I
AE
S In
t
er
na
t
io
na
l J
o
urna
l o
f
Art
if
icia
l In
t
ellig
ence
(
I
J
-
AI
)
Vo
l.
1
5
,
No
.
2
,
A
p
r
il 2
0
2
6
,
p
p
.
1
5
4
8
~
1
5
6
5
I
SS
N:
2
2
5
2
-
8
9
3
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijai.v
15
.i
2
.
p
p
1
5
4
8
-
1
5
6
5
1548
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
a
i
.
ia
esco
r
e.
co
m
A nov
el appro
a
ch
t
o
det
ect
t
o
ma
to l
ea
f
disea
se using
v
isio
n
trans
former
Sa
nje
ela
Sa
g
a
r
1
,
J
a
s
wind
er
Sin
g
h
2
1
D
e
p
a
r
t
me
n
t
o
f
I
n
f
o
r
mat
i
o
n
Te
c
h
n
o
l
o
g
y
a
n
d
D
a
t
a
S
c
i
e
n
c
e
,
V
i
d
y
a
l
a
n
k
a
r
S
c
h
o
o
l
o
f
I
n
f
o
r
mat
i
o
n
Te
c
h
n
o
l
o
g
y
,
M
u
mb
a
i
,
I
n
d
i
a
2
D
e
p
a
r
t
me
n
t
o
f
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
a
n
d
E
n
g
i
n
e
e
r
i
n
g
,
G
u
r
u
J
a
mb
h
e
sh
w
a
r
U
n
i
v
e
r
s
i
t
y
o
f
S
c
i
e
n
c
e
a
n
d
T
e
c
h
n
o
l
o
g
y
,
H
i
sar,
I
n
d
i
a
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
J
u
l 1
1
,
2
0
2
5
R
ev
is
ed
Dec
2
0
,
2
0
2
5
Acc
ep
ted
J
an
2
5
,
2
0
2
6
To
m
a
to
e
s
a
re
o
n
e
o
f
th
e
m
o
st
wi
d
e
ly
c
o
n
s
u
m
e
d
v
e
g
e
tab
les
a
c
ro
ss
th
e
wo
r
ld
.
Ho
we
v
e
r,
t
o
m
a
to
e
s
a
re
p
ro
n
e
to
d
ise
a
se
s.
Re
c
o
g
n
izin
g
a
n
d
c
las
sify
in
g
to
m
a
to
lea
f
d
ise
a
se
s
is
c
ru
c
ial
t
a
sk
.
Va
rio
u
s
d
e
e
p
lea
rn
i
n
g
(DL
)
m
e
th
o
d
s
h
a
v
e
b
e
e
n
d
e
v
e
lo
p
e
d
b
y
se
v
e
ra
l
re
se
a
rc
h
e
rs,
b
u
t
th
e
y
h
a
v
e
so
m
e
c
o
m
p
lex
issu
e
s
li
k
e
n
o
ise
in
ima
g
e
s,
h
i
g
h
c
o
m
p
u
tatio
n
a
l
c
o
m
p
lex
i
ty
,
p
o
o
r
a
c
c
u
ra
c
y
,
a
n
d
li
m
it
e
d
fe
a
tu
re
se
lec
ti
o
n
.
Th
e
m
a
in
g
o
a
l
o
f
t
h
is
re
se
a
rc
h
is
to
p
re
se
n
t
n
o
v
e
l
DL
b
a
se
d
t
o
m
a
to
lea
f
d
is
e
a
se
c
la
ss
ifi
c
a
ti
o
n
fra
m
e
wo
rk
with
n
e
u
ra
l
ne
two
rk
b
a
se
d
g
a
ted
v
isi
o
n
tra
n
sfo
rm
e
r
(G
-
ViT)
m
o
d
e
l
a
ss
isted
a
tt
e
n
ti
o
n
m
e
c
h
a
n
ism
.
Th
e
p
r
o
p
o
se
d
fra
m
e
wo
rk
u
se
s
d
i
late
d
c
o
n
v
o
lu
t
io
n
with
b
id
irec
ti
o
n
a
l
l
o
n
g
sh
o
rt
-
term
m
e
m
o
ry
(Bi
-
D
LS
TM
)
u
se
d
fo
r
e
ffici
e
n
t
fe
a
tu
re
e
x
trac
ti
o
n
t
o
e
n
h
a
n
c
e
th
e
c
las
sifica
ti
o
n
.
An
e
ffe
c
ti
v
e
c
h
a
o
ti
c
sp
i
d
e
r
wa
sp
o
p
ti
m
iza
ti
o
n
(CS
WO)
is
u
se
d
fo
r
fe
a
tu
re
se
lec
ti
o
n
.
F
u
r
th
e
r,
n
o
v
e
l
a
tt
e
n
ti
o
n
b
a
se
d
g
a
ted
v
isi
o
n
tran
sf
o
rm
e
r
(A
-
G
VT)
is
u
se
d
to
c
las
sify
to
m
a
to
lea
f
d
ise
a
se
s
wh
ich
in
teg
ra
tes
stre
n
g
th
s
o
f
a
tt
e
n
t
io
n
m
e
c
h
a
n
ism
a
n
d
G
-
ViT
m
o
d
e
ls.
F
u
rt
h
e
r,
to
imp
r
o
v
e
t
h
e
g
e
n
e
ra
li
z
a
b
il
it
y
o
f
c
las
sifica
ti
o
n
m
o
d
e
l,
it
s
p
a
ra
m
e
ters
a
r
e
tu
n
e
d
with
b
lac
k
wid
o
w
o
p
ti
m
iza
ti
o
n
(BWO)
a
lg
o
rit
h
m
.
Th
e
e
x
p
e
rime
n
tal
fin
d
in
g
s
sh
o
ws
th
a
t
p
ro
p
o
se
d
fra
m
e
wo
rk
o
u
t
p
e
rfo
rm
e
d
p
re
v
io
u
s
stu
d
ies
o
n
t
o
m
a
to
lea
f
d
ise
a
se
id
e
n
ti
fica
ti
o
n
a
n
d
c
la
ss
ifi
c
a
ti
o
n
m
o
d
e
ls
i
n
term
s
o
f
a
c
c
u
ra
c
y
,
p
r
e
c
isio
n
,
re
c
a
ll
,
F
1
-
sc
o
re
,
s
p
e
c
ifi
c
it
y
,
m
e
a
n
a
b
so
lu
te
e
rro
r
(M
AE),
a
n
d
ro
o
t
m
e
a
n
sq
u
a
re
e
rro
r
(RM
S
E)
with
9
9
.
7
%
,
9
8
.
2
9
%
,
9
8
.
2
2
%
,
9
8
.
2
5
%
,
9
9
.
1
9
%
,
0
.
0
3
,
a
n
d
0
.
2
5
re
sp
e
c
ti
v
e
ly
.
Th
e
p
ro
p
o
se
d
st
u
d
y
c
a
n
p
a
v
e
a
wa
y
f
o
r
n
e
w ag
ricu
lt
u
ra
l
re
v
o
lu
ti
o
n
.
K
ey
w
o
r
d
s
:
Atten
tio
n
m
ec
h
an
is
m
Dee
p
lear
n
in
g
R
ec
u
r
r
en
t n
eu
r
al
n
etwo
r
k
T
o
m
ato
leaf
d
is
ea
s
e
class
if
icatio
n
Vis
io
n
tr
an
s
f
o
r
m
er
T
h
is i
s
a
n
o
p
e
n
a
c
c
e
ss
a
rticle
u
n
d
e
r th
e
CC B
Y
-
SA
li
c
e
n
se
.
C
o
r
r
e
s
p
o
nd
ing
A
uth
o
r
:
San
jeela
Sag
ar
Dep
ar
tm
en
t o
f
I
n
f
o
r
m
atio
n
T
e
ch
n
o
lo
g
y
an
d
Data
Scien
ce
,
V
id
y
alan
k
ar
Sch
o
o
l o
f
I
n
f
o
r
m
at
io
n
T
ec
h
n
o
lo
g
y
San
g
am
Nag
ar
,
W
ad
ala
E
ast,
Mu
m
b
ai
3
7
,
I
n
d
ia
E
m
ail:
s
an
jeela
.
s
ag
ar
@
v
s
it.e
d
u
.
in
1.
I
NT
RO
D
UCT
I
O
N
Ag
r
icu
ltu
r
e
aid
s
t
h
e
n
atio
n
’
s
d
ev
elo
p
m
e
n
t.
T
h
e
ec
o
n
o
m
y
o
f
th
e
n
atio
n
m
ain
ly
d
ep
en
d
s
o
n
ag
r
icu
ltu
r
e
[
1
]
.
T
h
e
r
o
le
o
f
a
g
r
icu
ltu
r
e
h
as
ex
p
an
d
e
d
with
th
e
ac
ce
p
tan
ce
o
f
tech
n
o
lo
g
ie
s
lik
e
g
r
ee
n
en
er
g
y
s
u
ch
as
b
io
f
u
els.
Fu
r
t
h
er
m
o
r
e,
ag
r
icu
ltu
r
e
is
th
e
s
o
u
r
ce
o
f
r
a
w
m
ater
ials
in
m
ak
in
g
ch
em
ic
als
an
d
m
ed
icin
es
[
2
]
.
Acc
o
r
d
in
g
to
th
e
r
ep
o
r
t
o
f
f
o
o
d
an
d
ag
r
icu
ltu
r
al
o
r
g
a
n
izatio
n
(
FAO)
o
f
th
e
Un
ited
States
,
th
e
r
ate
o
f
h
u
n
g
e
r
is
g
r
ad
u
ally
in
cr
ea
s
in
g
s
in
ce
2
0
1
4
[
3
]
.
Plan
ts
p
r
o
v
id
e
f
o
o
d
f
o
r
all
liv
in
g
b
ein
g
s
a
n
d
life
is
n
o
t
p
o
s
s
ib
le
with
o
u
t
p
lan
ts
.
Alth
o
u
g
h
p
la
n
ts
ar
e
ess
en
tial
f
o
r
life
,
th
ei
r
p
r
o
d
u
ctio
n
is
af
f
ec
ted
b
y
s
ev
er
al
d
is
ea
s
es
[
4
]
.
T
o
m
ato
is
a
k
in
d
o
f
f
r
u
it
o
r
v
eg
etab
le,
wh
ich
is
wid
ely
cu
lt
iv
ated
in
I
n
d
ia.
T
h
e
r
e
ar
e
m
a
n
y
ty
p
es
o
f
to
m
ato
d
is
ea
s
es a
n
d
p
ests
in
th
e
wh
o
le
g
r
o
wth
o
f
to
m
at
o
life
cy
cle
[
5
]
.
T
o
m
ato
es
ar
e
r
ich
in
n
u
tr
ien
ts
h
ig
h
ly
co
n
s
u
m
e
d
b
y
h
u
m
an
s
[
6
]
.
T
o
m
ato
es
ca
n
g
r
o
w
in
an
y
ty
p
e
o
f
s
o
il if
it h
as a
n
ad
eq
u
ate
d
r
ain
ag
e
[
7
]
.
T
h
e
d
em
an
d
o
f
th
e
to
m
ato
is
h
ig
h
d
u
e
to
its
p
h
ar
m
a
co
lo
g
ical
p
r
o
p
er
ties
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
A
n
o
ve
l a
p
p
r
o
a
c
h
to
d
etec
t to
ma
to
lea
f d
is
ea
s
e
u
s
in
g
visi
o
n
tr
a
n
s
fo
r
mer (
S
a
n
jeela
S
a
g
a
r
)
1549
s
u
ch
as
r
ed
u
cin
g
h
y
p
e
r
ten
s
io
n
,
g
in
g
i
v
al
b
leed
in
g
,
an
d
h
e
p
atitis
[
8
]
.
T
h
ese
p
lan
ts
ar
e
m
o
r
e
s
u
s
ce
p
tib
le
to
d
is
ea
s
es
an
d
ca
u
s
es
f
in
a
n
cia
l
lo
s
s
in
a
g
r
icu
ltu
r
al
s
ec
to
r
[
9
]
.
Hen
ce
,
id
en
tify
i
n
g
t
h
e
p
lan
t
d
is
ea
s
es
ar
e
im
p
o
r
tan
t to
lig
h
ten
u
p
th
e
y
ie
ld
lo
s
s
es
[
1
0
]
.
T
r
ad
itio
n
al
d
iag
n
o
s
tic
m
eth
o
d
s
ar
e
ex
p
en
s
iv
e,
tim
e
co
n
s
u
m
in
g
an
d
lab
o
r
-
in
te
n
s
iv
e.
Vis
u
al
an
aly
s
is
an
d
ch
em
ical
test
in
g
ar
e
t
h
e
m
o
s
t
u
s
ed
tr
ad
itio
n
al
m
eth
o
d
s
f
o
r
to
m
ato
leaf
d
is
ea
s
e
d
iag
n
o
s
is
[
1
1
]
.
Far
m
er
s
m
ain
ly
u
s
e
tr
ad
itio
n
al
m
eth
o
d
s
o
f
id
en
tify
in
g
d
is
ea
s
es b
y
s
h
ar
in
g
s
am
p
les with
ex
p
er
ts
to
i
d
en
tify
th
e
d
is
ea
s
e.
E
x
p
er
ts
m
o
s
tly
d
o
n
ak
e
d
ey
e
o
b
s
er
v
atio
n
s
f
o
r
d
is
ea
s
e
id
en
tific
atio
n
,
an
d
th
is
lead
s
to
m
o
r
e
co
m
p
licatio
n
s
in
ter
m
s
o
f
d
is
ea
s
e
id
en
tific
atio
n
[
1
2
]
.
Acc
o
r
d
in
g
to
s
tatis
tical
d
ata,
m
o
r
e
th
an
2
0
to
m
ato
d
is
ea
s
es
af
f
ec
ts
th
e
to
m
ato
cr
o
p
p
r
o
d
u
ctio
n
ac
r
o
s
s
th
e
g
lo
b
e
[
1
3
]
.
W
ith
th
e
d
ev
elo
p
m
en
t
o
f
co
m
p
u
ter
s
an
d
in
ter
n
et
o
f
th
in
g
s
(
I
o
T
)
tec
h
n
o
lo
g
ies,
th
e
ag
r
ic
u
ltu
r
al
d
is
ea
s
e
d
iag
n
o
s
is
h
as e
n
h
an
ce
d
[
1
4
]
.
So
m
e
o
f
th
e
p
lan
t
d
is
ea
s
es
ca
n
b
e
d
iag
n
o
s
ed
b
y
ex
am
in
i
n
g
th
e
n
u
cleu
s
.
Hen
ce
,
ce
ll
s
eg
m
en
tatio
n
is
an
im
p
o
r
tan
t
asp
ec
t
o
f
t
o
m
ato
p
lan
t
d
is
ea
s
e
id
en
tific
atio
n
[
1
5
]
.
Dee
p
lear
n
in
g
(
DL
)
tech
n
iq
u
es
ar
e
co
m
m
o
n
ly
u
s
ed
f
o
r
s
eg
m
en
tatio
n
a
n
d
cl
ass
if
icatio
n
o
f
d
is
ea
s
e
d
iag
n
o
s
is
[
1
6
]
.
DL
with
co
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
s
(
C
NN)
en
h
an
ce
s
th
e
d
iag
n
o
s
is
an
d
class
if
icatio
n
o
f
to
m
ato
leaf
d
is
ea
s
es.
C
NN
ex
tr
ac
ts
th
e
f
ea
tu
r
es
an
d
en
s
u
r
e
r
eliab
ilit
y
,
au
th
en
ticity
an
d
v
alid
atio
n
o
f
th
e
m
o
d
el
[
1
7
]
.
C
NN
h
as
s
ev
er
al
ar
c
h
ite
ctu
r
es
an
d
m
o
s
t
o
f
th
em
h
av
e
lar
g
e
n
u
m
b
er
o
f
d
e
ep
lay
er
s
with
lar
g
e
n
u
m
b
e
r
o
f
p
ar
am
eter
s
[
1
8
]
.
I
n
r
e
c
en
t
t
i
m
e
s
,
s
e
v
er
a
l
s
tu
d
i
e
s
w
e
r
e
e
s
t
ab
l
i
s
h
ed
to
an
a
ly
z
e
th
e
d
i
s
e
a
s
e
s
o
f
t
o
m
a
t
o
le
a
v
e
s
.
DL
m
e
t
h
o
d
s
ar
e
th
e
m
o
s
t
g
e
n
e
r
a
l
m
e
t
h
o
d
u
s
e
d
f
o
r
d
e
t
ec
t
i
o
n
o
f
t
o
m
a
t
o
l
e
a
f
d
i
s
e
as
e
s
a
u
t
o
m
a
t
i
c
a
l
ly
.
N
e
v
e
r
th
e
l
e
s
s
,
m
an
y
o
f
th
e
D
L
a
r
ch
i
t
e
c
t
u
r
e
s
h
a
s
s
o
m
e
c
o
m
p
l
i
c
a
t
i
o
n
s
l
i
k
e
c
o
m
p
u
ta
ti
o
n
a
l
co
m
p
l
e
x
i
ty
,
u
p
d
a
t
i
n
g
th
e
p
ar
a
m
e
t
er
s
an
d
s
o
o
n
wh
i
c
h
l
e
ad
s
to
c
la
s
s
i
f
i
ca
t
i
o
n
c
o
m
p
l
ex
i
t
y
.
T
o
d
e
a
l
w
i
th
th
e
s
e
l
im
i
t
a
t
i
o
n
s
t
h
e
p
r
o
p
o
s
ed
s
t
u
d
y
i
n
tr
o
d
u
ce
s
a
n
o
v
e
l
c
l
a
s
s
i
f
i
c
a
t
io
n
a
p
p
r
o
a
ch
f
o
r
t
o
m
a
t
o
l
e
a
f
d
i
s
e
a
s
e
r
e
co
g
n
i
t
i
o
n
w
i
th
a
t
t
e
n
t
i
o
n
b
a
s
ed
g
a
t
ed
v
i
s
i
o
n
tr
a
n
s
f
o
r
m
e
r
(A
-
G
V
T
)
.
Ma
j
o
r
c
o
n
tr
i
b
u
t
i
o
n
s
o
f
t
h
e
p
r
o
p
o
s
e
d
r
e
s
e
ar
ch
ar
e
l
i
s
t
e
d
a
s
f
o
l
lo
w
s
:
i)
t
o
e
x
t
r
a
c
t
th
e
f
e
a
tu
r
e
s
o
f
to
m
a
to
le
a
f
d
i
s
e
a
s
e
s
a
d
i
l
a
t
ed
c
o
n
v
o
l
u
t
io
n
al
b
id
i
r
e
c
t
io
n
a
l
lo
n
g
s
h
o
r
t
-
t
e
r
m
m
em
o
r
y
(
Bi
-
D
L
ST
M
)
n
e
t
w
o
r
k
i
s
em
p
lo
y
ed
;
i
i
)
t
o
i
n
cr
e
a
s
e
th
e
a
cc
u
r
a
c
y
o
f
cl
a
s
s
i
f
i
c
a
t
i
o
n
a
n
d
t
o
s
e
l
e
c
t
t
h
e
o
p
t
i
m
a
l
f
e
a
tu
r
e
s
a
ch
a
o
t
i
c
s
p
i
d
e
r
wa
s
p
o
p
t
im
i
z
a
t
io
n
(
C
S
W
O
)
i
s
u
t
i
l
i
ze
d
;
ii
i
)
t
o
a
c
c
u
r
a
t
e
l
y
c
l
a
s
s
i
f
y
t
h
e
t
o
m
a
t
o
l
ea
f
d
i
s
e
a
s
e
s
,
a
n
at
t
e
n
t
i
o
n
b
a
s
ed
g
a
te
d
v
i
s
i
o
n
tr
a
n
s
f
e
r
m
o
d
e
l
i
s
d
ep
l
o
y
ed
;
a
n
d
i
v
)
t
o
r
e
d
u
c
e
t
h
e
t
o
m
a
to
d
i
s
e
a
s
e
c
la
s
s
i
f
i
c
a
t
i
o
n
c
o
m
p
l
ex
i
t
y
,
p
ar
a
m
e
t
e
r
s
o
f
t
h
e
s
e
m
o
d
e
l
s
a
r
e
t
u
n
e
d
u
s
i
n
g
B
l
a
ck
w
i
d
o
w
o
p
t
i
m
iz
a
t
i
o
n
a
l
g
o
r
i
th
m
.
R
e
m
a
i
n
i
n
g
p
ar
t
o
f
th
i
s
a
r
t
i
c
l
e
i
s
o
r
g
an
i
z
e
d
in
t
h
e
s
u
b
s
e
q
u
en
t
o
r
d
e
r
:
t
h
e
f
o
l
lo
w
i
n
g
s
e
c
t
i
o
n
p
r
e
s
en
t
s
s
e
v
er
a
l
l
i
ter
a
t
u
r
e
r
e
v
i
e
w
s
b
a
s
e
d
o
n
t
h
e
c
l
a
s
s
i
f
i
c
a
t
i
o
n
o
f
to
m
a
to
l
ea
v
e
s
.
I
n
s
e
c
t
io
n
2
,
p
r
o
p
o
s
ed
m
e
th
o
d
o
lo
g
y
i
s
e
x
p
l
a
i
n
ed
.
S
e
c
t
io
n
3
ev
a
l
u
a
te
s
e
x
p
er
i
m
en
t
a
l
f
in
d
in
g
s
an
d
s
ec
t
i
o
n
4
s
u
m
m
a
r
i
ze
s
t
h
e
w
h
o
l
e
r
e
s
e
ar
c
h
a
s
c
o
n
c
lu
s
i
o
n
.
T
h
e
v
ar
io
u
s
ex
is
tin
g
s
tu
d
ies
ass
o
ciate
d
with
d
is
ea
s
e
d
iag
n
o
s
is
o
f
to
m
ato
leav
es
ar
e
s
u
r
v
e
y
ed
in
th
is
p
ar
t.
C
o
m
p
u
ter
v
is
io
n
m
eth
o
d
o
lo
g
y
u
tili
ze
s
a
ca
p
s
u
le
n
eu
r
al
n
etwo
r
k
n
a
m
ed
C
ap
s
Net
is
d
ev
elo
p
ed
b
y
Ab
o
u
elm
ag
d
et
a
l.
[
1
9
]
to
o
v
er
co
m
e
th
e
r
elativ
e
s
p
a
tial
an
d
o
r
ien
tatio
n
r
elatio
n
s
h
ip
s
in
d
atasets
.
Pre
-
p
r
o
ce
s
s
in
g
an
d
d
ata
au
g
m
en
tatio
n
m
et
h
o
d
s
ar
e
u
tili
ze
d
to
o
v
er
c
o
m
e
th
e
o
v
er
f
itti
n
g
is
s
u
es.
A
s
tan
d
ar
d
d
ataset
n
am
ed
p
lan
t d
is
ea
s
e
i
m
ag
es o
f
to
m
ato
f
r
o
m
Kag
g
le
is
u
s
ed
an
d
co
n
s
is
ts
o
f
alm
o
s
t
7
0
,
8
0
0
im
ag
es.
T
h
e
d
ataset
co
n
tain
s
1
0
lab
ele
d
cla
s
s
es
wh
ich
u
n
d
er
g
o
es
tr
ain
i
n
g
an
d
test
in
g
.
C
lass
if
icatio
n
m
etr
ics
lik
e
a
cc
u
r
ac
y
,
r
ec
all,
F1
-
s
co
r
e
an
d
p
r
ec
is
io
n
ar
e
e
v
alu
ated
f
o
r
p
e
r
f
o
r
m
an
ce
m
ea
s
u
r
em
e
n
t.
C
ap
s
Net
attain
ed
9
6
.
3
9
%
ac
cu
r
ac
y
with
th
e
r
ate
o
f
l
o
s
s
at
0
.
2
2
1
.
T
h
is
ap
p
r
o
ac
h
f
ailed
to
co
n
s
id
er
th
e
u
n
m
an
n
e
d
ae
r
ial
v
eh
icle
(
UAV)
to
g
ath
er
p
lan
t le
af
im
ag
es.
T
o
r
ed
u
ce
th
e
to
m
ato
p
lan
t
lo
s
s
es,
a
tr
an
s
f
er
lear
n
in
g
b
ased
C
NN
o
n
to
m
ato
leaf
d
is
ea
s
e
d
iag
n
o
s
is
is
p
r
esen
ted
b
y
Saee
d
et
a
l.
[
2
0
]
.
T
h
e
to
m
ato
leaf
d
is
ea
s
e
class
if
icatio
n
is
ca
r
r
ied
o
u
t
b
y
two
m
o
d
els
wh
ich
a
r
e
p
r
etr
ain
ed
–
I
n
ce
p
tio
n
-
V3
,
I
n
ce
p
tio
n
-
R
esNet
-
V2
,
an
d
th
ese
m
o
d
els
wer
e
tr
ai
n
ed
u
n
d
er
Pl
an
tVillag
e
d
atasets
an
d
r
ea
l
-
tim
e
ca
p
tu
r
ed
im
ag
e
s
o
f
to
tal
5
,
2
2
5
im
ag
es.
T
h
is
s
tu
d
y
an
aly
s
es
o
n
ly
two
ty
p
es
o
f
leaf
d
is
ea
s
es
o
f
to
m
ato
–
ea
r
ly
b
lig
h
t
o
f
to
m
at
o
an
d
y
ello
w
leaf
cu
r
l
v
ir
u
s
.
T
h
is
ap
p
r
o
a
ch
c
o
n
s
is
ts
o
f
d
at
a
co
llectio
n
,
im
a
g
e
p
r
e
-
p
r
o
ce
s
s
in
g
,
au
g
m
en
tatio
n
,
m
o
d
el
with
tr
ain
in
g
(
8
0
%),
v
alid
atio
n
(
1
0
%)
,
an
d
test
in
g
(
1
0
%)
o
f
d
atasets
.
T
h
en
tr
an
s
f
er
r
ed
to
lear
n
i
n
g
m
o
d
els
f
o
r
class
if
icatio
n
an
d
ev
alu
atio
n
o
f
d
is
ea
s
es.
T
h
e
ap
p
r
o
ac
h
p
r
o
p
o
s
ed
in
th
e
s
tu
d
y
attain
ed
9
9
.
2
2
% a
cc
u
r
ac
y
.
T
o
av
o
id
t
h
e
lo
s
s
in
th
e
ag
r
icu
ltu
r
ally
b
ased
ec
o
n
o
m
y
Ash
o
k
et
a
l
.
[
2
1
]
p
r
esen
ted
a
DL
-
b
ased
im
ag
e
p
r
o
ce
s
s
in
g
tech
n
i
q
u
e
f
o
r
t
o
m
a
to
leaf
d
is
ea
s
e
d
iag
n
o
s
is
.
T
h
e
tech
n
iq
u
e
is
estab
lis
h
ed
b
ased
o
n
s
eg
m
en
tatio
n
,
o
p
en
-
s
o
u
r
ce
alg
o
r
ith
m
s
an
d
c
lu
s
ter
in
g
f
o
r
th
e
ac
c
u
r
ate
p
r
e
d
ictio
n
o
f
to
m
ato
leaf
d
is
ea
s
es.
T
h
e
im
ag
es
ar
e
co
llected
d
ir
ec
tly
f
r
o
m
p
la
n
ts
an
d
t
h
ese
im
ag
es
u
n
d
er
g
o
es
p
r
e
-
p
r
o
ce
s
s
in
g
,
ex
tr
a
ctio
n
o
f
f
ea
tu
r
es,
s
eg
m
en
tatio
n
an
d
ca
teg
o
r
izatio
n
u
s
in
g
C
NN
.
T
h
is
s
tu
d
y
an
aly
ze
s
o
n
ly
Ph
o
m
a
r
o
t,
leaf
m
in
er
an
d
tar
g
et
s
p
o
t
to
m
ato
leaf
d
is
ea
s
es.
T
h
e
p
r
es
en
ted
s
tu
d
y
’
s
p
e
r
f
o
r
m
an
ce
is
ev
alu
ated
u
s
in
g
m
etr
ic
ac
cu
r
a
cy
an
d
a
n
ac
c
u
r
ac
y
o
f
9
8
.
1
2
%
is
ac
h
iev
e
d
.
A
C
N
N
b
ased
to
m
ato
lea
f
d
is
ea
s
e
d
etec
tio
n
is
d
ev
elo
p
ed
b
y
Ag
ar
wal
et
a
l.
[
2
2
]
t
o
in
cr
ea
s
e
th
e
p
r
o
d
u
ctiv
ity
o
f
to
m
ato
es.
T
h
e
im
ag
es
o
f
to
m
ato
d
is
ea
s
es
ar
e
co
llected
f
r
o
m
a
p
u
b
lic
r
ep
o
s
ito
r
y
–
Plan
t
Villag
e
an
d
co
n
s
is
ts
o
f
t
h
e
f
o
llo
win
g
class
es:
h
ea
lth
y
,
b
ac
ter
ial
s
p
o
t,
late
b
lig
h
t,
ea
r
ly
b
lig
h
t,
leaf
m
o
ld
,
two
-
s
p
o
tted
s
p
id
er
m
ite,
s
ep
to
r
ia
leaf
s
p
o
t,
tar
g
et
s
p
o
t
,
y
ello
w
leaf
cu
r
l v
ir
u
s
,
an
d
s
ep
to
r
ia
leaf
s
p
o
t.
T
h
e
C
NN
m
o
d
el
f
o
r
ca
teg
o
r
izatio
n
co
n
s
is
ts
o
f
th
r
ee
lay
er
s
o
f
co
n
v
o
l
u
tio
n
with
o
n
e
m
ax
p
o
o
lin
g
lay
er
an
d
two
f
u
lly
co
n
n
ec
ted
la
y
er
s
.
T
h
e
m
ai
n
li
m
itatio
n
o
f
th
e
s
tu
d
y
is
th
at
a
s
m
all
n
u
m
b
er
o
f
im
ag
es we
r
e
tr
ain
ed
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
1
5
,
No
.
2
,
Ap
r
il 2
0
2
6
:
1
5
4
8
-
1
5
6
5
1550
A
DL
b
ased
ag
r
o
d
ee
p
m
o
b
ile
ap
p
licatio
n
is
d
e
v
elo
p
ed
b
y
Pay
m
o
d
e
et
a
l.
[
2
3
]
to
d
etec
t
t
h
e
to
m
at
o
leaf
d
is
ea
s
es
in
ea
r
ly
s
tag
es.
T
h
e
m
o
d
el
tr
ain
i
n
g
t
o
o
k
p
lace
with
Plan
tVillag
e
d
ataset.
T
h
e
d
ataset
co
n
s
is
ts
o
f
a
to
tal
o
f
1
2
,
0
0
0
im
a
g
es
with
ten
class
e
s
.
T
h
ese
im
ag
es
we
r
e
f
ed
in
to
au
g
m
e
n
tatio
n
p
r
o
c
ess
to
in
cr
ea
s
e
th
e
p
er
f
o
r
m
an
ce
a
n
d
to
r
ed
u
ce
o
v
er
f
itti
n
g
.
T
h
e
C
NN
m
o
d
el
co
n
s
is
ts
o
f
in
p
u
t
lay
er
s
,
h
id
d
en
lay
er
s
,
p
o
o
lin
g
lay
er
s
with
f
u
lly
c
o
n
n
ec
te
d
l
ay
er
s
.
T
h
e
p
e
r
f
o
r
m
an
ce
ev
al
u
atio
n
o
f
th
is
class
if
icatio
n
m
o
d
el
o
b
tain
ed
an
ac
cu
r
ac
y
o
f
9
7
%.
I
n
ad
d
itio
n
to
th
at,
th
e
s
eg
m
en
tatio
n
p
er
f
o
r
m
an
ce
p
ar
am
eter
s
u
s
ed
wer
e
F1
-
s
co
r
e,
p
r
ec
is
io
n
,
r
ec
all,
G
-
m
ea
n
s
,
an
d
s
en
s
itiv
ity
.
A
d
ee
p
co
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
(
DC
NN)
in
teg
r
ated
with
atten
tio
n
m
ec
h
an
is
m
is
d
ev
elo
p
ed
b
y
Z
h
ao
et
a
l.
[
2
4
]
f
o
r
th
e
ac
cu
r
ate
p
r
ed
ictio
n
o
f
v
ar
iety
o
f
t
o
m
ato
d
is
ea
s
es.
Plan
t
Vil
lag
e
d
ataset
co
n
s
is
ts
o
f
4
,
5
8
5
im
ag
es
with
1
0
class
es
u
tili
ze
d
.
Af
ter
th
e
d
ata
a
u
g
m
e
n
tatio
n
p
r
o
ce
s
s
2
2
,
9
2
5
im
a
g
es
wer
e
o
b
tain
ed
an
d
80%
-
2
0
%
tr
ain
-
test
s
p
lit
was
u
s
ed
.
T
h
e
p
er
f
o
r
m
a
n
ce
o
f
t
h
e
m
o
d
el
was
ev
alu
ated
u
s
in
g
s
o
m
e
m
etr
ics
lik
e
ac
cu
r
ac
y
,
r
ec
all,
p
r
ec
is
io
n
,
F1
-
s
co
r
e
an
d
o
b
tain
ed
th
e
h
i
g
h
est
v
alu
es
o
f
9
6
.
8
1
%,
9
6
.
7
7
%,
9
6
.
8
1
%
,
an
d
9
6
.
7
9
% r
esp
ec
tiv
el
y
.
T
o
im
p
r
o
v
e
t
h
e
p
er
f
o
r
m
a
n
ce
o
f
to
m
ato
leaf
d
is
ea
s
e
d
iag
n
o
s
is
an
d
to
r
ed
u
ce
th
e
im
p
a
ct
o
n
cr
o
p
p
r
o
d
u
ctio
n
,
Z
h
o
u
et
a
l
.
[
2
5
]
p
r
esen
ted
a
r
ec
o
n
s
tr
u
cted
d
ee
p
r
esid
u
al
d
en
s
e
n
etwo
r
k
(
R
R
DN)
.
T
h
e
d
ataset
o
f
AI
C
HAL
L
E
NGE
R
2
0
1
8
f
o
r
to
m
ato
leaf
d
is
ea
s
e
is
tak
en
i
n
th
is
s
tu
d
y
a
n
d
with
1
3
,
1
8
5
im
ag
es
ca
teg
o
r
ize
d
in
to
9
class
es.
T
h
e
d
ataset
was
d
iv
id
ed
in
to
tr
ai
n
in
g
,
test
,
an
d
v
alid
atio
n
with
a
r
atio
o
f
6
:2
:2
.
T
h
e
R
R
DN
m
o
d
el
im
p
r
o
v
es
th
e
d
en
o
is
in
g
is
s
u
es
in
th
e
im
a
g
es.
T
h
is
p
a
p
er
h
i
g
h
lig
h
ts
th
e
tr
ain
in
g
a
n
d
test
in
g
ac
cu
r
ac
y
o
f
R
R
D
N
m
o
d
els
with
ad
v
an
ce
d
m
eth
o
d
s
.
On
c
o
m
p
ar
i
n
g
wi
th
th
o
s
e
tr
ad
itio
n
al
ap
p
r
o
ac
h
es,
R
R
DN
attain
ed
h
ig
h
est ac
cu
r
ac
y
o
f
9
5
% in
to
m
ato
leaf
d
is
ea
s
e
id
en
tific
atio
n
.
T
o
m
ato
leaf
d
is
ea
s
e
d
etec
tio
n
u
s
in
g
th
e
th
r
ee
-
m
ac
h
i
n
e
lear
n
in
g
(
ML
)
m
eth
o
d
s
lik
e
C
NN,
k
-
n
ea
r
est
n
eig
h
b
o
r
(
KNN)
,
an
d
s
u
p
p
o
r
t
v
ec
to
r
m
ac
h
in
e
(
SVM)
was d
ev
elo
p
ed
b
y
L
ek
h
a
et
a
l
.
[
2
6
]
.
T
h
e
im
ag
es u
s
ed
in
th
is
p
ap
er
wer
e
tak
en
f
r
o
m
Pl
an
tVillag
e
im
ag
e
d
ataset
with
a
to
tal
o
f
1
6
,
1
0
2
im
ag
es
with
all
1
0
class
es.
8
0
%
o
f
th
e
im
ag
es
u
n
d
er
wen
t
tr
ain
in
g
an
d
2
0
%
o
f
im
ag
es
wer
e
f
ed
in
to
v
alid
atio
n
.
T
r
ain
i
n
g
a
n
d
test
in
g
ac
cu
r
ac
y
ar
e
ev
alu
ated
with
4
0
ep
o
ch
s
o
n
C
NN
m
o
d
el.
E
v
alu
atin
g
th
e
ac
cu
r
ac
y
o
f
th
ese
ML
m
o
d
el
s
,
C
N
N
attain
ed
th
e
h
ig
h
est ac
cu
r
ac
y
o
f
7
9
.
1
4
% c
o
m
p
ar
ed
with
SVM
(
6
8
.
2
2
%)
an
d
KNN
(
7
4
.
5
6
%).
T
ar
g
et
id
en
tific
atio
n
alg
o
r
it
h
m
im
p
r
o
v
e
d
th
e
d
etec
tio
n
o
f
to
m
ato
leaf
d
is
ea
s
e,
an
d
h
as
s
o
m
e
lim
itatio
n
s
s
u
ch
as
d
iv
er
s
ity
o
f
co
m
p
lex
b
ac
k
g
r
o
u
n
d
s
,
g
en
e
r
aliza
tio
n
ab
ilit
y
an
d
r
ed
u
ce
d
a
cc
u
r
ac
y
.
Hen
ce
to
o
v
er
co
m
e
th
ese
s
h
o
r
tco
m
in
g
s
,
W
an
g
et
a
l
.
[
2
7
]
p
r
esen
te
d
a
tar
g
et
d
etec
tio
n
m
o
d
el
b
ased
o
n
im
p
r
o
v
e
d
YOL
O
-
v
5
.
T
h
e
im
ag
es
u
s
ed
i
n
th
ese
ex
p
er
im
en
ts
wer
e
co
ll
ec
ted
f
r
o
m
in
ter
n
et
s
o
u
r
ce
s
,
i
m
ag
e
s
y
n
th
esis
an
d
Kag
g
le,
co
n
tain
in
g
a
to
tal
o
f
4
,
5
7
8
im
a
g
es.
T
h
e
d
ataset
co
n
s
is
ts
o
f
f
o
u
r
class
es
lik
e
leaf
m
o
ld
,
ea
r
ly
b
lig
h
t,
h
ea
lth
y
leav
es
an
d
late
b
lig
h
t.
T
ar
g
et
d
etec
tio
n
with
g
lo
b
al
atten
tio
n
m
ec
h
an
is
m
(
T
D
GA)
was
d
ev
elo
p
ed
b
ased
o
n
YOL
O
-
v
5
.
T
h
is
T
D
GA
m
o
d
el
attain
ed
9
1
.
4
0
% o
f
ac
cu
r
ac
y
wh
ich
was m
u
ch
b
et
ter
th
an
th
e
o
r
ig
in
al
YOL
O
-
v
5
m
o
d
el.
T
o
id
en
tif
y
th
e
leaf
d
is
ea
s
es
ac
cu
r
ately
,
ea
r
ly
C
NN
b
ased
Alex
Net
an
d
VGG
-
1
6
a
r
e
u
tili
ze
d
b
y
Sh
an
th
i
et
a
l
.
[
2
8
]
.
T
h
e
Plan
tVillag
e
d
ataset
co
llected
f
r
o
m
Kag
g
le
co
n
s
is
ts
o
f
th
e
f
o
llo
win
g
class
es:
h
ea
lth
y
leav
es,
B
ac
ter
ial
s
p
o
t,
ea
r
ly
b
lig
h
t,
y
ello
w
leaf
cu
r
l,
leaf
m
o
ld
,
an
d
s
ep
to
r
ia
leaf
s
p
o
t
leav
es.
T
h
e
d
ataset
im
ag
es
u
n
d
er
g
o
tr
ain
,
test
an
d
v
alid
atio
n
p
h
ase
in
th
e
r
atio
n
o
f
8
:1
:1
.
T
h
e
im
ag
es
ar
e
p
r
e
-
p
r
o
ce
s
s
ed
,
n
o
r
m
alize
d
a
n
d
a
u
g
m
e
n
ted
f
o
r
b
etter
p
e
r
f
o
r
m
an
ce
.
T
h
is
s
tu
d
y
h
ig
h
lig
h
ted
th
e
ac
cu
r
ac
y
p
er
f
o
r
m
a
n
ce
f
o
r
ea
ch
class
an
d
th
e
m
o
d
el
attain
ed
a
n
ac
cu
r
ac
y
o
f
9
1
.
2
%.
A
DL
-
b
ased
C
NN
m
o
d
el
was
d
ev
elo
p
ed
b
y
I
b
a
ñ
ez
an
d
Mu
ñ
o
z
[
2
9
]
to
m
o
n
ito
r
an
d
class
if
y
th
e
to
m
ato
leaf
d
is
ea
s
es
u
s
in
g
a
p
u
b
licly
av
ailab
le
d
ataset
o
f
1
1
,
0
0
0
im
ag
es.
T
h
e
d
ataset
is
ac
co
m
p
lis
h
ed
wit
h
2
,
5
0
0
im
a
g
es
g
ath
er
ed
f
r
o
m
t
h
e
Me
x
ican
f
ield
s
.
Fo
r
f
ea
tu
r
e
s
elec
tio
n
,
th
e
C
NN
m
o
d
el
co
n
s
is
ts
o
f
k
er
n
el
s
ize
o
f
3
×
3
with
f
o
u
r
c
o
n
v
o
lu
tio
n
al
lay
er
s
.
Af
ter
t
h
at,
th
e
cl
ass
if
icatio
n
m
o
d
el
co
m
p
r
is
ed
o
f
g
lo
b
al
a
v
er
ag
e
p
o
o
lin
g
lay
e
r
,
So
f
tMa
x
ac
tiv
atio
n
f
o
r
d
e
n
s
e
lay
er
.
T
o
an
a
ly
ze
th
e
p
er
f
o
r
m
an
ce
o
f
th
e
m
o
d
el
class
if
icatio
n
m
etr
ics is
o
b
s
er
v
ed
.
T
h
e
h
ig
h
e
s
t a
cc
u
r
ac
y
o
f
9
9
.
6
% is
attain
ed
b
y
c
u
s
to
m
ized
C
NN
m
o
d
el.
A
DL
b
ased
Den
s
eNe
t
ar
ch
itectu
r
e
was
im
p
lem
en
ted
b
y
Y
u
lita
et
a
l.
[
3
0
]
to
d
etec
t
th
e
to
m
ato
leaf
d
is
ea
s
es
an
d
it
was
im
p
lem
en
ted
in
a
m
o
b
ile
ap
p
licatio
n
.
T
h
e
im
ag
e
d
ata
f
o
r
to
m
ato
leaf
d
is
ea
s
e
was
d
o
wn
lo
ad
e
d
f
r
o
m
Ka
g
g
le
w
ith
1
0
,
0
0
0
im
ag
es.
T
h
e
im
ag
e
d
ata
was
f
e
d
to
p
r
e
-
p
r
o
ce
s
s
in
g
an
d
d
ata
au
g
m
en
tatio
n
was
u
s
ed
f
o
r
o
b
tain
in
g
q
u
ality
im
ag
e
a
n
d
to
a
v
o
id
o
v
e
r
f
itti
n
g
.
T
h
en
class
if
icatio
n
was
ca
r
r
ied
o
u
t
b
y
Den
s
eNe
t
ar
ch
itectu
r
e
with
1
0
d
en
s
e
u
n
its
.
Af
ter
t
h
at
th
e
m
o
b
ile
ap
p
licatio
n
’
s
p
r
im
ar
y
g
o
al
was
to
d
etec
t
th
e
to
m
ato
lea
f
d
is
ea
s
es
f
o
r
I
n
d
o
n
esian
f
ar
m
er
s
.
T
h
e
e
f
f
ec
tiv
en
ess
o
f
th
e
m
o
d
el
was
ass
es
s
ed
b
ased
o
n
class
if
icatio
n
m
etr
ic.
A
DC
N
N
ap
p
r
o
ac
h
was
estab
lis
h
ed
b
y
L
in
g
wal
et
a
l.
[
3
1
]
to
d
etec
t
to
m
ato
p
lan
t
leaf
d
is
ea
s
es.
T
h
e
d
ata
was
co
llected
f
r
o
m
th
e
p
u
b
lic
r
ep
o
s
ito
r
y
,
n
am
ely
,
Plan
tVillag
e
d
ataset.
T
h
e
im
ag
e
d
ata
was
d
iv
id
ed
in
to
tr
ain
in
g
,
v
alid
atio
n
an
d
test
in
g
in
a
r
atio
o
f
7
0
:1
0
:2
0
.
T
h
e
e
f
f
icien
cy
o
f
th
e
Den
s
eNe
t,
I
n
c
ep
tio
n
-
V3
,
R
esNet,
VGG
-
16
,
an
d
b
ase
C
NN
was
ev
alu
ated
f
o
r
m
etr
ic
p
er
f
o
r
m
a
n
ce
s
.
T
h
e
v
alid
atio
n
an
d
tr
ai
n
in
g
ac
cu
r
ac
y
o
f
all
th
ese
m
o
d
els
wer
e
e
v
alu
ated
f
o
r
p
er
f
o
r
m
an
ce
an
aly
s
is
.
Am
o
n
g
all
t
h
ese
m
eth
o
d
s
VGG
-
1
6
p
er
f
o
r
m
e
d
b
etter
with
9
0
.
8
4
% a
cc
u
r
ac
y
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
A
n
o
ve
l a
p
p
r
o
a
c
h
to
d
etec
t to
ma
to
lea
f d
is
ea
s
e
u
s
in
g
visi
o
n
tr
a
n
s
fo
r
mer (
S
a
n
jeela
S
a
g
a
r
)
1551
T
o
m
ato
leaf
d
is
ea
s
e
d
etec
tio
n
f
o
r
ag
r
o
b
ased
in
d
u
s
tr
ies
was
d
ev
elo
p
ed
b
y
R
o
y
et
a
l.
[
3
2
]
b
ased
o
n
n
o
v
el
PC
A
Dee
p
Net.
T
h
e
to
m
ato
im
ag
e
d
ata
was
g
ath
er
ed
f
r
o
m
p
u
b
lic
r
ep
o
s
ito
r
y
d
a
taset
co
n
s
is
tin
g
o
f
1
8
,
1
2
8
im
a
g
es
o
f
to
m
ato
leav
es.
T
h
e
im
a
g
e
d
ata
was
s
p
lit
in
to
7
:
3
r
atio
f
o
r
tr
ain
in
g
a
n
d
v
alid
atio
n
.
T
h
is
d
ataset
u
n
d
er
we
n
t
d
ata
en
g
i
n
e
er
in
g
,
f
ea
tu
r
e
s
elec
tio
n
an
d
cla
s
s
if
icatio
n
.
T
h
e
PC
A
Dee
p
Net
m
o
d
el
attain
ed
an
ac
cu
r
ac
y
o
f
9
9
.
6
%,
r
ec
all
o
f
9
8
.
4
9
%,
p
r
ec
is
io
n
o
f
9
8
.
5
%,
an
d
F1
-
m
ea
s
u
r
e
o
f
9
8
.
5
%.
T
o
o
v
er
co
m
e
th
e
co
m
p
lex
b
a
ck
g
r
o
u
n
d
o
f
th
e
im
ag
e,
Pen
g
et
a
l.
[
3
3
]
estab
lis
h
ed
a
Den
s
e
I
n
ce
p
tio
n
Mo
b
ileNet
-
V2
p
ar
allel
co
n
v
o
lu
tio
n
al
b
lo
c
k
atten
tio
n
n
etwo
r
k
f
o
r
th
e
d
iag
n
o
s
is
o
f
to
m
ato
leaf
d
is
ea
s
e.
T
h
e
s
tu
d
y
u
tili
ze
d
t
h
e
to
m
ato
lea
f
d
is
ea
s
e
d
ataset
o
f
1
,
2
5
6
im
ag
es
with
f
iv
e
leaf
d
is
ea
s
es
an
d
th
e
n
u
m
b
er
o
f
im
ag
es we
r
e
ex
p
an
d
ed
to
8
,
1
9
0
im
ag
es u
s
in
g
im
ag
e
e
n
h
an
ce
m
en
t.
C
lass
if
icatio
n
m
etr
ic
was e
v
alu
ated
f
o
r
t
h
e
an
aly
s
is
o
f
p
er
f
o
r
m
a
n
ce
.
T
h
e
ac
cu
r
ac
y
attain
ed
b
y
t
h
is
m
o
d
el
was
9
4
.
4
4
%.
Data
a
u
g
m
en
tatio
n
was
n
o
t
m
an
ag
ed
p
r
o
p
er
ly
,
w
h
ich
is
th
e
lim
itatio
n
o
f
th
is
m
o
d
el.
A
to
m
ato
leaf
d
is
ea
s
e
d
etec
tio
n
in
ea
r
ly
s
tag
es
is
q
u
ite
n
ec
ess
ar
y
to
d
etec
t
th
e
to
m
ato
leaf
d
is
ea
s
es.
Hen
ce
Gaa
s
h
an
i
et
a
l.
[
3
4
]
d
ev
elo
p
ed
a
to
m
at
o
leaf
d
is
ea
s
e
class
if
icatio
n
m
o
d
el
b
ased
o
n
tr
an
s
f
e
r
lear
n
i
n
g
an
d
f
ea
tu
r
e
in
teg
r
atio
n
.
T
h
e
s
t
u
d
y
u
s
ed
a
d
ataset
wh
ich
was
g
ath
er
ed
f
r
o
m
a
p
u
b
licly
a
v
ailab
le
d
ataset
n
am
ed
Plan
tVillag
e
an
d
u
s
ed
1
,
1
5
2
im
ag
es
f
o
r
f
u
r
t
h
er
p
r
o
ce
s
s
es.
T
h
is
m
o
d
el
was
ev
alu
ated
b
y
th
r
ee
p
o
p
u
lar
tr
ad
itio
n
al
ML
class
if
ier
s
:
SV
M
,
r
an
d
o
m
f
o
r
est,
an
d
m
u
ltin
o
m
ial
lo
g
is
tic
r
eg
r
ess
io
n
,
an
d
m
u
ltin
o
m
ial
lo
g
is
tic
r
eg
r
ess
io
n
o
u
t
p
er
f
o
r
m
ed
with
an
av
e
r
ag
e
ac
c
u
r
ac
y
o
f
9
7
%
.
T
h
e
m
aj
o
r
is
s
u
e
is
th
at
th
e
h
ig
h
ac
cu
r
ac
y
i
n
a
p
ar
ticu
lar
d
ata
s
et
m
ay
lead
to
o
v
er
f
itti
n
g
.
T
h
e
to
m
ato
d
is
ea
s
e
class
if
icatio
n
is
ca
r
r
ied
o
u
t
b
y
f
in
e
-
tu
n
ed
r
esid
u
al
n
eu
r
al
n
etwo
r
k
m
o
d
els.
Kan
d
a
et
a
l
.
[
3
5
]
d
ev
elo
p
ed
an
in
tellig
en
t
b
ased
DL
m
eth
o
d
f
o
r
r
ec
o
g
n
izin
g
th
e
c
o
m
m
o
n
d
is
ea
s
es
in
to
m
ato
es.
T
h
e
t
o
m
ato
leaf
d
is
ea
s
e
d
ataset
co
n
s
is
ted
o
f
1
8
,
1
6
0
im
ag
es
a
n
d
d
iv
id
ed
in
to
t
r
a
in
in
g
an
d
v
alid
atio
n
p
h
ase
an
d
attain
ed
b
est
s
o
lu
tio
n
in
th
e
r
atio
o
f
8
:2
.
T
h
e
cla
s
s
if
icatio
n
ap
p
r
o
ac
h
was
ca
r
r
ied
o
u
t
b
y
a
tr
an
s
f
e
r
lear
n
in
g
m
o
d
el
f
o
llo
wed
b
y
a
C
NN
with
s
ev
er
al
lay
er
s
.
T
h
is
s
tu
d
y
co
n
d
u
cted
an
ex
p
er
im
en
tal
ev
alu
atio
n
b
ased
o
n
v
ar
i
o
u
s
tr
ain
an
d
v
alid
atio
n
r
atio
,
d
i
f
f
er
en
t
b
a
tch
s
izes.
C
o
n
f
u
s
io
n
m
atr
ix
was
ev
alu
ated
f
o
r
p
er
f
o
r
m
an
ce
v
alid
ati
o
n
.
An
e
x
p
lai
n
ab
le
d
r
i
v
e
n
D
L
f
r
am
e
wo
r
k
f
o
r
t
o
m
at
o
le
af
d
is
ea
s
e
c
lass
i
f
i
ca
t
io
n
was
p
e
r
f
o
r
m
e
d
wit
h
tr
a
n
s
f
e
r
le
ar
n
i
n
g
b
ase
d
E
f
f
i
ci
en
t
Net
-
B
5
m
o
d
el
w
h
i
c
h
was
d
e
m
o
n
s
tr
ate
d
b
y
B
h
a
n
d
a
r
i
e
t
a
l
.
[
3
6
]
.
T
h
e
s
tu
d
y
u
s
e
d
t
h
e
to
m
a
to
lea
f
d
is
e
ase
d
at
ase
t
f
r
o
m
K
a
g
g
le
.
T
h
e
d
a
t
aset
i
m
a
g
es
we
r
e
d
i
v
i
d
e
d
i
n
t
o
9
0
:1
0
f
o
r
tr
ai
n
i
n
g
-
test
i
n
g
s
p
li
t
a
n
d
f
u
r
t
h
e
r
m
o
r
e
,
t
h
e
1
0
%
im
a
g
es f
r
o
m
tr
ai
n
i
n
g
wer
e
als
o
c
o
n
s
i
d
e
r
e
d
f
o
r
v
a
li
d
ati
o
n
p
h
ase
.
I
n
iti
all
y
th
e
d
a
tase
t
im
a
g
es
we
r
e
au
g
m
e
n
te
d
t
o
a
v
o
i
d
o
v
e
r
f
itt
in
g
,
t
h
e
n
t
h
e
i
m
a
g
es
w
er
e
f
ed
i
n
t
o
E
f
f
ic
ie
n
tN
et
-
B
5
t
o
in
c
r
ea
s
e
e
f
f
ici
en
c
y
a
n
d
a
cc
u
r
a
cy
.
T
h
e
n
f
o
r
s
t
atis
t
ic
al
an
aly
s
is
e
x
p
l
ai
n
a
b
le
a
r
ti
f
i
cia
l
i
n
te
lli
g
e
n
ce
(
XA
I
)
tec
h
n
i
q
u
e
was
u
tili
ze
d
.
A
ll
cl
a
s
s
if
i
ca
t
io
n
m
et
r
ics
we
r
e
ev
al
u
a
ted
f
o
r
t
h
e
p
e
r
f
o
r
m
an
ce
e
v
a
lu
a
tio
n
.
An
au
to
m
atic
s
eg
m
en
tatio
n
(
AS)
b
ased
to
m
ato
leaf
d
is
ea
s
e
class
if
icatio
n
was
p
er
f
o
r
m
e
d
[
3
7
]
to
av
o
id
o
v
er
f
itti
n
g
is
s
u
es
in
t
h
e
DL
ap
p
r
o
ac
h
es.
T
h
e
to
m
at
o
Plan
tVillag
e
d
ataset
with
1
8
,
1
6
1
im
ag
es
wer
e
u
tili
ze
d
to
au
g
m
en
t th
e
im
a
g
e
s
.
T
h
e
m
o
d
if
ied
UNe
t m
o
d
el
was u
tili
ze
d
to
s
eg
m
en
t th
e
in
p
u
t im
ag
es.
T
h
en
t
h
e
o
p
tim
izatio
n
with
h
y
p
e
r
p
ar
a
m
eter
s
b
ased
o
n
ar
tific
ial
r
ab
b
it
alg
o
r
ith
m
(
HPOAR
A)
is
d
em
o
n
s
tr
ated
to
class
if
y
to
m
ato
leaf
d
is
ea
s
es
as
h
ea
lth
y
o
r
u
n
h
ea
lth
y
.
C
lass
if
icatio
n
m
etr
ic
was
v
alid
ated
f
o
r
th
e
p
er
f
o
r
m
an
ce
ev
alu
atio
n
.
Var
io
u
s
m
eta
-
h
eu
r
is
tic
ap
p
r
o
ac
h
es we
r
e
an
aly
ze
d
to
in
cr
ea
s
e
class
if
icatio
n
ac
c
u
r
ac
y
.
A
co
m
b
in
atio
n
o
f
C
NN,
co
n
v
o
lu
tio
n
al
b
lo
ck
atten
tio
n
m
o
d
u
le
(
C
B
AM
)
an
d
S
VM
wer
e
u
tili
ze
d
to
class
if
y
an
d
id
en
tify
th
e
t
o
m
a
to
leaf
d
is
ea
s
e
b
y
Altalak
et
a
l.
[
3
8
]
.
T
o
m
ato
leaf
d
is
ea
s
e
d
ataset
im
ag
es
wer
e
p
r
e
-
p
r
o
ce
s
s
ed
an
d
au
g
m
en
ted
f
o
r
f
u
r
t
h
er
class
if
icatio
n
p
r
o
ce
s
s
.
I
m
ag
es
wer
e
zo
o
m
ed
,
f
lip
p
ed
h
o
r
izo
n
tally
a
n
d
v
e
r
t
i
c
a
l
l
y
,
r
o
t
a
t
e
d
,
s
h
i
f
t
e
d
i
n
w
i
d
t
h
a
n
d
h
e
i
g
h
t
a
n
d
t
h
e
n
u
s
e
d
f
o
r
t
h
e
a
u
g
m
e
n
t
a
t
i
o
n
.
C
N
N
b
a
s
e
d
R
e
s
N
e
t
-
50
was
u
tili
ze
d
f
o
r
f
ea
tu
r
e
ex
tr
a
ctio
n
.
Af
ter
th
at
t
h
e
b
ase
R
esNet
-
5
0
m
o
d
el
was
f
e
d
in
to
C
B
AM
an
d
SVM
class
if
icatio
n
m
o
d
el.
T
h
e
m
o
d
el
was tr
ain
ed
with
5
0
e
p
o
ch
s
an
d
attain
ed
9
7
.
2
% o
f
class
if
icatio
n
ac
cu
r
ac
y
.
T
o
m
ato
leaf
d
is
ea
s
e
id
en
tific
atio
n
p
lay
s
a
v
ital
r
o
le
in
a
g
r
icu
ltu
r
al
s
ec
to
r
.
So
,
it
is
n
e
ce
s
s
ar
y
to
id
en
tify
th
e
d
is
ea
s
es
wh
ich
af
f
ec
ts
th
e
p
r
o
d
u
ctio
n
o
f
t
o
m
ato
e
s
.
Sev
er
al
r
esear
ch
er
s
h
a
v
e
d
e
v
elo
p
ed
n
u
m
e
r
o
u
s
m
eth
o
d
s
to
id
en
tify
a
n
d
class
if
y
d
is
ea
s
es.
On
an
aly
zin
g
th
e
v
ar
io
u
s
s
tu
d
ies,
in
s
u
f
f
icien
t
d
ata
s
eg
m
en
tatio
n
,
lim
ited
n
u
m
b
e
r
o
f
tr
ain
in
g
a
n
d
test
in
g
ep
o
ch
s
,
ab
s
en
ce
o
f
m
eta
-
h
eu
r
is
tic
ap
p
r
o
ac
h
es,
h
ig
h
co
m
p
u
tatio
n
al
co
m
p
lex
ity
,
lim
ited
f
ea
tu
r
e
ex
tr
ac
tio
n
an
d
in
ef
f
icien
cy
o
f
p
r
e
-
p
r
o
ce
s
s
in
g
m
eth
o
d
s
ar
e
c
o
n
s
id
er
ed
as
th
e
m
ain
lim
itatio
n
s
o
f
ex
is
tin
g
ap
p
r
o
a
ch
es.
I
n
v
iew
o
f
th
ese
lim
itatio
n
s
th
e
s
u
g
g
ested
m
eth
o
d
o
lo
g
y
p
r
o
p
o
s
ed
a
n
o
v
el
to
m
ato
leaf
d
is
ea
s
e
class
if
icati
o
n
f
r
a
m
ewo
r
k
a
n
d
is
ex
p
lai
n
e
d
in
d
etail
in
th
e
f
o
llo
win
g
s
ec
tio
n
.
T
o
d
ea
l
with
th
ese
lim
itatio
n
s
th
e
p
r
o
p
o
s
ed
s
tu
d
y
in
tr
o
d
u
ce
s
a
n
o
v
el
class
if
icatio
n
ap
p
r
o
a
ch
f
o
r
leaf
d
is
ea
s
e
d
etec
tio
n
o
f
to
m
ato
e
s
b
ased
o
n
A
-
GVT
m
o
d
el.
T
h
e
f
o
llo
win
g
is
an
o
v
er
v
iew
ab
o
u
t
th
e
n
o
v
e
l
co
n
tr
ib
u
tio
n
s
o
f
th
e
p
r
o
p
o
s
ed
r
esear
ch
.
First,
t
o
ex
tr
ac
t
th
e
f
ea
tu
r
es
o
f
th
e
to
m
ato
leaf
d
is
ea
s
es
a
B
i
-
D
L
STM
n
etwo
r
k
is
em
p
lo
y
e
d
.
Seco
n
d
,
t
o
in
c
r
ea
s
e
th
e
ac
c
u
r
ac
y
o
f
class
if
icatio
n
an
d
to
s
elec
t
th
e
o
p
tim
al
f
ea
tu
r
es
a
C
S
W
O
is
u
tili
ze
d
.
T
h
ir
d
,
t
o
ac
cu
r
ately
class
if
y
th
e
to
m
at
o
leaf
d
is
ea
s
es,
an
atten
tio
n
b
ased
g
ated
v
is
io
n
tr
an
s
f
er
m
o
d
el
is
d
e
p
lo
y
ed
.
F
o
u
r
th
,
t
o
r
ed
u
ce
th
e
to
m
ato
d
is
ea
s
e
clas
s
if
icatio
n
co
m
p
lex
i
ty
,
p
ar
a
m
eter
s
ar
e
tu
n
ed
u
s
in
g
th
e
b
lack
wid
o
w
o
p
tim
izatio
n
(
B
W
O)
alg
o
r
ith
m
.
T
o
m
ato
leaf
d
is
ea
s
e
id
en
tific
atio
n
p
lay
s
a
v
ital
r
o
le
in
ag
r
icu
ltu
r
al
s
ec
to
r
.
I
t
is
n
ec
ess
ar
y
t
o
id
en
tify
th
e
d
is
ea
s
es
wh
ich
af
f
ec
ts
th
e
p
r
o
d
u
ctio
n
o
f
to
m
ato
es.
So
,
s
ev
er
al
r
esear
ch
er
s
h
av
e
d
ev
elo
p
ed
n
u
m
er
o
u
s
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
1
5
,
No
.
2
,
Ap
r
il 2
0
2
6
:
1
5
4
8
-
1
5
6
5
1552
r
esear
ch
es
in
th
e
n
am
e
o
f
to
m
ato
leaf
d
is
ea
s
e
d
etec
tio
n
,
id
en
tific
atio
n
an
d
class
if
icat
io
n
.
On
an
aly
zin
g
v
ar
io
u
s
p
r
e
v
io
u
s
s
tu
d
ies,
in
s
u
f
f
icien
t
d
ata
s
eg
m
en
tatio
n
,
lim
ited
n
u
m
b
er
o
f
tr
ain
in
g
a
n
d
test
in
g
ep
o
c
h
s
,
ab
s
en
ce
o
f
m
eta
-
h
eu
r
is
tic
a
p
p
r
o
ac
h
es,
h
ig
h
c
o
m
p
u
tatio
n
al
co
m
p
lex
it
y
,
lim
ited
f
ea
tu
r
e
ex
tr
ac
tio
n
,
a
n
d
in
ef
f
icien
cy
o
f
p
r
e
-
p
r
o
ce
s
s
in
g
m
eth
o
d
s
ar
e
c
o
n
s
id
er
ed
as
th
e
m
ain
lim
itatio
n
s
o
f
ex
is
tin
g
a
p
p
r
o
ac
h
es.
I
n
v
iew
o
f
th
ese
lim
itatio
n
s
th
e
s
u
g
g
ested
m
eth
o
d
o
lo
g
y
p
r
o
p
o
s
ed
a
n
o
v
el
to
m
ato
leaf
d
i
s
ea
s
e
class
if
ica
tio
n
f
r
am
ewo
r
k
is
d
ev
elo
p
e
d
an
d
is
ex
p
lain
ed
i
n
d
etail
in
th
e
f
o
llo
win
g
s
ec
tio
n
.
2.
M
E
T
H
O
D
T
o
m
ato
es
ar
e
u
n
iv
er
s
al
cr
o
p
with
h
ig
h
n
u
tr
itio
n
al
v
alu
es;
h
o
wev
er
,
it
is
p
r
o
n
e
to
v
ar
io
u
s
d
is
ea
s
e
s
,
an
d
p
r
ec
is
e
id
en
tific
atio
n
o
f
t
h
o
s
e
d
is
ea
s
es
ar
e
s
til
l
a
ch
alle
n
g
in
g
task
.
Hen
ce
,
th
is
p
r
esen
ted
r
esear
ch
u
tili
ze
d
a
n
o
v
el
an
d
e
f
f
ec
tiv
e
to
m
ato
l
ea
f
d
is
ea
s
e
class
if
icatio
n
u
s
in
g
DL
a
r
ch
itectu
r
e.
T
h
is
r
esear
ch
f
in
d
s
th
e
ex
p
e
r
ts
o
f
ag
r
icu
ltu
r
e
s
ec
to
r
f
o
r
an
al
y
zin
g
th
e
t
o
m
ato
lea
f
d
is
ea
s
es.
Ov
er
all
wo
r
k
f
lo
w
o
f
th
e
s
u
g
g
ested
to
m
ato
lea
f
d
is
ea
s
e
class
if
icatio
n
f
r
am
ewo
r
k
is
p
ictu
r
ed
i
n
Fig
u
r
e
1
.
T
h
is
r
esear
ch
b
eg
in
s
with
d
at
a
ac
q
u
is
itio
n
,
wh
ich
is
c
o
llected
f
r
o
m
Plan
tVillag
e
–
a
p
u
b
lic
r
ep
o
s
ito
r
y
an
d
p
r
e
-
p
r
o
ce
s
s
in
g
o
f
im
ag
es
to
im
p
r
o
v
e
th
e
q
u
ality
o
f
d
ata.
T
h
e
p
r
e
-
p
r
o
ce
s
s
ed
im
ag
es
ar
e
s
eg
m
en
ted
u
s
in
g
r
esid
u
al
atten
tio
n
UNe
t++
to
id
en
tify
v
a
r
io
u
s
p
a
r
ts
o
f
d
is
ea
s
ed
p
o
r
tio
n
s
.
Seco
n
d
ly
,
th
e
f
e
atu
r
es
ar
e
s
elec
ted
an
d
ex
tr
ac
ted
f
r
o
m
th
e
im
ag
es
th
at
ar
e
s
eg
m
en
ted
,
a
n
d
a
c
h
a
o
tic
s
p
id
er
wasp
o
p
tim
izer
alg
o
r
ith
m
(
C
SOA)
is
co
n
s
id
er
ed
to
s
elec
t
th
e
o
p
ti
m
al
f
ea
tu
r
es.
T
h
en
th
e
class
if
icatio
n
o
f
d
is
ea
s
es
o
f
to
m
ato
leav
es
is
ca
r
r
ied
o
u
t
u
s
in
g
A
-
GVT
to
ca
teg
o
r
ize
t
h
e
ty
p
es
o
f
d
is
ea
s
es.
I
n
ad
d
i
tio
n
to
th
at,
o
p
tim
izatio
n
is
u
s
ed
to
r
ef
in
e
th
e
p
ar
am
eter
s
f
o
r
en
h
a
n
ce
d
p
e
r
f
o
r
m
an
ce
o
f
class
if
icatio
n
m
o
d
e
l.
I
n
p
u
t i
mage
P
r
e
-
p
r
oc
e
s
s
i
n
g
A
da
pt
i
ve
rol
l
i
ng
gui
da
nc
e
w
i
e
ne
r
fi
l
t
e
r
S
e
gme
n
tati
on
Re
s
i
dua
l
a
t
t
e
nt
i
on
ba
s
e
d U
-
N
e
t
++
F
e
atu
r
e
e
xtr
ac
ti
on
C
l
as
s
i
fi
c
ati
on
Bl
a
c
k w
i
dow
opt
i
m
i
z
a
t
i
on
A
t
t
e
nt
i
on ba
s
e
d
ga
t
e
d vi
s
i
on
t
ra
ns
form
e
r m
ode
l
P
ar
ame
te
r
tu
n
i
n
g
M
os
a
i
c
S
e
pt
ori
a
l
l
e
a
f s
pot
T
a
rge
t
s
pot
Y
e
l
l
ow
l
e
a
f c
url
D
i
l
a
t
e
d
c
onvol
ut
i
ona
l
Bi
-
L
S
T
M
Ch
a
ot
i
c
s
pi
de
r
w
a
s
p opt
i
m
i
z
a
t
i
on
F
e
atu
r
e
s
e
l
e
c
ti
on
Fig
u
r
e
1
.
Ov
e
r
all
wo
r
k
f
l
o
w
o
f
p
r
o
p
o
s
ed
to
m
ato
leaf
d
is
ea
s
e
class
if
icatio
n
m
eth
o
d
2
.
1
.
P
re
-
pro
ce
s
s
ing
T
h
e
p
r
e
-
p
r
o
ce
s
s
in
g
was
p
er
f
o
r
m
ed
to
r
e
m
o
v
e
t
h
e
n
o
is
e
an
d
to
r
ec
o
v
e
r
th
e
ed
g
es
b
y
a
d
ap
t
iv
e
r
o
llin
g
g
u
id
an
ce
wien
er
f
ilter
in
g
tec
h
n
iq
u
e
(
AR
GW
F)
[
3
9
]
.
I
n
r
o
llin
g
g
u
id
an
ce
wien
er
f
ilter
,
th
e
g
u
id
e
d
f
ilter
in
g
is
u
s
ed
f
o
r
ed
g
e
r
ec
o
v
er
y
p
r
o
ce
s
s
.
T
h
e
g
u
id
e
d
f
ilter
im
p
r
o
v
e
s
th
e
r
em
o
v
al
o
f
s
m
all
s
tr
u
ctu
r
es,
en
h
a
n
ce
th
e
s
tr
u
ctu
r
e
an
d
ed
g
es
o
f
th
e
in
p
u
t
im
ag
es.
Fo
r
s
m
all
s
tr
u
ctu
r
al
r
em
o
v
al
th
e
wien
er
f
ilter
in
g
is
u
s
ed
to
s
m
o
o
th
th
e
in
p
u
t im
a
g
es,
an
d
th
e
f
ilter
ed
im
ag
e
is
r
ep
r
esen
te
d
as
(
1
)
.
=
(
,
,
,
)
(
1
)
W
h
er
e
d
en
o
tes th
e
s
m
o
o
th
ed
im
ag
e,
r
ep
r
esen
ts
in
p
u
t im
ag
e
,
f
ilter
r
ad
iu
s
is
r
ep
r
esen
ted
as
,
r
eg
u
lar
izatio
n
p
ar
am
eter
is
d
en
o
te
d
as
,
an
d
is
co
m
p
u
ted
f
o
r
t
h
e
s
ize
o
f
ev
er
y
l
o
ca
l
p
at
h
with
th
e
b
o
x
f
ilter
.
T
h
en
th
e
r
ec
o
v
er
y
o
f
ed
g
e
is
d
o
n
e
b
y
g
u
id
ed
f
ilter
,
an
d
it
is
an
i
ter
ativ
e
p
r
o
ce
s
s
.
T
h
e
ed
g
e
r
ec
o
v
er
y
p
r
o
ce
s
s
is
r
ep
r
esen
ted
as
(
2
)
.
+
1
=
(
,
,
,
2
)
(
2
)
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
A
n
o
ve
l a
p
p
r
o
a
c
h
to
d
etec
t to
ma
to
lea
f d
is
ea
s
e
u
s
in
g
visi
o
n
tr
a
n
s
fo
r
mer (
S
a
n
jeela
S
a
g
a
r
)
1553
Her
e,
g
u
id
e
d
f
ilter
in
g
o
p
er
atio
n
is
d
en
o
te
d
(
.
)
,
is
th
e
im
ag
e
wh
ich
is
s
m
o
o
th
ed
f
o
r
t
h
e
ℎ
iter
atio
n
,
2
an
d
ar
e
th
e
r
eg
u
lar
i
za
tio
n
p
ar
a
m
eter
an
d
s
tan
d
a
r
d
d
ev
iatio
n
r
esp
ec
tiv
ely
.
I
n
g
en
er
al,
AR
GW
F
is
s
im
p
ly
wr
itten
as
(
3
)
.
=
(
,
,
,
)
(
3
)
Her
e,
th
e
f
ilter
in
g
o
u
tp
u
t
is
r
ep
r
esen
ted
as
an
d
d
en
o
tes
t
h
e
iter
atio
n
n
u
m
b
er
.
T
h
e
AR
GW
F
s
m
o
o
th
th
e
in
p
u
t
im
ag
es,
a
n
d
th
e
f
ilter
ed
im
ag
es
ed
g
es
ar
e
m
ain
ly
b
l
u
r
r
ed
t
o
f
o
c
u
s
o
n
lar
g
e
-
s
ca
le
ed
g
e
d
etails
in
th
e
s
u
cc
ess
iv
e
ed
g
e
r
ec
o
v
er
y
p
r
o
c
ess
.
2
.
2
.
Seg
m
ent
a
t
i
o
n
T
h
e
s
eg
m
en
tatio
n
o
f
t
h
e
p
r
e
-
p
r
o
ce
s
s
ed
im
ag
es
ar
e
d
o
n
e
b
y
r
esid
u
al
atten
tio
n
b
ased
U
Net+
+
[
4
0
]
m
o
d
el
to
s
ig
n
if
ican
tly
id
en
tify
th
e
d
is
ea
s
ed
p
ar
ts
o
f
th
e
to
m
ato
p
lan
t.
R
esid
u
al
atten
tio
n
b
ased
UNe
t++
i
s
th
e
co
m
b
in
atio
n
o
f
r
esid
u
al
n
etwo
r
k
,
atten
tio
n
m
ec
h
an
is
m
an
d
UNe
t++
.
T
h
e
in
teg
r
atio
n
o
f
th
ese
n
etwo
r
k
s
b
r
in
g
s
s
o
m
e
b
en
ef
its
in
clu
d
in
g
UNe
t
++
im
p
r
o
v
is
es th
e
s
em
an
tic
g
ap
b
etwe
en
th
e
en
co
d
er
an
d
d
e
co
d
er
n
etwo
r
k
s
an
d
f
ea
tu
r
e
m
ap
s
,
r
esid
u
al
u
n
it
ad
d
r
ess
es
th
e
d
eg
r
ad
atio
n
p
r
o
b
lem
an
d
ea
s
es
th
e
tr
ain
in
g
n
etwo
r
k
,
an
d
th
e
atten
tio
n
m
ec
h
an
is
m
s
u
p
p
r
ess
th
e
ir
r
elev
a
n
t
s
eg
m
en
tatio
n
b
ac
k
g
r
o
u
n
d
a
r
ea
an
d
also
in
cr
e
ases
th
e
weig
h
t
o
f
th
e
tar
g
et
ar
ea
.
T
h
e
UNe
t++
co
n
n
ec
ts
th
e
en
co
d
er
a
n
d
d
ec
o
d
er
n
etwo
r
k
th
r
o
u
g
h
r
ev
is
ed
s
k
ip
p
ath
way
s
.
T
h
e
en
co
d
er
n
etwo
r
k
s
f
ea
tu
r
e
m
ap
is
m
ap
p
ed
to
t
h
e
d
ec
o
d
er
n
et
wo
r
k
with
t
h
e
h
elp
o
f
d
en
s
e
c
o
n
v
o
lu
ti
o
n
al
b
l
o
ck
s
.
T
h
e
s
k
ip
p
ath
wa
y
is
f
o
r
m
u
lated
in
(
4
)
.
,
=
{
{
−
1
,
}
,
{
[
∫
−
1
=
0
(
,
)
,
(
+
1
,
−
1
)
]
}
}
,
=
0
>
0
(
4
)
W
h
er
e
,
r
ep
r
esen
ts
th
e
o
u
tp
u
t
o
f
th
e
n
o
d
e
,
,
in
d
icate
s
th
e
la
y
er
(
d
o
wn
s
am
p
lin
g
)
ac
co
r
d
i
n
g
to
th
e
en
co
d
er
s
u
b
n
etwo
r
k
,
th
e
co
n
v
o
lu
tio
n
al
lay
er
o
f
th
e
b
lo
ck
(
d
en
s
e)
alo
n
g
with
s
k
ip
p
ath
w
ay
s
an
d
ar
e
d
en
o
ted
as
,
a
co
n
v
o
l
u
tio
n
al
o
p
er
atio
n
f
o
llo
wed
b
y
r
ec
tifie
d
lin
ea
r
u
n
it
(
R
eL
U
)
ac
tiv
atio
n
f
u
n
ct
io
n
is
r
ep
r
esen
ted
as
{
.
}
,
an
d
[]
r
ep
r
esen
ts
th
e
at
ten
tio
n
g
ate
an
d
co
n
ca
ten
atio
n
lay
er
r
esp
ec
tiv
ely
.
{
.
}
r
ep
r
ese
n
ts
th
e
up
-
s
am
p
lin
g
o
p
er
atio
n
s
.
2
.
3
.
F
e
a
t
ure
ex
t
r
a
ct
io
n
Af
ter
s
eg
m
en
tatio
n
,
s
eg
m
en
te
d
im
ag
es
g
o
th
r
o
u
g
h
f
ea
t
u
r
e
ex
tr
ac
tio
n
p
r
o
ce
s
s
.
Fo
r
e
f
f
icie
n
t
to
m
ato
leaf
d
is
ea
s
e
id
en
tific
atio
n
,
a
n
ef
f
e
ctiv
e
f
ea
t
u
r
e
ex
tr
ac
tio
n
is
n
ec
ess
ar
y
.
Fo
r
th
at
p
u
r
p
o
s
e,
th
e
p
r
o
p
o
s
ed
ap
p
r
o
ac
h
u
tili
ze
s
Bi
-
DL
ST
M
m
o
d
el.
C
o
m
p
ar
e
d
to
tr
a
d
itio
n
al
f
ea
tu
r
e
e
x
tr
ac
tio
n
a
p
p
r
o
ac
h
es
in
clu
d
in
g
lo
n
g
s
h
o
r
t
-
ter
m
m
em
o
r
y
(
L
S
T
M
)
an
d
C
NNs,
th
e
p
r
o
p
o
s
e
d
ap
p
r
o
ac
h
o
f
f
er
s
a
h
y
b
r
id
iz
ed
ap
p
r
o
ac
h
w
h
ich
in
teg
r
ates
th
e
b
en
e
f
its
o
f
b
o
th
b
id
ir
ec
tio
n
al
lo
n
g
s
h
o
r
t
-
ter
m
m
em
o
r
y
(
B
i
-
L
STM
)
m
o
d
el
an
d
d
ilate
d
co
n
v
o
l
u
tio
n
s
.
T
h
e
a
r
ch
itectu
r
e
o
f
Bi
-
DL
STM
is
v
is
u
alize
d
in
Fig
u
r
e
2
.
W
h
ile
u
s
in
g
r
ec
u
r
r
e
n
t
n
e
u
r
al
n
etwo
r
k
s
(
R
NN)
f
o
r
ex
tr
ac
ti
n
g
lo
n
g
f
ea
tu
r
es,
m
ain
tain
in
g
m
id
-
an
d
s
h
o
r
t
-
ter
m
m
em
o
r
y
is
q
u
ite
ch
allen
g
in
g
.
T
o
o
v
er
c
o
m
e
th
is
is
s
u
e
an
d
to
en
h
an
ce
th
e
ex
tr
ac
tio
n
ca
p
ab
ilit
y
a
n
ew
d
ilated
L
STM
h
as
b
ee
n
i
n
tr
o
d
u
ce
d
an
d
it
h
as
o
n
e
d
ilated
r
ec
u
r
r
e
n
t
lay
er
with
f
ix
ed
d
ilatio
n
s
.
Alth
o
u
g
h
a
s
tan
d
ar
d
L
STM
s
o
lv
es
th
e
v
an
is
h
in
g
g
r
ad
ien
t
d
escen
t
p
r
o
b
lem
,
it
h
as
s
o
m
e
lim
itati
o
n
s
wh
ile
lear
n
i
n
g
lo
n
g
-
ter
m
d
ep
e
n
d
en
cies.
T
h
e
m
ain
is
s
u
e
in
s
tan
d
ar
d
L
STM
is
th
e
ch
an
g
e
o
f
ce
ll
s
tate,
wh
en
n
ew
f
u
n
ctio
n
s
ar
e
ad
d
ed
t
o
th
e
in
p
u
ts
.
T
h
ese
is
s
u
es
ar
e
ad
d
r
ess
ed
b
y
u
s
in
g
th
e
d
ilatio
n
s
an
d
s
k
ip
p
in
g
co
n
n
ec
tio
n
s
in
th
e
d
ilated
L
STM
.
A
r
ec
u
r
r
en
t
s
k
i
p
co
n
n
ec
tio
n
with
d
ilatio
n
s
in
cr
ea
s
es
ex
p
o
n
en
tially
i
n
a
m
u
lti
-
lay
er
ed
lear
n
i
n
g
m
o
d
el
im
p
lem
en
tatio
n
is
u
tili
ze
d
f
o
r
th
e
d
ilated
L
STM
im
p
lem
en
tatio
n
.
T
h
e
co
m
p
le
x
d
ata
ar
e
lear
n
ed
o
n
d
if
f
er
en
t
lay
er
s
o
f
d
ilated
L
S
T
M.
T
h
e
d
ilated
r
ec
u
r
r
e
n
t
s
k
ip
co
n
n
ec
tio
n
in
th
e
L
STM
ce
l
l
ca
n
b
e
r
ep
r
esen
ted
as
in
(
5
)
.
(
)
=
(
(
)
,
−
(
)
)
(
5
)
Her
e,
(
)
r
ep
r
esen
ts
th
e
ce
ll
in
la
y
er
at
a
tim
e
,
(
)
r
ep
r
esen
ts
th
e
len
g
th
o
f
s
k
ip
o
f
lay
e
r
an
d
th
e
L
STM
in
p
u
t la
y
er
at
tim
e
is
d
en
o
ted
as
(
)
.
T
h
e
d
ilatio
n
o
f
th
e
l
-
th
la
y
er
(
)
is
ca
lcu
lated
a
s
in
(
6
)
.
(
)
=
(
−
1
)
,
=
1
,
2
,
…
…
(
6
)
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
1
5
,
No
.
2
,
Ap
r
il 2
0
2
6
:
1
5
4
8
-
1
5
6
5
1554
W
h
er
e
an
d
im
p
lies
th
e
d
ilat
i
o
n
s
at
d
if
f
er
en
t la
y
er
s
.
T
h
e
co
m
b
in
atio
n
o
f
d
ilated
r
ec
u
r
r
en
t
lay
er
s
b
r
in
g
s
o
u
t
s
o
m
e
b
en
ef
its
v
ia
wh
ich
d
if
f
er
en
t
lay
er
s
co
n
ce
n
tr
ate
o
n
d
if
f
e
r
en
t
r
eso
lu
tio
n
s
an
d
en
ab
les
th
e
n
etwo
r
k
to
lear
n
m
o
r
e
lo
n
g
-
ter
m
d
ep
en
d
en
cies.
No
t o
n
ly
th
at,
b
u
t it
also
r
e
d
u
ce
s
th
e
p
ath
b
etwe
en
d
if
f
er
e
n
t
tim
e
s
tep
s
.
L
S
T
M
L
S
T
M
L
S
T
M
L
S
T
M
L
S
T
M
L
S
T
M
L
S
T
M
L
S
T
M
C
onc
a
t
e
na
t
i
on l
a
ye
r
Dila
ti
on
1
Dila
ti
on
2
Dil
a
t
i
on out
put
s
S
of
t
m
a
x l
a
ye
r
O
u
t
p
u
t
Fig
u
r
e
2
.
Ar
c
h
itectu
r
e
o
f
B
i
-
DL
STM
2
.
4
.
F
e
a
t
ure
s
elec
t
io
n
I
t is a
p
r
o
ce
s
s
o
f
s
elec
tin
g
ef
f
icien
t a
n
d
o
p
tim
al
f
ea
tu
r
es f
r
o
m
th
e
ex
tr
ac
ted
f
ea
tu
r
es.
I
n
th
i
s
p
r
o
p
o
s
ed
ap
p
r
o
ac
h
,
f
ea
tu
r
e
s
elec
tio
n
is
d
o
n
e
b
y
C
SW
O
alg
o
r
ith
m
.
I
t
is
an
im
p
r
o
v
ed
v
er
s
io
n
o
f
s
p
i
d
er
wasp
o
p
tim
izer
(
SW
O)
[
4
1
]
wh
ic
h
is
a
m
eta
-
h
eu
r
is
tic
ap
p
r
o
ac
h
u
s
ed
t
o
a
d
d
r
ess
th
e
co
n
tin
u
o
u
s
o
p
tim
i
za
tio
n
is
s
u
es.
T
h
e
f
ea
tu
r
e
s
elec
tio
n
b
ased
C
SW
O
alg
o
r
ith
m
is
ca
teg
o
r
ized
i
n
to
s
ev
er
al
ac
tiv
ities
lik
e
in
i
tializatio
n
,
f
itn
ess
ev
alu
atio
n
,
s
p
id
e
r
m
o
v
em
en
t,
wasp
m
o
v
em
en
t
,
a
n
d
c
h
ao
tic
m
ap
p
in
g
.
2
.
4
.
1
.
I
nitia
liza
t
io
n o
f
pa
ra
met
er
s
I
n
itializatio
n
o
f
p
ar
a
m
eter
s
r
a
n
d
o
m
ly
g
e
n
er
ates
p
o
p
u
latio
n
o
f
s
p
id
er
s
an
d
wasp
s
with
in
t
h
e
s
ea
r
ch
s
p
ac
e
an
d
th
e
s
ize
o
f
th
e
s
ea
r
ch
s
p
ac
e
is
d
ef
in
e
d
as
.
/
(
/
)
∗
(
.
(
)
)
.
T
h
e
wasp
an
d
s
p
id
e
r
p
o
p
u
latio
n
ar
e
d
escr
ib
e
d
as
(
7
)
an
d
(
8
)
.
=
[
1
,
1
⋯
1
,
/
2
⋯
1
,
⋮
⋱
⋮
/
⋮
/
2
,
1
/
2
,
/
2
/
2
,
⋮
/
⋮
⋱
⋮
,
1
⋯
,
/
2
⋯
,
]
×
(
7
)
=
[
1
,
1
⋯
1
,
/
2
⋯
1
,
⋮
⋱
⋮
/
⋮
/
2
,
1
/
2
,
/
2
/
2
,
⋮
/
⋮
⋱
⋮
,
1
⋯
,
/
2
⋯
,
]
×
(
8
)
2
.
4
.
2
.
F
it
nes
s
ev
a
lua
t
io
n
T
h
e
f
itn
ess
o
f
ea
ch
s
u
b
s
et
o
f
f
ea
tu
r
es
(
ea
ch
r
o
w
o
f
wasp
s
an
d
s
p
id
er
s
)
,
is
ca
lc
u
lated
b
y
p
a
s
s
in
g
ea
ch
s
u
b
s
et
in
to
th
e
f
itn
ess
f
u
n
ctio
n
.
T
h
e
r
esu
lts
ar
e
s
to
r
ed
as
wasp
f
itn
ess
an
d
s
p
id
er
f
itn
ess
.
Af
ter
th
at,
th
e
n
ew
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
A
n
o
ve
l a
p
p
r
o
a
c
h
to
d
etec
t to
ma
to
lea
f d
is
ea
s
e
u
s
in
g
visi
o
n
tr
a
n
s
fo
r
mer (
S
a
n
jeela
S
a
g
a
r
)
1555
p
o
p
u
latio
n
s
s
p
id
er
s
o
r
ted
an
d
wasp
s
o
r
ted
ar
e
cr
ea
ted
,
wh
ic
h
s
to
r
es
th
e
s
u
b
s
et
o
f
f
ea
tu
r
es
f
o
r
th
e
s
p
id
er
an
d
wasp
p
o
p
u
latio
n
s
in
ascen
d
in
g
o
r
d
e
r
o
f
er
r
o
r
o
r
f
itn
ess
v
alu
e.
Sp
id
er
to
p
an
d
wasp
t
o
p
ar
e
th
e
n
ewly
d
e
f
in
ed
p
o
p
u
latio
n
s
to
s
to
r
e
th
o
s
e
p
o
p
u
latio
n
s
f
o
r
t
h
e
cu
r
r
en
t
iter
atio
n
r
esp
ec
tiv
ely
.
T
h
e
f
o
u
r
f
itn
ess
f
u
n
ctio
n
s
s
u
ch
as
d
ec
is
io
n
tr
ee
(
DT
)
,
n
aïv
e
B
ay
es
(
NB
)
,
KNN
,
an
d
lin
ea
r
d
is
cr
im
in
an
t
an
aly
s
is
(
L
DA)
u
s
e
d
in
th
is
m
eth
o
d
a
r
e
p
ass
ed
in
to
SW
O
f
u
n
ctio
n
f
o
r
th
e
r
esu
lts
.
T
h
e
ca
lcu
latio
n
o
f
f
itn
ess
f
o
r
ea
ch
s
elec
ted
s
u
b
s
et
o
f
f
ea
tu
r
es
th
en
b
e
c
o
m
p
u
te
d
b
y
(
9
)
an
d
(
1
0
)
.
=
∑
=
1
(
9
)
=
[
′
≠
]
(
1
0
)
W
h
er
e
th
e
d
if
f
er
en
ce
b
etwe
en
th
e
o
r
ig
in
al
a
n
d
p
r
ed
icted
r
es
u
lts
ar
e
d
en
o
ted
b
y
⥂
an
d
is
u
s
ed
to
r
ep
r
esen
t th
e
to
tal
n
u
m
b
e
r
in
s
tan
ce
s
.
2
.
4
.
3
.
Sp
ider
m
o
v
em
ent
T
h
e
s
p
id
er
m
o
v
e
m
en
t
o
r
cr
o
s
s
o
v
er
f
u
n
ctio
n
is
p
er
f
o
r
m
e
d
b
y
co
n
ca
ten
atin
g
s
o
m
e
f
e
atu
r
es
f
r
o
m
t
h
e
s
p
id
er
to
p
(
)
an
d
wasp
to
p
(
)
p
o
p
u
latio
n
w
h
ich
ar
e
d
ep
icted
a
s
in
(
1
1
)
a
n
d
(
1
2
)
.
=
[
(
,
2
)
]
[
:
2
]
(
1
1
)
=
[
(
,
2
)
]
[
2
:
]
(
1
2
)
Her
e
an
d
ar
e
th
e
f
ea
tu
r
es
s
elec
ted
f
r
o
m
an
d
f
o
r
th
e
iter
a
tio
n
.
r
ep
r
esen
ts
th
e
to
tal
n
u
m
b
e
r
o
f
f
ea
tu
r
es
av
ai
lab
le,
an
d
d
en
o
tes
th
e
n
u
m
b
er
o
f
s
p
id
er
s
an
d
wasp
s
r
esp
ec
tiv
ely
.
T
o
r
etr
iev
e
th
e
o
r
ig
in
al
n
u
m
b
e
r
o
f
f
ea
tu
r
es,
s
p
id
er
,
a
n
d
wasp
f
ea
tu
r
es
ar
e
co
n
ca
ten
ated
a
n
d
a
n
ew
s
u
b
s
et
i
s
f
o
r
m
ed
f
r
o
m
wh
er
e
f
ea
tu
r
es f
o
r
ea
ch
iter
atio
n
is
ap
p
lied
to
a
n
ew
p
o
p
u
latio
n
n
a
m
ed
.
2
.
4
.
4
.
Wa
s
p m
o
v
e
m
ent
T
h
e
wasp
m
o
v
em
e
n
t
o
r
m
u
t
atio
n
is
p
e
r
f
o
r
m
ed
b
y
r
an
d
o
m
ly
ch
a
n
g
in
g
th
e
s
elec
tio
n
o
f
ce
r
tain
f
ea
tu
r
es
in
th
e
wasp
to
p
p
o
p
u
l
atio
n
.
T
h
e
n
ewly
f
o
r
m
e
d
f
ea
tu
r
es
ar
e
s
to
r
ed
i
n
a
n
ew
p
o
p
u
lat
io
n
n
am
e
d
.
T
h
e
m
u
tated
v
alu
e
f
o
r
th
e
n
ew
p
o
p
u
latio
n
is
ca
lcu
lated
as
in
(
1
3
)
.
=
[
(
,
2
)
]
[
]
×
[
(
0
,
1
)
]
(
1
3
)
Her
e
r
ep
r
esen
ts
th
e
m
u
tated
v
alu
e
f
o
r
th
e
f
ea
tu
r
e
f
o
r
n
u
m
b
er
o
f
iter
atio
n
an
d
(
,
)
r
etu
r
n
s
an
in
teg
er
with
t
h
e
b
o
u
n
d
s
o
f
an
d
.
T
h
e
n
th
e
in
itial
s
p
id
er
a
n
d
wasp
p
o
p
u
latio
n
s
ar
e
c
o
n
ca
ten
ated
with
th
e
n
ew
p
o
p
u
latio
n
an
d
all
th
e
f
itn
ess
f
ea
tu
r
es
ar
e
co
m
p
ar
ed
to
th
e
b
est
g
lo
b
al
f
itn
ess
v
alu
e
f
o
r
all
iter
atio
n
s
.
T
h
e
g
lo
b
al
f
itn
ess
an
d
th
e
f
ea
tu
r
es
ar
e
u
p
d
ate
d
o
n
ly
wh
en
th
e
l
o
ca
l
f
itn
ess
v
alu
e
ex
ce
e
d
s
a
g
lo
b
al
f
itn
ess
v
alu
e
.
T
h
e
en
tire
p
r
o
ce
s
s
will
b
e
r
ep
ea
ted
u
n
til
th
e
m
ax
im
u
m
n
u
m
b
er
o
f
iter
atio
n
s
is
ac
h
iev
ed
.
T
h
en
o
n
ly
th
e
f
in
al
v
ar
iab
les ar
e
r
etu
r
n
ed
to
th
e
g
l
o
b
al
b
est f
itn
ess
an
d
g
l
o
b
al
b
e
s
t f
ea
tu
r
es.
2
.
4
.
5
.
Cha
o
t
ic
im
pro
v
em
ent
Fo
r
o
p
tim
al
f
ea
tu
r
e
ex
tr
ac
tio
n
,
th
e
m
o
d
el
is
im
p
r
o
v
ed
with
a
n
o
p
tim
izatio
n
alg
o
r
ith
m
.
T
h
e
alg
o
r
ith
m
is
u
p
d
ated
with
C
h
eb
y
s
h
e
v
c
h
ao
tic
m
ap
[
4
2
]
,
wh
ich
im
p
r
o
v
es
th
e
ef
f
icien
c
y
o
f
th
e
alg
o
r
ith
m
an
d
p
er
f
o
r
m
well
o
n
f
ea
tu
r
e
e
x
tr
ac
tio
n
p
r
o
c
ess
.
T
h
e
u
p
d
ated
c
h
ao
tic
m
ap
r
ep
r
esen
tatio
n
is
d
escr
ib
ed
as in
(
1
4
)
.
+
1
=
(
−
1
(
)
)
(
14)
T
h
e
p
r
esen
ted
f
ea
tu
r
e
s
elec
tio
n
alg
o
r
ith
m
co
m
b
in
es
th
e
s
tr
e
n
g
th
o
f
b
o
th
SW
O
an
d
ch
a
o
tic
m
ap
p
in
g
.
T
h
is
C
h
eb
y
s
h
ev
ch
ao
tic
m
ap
p
in
g
en
h
an
ce
s
th
e
s
ea
r
ch
ca
p
ab
ilit
y
o
f
SW
O
wh
ich
h
elp
s
to
f
in
d
b
etter
f
ea
tu
r
es
th
at
h
elp
s
th
e
m
o
d
el
f
o
r
ac
cu
r
ate
class
if
icatio
n
.
I
n
s
u
m
m
ar
y
,
th
e
p
r
esen
ted
f
ea
tu
r
e
s
elec
tio
n
alg
o
r
ith
m
ef
f
ec
tiv
ely
ca
p
tu
r
es th
e
m
o
s
t r
elev
an
t f
ea
tu
r
es f
o
r
to
m
at
o
lea
f
d
is
ea
s
e
class
if
icatio
n
.
2
.
5
.
Cla
s
s
if
ica
t
io
n
Af
ter
th
e
e
x
tr
ac
tio
n
o
f
f
ea
tu
r
es
an
d
s
elec
tio
n
o
f
o
p
tim
al
f
ea
tu
r
es,
class
if
icatio
n
o
f
to
m
ato
leaf
d
is
ea
s
es
i
s
ca
r
r
ied
o
u
t
b
y
A
-
GVT
m
o
d
el.
T
h
e
class
if
icatio
n
m
o
d
el
ca
teg
o
r
izes
th
e
f
ea
tu
r
es
b
ased
o
n
th
e
ty
p
es
o
f
d
is
ea
s
es.
T
h
e
ar
ch
itectu
r
e
o
f
A
-
GVT
is
d
ep
icted
as Fig
u
r
e
3
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
1
5
,
No
.
2
,
Ap
r
il 2
0
2
6
:
1
5
4
8
-
1
5
6
5
1556
F
e
a
tur
e
s
+
Output
+
Convol
ut
i
ona
l
bl
oc
k
E
nc
ode
r bl
oc
k
D
e
c
ode
r bl
oc
k
A
dd
1
x
1
Conv
Fig
u
r
e
3
.
Ar
c
h
itectu
r
e
o
f
A
-
G
VT
s
T
o
o
v
e
r
co
m
e
th
e
c
o
m
p
u
tatio
n
al
co
m
p
lex
ity
o
f
af
f
in
ities
,
s
elf
-
atten
tio
n
[
4
3
]
is
d
ec
o
m
p
o
s
e
d
in
to
two
s
elf
-
atten
tio
n
m
o
d
u
les.
T
h
e
f
ir
s
t
m
o
d
u
le
em
p
lo
y
s
th
e
h
eig
h
t
ax
is
o
f
f
ea
tu
r
e
m
a
p
,
an
d
th
e
s
ec
o
n
d
m
o
d
u
le
wo
r
k
s
o
n
wid
th
ax
is
.
T
h
e
s
elf
-
atten
tio
n
lay
er
is
ca
lcu
lated
wi
th
th
e
h
elp
o
f
p
r
o
jecte
d
i
n
p
u
t,
an
d
it
is
ex
p
r
ess
ed
as
in
(
1
5
)
.
=
∑
∑
(
ℎ
)
ℎ
=
1
ℎ
=
1
(
1
5
)
Her
e
q
u
er
ies
ar
e
r
ep
r
esen
te
d
as
,
th
e
k
e
y
s
ar
e
d
e
n
o
ted
as
a
n
d
th
e
v
alu
es
ar
e
d
escr
ib
ed
as
.
W
an
d
H
ar
e
th
e
wid
th
an
d
h
eig
h
t
o
f
th
e
f
ea
tu
r
e
m
ap
s
.
T
h
e
s
elf
-
atten
tio
n
m
ec
h
an
is
m
g
ath
e
r
s
lo
ca
l
in
f
o
r
m
atio
n
f
r
o
m
th
e
f
ea
tu
r
e
m
ap
.
C
alcu
latin
g
th
ese
af
f
in
ities
is
co
m
p
u
tatio
n
ally
ex
p
en
s
iv
e
an
d
s
elf
-
atten
tio
n
la
y
er
d
o
es
n
o
t
u
tili
ze
p
o
s
itio
n
al
in
f
o
r
m
atio
n
.
I
t
is
u
s
ef
u
l
in
v
is
io
n
tr
a
n
s
f
o
r
m
e
r
(
Vi
T
)
m
o
d
el
t
o
d
etec
t
th
e
s
tr
u
ct
u
r
e
o
f
th
e
o
b
ject.
T
h
e
ax
ial
atten
tio
n
an
d
p
o
s
itio
n
al
em
b
ed
d
in
g
ar
e
co
m
b
in
ed
to
u
tili
ze
it
f
o
r
all
th
e
q
u
er
ies,
k
e
y
s
an
d
v
alu
es.
T
h
e
s
elf
-
atten
tio
n
m
ec
h
an
is
m
with
tr
an
s
f
o
r
m
er
m
o
d
el
an
d
p
o
s
itio
n
al
en
co
d
i
n
g
s
f
o
r
an
y
in
p
u
t
f
ea
tu
r
e
m
ap
ca
n
b
e
wr
itten
as
in
(
1
6
)
.
=
∑
(
+
+
)
=
1
(
+
)
(
1
6
)
W
h
er
e
,
,
all
th
ese
m
o
d
u
les
b
el
o
n
g
to
a
wid
th
wis
e
atten
tio
n
m
o
d
el.
I
n
(
1
6
)
d
escr
ib
es
th
e
atten
tio
n
alo
n
g
with
wid
th
ax
is
,
s
im
ilar
ly
atten
tio
n
with
h
eig
h
t
ax
is
is
also
ex
p
r
ess
ed
.
On
co
m
b
in
in
g
th
ese
two
s
elf
-
atten
tio
n
m
o
d
els,
a
n
ew
c
o
m
p
u
tatio
n
ally
ef
f
icien
t
s
in
g
le
s
elf
-
atten
tio
n
m
o
d
el
is
f
o
r
m
ed
.
T
h
e
s
elf
-
atten
tio
n
lay
er
with
p
o
s
itio
n
al
en
co
d
i
n
g
ca
n
ca
lcu
late
n
o
n
-
lo
ca
l
c
o
n
tex
t
b
y
u
s
in
g
co
m
p
u
tatio
n
al
ef
f
icien
cy
.
T
h
e
p
r
o
p
o
s
ed
ap
p
r
o
ac
h
u
tili
ze
s
s
m
all
s
ca
le
d
atasets
.
T
o
s
o
lv
e
th
e
is
s
u
e
o
f
ac
cu
r
ate
p
o
s
itio
n
al
b
ias,
an
d
to
co
n
tr
o
l
th
e
p
o
s
itio
n
al
b
ias,
a
m
o
d
if
ied
atten
tio
n
b
lo
ck
is
d
ev
elo
p
ed
wh
ich
ca
n
b
e
em
p
l
o
y
ed
in
th
e
en
co
d
in
g
o
f
non
-
lo
ca
l c
o
n
tex
t.
A
r
ch
itectu
r
e
o
f
g
ated
atten
tio
n
m
ec
h
an
is
m
is
illu
s
tr
ated
in
Fig
u
r
e
4
.
T
h
e
s
elf
-
atten
tio
n
m
o
d
el
is
u
p
d
ated
with
G
-
ViT
m
o
d
el
ap
p
lied
o
n
th
e
wid
th
ax
is
is
ex
p
r
ess
ed
as
in
(
1
7
)
.
=
∑
(
+
+
)
=
1
(
1
+
2
)
(
1
7
)
W
h
er
e
th
e
s
elf
-
atten
tio
n
f
o
r
m
u
la
f
o
llo
ws
with
g
ated
m
ec
h
an
is
m
.
,
,
1
,
2
ar
e
p
ar
am
eter
s
,
an
d
th
ey
to
g
eth
er
c
r
ea
te
g
ated
m
e
ch
an
is
m
wh
ich
in
f
lu
en
ce
th
e
p
o
s
itio
n
al
en
co
d
in
g
with
n
o
n
-
lo
ca
l
co
n
tex
t.
T
h
e
g
ated
m
ec
h
a
n
is
m
will
ass
ig
n
an
ac
cu
r
ate
p
o
s
itio
n
al
e
n
co
d
in
g
a
h
ig
h
weig
h
t
co
m
p
ar
e
d
with
in
ac
cu
r
ate
lear
n
in
g
m
o
d
els.
T
h
e
p
r
o
p
o
s
ed
A
-
GVT
u
s
es
g
ated
s
elf
-
atten
tio
n
lay
er
as
th
e
b
asic
b
u
i
ld
in
g
b
lo
ck
o
f
th
e
class
if
icatio
n
m
o
d
el.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
A
n
o
ve
l a
p
p
r
o
a
c
h
to
d
etec
t to
ma
to
lea
f d
is
ea
s
e
u
s
in
g
visi
o
n
tr
a
n
s
fo
r
mer (
S
a
n
jeela
S
a
g
a
r
)
1557
+
X
X
X
X
X
X
X
X
X
X
+
+
X
G
a
t
e
s
P
os
i
t
i
ona
l
e
m
be
ddi
ng
W
e
i
ght
s
A
ddi
t
i
on
M
a
t
ri
x
m
ul
t
i
pl
i
c
a
t
i
on
Fig
u
r
e
4
.
Ar
c
h
itectu
r
e
o
f
g
ate
d
atten
tio
n
lay
er
2
.
6
.
B
la
c
k
wido
w
o
ptim
iza
t
i
o
n a
lg
o
rit
hm
T
h
e
B
W
O
is
a
m
eta
-
h
eu
r
is
tic
alg
o
r
ith
m
d
ev
elo
p
e
d
to
s
o
lv
e
co
m
p
lex
n
u
m
er
ical
o
p
t
im
izatio
n
p
r
o
b
lem
s
.
T
h
e
B
W
O
d
eliv
er
f
ast
co
n
v
er
g
en
c
e,
av
o
i
d
o
p
tim
a
l
lo
ca
l
s
o
lu
tio
n
s
an
d
b
ala
n
ce
t
h
e
ex
p
lo
r
atio
n
an
d
ex
p
lo
itatio
n
p
h
ase.
Hen
ce
,
B
W
O
i
s
a
g
o
o
d
m
eth
o
d
to
s
o
lv
e
o
p
tim
izatio
n
p
r
o
b
lem
s
an
d
to
tu
n
e
th
e
p
ar
am
eter
s
.
T
h
e
o
p
tim
izatio
n
alg
o
r
ith
m
in
itialize
s
r
an
d
o
m
weig
h
ts
an
d
u
s
es
m
u
tatio
n
to
s
elec
t
r
an
d
o
m
f
ea
tu
r
es
f
r
o
m
th
e
ex
tr
ac
te
d
i
m
ag
es.
Af
ter
t
h
at
th
e
o
b
tain
e
d
f
ea
tu
r
es
ar
e
s
to
r
e
d
as
n
ew
f
ea
tu
r
es
an
d
attain
s
o
p
tim
al
weig
h
ts
.
A
s
u
m
m
ar
y
o
f
B
W
O
o
p
tim
izatio
n
is
ex
p
r
e
s
s
ed
as f
o
llo
ws
.
i)
I
n
itializatio
n
:
t
h
e
n
u
m
b
er
o
f
wid
o
ws
with
s
ize
N
r
ep
r
esen
ts
it
s
p
o
p
u
latio
n
,
an
d
ea
ch
w
id
o
w
ca
n
b
e
co
n
s
titu
ted
as
an
ar
r
a
y
as
=
(
1
,
2
,
…
…
,
)
,
wh
er
e
is
th
e
d
im
en
s
io
n
o
f
th
e
o
p
tim
izatio
n
p
r
o
b
lem
.
T
h
e
f
itn
ess
o
f
th
e
wi
d
o
w
is
ca
lcu
lated
b
y
ev
alu
atin
g
th
e
f
itn
ess
f
u
n
ctio
n
o
f
ea
ch
wid
o
w
in
th
e
ar
r
ay
.
T
h
e
f
itn
ess
ca
n
b
e
r
ep
r
e
s
en
ted
as
=
(
1
,
2
,
…
…
,
)
.
ii)
Pro
cr
ea
te:
i
n
th
is
th
e
p
ar
en
ts
an
d
o
f
f
s
p
r
in
g
ar
e
c
o
m
b
in
e
d
;
th
e
ev
alu
atio
n
o
f
c
r
o
s
s
o
v
er
r
esu
l
t is st
o
r
ed
an
d
r
ep
r
esen
ted
as
in
(
1
8
)
.
1
=
×
1
+
(
1
−
)
×
2
an
d
2
=
×
2
+
(
1
−
)
×
1
(
1
8
)
iii)
C
an
n
ib
alis
m
:
a
f
ter
ca
n
n
ib
alis
m
,
a
n
ew
p
o
p
u
latio
n
is
ass
ess
e
d
an
d
s
to
r
e
d
in
a
v
ar
iab
le
p
o
p
2
.
iv
)
Mu
tatio
n
:
r
an
d
o
m
s
elec
tio
n
o
f
f
ea
tu
r
es
f
r
o
m
th
e
p
o
p
u
latio
n
is
m
u
tated
an
d
f
o
r
m
s
a
n
ew
p
o
p
u
latio
n
a
n
d
is
s
to
r
ed
in
a
v
ar
iab
le
n
am
ed
p
o
p
3
.
T
h
en
th
e
p
o
p
3
is
s
o
r
ted
to
r
etu
r
n
b
est
wid
o
w
th
r
esh
o
ld
v
alu
es.
Af
ter
all
th
is
,
th
e
p
ar
a
m
eter
s
ar
e
tu
n
ed
b
ased
o
n
th
e
b
est
weig
h
t
s
.
T
h
e
alg
o
r
ith
m
o
f
BWO
is
d
escr
ib
ed
as
A
lg
o
r
ith
m
1
.
Alg
o
r
ith
m
1
: T
h
e
B
W
O
alg
o
r
ith
m
I
n
itialize:
m
ax
im
u
m
n
o
.
o
f
iter
atio
n
s
,
p
r
o
c
r
ea
tin
g
r
ate,
ca
n
n
i
b
alis
m
r
ate,
m
u
tatio
n
r
ate.
wh
ile
Sto
p
co
n
d
itio
n
n
o
t
m
et
d
o
f
o
r
i=
1
to
n
r
d
o
s
elec
t 2
s
o
lu
tio
n
s
as p
ar
e
n
t
s
f
r
o
m
p
o
p
1
r
an
d
o
m
ly
.
C
r
ea
te
D
ch
ild
r
e
n
On
th
e
b
asis
o
f
r
ate
o
f
ca
n
n
ib
alis
m
,
k
ill a
f
ew
ch
ild
r
e
n
Evaluation Warning : The document was created with Spire.PDF for Python.