggregating) (Breiman, 1996), where the final forecast is performed by the average B predictions (Equation 2) or by taking the majority vote (Equation 1) (Goldstein, Polley, & Briggs, 2011).
jas.ccsenet.
where, ft(x
where, f*b
(
The basic Olshen, &capacity o(Equation without pr
where, N: examples.
where, Pi:
The TWSreduction. delimitinginto four pof image d
where, p: p
The propotraining ofregistration
The tetrazknown as The stocksuggested subsampleconditionsseeds wereAfterward
where, TS
After the ithrough th
org
x): function wit
(x): function w
algorithm of tr& Stone, 1984)of generalizatio
4) indices (Hruning procedu
set examples;
: frequency of
plugin allowSATTz extra
a radius (1 toparts (32, 64, 1data. This proc
probability of
osed system (Ff a classifier tn and manipul
olium test wascarriers of spe
k solution at 1by Associatio
es of 50 seedss for 16 hourse put in plastic
ds, the tempera
: Tetrazolium
ideal staining, he embryonic a
th the characte
ith the charact
ree constructio), which has aon, using as c
Han, Pei, & Kaure.
; M-number o
each class in e
ws the extractioacts the imag
o 8 pixels) arou28, 256 interv
cedure is param
each histogram
Figure 2) is cthrough the TWlation of the da
Figure 2
s carried out oecific damages1.0% was prepon of Official s each). The s in a BOD (B
c bags to receivture was betwe
solution; SS: S
the seeds wereaxis, with a sc
Journal of A
fmedium
eristic to be stu
fbag x
teristic to be st
on is the Classas particularitycriteria for theamber, 2003).
Impurity N =f classes; ωi: n
Gini
each knot; M:
on of image ges characterisund each pixe
vals in the histomount in the cl
E
m part.
composed of aWS interface ata on batches,
2. Analytical sy
on soybean sees. They were ppared by mixiSeed Analystsseeds were pa
Biochemical Ove 0.075% (Eqeen 35 and 40
TS0.075%
Stock solution
e removed fromcalpel. The cot
Agricultural Sci
189
m(x)= 1
T∑ f
tTt=1 (
udied; T: Numb= 1
B∑ f
*bBb=1 (
tudied; B: Num
sification and Ry the ability toe best feature In the forests
= -∑ Pω
i
tlM
i=1
number of exa
(N)=1– ∑Mi=1
number of cla
characteristicsstics using TWl of the imageogram). Then, assifier learnin
= ∑ – plog2(p
a graphical usand the printin, samples, repl
ystem manage
eds of cultivarprovided by aning 10.0 g of s [AOSA] (19acked in mois
Oxygen Demanquation 6) tetra°C for 150-18
= SS75ml + D
n; DW: Distille
m the solutiontyledons were
ience
(x
ber of training
(x)
mber of bootstr
Regression Treo generate higselection the Is, the CARTs
log 2 Pω
i
t
amples at ith p
Pi2
1
asses.
s by edge deteWS/FIJI core,e to obtain the the system cal
ng process.
p)
ser interface, wng of the tetralications, and s
ement framewo
s from agriculn official seed f tetrazolium s83) 100 seeds
sture germinatnd) type oven azolium solutio80 minutes (2.5
DW925ml
ed water.
n, washed with arranged on a
V
g samples.
rap samples.
ees (CART) (Bgh-performancImpurity (Equare expanded
partition; t: tota
ectors, texture, by HSB cohistogram of
lculates the En
which allows, azolium test eseeds.
ork
ltural propertieanalysis laborasalt in 1.0 L s of each samption paper and
at 25 °C. Afton to stay com5 to 3 hours) u
tap water, anda blue and squ
Vol. 11, No. 13;
Breiman, Freadce trees, with uation 3) and Gd to the maxim
al number of N
e filters, and nolor attributes,
that circle, divntropy (Equatio
in addition tovaluation card
es, as well as satory in the redistilled waterple were used d kept under fter this period
mpletely submeuntil the stainin
d cut longitudiuared sheet of
2019
(1)
(2)
dman, great GINI mum,
(3)
N set
(4)
noise and vided on 5)
(5)
o the d, the
seeds gion. r. As (two these
d, the ersed. ng.
(6)
nally EVA
jas.ccsenet.
(Ethyl Vinnecessary embryos s(bedbug, mtraining imof repetiticovering m
The imageExmor R(Arganda-related to t
The characused classFastRandooptimizes images usethe previou
3. ResultsThe set upresented humidity d
During thidamage (4humidity ccreating a classifier p
org
nyl Acetate) (Fto previously
samples (in remechanic, and
mage (Figure 3on I (Figure 3
mechanic, hum
(a) Repet
es were obtaineR™ CMOS S
Carreras, 201the tetrazolium
cteristics of thsifier was the omForest is aspeed and memed in the systeus process.
and Discussised has severbackground p
damage, and ti
is process, 39 r48.005 pixels),category (30.3training data s
pattern and, as
Figure 3). Eachtrain a classif
eplication imad humidity) anc). It was obta3a) and II (Fi
midity, and bed
tition I
Figure
ed using a SonSensor, witho6) was used,
m test.
e analyzed imaFastRandomF
a multitaskingmory usage (Aem and the bu
on ral damages thpatterns were ssue without l
regions from t, two regions i354 pixels) andset with 143,29a result, the cl
Journal of A
h cotyledon pafication modelages) were send healthy seeained from the igure 3b) that
dbugs’ damage
(b)
3. Arrangemen
ny HX200V 18out flash. Fo
allowing cha
ages are color Forest beginning reimplementArganda-Carreuilding of deci
hat, in some identified in esions.
the bottom clain mechanic dd eight region91 instances tolassified traini
Agricultural Sci
190
air was placed as the TWS ulected, represe
eds that take pcomposition osatisfactorily
s, and healthy
Repetition II
nt of embryos
8.2 Mega Pixeor the traininaracteristics ex
in HSB patterng with 200 trtation of Raneras, 2016). Thision trees, bas
cases, were fothe training i
ss of image (6damage class (ns related to tho describe the cing image was
ience
in a rectangleuses superviseenting the mapart on the tesof 16 embryosdescribed theseeds.
in EVA patter
els camera withng stage of xtraction and
rn and Entropyrees and two
ndom Forest, his process aimsed on the cha
found in the simage, bed bu
68.236 pixels),(470 pixels), 5he class of damclasses. The trobtained (Figu
V
e (2.5 × 2.8 cmed classificatioajor informatist to create a s image, selectee pattern in th
(c) Trainin
rn
h Charge-Coupthe classifierclassifier trai
y with rmax=8random characreated by F
ms to extract tharacteristic vec
same embryosug damage, de
, 141 regions o59 regions in dmage per bedbraining was perure 4).
Vol. 11, No. 13;
m) of pattern. Iton technique. Ton about damnew image, ced from the imhe tetrazolium
ng
pled Device (Cr, TWS inteining for attrib
8 and rmin=1acteristics per Fran Supek, whe characteristictor constructe
s. The regionseterioration du
of the class wideterioration dbug (2.775 pixrformed to obt
2019
t was Thus,
mages called mages
test,
CMD) rface butes
. The knot.
which ics of ed by
s that ue to
thout ue to xels), tain a
jas.ccsenet.
Figure 4.and em
org
(
(c
(
(
(i
(
(m
(
. Training imagmbryos withou
deteriorat
(a) MD
c) BBD
(e) HD
(g) HD
i) BBD
(k) HD
m) HD
(o) MD
ges with mechut damage (WDtion; Green-be
Journal of A
hanic damage (D), with respecedbug damage
Agricultural Sci
191
(MD), humiditctive segmente; Blue-mechan
ience
ty deteriorationed images: Grenic damage; Ye
V
(b) HD
(d) BBD
(f) BBD
(h) WD
(j) HD
(l) HD
(n) HD
(p) HD
n (HD), bedbuey-backgroundellow-vigor tis
Vol. 11, No. 13;
ug damage (BBd; Red-humiditssue
2019
BD), y
jas.ccsenet.
The registcarried outinference informatioanalyzed aboundariesthe inform
The informdefinition bedbug daalso possibanalyst’s in
org
tration of batct through the bin any soybea
on was extracas from the cs where there w
mation extractio
mation extractto the seed)
amage, and humble to make a nference proce
ch samples anbatch creationan cotyledons
cted from cotyclassification was a seed coton interface to
Figure
tion interface from the cotymidity deteriovisual comparess or to verify
(a) Emb
Figure
Journal of A
nd their replicn interface, afte
sample. Oncyledons containterface (Figtyledons pair, help on the cl
5. Soybean em
of embryos (Fyledons image
oration) as welrison between y if the classifi
bryos informatio
6. Embryos in
Agricultural Sci
192
ations (consiser obtaining thce the batch wained in their gure 5), usinghence obtaininassification pr
mbryos classifi
Figure 6) alloe segmented il as the percensegmented an
ication has bee
on
nformation extr
ience
sting of two imhe classificatiowas registered
replications. g the system ng informationrocess of the te
ication interfac
ows the decisiointo classes (hntages of pixelnd original emben correctly pe
raction interfac
V
mages of tetraon pattern that
with its respThe replicatioindividual seln about the emetrazolium test
ce
on-making (vihealthy seed, ls classified inbryos (Figure rformed.
(b) Origin
ce
Vol. 11, No. 13;
azolium test) serves as basi
pective sampleons I and II lection tool in
mbryos display.
igor (Note 1)mechanic dam
n these classes6b) to improv
nal image
2019
were is for e, the were
n the ed in
class mage, . It is
ve the
jas.ccsenet.
To improvthe test, thclassificati
The imagehealthy seare uniformCosta, 199
It can be oand 0.67%of a more Figure 7b deteriorati
The interpstriations (function oand lesion
Areas of mthey are a 1998).
It can alsopattern, are
org
ve the predictivhe characteristiion process, th
es labeled by ed was higherm and superfic98).
(a) Healt
observed that % with humidit
intense red codid not show
on class (0.35%
pretation of hu(evidenced by f expansion (hin cotyledons
more intense rresult of smal
o be seen in Fe those that giv
ve properties oics extraction
he data set reco
the pattern arr than 99.33% cial, indicating
thy seeds (99.33
Figure 7. S
in Figure 7a, ty deteriorationolor in one ofw any deterior%).
umidity deterithe test in dar
hydration) and . This kind of
red or milky wll striations inc
Figures 8a andve to pixel hum
Journal of A
of the classifierwas carried ou
orded 230.388
re shown in Fand represente
g slow penetra
3%)
Segmented ima
the soybean en. When compf the cotyledonration. Howev
ioration in tetrk red carminecontraction (dpattern may be
white staining crease and cau
d 8b images wmidity deterior
Agricultural Sci
193
r pattern, durinut in some iminstances (add
Figures 7a anded seeds of vigation of the tet
ages classified
embryos recordpared with the ns, originatingver, some pix
trazolium test e), caused by dehydration) cye found in emb
are also charause a greater d
whose most appration patterns
ience
ng the classificmages to feed thding 87.097 ins
d 7b, whose pgor class 1. Thtrazolium salt
(b) Hea
d as without da
ded 99.33% ooriginal image
g from humiditxels were clas
can be descrcompression oycles, whose rbryonic axis an
acteristic patterdamage (Franç
parent charact.
V
cation process he training setstances to the i
ercentage of pheir external c(França Neto
althy seeds (99.6
mage
f pixels labelee, there were tty deterioratiosified as belo
ribed by someof integument esults are wrinnd cotyledons.
rns of this detça Neto, Krzyz
teristics, labele
Vol. 11, No. 13;
of the 100 seet. At the end oinitial tests set
pixels classifiecolor character, Krzyzanowsk
62%)
ed as healthy stwo small striaon. The embryonging to hum
e patterns, sucon cotyledons
nkles in integu.
terioration. Mozanowski, & C
ed by the clas
2019
eds in of the t).
ed as istics ki, &
seeds ations os of
midity
ch as s as a ment
ostly, Costa,
sifier
jas.ccsenet.
In Figuresclustered iin these re
In Figures with the ostriations adeterioratisegmentatiimages.
The mechaprocessingWhen recein cotyledCharacteri
Pixels clasadjacent toand its oppimage, tha
In Figure 9(vascular rAnd whenhappened humidity i
In general,
org
(a) Humidity deregion
Fig
s 8a, it is possin the oppositegions, charact
8b, it is obseroriginal image and bands in don. When loion generated
anic damage rg process identent, they are codons as in emistic embryos w
ssified as meco the attachmenposite region; oat the mechanic
9b some pixelsregion), plus an analyzed theduring some
in its cotyledon
, pixels with th
eterioration on thto embryonic ax
ure 8. Segmen
ible to observe region to emberistics of hum
rved that the em(without dama
darker red carmoking at imaby the system
result from imtified in the tetolored in dark
mbryonic axis with this kind o
chanic damagnt point. Thereon the other hac damage cam
s classified as a substantial exe original imagof the produc
ns.
his pattern of d
Journal of A
he opposite xis
nted images wi
ve the number bryonic axis. I
midity deteriora
mbryonic axisage), it is possmine color an
ages set that corroborates w
mpacts, cuts, antrazolium test
k red carmine, (Bewley & Bof damage are
e can be obsee has been recoand, it is not de from an abra
mechanic damxtent of deterige, it was recction processe
damage were w
Agricultural Sci
194
(b
ith humidity de
of pixels clasn the non-segmation.
also presentedsible to observ
nd whitish partpresented humwith the real lo
nd abrasions arpresents charaor white in ca
Black, 1994; F presented in F
erved in one orded humidityistributed throasion in seed.
mage can be seioration due toorded that me
es, and that th
well identified
ience
b) Humidity det
eterioration ch
ssified as humimented image,
d pixels classifve, both in thets (dead tissuemidity deterioocation and ex
rising from thacteristic patterase they have nFrança Neto, KFigures 9a to 9
cotyledon (Figy deteriorationughout its leng
een in the regioo humidity disechanic damaghe seed also h
and classified
V
terioration on em
haracteristics
idity deteriora, striations and
fied in this wae cotyledons ane), which are aoration, it is xtension of dam
e harvesting, trns such as cranot been recenKrzyzanowski
9b.
gure 9a), in thn in cotyledonsgth. It can be s
on just above hstributed betwege is due to sehas deterioratio
.
Vol. 11, No. 13;
mbryonic axis
ation in cotyledd bands can be
ay. When compnd embryonic aspects of hum
observed thamage in the ori
trail, transportacks and abrasntly identified, i, & Costa, 1
he vascular res in embryonicseen, in the ori
half of the embeen the cotyledeeds kneadingon patterns du
2019
dons, seen
pared axis,
midity t the ginal
, and sions.
both 998).
egion c axis ginal
bryos dons. g that ue to
jas.ccsenet.
The damagyellowish The embry
Pixels claspart of theobserved areach a noembryonicoriginal imsome case
After the interface. TThe systemvigor percmost appar
After selecresults (Fig
org
(a) Mechanic da
F
ge caused by bcolor, as well
yos that were s
(a) Sting bed
ssified as bedbe soybean embaround this daon-vascularizedc axis. It is pomages of theses the red ring a
embryos wereThis data sheem generates thcentages are rerent ones.
cting the batchgure 11).
amage in vascul
Figure 9. Segm
bedbug sting i as an intense
segmented with
dbug in vascular
Figure 10. Se
bug damage arbryo in a cotyleamage are alsod region of coossible to obsee segmentationaround the dam
e classified, thet is based on this data file auegistered. It als
h and sample,
Journal of A
lar region
mented images
is easily identired carmine r
h this characte
r region
egmented imag
re found in cotedon and a smo classified as otyledons, as cerve that pixens, both for bmage was iden
he results interthe tetrazolium
utomatically, coso identifies an
the system gen
Agricultural Sci
195
with mechanic
ified by typicaring around it eristic pattern a
ges with bedbu
tyledons, as obmall part in the
humidity detecan be observels were correc
bedbug damagntified as a dete
rface of the tem test data file onsidering thend quantifies t
nerates a Porta
ience
(b) Mechanic d
c damage char
al lesions, with(França Neto,
are shown in F
(b) Sting bedb
ug sting charac
bserved in Figue central positierioration. Pixed in Figure 1ctly classified e and humiditerioration patte
etrazolium test(França Neto
e previous clasthe kinds of da
able Documen
V
damage in centra
racteristics
h circular patte, Krzyzanowsk
Figures 10a and
bug below vascu
cteristics
ure 10a. They ion of the othexels classified 0b, in the regby the visual
ty deterioratioern by humidit
t was generate, Krzyzanowsk
ssification, in wamages, as we
nt Format (PDF
Vol. 11, No. 13;
al portion
erns and whitiki, & Costa, 1d 10b.
ular region
are in the super cotyledon. Pas bedbug damion opposite tl analysis fromon. Even thougty.
ed by the repoki, & Costa, 1which viabilityell as highlight
F) file with the
2019
sh or 998).
perior Pixels mage o the
m the gh in
orting 998). y and ts the
e test
jas.ccsenet.
The reportgerminatiothe sum oestablished
4. ConclusConcerninseeds clasthe segmeof tetrazolshows itse
AcknowleAuthors aCoordinatisupporting
ReferenceAlekseev,
MethoConfe
Arganda-Csegme
Arganda-CTrain
org
t determines ton potential upof the levels d.
sion ng the developesification for tnted images pium test repor
elf as a potentia
edgements are thankful tion for the Img this research.
es A. V., Orlova
od for Conteference JCKBS
Carreras, I., Centation. Retri
Carreras, I., Knable_Segment
the percentagep from the sum
from 1 to 3
ed software, ittetrazolium tes
provides valuabrt from the clasal automation t
to Western Pmprovement o.
a, Y. A., Rozaext-Sensitive ISE, Volgograd,
Cardona, A., ieved from http
Kaynig, V., tation: Release
Journal of A
Figure 11. T
e of seeds, clam of estimated v
, following w
t can be conclust, since perceble informatiossified sampletool in researc
Paraná State Uof Higher Ed
aliev, V. L., &Image Analys, Russia. https
Kaynig, V., ps://imagej.net
Rueden, C., e v3.1.2 Retrie
Agricultural Sci
196
etrazolium tes
assified in eacvalues in the c
what França N
uded it presententages extracton in the deciss saves time in
ch and commer
University, Feducation Perso
& Zaboleeva-Zis (pp. 331-3://doi.org/10.1
Rueden, C., t/Trainable_W
Schindelin, eved from https
ience
t report
ch viability clclasses from 1 Neto, Krzyzan
ts an alternativtion of classifision making. Mn the extractinrcial environm
ederal Univeronnel/Araucári
Zotova, A. V. (340). Paper p007/978-3-319
& SchindelinWeka_Segment
J., Cardona, s://zenodo.org
V
lass, and the sto 5, and the v
nowski, and C
ve to performinfied pixels for Moreover, Autg process of te
ments.
sity of Technia Foundation
(2014). Two-Spresented at t9-11854-3_28
n, J. (2011). ation
A., & Seun/record/59290
Vol. 11, No. 13;
software calcuvigor is obtaineCosta (1998)
ng a less subjedamage classetomated generest information
nology-Paranán (CAPES/FA
Stage Segmentthe eleventh
Trainable W
ng, H. S. (2#.XO6-cohKiU
2019
ulates ed by have
ective es by ration n and
and ) for
ation Joint
WEKA
2016) Uk
jas.ccsenet.org Journal of Agricultural Science Vol. 11, No. 13; 2019
197
Arganda-Carreras, I., Kaynig, V., Schindelin, J., Cardona A. H., & Seung, S. (2014), Trainable Weka Segmentation: A machine learning tool for microscopy image segmentation. Retrieved from http://bit.do/eTHEL
Association of Official Seed Analysts. (2008). Seed vigor testing handbook. East Lansing: AOSA.
Bewley J. D., & Black M. (1994). Seeds. Physiology of development and germination (2nd ed). New York, NY: Plenum Press. https://doi.org/10.1007/978-1-4899-1002-8
Breiman, L. (1996). Bagging predictors. Machine Learning, 24(2), 123-140. https://doi.org/10.1023/A:101805 4314350
Breiman, L. (2001). Random forests. Machine Learning, 45(1), 5-32. https://doi.org/10.1023/A:1010933404324
Breiman, L., Freadman, J., Olshen, R., & Stone, C. (1984). Classification and regression trees. Grove, CA: Wadsworth International.
Carvalho, M. L. M. (2010). Utilização da análise de imagem-conceitos, metodologias e usos. Informativo ABRATES, 20(3), 45-47.
Cícero, S. M., Banzatto-Junior, H. L. (2003). Avaliação do relacionamento entre danos mecânicos e vigor, em sementes de milho, por meio da análise de imagens. Revista Brasileira de Sementes, 25(1), 29-36. https://doi.org/10.1590/S0101-31222003000100006
Corrêa, P. D., Cicero, S. M., & Abud, H. F. (2019). Evaluation cottonseed vigor through computer image analysis of the seedlings. Journal of Seed Science, 41(1), 54-59. https://doi.org/10.1590/2317-1545 v41n1200649
Dall’Gnol, A. (2000). The impact of soybeans on the Brazilian economy. Technical information for agriculture. São Paulo: Máquinas Agrícolas Jacto.
Deswal, D. P., & Chand, U. (1997). Standardization of thetetrazolium test for viability estimation in rice bean (Vigna umbellata (Thunb.) Ohwi & Ohashi) seeds. Seed Science and Technology, 25(3), 409-417.
FAO (Food and Agriculture Organization of the United Nations). (2010). Seeds in emergencies: A technical handbook. FAO Plant Production and Protection Paper 202. FAO, Rome, Italy.
França Neto, J. B., Krzyzanowski, F. C., & Costa, N. P. da. (1998). The tetrazolium test for soybean seeds. Londrina, PR: EMBRAPA-CNPSo.
Goldstein, B. A., Polley, E. C., & Briggs, F. B. (2011). Random forests for genetic association studies. Statistical Applications in Genetics and Molecular Biology, 10(1), 32. https://doi.org/10.2202/1544-6115.1691
Hall, M., Frank, E., Holmes, G., Pfahringer, B., Reutemann, P., & Witten, I. H. (2009). The WEKA data mining software: An update. ACM SIGKDD Explorations Newsletter, 11(1), 10-18. https://doi.org/10.1145/ 1656274.1656278
Han, J., Pei, J., & Kamber, M. (2011). Data mining: Concepts and techniques. Waltham, MA: Elsevier.
ISTA (International Seed Testing Association). (1985). International rules for seed testing. Seed Science and Technology, 13, 229-526.
Marcos-Filho, J. (2015). Seed vigor testing: an overview of the past, present and future perspective. Scientia Agricola, 72(4), 363-374. https://dx.doi.org/10.1590/0103-9016-2015-0007
Moore, R. P. (1973). Tetrazolium staining for assessing seed quality. In W. Heydecker (Ed.), Seed ecology (pp. 347-366). London, Butterworth.
O’Neil, E. J. (2008). Object/relational mapping 2008: Hibernate and the entity data model. Paper presented at ACM SIGMOD international conference on management of data (pp. 1351-1356). ACM, New York. https://doi.org/10.1145/1376616.1376773
Pereira, D. F., Saito, P. T. M., & Bugatti, P. H. (2016). An image analysis framework for effective classification of seed damages (pp. 61-66). Paper presented at the 31st Annual ACM Symposium on Applied Computing. ACM, New York, NY. https://doi.org/10.1145/2851613.2851637
Pinto, T. L. F., Cicero, S. M., França Neto, J. B., & Forti, V. A. (2009). An assessment of mechanical and stink bug damage in soybean seed using X-ray analysis test. Seed Science and Technology, 37(1), 110-220. https://doi.org/10.15258/sst.2009.37.1.13
jas.ccsenet.org Journal of Agricultural Science Vol. 11, No. 13; 2019
198
Roessing, A. C., Dall’Agnol, A., Lazzarotto, J. J., Hirakuri, M. H., & Oliveira, A. B. (2007). O complexo agroindustrial da soja brasileira. Circular Técnica, 43.
Schindelin J., Arganda-Carreras I., Frise, E., Kaynig, V., Longair, M., Pietzsch, T., … Cardona A. (2012). Fiji: An open-source platform for biological-image analysis. Nature Methods, 9(7), 676-682. https://doi.org/ 10.1038/nmeth.2019
Schindelin, J., Rueden, C. T., Hiner, M. C., Eliceiri, K. W. (2015). The ImageJ ecosystem: An open platform for biomedical image analysis. Molecular Reproduction and Development, 82(7), 518-529. https://doi.org/ 10.1002/mrd.22489
Silva, P. P., Barros, A. C. S. A., Marcos-Filho, J., Gomes Junior, F. G., & Nascimento, W. M. (2017). Assessment of squash seed vigor using computerized image analysis. Journal of Seed Science, 39(2), 159-165. https://dx.doi.org/10.1590/2317-1545v39n2171177
Notes Note 1. The assignments of seed vigor classifications based on the information extracted by the system were according to the methodology of tetrazolium test for soybean seeds (França Neto, Krzyzanowski, & Costa, 1998).
Copyrights Copyright for this article is retained by the author(s), with first publication rights granted to the journal.
This is an open-access article distributed under the terms and conditions of the Creative Commons Attribution license (http://creativecommons.org/licenses/by/4.0/).