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.
15
,
No
.
3
,
J
u
n
e
20
26
,
p
p
.
2956
~
2
9
6
9
I
SS
N:
2
2
5
2
-
8
9
3
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijai.v
15
.i
3
.
p
p
2
9
5
6
-
2
9
6
9
2956
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
a
i
.
ia
esco
r
e.
co
m
Ada
ptive mul
ti
-
sc
a
le conv
o
lutiona
l net
wo
rk
for pla
nt
leaf
disea
se detec
tion
a
nd clas
sifica
tion
T
ej
a
s
hwin
i C
.
G
a
da
g
,
D
.
R
.
K
um
a
r
Ra
j
a
S
c
h
o
o
l
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
,
R
EV
A
U
n
i
v
e
r
si
t
y
,
B
e
n
g
a
l
u
r
u
,
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
an
3
,
2
0
2
5
R
ev
is
ed
Feb
12
,
2
0
2
6
Acc
ep
ted
Ap
r
20
,
2
0
2
6
P
lan
t
d
ise
a
se
d
e
tec
ti
o
n
is
a
c
ri
ti
c
a
l
tas
k
in
m
o
d
e
rn
a
g
ricu
lt
u
re
,
d
irec
tl
y
imp
a
c
ti
n
g
c
r
o
p
y
iel
d
,
f
o
o
d
se
c
u
rit
y
,
a
n
d
su
sta
i
n
a
b
le
fa
rm
in
g
p
ra
c
ti
c
e
s.
Trad
it
io
n
a
l
m
e
th
o
d
s
re
ly
o
n
e
x
p
e
rt
v
is
u
a
l
in
sp
e
c
ti
o
n
,
wh
ic
h
is
ti
m
e
-
c
o
n
su
m
in
g
,
i
n
c
o
n
siste
n
t,
a
n
d
i
n
a
c
c
e
ss
ib
le
in
re
m
o
te
a
re
a
s.
Th
is
stu
d
y
in
tro
d
u
c
e
s
a
d
v
a
n
c
e
d
d
e
e
p
lea
rn
i
n
g
(DL)
fra
m
e
wo
rk
,
t
h
e
a
d
a
p
ti
v
e
m
u
lt
i
-
sc
a
le
c
o
n
v
o
lu
ti
o
n
a
l
n
e
two
r
k
(AMS
-
C
o
n
v
Ne
t)
o
p
ti
m
ize
d
fo
r
a
c
c
u
ra
te
a
n
d
e
fficie
n
t
p
lan
t
d
ise
a
se
id
e
n
ti
fica
ti
o
n
.
Hie
ra
rc
h
ica
l
fe
a
tu
re
e
x
trac
ti
o
n
n
e
tw
o
r
k
(
HFE
N
)
in
teg
ra
tes
t
h
e
m
u
l
ti
-
d
o
m
a
in
a
tt
e
n
t
io
n
fra
m
e
wo
rk
(M
DA
F
)
a
n
d
a
d
a
p
ti
v
e
sc
a
le
fu
sio
n
m
o
d
u
le
(ASF
M
)
to
e
n
h
a
n
c
e
fe
a
tu
re
e
x
trac
ti
o
n
a
n
d
a
d
d
re
ss
c
h
a
ll
e
n
g
e
s
su
c
h
a
s
c
o
m
p
lex
n
a
tu
ra
l
b
a
c
k
g
ro
u
n
d
s,
n
o
n
-
u
n
ifo
rm
lea
f
stru
c
tu
re
s,
a
n
d
v
a
ry
in
g
e
n
v
ir
o
n
m
e
n
tal
c
o
n
d
it
i
o
n
s.
Th
e
p
r
o
p
o
se
d
fra
m
e
wo
rk
e
m
p
lo
y
s
p
re
-
train
e
d
k
n
o
wle
d
g
e
a
d
a
p
tatio
n
(P
TKA)
tec
h
n
iq
u
e
s
to
im
p
ro
v
e
g
e
n
e
ra
li
z
a
ti
o
n
a
n
d
o
v
e
rc
o
m
e
d
a
ta
sc
a
rc
it
y
.
Co
m
p
re
h
e
n
siv
e
e
v
a
l
u
a
ti
o
n
s
o
n
m
u
lt
ip
l
e
d
a
tas
e
ts
d
e
m
o
n
stra
te
th
e
m
o
d
e
l'
s
b
e
tt
e
r
p
e
rfo
rm
a
n
c
e
,
a
c
h
iev
i
n
g
sta
te
-
of
-
th
e
-
a
rt
m
e
tri
c
s
in
p
re
c
isio
n
,
re
c
a
ll
,
F
1
-
sc
o
re
,
a
n
d
a
c
c
u
ra
c
y
.
F
u
rth
e
r
m
o
re
,
th
i
s
a
p
p
ro
a
c
h
e
n
su
re
s
sc
a
lab
il
it
y
a
n
d
a
d
a
p
tab
il
it
y
,
m
a
k
in
g
it
su
it
a
b
le
fo
r
re
a
l
-
field
c
o
n
d
it
io
n
s.
T
h
e
stu
d
y
e
m
p
h
a
siz
e
s
th
e
imp
o
rtan
c
e
o
f
ro
b
u
st
,
a
u
to
m
a
ted
so
lu
ti
o
n
s
in
m
i
n
imiz
in
g
c
ro
p
lo
ss
e
s,
re
d
u
c
in
g
lab
o
r
c
o
sts,
a
n
d
e
n
h
a
n
c
in
g
a
g
ricu
lt
u
ra
l
su
sta
in
a
b
il
it
y
th
r
o
u
g
h
p
re
c
isio
n
d
ise
a
se
m
a
n
a
g
e
m
e
n
t.
K
ey
w
o
r
d
s
:
Ad
ap
tiv
e
m
u
lti
-
s
ca
le
co
n
v
o
l
u
tio
n
al
n
etwo
r
k
Ad
ap
tiv
e
s
ca
le
f
u
s
io
n
m
o
d
u
le
Ag
r
icu
ltu
r
e
Mu
lti
-
d
o
m
ain
atten
tio
n
f
r
am
ewo
r
k
Plan
t d
is
ea
s
e
d
etec
tio
n
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
:
T
ejash
win
i C
.
Gad
ag
Sch
o
o
l o
f
C
o
m
p
u
ter
Scien
ce
a
n
d
E
n
g
in
ee
r
in
g
,
R
E
VA
Un
iv
er
s
ity
B
en
g
alu
r
u
,
I
n
d
ia
E
m
ail: te
jash
win
icg
_
1
2
@
r
ed
if
f
m
ail.
co
m
1.
I
NT
RO
D
UCT
I
O
N
Fin
d
in
g
an
d
d
iag
n
o
s
in
g
p
lan
t
leaf
d
is
ea
s
es
is
e
s
s
en
tial
to
m
a
in
tain
in
g
cr
o
p
r
esil
ien
ce
an
d
p
r
o
d
u
ctio
n
in
co
n
tem
p
o
r
a
r
y
h
o
r
ticu
ltu
r
e
an
d
ag
r
icu
lt
u
r
e
[
1
]
.
C
o
m
p
u
ter
v
is
io
n
a
n
d
m
ac
h
in
e
lear
n
in
g
(
ML
)
ar
e
two
ex
am
p
les
o
f
ad
v
a
n
ce
d
tech
n
o
l
o
g
ies
th
at
p
r
o
v
id
e
u
s
ef
u
l
to
o
ls
f
o
r
p
r
ec
is
e
illn
ess
id
en
tific
ati
o
n
.
I
n
r
ec
en
t
y
ea
r
s
,
v
is
u
al
s
y
m
p
to
m
s
o
n
leav
es,
s
tem
s
,
an
d
f
r
u
its
h
av
e
b
ee
n
an
aly
ze
d
u
s
in
g
au
t
o
m
ated
im
ag
in
g
s
y
s
tem
s
an
d
s
m
ar
tp
h
o
n
e
ap
p
licatio
n
s
th
at
u
s
e
im
ag
e
r
ec
o
g
n
itio
n
alg
o
r
ith
m
s
f
o
r
q
u
ick
p
atter
n
id
en
tific
a
tio
n
.
I
m
p
r
o
v
em
en
ts
in
d
ee
p
lear
n
in
g
(
DL
)
an
d
p
r
e
-
tr
ain
ed
k
n
o
wled
g
e
ad
a
p
tatio
n
(
PTKA
)
m
o
d
els
h
a
v
e
en
h
an
ce
d
th
e
ca
p
ac
ity
t
o
id
en
tify
illn
ess
es
in
p
la
n
t
leav
es.
T
h
e
s
tu
d
y
'
s
m
ain
f
o
c
u
s
is
o
n
th
e
d
ec
lin
e
in
ag
r
icu
ltu
r
al
p
r
o
d
u
ctio
n
.
A
p
lan
t
ev
en
tu
ally
d
ies
wh
en
a
d
is
ea
s
e
o
r
o
th
er
co
n
d
itio
n
p
r
ev
en
ts
i
ts
leav
es
f
r
o
m
ca
r
r
y
in
g
o
u
t
p
h
o
to
s
y
n
th
esis
,
wh
ich
is
n
ec
ess
ar
y
f
o
r
th
e
p
r
o
d
u
cti
o
n
o
f
c
h
lo
r
o
p
h
y
ll.
Vis
u
al
ev
alu
atio
n
s
o
f
d
is
o
r
d
e
r
s
ar
e
p
e
r
f
o
r
m
e
d
b
y
e
x
p
er
ts
ac
co
r
d
in
g
to
r
ec
o
g
n
ize
d
s
tan
d
ar
d
s
[
2
]
.
W
h
en
o
p
p
o
s
ed
to
au
to
m
ated
p
r
o
ce
d
u
r
es,
th
e
id
en
tific
atio
n
o
f
th
e
ailm
en
t
tak
es
m
o
r
e
tim
e.
E
x
p
er
ts
co
u
ld
o
n
ly
b
e
av
ailab
le
in
ce
r
tain
n
atio
n
s
.
I
m
ag
e
an
aly
s
is
-
b
ased
au
to
m
ated
p
lan
t d
is
ea
s
e
d
etec
tio
n
s
y
s
tem
s
m
u
s
t b
e
p
u
t in
p
lace
to
s
o
lv
e
th
is
p
r
o
b
lem
.
A
c
o
m
p
r
e
h
en
s
i
v
e
ev
alu
atio
n
o
f
th
e
co
n
d
itio
n
'
s
s
ev
er
ity
is
n
ec
ess
a
r
y
f
o
r
b
o
t
h
ac
cu
r
ate
y
ield
p
r
e
d
ictio
n
an
d
th
e
s
elec
tio
n
o
f
th
e
b
est tr
ea
tm
en
t.
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
d
a
p
tive
mu
lti
-
s
ca
le
co
n
v
o
lu
ti
o
n
a
l n
etw
o
r
k
fo
r
p
la
n
t le
a
f d
is
ea
s
e
d
etec
tio
n
a
n
d
… (
Teja
s
h
w
in
i C.
Ga
d
a
g
)
2957
I
n
ter
n
al
ca
u
s
es
o
f
p
lan
t
illn
ess
es
in
clu
d
e
p
ath
o
g
e
n
s
,
wh
ich
in
clu
d
e
m
o
l
d
s
,
v
ir
u
s
es,
an
d
f
u
n
g
u
s
.
On
th
e
o
th
e
r
h
an
d
,
en
v
ir
o
n
m
en
tal
elem
en
ts
in
clu
d
i
n
g
p
r
ec
i
p
itatio
n
,
h
u
m
id
ity
,
an
d
tem
p
er
atu
r
e
m
ay
also
co
n
tr
ib
u
te
to
th
eir
o
cc
u
r
r
e
n
c
e
[
3
]
.
T
h
ese
d
is
ea
s
es
ca
n
h
av
e
a
s
u
b
s
tan
tial
f
in
an
cial
i
m
p
ac
t
o
n
f
ar
m
er
s
'
liv
elih
o
o
d
s
.
C
r
o
p
s
ar
e
s
er
io
u
s
ly
th
r
ea
ten
e
d
b
y
p
lan
t
d
is
ea
s
es,
wh
ich
m
ig
h
t
r
esu
lt
in
a
f
o
o
d
ca
tast
r
o
p
h
e.
T
h
e
s
ec
u
r
ity
an
d
s
u
s
tain
ab
ilit
y
o
f
h
u
m
a
n
f
o
o
d
s
y
s
tem
s
ar
e
i
m
p
ac
ted
b
y
th
is
is
s
u
e.
Fo
r
ag
r
icu
ltu
r
al
d
is
ea
s
e
m
an
ag
em
en
t
an
d
in
ter
v
en
tio
n
to
b
e
ef
f
ec
tiv
e,
p
lan
t
illn
e
s
s
es
m
u
s
t
b
e
id
en
tifie
d
p
r
o
m
p
tly
.
Plan
t
d
is
ea
s
e
d
iag
n
o
s
is
r
eq
u
i
r
es
h
u
m
a
n
s
k
il
l,
b
u
t
in
r
u
r
al
an
d
is
o
lated
p
a
r
ts
o
f
d
ev
el
o
p
in
g
n
atio
n
s
,
th
is
in
f
o
r
m
atio
n
h
as
to
b
e
im
p
r
o
v
ed
an
d
s
tan
d
ar
d
ized
.
T
h
e
d
e
v
elo
p
m
e
n
t
o
f
q
u
i
ck
an
d
p
r
ec
is
e
tech
n
iq
u
es
f
o
r
id
en
tify
i
n
g
p
la
n
t
d
is
ea
s
es
i
s
m
ad
e
ea
s
ier
b
y
ar
tific
ial
in
tellig
en
ce
(
AI
)
tech
n
o
lo
g
y
.
Mo
d
er
n
im
a
g
e
p
r
o
ce
s
s
in
g
an
d
p
atter
n
r
ec
o
g
n
itio
n
alg
o
r
ith
m
s
allo
w
f
ar
m
er
s
a
n
d
ag
r
icu
ltu
r
al
ex
p
er
ts
to
s
p
o
t
d
is
ea
s
es
u
s
in
g
a
v
a
r
iety
o
f
tech
n
i
q
u
es.
Pla
n
t d
is
e
ases
h
av
e
b
e
en
i
d
e
n
t
if
i
ed
u
s
in
g
a
v
a
r
ie
ty
o
f
m
e
th
o
d
s
[
4
]
.
B
y
u
s
i
n
g
ML
a
n
d
DL
te
ch
n
i
q
u
es to
a
n
al
y
ze
v
is
u
al
d
at
a,
m
o
d
els
m
ay
b
e
cr
ea
te
d
t
o
d
ete
ct
p
l
a
n
t
i
ll
n
ess
es.
R
ec
e
n
t
d
e
v
el
o
p
m
e
n
ts
in
a
g
r
i
c
u
lt
u
r
e
h
av
e
co
n
c
e
n
tr
at
ed
o
n
u
s
i
n
g
d
i
f
f
e
r
e
n
t
D
L
tec
h
n
i
q
u
es
t
o
tac
k
l
e
a
r
an
g
e
o
f
p
r
o
b
le
m
s
.
I
n
s
ec
t d
ete
cti
o
n
,
f
r
u
i
t a
n
d
d
is
e
ase
id
e
n
t
if
ica
ti
o
n
,
p
la
n
t
le
af
c
ate
g
o
r
iz
ati
o
n
,
a
n
d
l
ea
f
d
is
ea
s
e
i
d
e
n
ti
f
i
ca
t
io
n
a
r
e
a
m
o
n
g
t
h
e
d
i
f
f
i
cu
l
ties
.
W
h
e
n
u
s
in
g
co
n
v
e
n
ti
o
n
a
l
ML
t
ec
h
n
i
q
u
es
f
o
r
r
ea
l
-
tim
e
d
is
e
ase
d
ia
g
n
o
s
i
s
,
a
n
u
m
b
e
r
o
f
c
h
a
lle
n
g
es
n
e
ed
t
o
b
e
o
v
e
r
c
o
m
e.
E
n
g
i
n
e
e
r
s
ca
n
th
u
s
o
v
er
co
m
e
t
h
es
e
o
b
s
ta
cles
a
n
d
d
ev
el
o
p
t
h
e
a
g
r
ic
u
lt
u
r
e
i
n
d
u
s
tr
y
t
h
a
n
k
s
t
o
DL
tec
h
n
i
q
u
es
.
I
t is cr
u
cial
to
g
ath
er
a
r
an
g
e
o
f
p
h
o
to
s
s
h
o
win
g
v
ar
io
u
s
p
lan
t sectio
n
s
in
o
r
d
er
to
cr
ea
te
a
m
o
d
el
th
at
ca
n
r
ec
o
g
n
ize
p
lan
t
illn
ess
es
[
5
]
.
T
h
e
leav
es
o
f
a
p
lan
t
ar
e
t
h
e
m
ain
ar
ea
u
s
ed
to
d
eter
m
in
e
wh
eth
er
d
is
ea
s
e
is
p
r
esen
t.
Var
iatio
n
s
in
leaf
s
h
a
p
e,
tex
tu
r
e,
co
lo
r
,
p
ictu
r
e
n
o
is
e,
an
d
o
t
h
er
c
h
ar
ac
ter
is
tics
m
ig
h
t
co
n
f
o
u
n
d
im
ag
e
p
r
o
ce
s
s
in
g
tech
n
iq
u
es,
w
h
ich
ar
e
u
s
ef
u
l
f
o
r
r
ec
o
g
n
izin
g
p
l
an
t
illn
ess
es.
I
t
is
p
o
s
s
ib
le
to
id
en
tify
th
e
m
ain
s
ig
n
s
o
f
illn
ess
b
y
m
eth
o
d
ically
ex
am
in
in
g
p
h
o
to
s
o
f
p
lan
t
l
ea
v
es.
Sev
er
al
ML
alg
o
r
ith
m
s
h
av
e
s
h
o
wn
to
b
e
an
ef
f
ec
tiv
e
m
eth
o
d
f
o
r
au
t
o
m
ated
p
lan
t
d
is
ea
s
e
id
en
tific
atio
n
in
p
r
ec
is
io
n
ag
r
icu
ltu
r
e.
Plan
t
d
is
ea
s
e
s
h
av
e
b
ee
n
ca
teg
o
r
ized
an
d
id
en
tifi
ed
u
s
in
g
a
v
ar
iety
o
f
ML
tech
n
iq
u
es,
s
u
ch
as
s
u
p
p
o
r
t
v
e
cto
r
m
ac
h
in
es
an
d
K
-
m
ea
n
s
clu
s
ter
in
g
.
Po
ten
tial
p
r
o
b
lem
s
in
th
e
p
r
e
-
p
r
o
ce
s
s
in
g
an
d
f
ea
tu
r
e
ex
tr
ac
tio
n
p
h
ases
r
estrict
th
e
ef
f
icac
y
o
f
th
ese
tech
n
i
q
u
es
i
n
r
ea
l
-
tim
e
illn
ess
d
iag
n
o
s
is
.
C
h
an
g
es
to
t
r
ad
itio
n
al
ML
tec
h
n
iq
u
es
a
r
e
r
eq
u
ir
e
d
d
u
e
to
th
e
in
tr
icac
y
an
d
u
n
p
r
e
d
ictab
ilit
y
o
f
r
ea
l
-
wo
r
ld
d
etec
tin
g
s
itu
atio
n
s
.
A
v
ar
iety
o
f
il
ln
ess
es
m
ay
af
f
ec
t
cr
o
p
s
,
an
d
ea
c
h
o
n
e
h
as
its
o
wn
ch
ar
ac
ter
is
tics
.
T
h
er
e
ar
e
s
ev
er
al
o
b
s
tacle
s
in
th
e
way
o
f
cr
ea
tin
g
a
r
eliab
le
s
y
s
tem
th
at
ca
n
d
etec
t
d
if
f
er
e
n
t
ty
p
es
o
f
illn
ess
es
in
a
v
ar
iet
y
o
f
cr
o
p
s
p
ec
ies.
I
t
is
cr
u
cial
to
p
er
s
o
n
ally
c
h
ec
k
lar
g
e
ag
r
icu
ltu
r
al
f
ield
s
f
o
r
s
ig
n
s
o
f
illn
ess
o
n
a
r
e
g
u
lar
b
asis
[
6
]
.
C
itru
s
,
r
ice,
ca
s
s
av
a,
Plan
tVil
lag
e,
iB
ea
n
,
an
d
AI
C
h
allen
g
er
2
0
1
8
ar
e
th
e
m
o
s
t
well
-
k
n
o
wn
p
lan
t
d
is
ea
s
e
d
ataset
s
th
at
co
n
tain
lab
o
r
ato
r
y
im
ag
es.
T
h
e
d
atasets
wer
e
u
s
ed
to
tr
ain
th
e
co
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
(
C
NN)
,
wh
ich
m
ad
e
it
ea
s
ier
to
id
e
n
tify
a
n
d
ca
teg
o
r
ize
p
lan
t
illn
ess
es.
T
h
e
n
eu
r
a
l
n
etwo
r
k
s
u
s
ed
o
n
th
ese
d
atasets
d
em
o
n
s
tr
ated
a
n
o
tab
le
d
e
g
r
ee
o
f
class
if
icatio
n
ac
cu
r
ac
y
th
r
o
u
g
h
o
u
t
th
e
tr
ain
in
g
s
tag
e.
Ho
wev
er
,
wh
en
test
ed
in
r
ea
l
-
wo
r
ld
s
ce
n
ar
io
s
,
th
ese
g
ad
g
ets'
ef
f
icac
y
d
r
am
atica
lly
d
ec
lin
ed
.
Un
lik
e
lab
o
r
ato
r
y
p
h
o
to
g
r
a
p
h
y
,
f
ield
p
h
o
to
g
r
ap
h
y
h
as
m
o
r
e
in
tr
icat
e
b
ac
k
g
r
o
u
n
d
co
m
p
o
n
e
n
ts
in
clu
d
in
g
ex
tr
a
leav
es,
s
tem
s
,
f
r
u
its
,
s
o
il,
an
d
m
u
lch
.
T
h
e
m
ain
ca
u
s
e
o
f
d
ec
lin
in
g
p
er
f
o
r
m
a
n
ce
in
o
u
t
d
o
o
r
p
h
o
to
g
r
ap
h
y
,
ac
co
r
d
in
g
to
r
esear
ch
,
is
co
m
p
licated
b
ac
k
d
r
o
p
s
.
E
x
clu
d
i
n
g
t
h
ese
b
ac
k
d
r
o
p
s
h
as
b
ee
n
s
h
o
wn
t
o
im
p
r
o
v
e
th
e
ac
cu
r
ac
y
o
f
s
ick
n
ess
d
etec
tio
n
[
7
]
.
7
5
9
p
h
o
to
s
o
f
b
o
th
h
ea
lth
y
an
d
s
ick
c
itru
s
f
r
u
its
an
d
f
o
liag
e
m
ak
e
u
p
th
e
citr
u
s
d
ataset.
T
h
e
in
f
o
r
m
atio
n
in
clu
d
es
d
is
ea
s
es
in
clu
d
in
g
m
elan
o
s
e,
ca
n
k
er
,
s
ca
b
,
g
r
ee
n
in
g
,
a
n
d
b
la
ck
s
p
o
t
t
h
at
im
p
ac
t
citr
u
s
p
lan
ts
.
u
s
in
g
th
e
h
elp
o
f
an
ex
p
e
r
t,
th
e
p
h
o
to
g
r
a
p
h
s
i
n
th
e
d
ataset
wer
e
m
a
n
u
ally
tak
en
u
s
in
g
a
d
ig
ital
s
in
g
le
-
len
s
r
ef
lex
(
DSLR).
Acc
o
r
d
in
g
to
r
ec
en
t
r
esear
ch
,
th
e
n
ew
DL
ap
p
r
o
ac
h
o
u
tp
er
f
o
r
m
s
co
n
v
en
tio
n
al
h
an
d
-
e
n
g
in
ee
r
ed
f
ea
tu
r
e
ap
p
r
o
ac
h
es
wh
en
d
ata
v
o
lu
m
e
e
x
ce
ed
s
a
ce
r
tain
t
h
r
e
s
h
o
ld
.
T
h
e
u
s
e
o
f
C
NN
tech
n
i
q
u
es
m
ig
h
t
m
a
k
e
it
ea
s
ier
to
ex
clu
d
e
im
p
o
r
ta
n
t
s
o
u
r
ce
s
th
at
s
ig
n
if
ican
tly
af
f
ec
t
t
h
e
m
ajo
r
elem
en
ts
s
h
o
wn
in
p
ictu
r
es.
I
n
o
r
d
e
r
to
p
r
o
p
er
l
y
d
iag
n
o
s
e
a
v
a
r
iety
o
f
p
lan
t
d
is
ea
s
es,
r
ec
en
t
r
esear
ch
h
as
f
o
cu
s
ed
o
n
d
e
v
elo
p
m
en
ts
in
DL
,
im
ag
e
f
u
s
io
n
tech
n
iq
u
es,
en
co
d
er
-
d
ec
o
d
er
n
etwo
r
k
d
esig
n
s
,
an
d
co
m
p
u
ter
v
is
io
n
ap
p
r
o
ac
h
es
[
8
]
.
Dete
ctio
n
tech
n
iq
u
es
b
ased
o
n
v
is
io
n
im
p
r
o
v
e
p
r
ec
is
io
n
an
d
d
e
p
en
d
a
b
ilit
y
in
ag
r
icu
ltu
r
al
ap
p
licatio
n
s
.
I
t
is
an
ticip
ated
th
at
th
e
a
p
p
licatio
n
o
f
th
ese
te
ch
n
iq
u
es
will
im
p
r
o
v
e
cr
o
p
q
u
ality
,
r
e
d
u
ce
la
b
o
r
ex
p
e
n
s
es,
an
d
elim
in
ate
tim
e
in
ef
f
icien
cies.
T
h
e
ap
p
licatio
n
o
f
s
u
itab
le
m
a
n
ag
em
en
t
tec
h
n
iq
u
es
im
p
r
o
v
es
th
e
ef
f
icac
y
o
f
d
is
ea
s
e
co
n
tr
o
l,
esp
ec
ially
wh
en
g
u
id
ed
b
y
p
r
io
r
ex
p
er
ien
ce
s
in
an
i
n
f
ec
te
d
s
ettin
g
.
Plan
t
d
is
ea
s
e
id
en
t
if
icatio
n
b
y
im
ag
e
an
aly
s
is
is
a
m
u
lti
-
s
tep
p
r
o
ce
d
u
r
e
with
a
n
u
m
b
er
o
f
o
b
s
tacle
s
to
o
v
er
co
m
e.
T
h
ese
p
r
o
b
lem
s
in
clu
d
e
illn
ess
e
s
with
s
im
ilar
s
y
m
p
to
m
s
,
th
e
p
o
s
s
ib
ilit
y
o
f
m
an
y
d
is
ea
s
es
af
f
ec
tin
g
th
e
s
am
e
c
r
o
p
,
a
n
d
s
y
m
p
to
m
s
with
d
if
f
er
en
t
v
is
u
al
f
ea
tu
r
es
[
9
]
.
Pictu
r
e
p
r
o
ce
s
s
in
g
m
eth
o
d
s
in
clu
d
in
g
f
ea
t
u
r
e
e
n
g
in
ee
r
i
n
g
,
p
r
ep
r
o
ce
s
s
in
g
,
class
if
icatio
n
,
an
d
p
ictu
r
e
s
eg
m
en
tatio
n
h
a
v
e
b
ee
n
u
s
ed
i
n
r
ec
en
t
b
r
ea
k
th
r
o
u
g
h
s
in
th
e
d
iag
n
o
s
is
o
f
p
lan
t
d
is
ea
s
es.
Pictu
r
e
f
u
s
io
n
alg
o
r
ith
m
s
h
av
e
b
ee
n
in
v
esti
g
ated
b
y
r
esear
ch
er
s
as
a
p
o
ten
tia
l
d
is
ea
s
e
d
iag
n
o
s
is
tech
n
iq
u
e.
Fu
r
th
er
m
o
r
e,
s
o
p
h
i
s
ticated
C
NN
-
b
ased
DL
tech
n
iq
u
es
wer
e
wid
ely
u
s
ed
b
y
r
es
ea
r
ch
er
s
to
id
e
n
tify
p
lan
t
illn
ess
es
[
1
0
]
.
T
h
e
s
tu
d
y
team
f
o
cu
s
ed
o
n
im
p
r
o
v
in
g
th
e
g
r
o
u
p
'
s
ca
teg
o
r
izatio
n
ac
cu
r
ac
y
,
wh
ich
p
r
o
d
u
ce
d
f
a
v
o
r
a
b
le
r
esu
lts
.
On
th
e
o
th
er
h
an
d
,
it
will
class
if
y
th
e
d
if
f
e
r
en
t
illn
ess
es
th
at
af
f
ec
t
a
s
in
g
le
leaf
o
r
s
ev
er
al
ty
p
es.
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.
15
,
No
.
3
,
J
u
n
e
20
26
:
2
9
5
6
-
2
9
6
9
2958
T
h
e
m
o
tiv
atio
n
f
o
r
th
is
r
esear
ch
lies
in
ad
d
r
ess
in
g
th
e
p
r
ess
in
g
ch
allen
g
es
o
f
p
lan
t
d
is
ea
s
e
d
etec
tio
n
,
wh
ich
is
cr
itical
f
o
r
en
s
u
r
i
n
g
cr
o
p
r
esil
ien
ce
,
p
r
o
d
u
cti
v
ity
,
an
d
f
o
o
d
s
ec
u
r
ity
in
m
o
d
er
n
ag
r
ic
u
ltu
r
e.
T
r
ad
itio
n
al
d
ia
g
n
o
s
tic
m
eth
o
d
s
ar
e
co
n
s
tr
ain
ed
b
y
lim
itatio
n
s
s
u
ch
as
tim
e
-
in
ten
s
iv
e
p
r
o
ce
s
s
es,
d
ep
en
d
en
c
y
o
n
ex
p
er
t
a
v
ailab
ilit
y
,
an
d
in
co
n
s
is
ten
t
o
u
tco
m
es,
p
ar
ticu
l
ar
ly
in
r
em
o
te
o
r
r
eso
u
r
ce
-
li
m
ited
r
eg
io
n
s
.
T
h
ese
lim
itatio
n
s
u
n
d
er
s
co
r
e
th
e
n
e
ed
f
o
r
a
d
v
an
ce
d
,
au
to
m
ated
,
an
d
s
ca
lab
le
s
o
lu
tio
n
s
.
L
ev
er
ag
in
g
cu
ttin
g
-
ed
g
e
d
ev
elo
p
m
e
n
ts
in
DL
an
d
co
m
p
u
ter
v
is
io
n
,
th
is
r
esear
ch
ai
m
s
to
o
v
er
co
m
e
c
h
allen
g
es
s
u
ch
as
n
o
n
-
u
n
if
o
r
m
leaf
s
tr
u
ctu
r
es,
co
m
p
lex
n
atu
r
al
b
ac
k
g
r
o
u
n
d
s
,
an
d
v
ar
y
in
g
en
v
ir
o
n
m
en
tal
co
n
d
itio
n
s
,
w
h
ich
o
f
ten
im
p
ed
e
r
ea
l
-
wo
r
ld
a
p
p
licab
ilit
y
o
f
d
is
ea
s
e
d
etec
tio
n
s
y
s
tem
s
.
T
h
is
s
tu
d
y
f
o
cu
s
es
o
n
th
e
d
ev
elo
p
m
en
t
o
f
a
r
o
b
u
s
t
an
d
o
p
tim
ized
DL
f
r
am
ewo
r
k
th
at
b
r
id
g
es th
e
g
ap
b
etwe
en
la
b
o
r
ato
r
y
-
g
r
ad
e
ac
cu
r
ac
y
an
d
f
ield
-
lev
el
d
ep
lo
y
m
en
t,
f
ac
ilit
atin
g
p
r
ec
is
e,
r
ea
l
-
tim
e
p
lan
t
d
is
ea
s
e
id
en
tific
atio
n
an
d
class
if
icatio
n
.
B
y
ad
d
r
ess
in
g
th
ese
ch
allen
g
es,
th
is
wo
r
k
co
n
tr
ib
u
tes
s
ig
n
if
ican
tly
to
th
e
ad
v
an
ce
m
e
n
t
o
f
p
r
ec
is
io
n
ag
r
icu
ltu
r
e,
en
a
b
lin
g
p
r
o
ac
tiv
e
d
is
ea
s
e
m
an
ag
em
en
t a
n
d
s
u
s
tain
ab
le
c
r
o
p
p
r
o
d
u
ctio
n
.
i)
Dev
elo
p
m
en
t
o
f
an
en
h
a
n
ce
d
DL
f
r
am
ewo
r
k
,
t
h
e
s
tu
d
y
in
tr
o
d
u
ce
s
a
r
o
b
u
s
t
ad
ap
ti
v
e
m
u
lti
-
s
ca
le
co
n
v
o
l
u
tio
n
al
n
etwo
r
k
(
AM
S
-
C
o
n
v
Net)
in
teg
r
ated
with
a
d
v
an
ce
d
atten
tio
n
m
ec
h
a
n
is
m
s
s
u
ch
as
th
e
m
u
lti
-
d
o
m
ain
atten
tio
n
f
r
am
e
wo
r
k
(
MD
AF)
a
n
d
t
h
e
ad
a
p
tiv
e
s
ca
le
f
u
s
io
n
m
o
d
u
le
(
A
SF
M)
.
T
h
ese
in
n
o
v
atio
n
s
s
ig
n
if
ican
tly
im
p
r
o
v
e
f
ea
tu
r
e
ex
tr
ac
tio
n
,
f
o
cu
s
in
g
o
n
d
iv
er
s
e
s
ca
les
o
f
lesi
o
n
c
h
ar
ac
ter
is
tics
an
d
ad
d
r
ess
in
g
ch
allen
g
es p
o
s
ed
b
y
c
o
m
p
lex
n
atu
r
al
b
ac
k
g
r
o
u
n
d
s
in
p
lan
t d
is
ea
s
e
d
etec
tio
n
.
ii)
I
n
teg
r
atio
n
o
f
ad
v
an
ce
d
im
ag
e
p
r
o
ce
s
s
in
g
tech
n
iq
u
es,
th
e
p
r
o
p
o
s
ed
m
eth
o
d
in
co
r
p
o
r
at
es
au
g
m
en
ted
co
n
v
o
l
u
tio
n
al
b
lo
c
k
atten
tio
n
m
o
d
u
les
an
d
d
ep
th
wis
e
s
ep
ar
a
b
le
co
n
v
o
lu
tio
n
s
,
o
p
tim
izin
g
c
o
m
p
u
tatio
n
al
ef
f
icien
cy
wh
ile
m
ain
tain
in
g
h
ig
h
ac
cu
r
ac
y
.
T
h
ese
m
o
d
if
icatio
n
s
ef
f
ec
tiv
ely
e
n
h
an
ce
d
is
ea
s
e
class
if
icatio
n
p
er
f
o
r
m
an
ce
,
es
p
ec
ially
in
r
ea
l
-
f
ield
co
n
d
itio
n
s
wh
er
e
ex
is
tin
g
ap
p
r
o
ac
h
es
o
f
ten
f
ail
d
u
e
to
en
v
ir
o
n
m
en
tal
co
m
p
lex
ities
.
iii)
E
x
ten
s
iv
e
ev
alu
atio
n
an
d
ad
a
p
tatio
n
f
o
r
p
r
ac
tical
d
ep
lo
y
m
en
t
,
th
e
p
r
o
p
o
s
ed
f
r
am
ewo
r
k
is
th
o
r
o
u
g
h
ly
v
alid
ated
o
n
m
u
ltip
le
d
atasets
,
s
h
o
wca
s
in
g
s
ig
n
if
ican
t
im
p
r
o
v
em
en
ts
in
p
r
ec
is
io
n
,
r
ec
all,
F1
-
s
co
r
e,
an
d
o
v
er
all
ac
cu
r
ac
y
c
o
m
p
ar
e
d
t
o
s
tate
-
of
-
th
e
-
ar
t
m
o
d
els.
Fu
r
th
er
m
o
r
e,
th
e
m
eth
o
d
is
d
esig
n
ed
with
s
ca
lab
ilit
y
an
d
r
o
b
u
s
tn
ess
,
e
n
s
u
r
in
g
its
ad
a
p
tab
ilit
y
to
r
e
al
-
f
ield
co
n
d
itio
n
s
,
th
e
r
eb
y
a
d
v
an
cin
g
th
e
p
r
ac
tical
im
p
lem
en
tatio
n
o
f
au
to
m
ated
p
lan
t
d
is
ea
s
e
d
etec
tio
n
s
y
s
tem
s
in
p
r
ec
is
io
n
ag
r
ic
u
ltu
r
e.
T
h
e
r
esear
ch
wo
r
k
is
o
r
g
an
i
ze
d
in
4
s
ec
tio
n
s
:
th
e
f
ir
s
t
s
ec
tio
n
g
iv
es
a
b
r
ief
d
escr
ip
ti
o
n
o
f
th
e
o
v
er
v
iew
an
d
b
ac
k
g
r
o
u
n
d
o
f
p
lan
t
d
is
ea
s
e
d
etec
tio
n
an
d
class
if
icatio
n
.
T
h
e
s
ec
o
n
d
s
e
ctio
n
g
iv
es
a
b
r
ief
d
escr
ip
tio
n
o
f
th
e
r
elate
d
wo
r
k
.
T
h
e
t
h
ir
d
s
ec
tio
n
d
is
cu
s
s
es
th
e
p
r
o
p
o
s
ed
m
et
h
o
d
o
lo
g
y
.
T
h
e
f
o
u
r
th
s
ec
tio
n
d
is
cu
s
s
es
th
e
p
er
f
o
r
m
an
ce
ev
a
lu
atio
n
wh
er
e
th
e
r
esu
lts
ar
e
d
is
p
lay
ed
in
th
e
f
o
r
m
o
f
g
r
a
p
h
an
d
tab
les.
T
h
e
last
s
ec
tio
n
co
n
clu
d
es th
is
wo
r
k
.
2.
RE
L
AT
E
D
WO
RK
T
h
er
e
h
as
b
ee
n
p
l
e
n
t
y
o
f
r
el
ate
d
w
o
r
k
d
e
v
e
lo
p
e
d
i
n
t
h
e
p
ast
f
o
r
i
n
d
i
v
i
d
u
al
le
af
d
is
ea
s
e
d
e
te
c
tio
n
u
s
i
n
g
th
e
DL
a
n
d
s
o
m
e
o
f
t
h
e
m
i
n
c
l
u
d
e
d
m
u
l
ti
p
le
p
l
a
n
ts
;
co
n
s
i
d
er
in
g
o
u
r
r
es
ea
r
c
h
s
o
m
e
o
f
th
e
m
ar
e
d
is
c
u
s
s
e
d
.
T
h
e
d
il
ate
d
att
e
n
ti
o
n
co
n
v
o
l
u
ti
o
n
-
Pad
d
l
ePa
d
d
le
y
o
u
o
n
l
y
lo
o
k
o
n
c
e
e
n
h
a
n
ce
d
p
l
u
s
(
DAC
-
PP
YOL
O
E
)
m
o
d
el
was
cr
ea
ted
b
y
[
1
1
]
to
in
cr
ea
s
e
th
e
ac
cu
r
ac
y
o
f
ap
p
le
p
est
d
iag
n
o
s
is
u
n
d
er
d
if
f
icu
lt
cir
cu
m
s
tan
ce
s
.
T
h
is
m
o
d
el
r
ec
o
g
n
izes
s
m
all
o
b
jects
u
s
in
g
b
o
t
h
s
h
allo
w
an
d
d
ee
p
f
ea
tu
r
e
m
ap
s
u
s
in
g
an
a
d
ap
tiv
e
f
ea
tu
r
e
f
u
s
io
n
m
eth
o
d
th
at
co
m
b
i
n
es
d
ee
p
s
ep
ar
ab
le
co
n
v
o
lu
tio
n
an
d
r
esid
u
al
co
n
n
ec
tiv
ity
.
Qi
et
a
l.
[
1
2
]
p
r
es
en
ted
th
e
en
h
a
n
ce
d
s
q
u
ee
ze
-
an
d
-
e
x
citatio
n
(
SE)
-
YOL
Ov
5
n
etwo
r
k
f
o
r
t
o
m
at
o
d
is
ea
s
e
an
d
p
est
d
etec
tio
n
.
T
h
e
SE
atten
tio
n
m
ec
h
an
is
m
o
v
er
co
m
es
th
e
li
m
itatio
n
s
o
f
ex
is
tin
g
f
ea
tu
r
e
s
cr
ee
n
in
g
a
n
d
m
o
d
el
g
e
n
er
aliza
tio
n
m
eth
o
d
s
b
y
en
h
an
cin
g
cr
itical
f
ea
tu
r
e
ex
t
r
ac
tio
n
.
De
n
g
et
a
l.
[
1
3
]
d
ev
elo
p
ed
a
n
e
n
h
an
ce
d
r
eg
io
n
-
b
ased
co
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
(
R
-
C
NN
)
m
o
d
el
em
p
lo
y
in
g
f
ed
er
ate
d
lear
n
in
g
(
FL)
to
s
o
lv
e
is
s
u
es
with
d
ata
im
b
alan
ce
,
v
ar
iety
,
a
n
d
c
o
m
p
lex
d
etec
tio
n
s
ettin
g
s
in
tr
a
d
itio
n
al
p
lan
t
d
is
ea
s
e
an
d
p
est
d
etec
tio
n
.
T
h
is
p
ar
ad
ig
m
r
ed
u
ce
s
co
m
m
u
n
icatio
n
an
d
d
ata
s
to
r
ag
e
ex
p
e
n
s
es
b
y
lev
er
ag
i
n
g
FL'
s
d
is
tr
ib
u
ted
co
m
p
u
ti
n
g
ch
ar
ac
ter
is
tics
.
Ad
d
itio
n
ally
,
R
esNet
-
1
0
1
en
h
an
ce
s
m
u
ltis
ca
le
d
etec
tio
n
ac
cu
r
ac
y
f
o
r
a
r
a
n
g
e
o
f
d
is
ea
s
es
an
d
p
ests
b
y
s
u
b
s
titu
tin
g
v
is
u
al
g
e
o
m
etr
y
g
r
o
u
p
-
1
6
(
VGG
-
16
)
in
th
e
co
n
v
o
lu
tio
n
al
lay
er
.
T
o
in
c
r
ea
s
e
th
e
d
etec
tio
n
ac
cu
r
ac
y
o
f
m
aize
d
is
ea
s
es,
I
r
ian
to
an
d
Fin
d
ley
[
1
4
]
p
r
o
p
o
s
ed
a
tech
n
i
q
u
e
f
o
r
c
o
r
n
leaf
d
is
ea
s
e
d
iag
n
o
s
is
u
tili
zin
g
f
u
zz
y
c
-
m
ea
n
s
(
F
C
M)
an
d
lo
n
g
s
h
o
r
t
-
ter
m
m
em
o
r
y
(
L
STM
)
alg
o
r
ith
m
s
.
T
h
e
g
r
ay
lev
el
co
-
o
cc
u
r
r
en
ce
m
atr
i
x
was
u
s
ed
to
ex
tr
ac
t
tex
tu
r
e
in
f
o
r
m
atio
n
f
r
o
m
illn
ess
p
ictu
r
es.
T
h
e
L
STM
alg
o
r
ith
m
was
th
en
u
s
ed
to
class
if
y
th
ese
t
ex
tu
r
e
p
r
o
p
er
ties
.
T
h
e
ac
cu
r
ac
y
r
atin
g
o
b
tain
ed
b
y
u
s
in
g
th
is
s
tr
ateg
y
was
8
0
.
2
4
%.
C
h
an
g
et
a
l.
[
1
5
]
c
r
e
ated
an
ed
g
e
f
ea
tu
r
e
g
u
id
an
ce
(
E
FG)
m
o
d
u
le
to
e
n
h
an
ce
th
e
m
o
d
el'
s
ab
ilit
y
to
ex
tr
ac
t
lo
ca
l
ed
g
e
f
ea
tu
r
es
in
o
r
d
e
r
to
g
et
o
v
e
r
th
e
c
h
allen
g
es
p
r
esen
ted
b
y
c
o
m
p
lex
p
lan
t
d
is
ea
s
e
ch
ar
ac
ter
is
tics
an
d
s
p
ar
s
e
d
at
asets
.
T
h
e
o
v
er
all
p
er
f
o
r
m
an
ce
o
f
t
h
e
m
o
d
el
m
ay
b
e
im
p
r
o
v
ed
b
y
in
clu
d
i
n
g
m
u
ltis
ca
le
f
ea
tu
r
es
an
d
e
d
g
e
in
f
o
r
m
atio
n
b
y
in
teg
r
atin
g
th
e
E
FG
m
o
d
u
le
in
to
v
is
io
n
tr
an
s
f
o
r
m
er
s
lik
e
ViT
an
d
Swin
.
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
d
a
p
tive
mu
lti
-
s
ca
le
co
n
v
o
lu
ti
o
n
a
l n
etw
o
r
k
fo
r
p
la
n
t le
a
f d
is
ea
s
e
d
etec
tio
n
a
n
d
… (
Teja
s
h
w
in
i C.
Ga
d
a
g
)
2959
Paca
l
et
a
l.
[
1
6
]
th
e
d
iag
n
o
s
is
o
f
p
lan
t
d
is
ea
s
es
h
as
b
ee
n
tr
an
s
f
o
r
m
ed
b
y
r
ec
e
n
t
ad
v
a
n
c
es
in
DL
,
wh
ich
h
av
e
also
in
tr
o
d
u
ce
d
n
ew
m
eth
o
d
s
f
o
r
a
v
ar
iety
o
f
cr
o
p
s
,
in
clu
d
in
g
m
aize
,
p
o
tato
es,
to
m
ato
es,
s
u
g
ar
ca
n
e,
a
n
d
g
r
o
u
n
d
n
u
ts
.
Acc
o
r
d
in
g
to
r
esear
ch
,
AI
-
p
o
wer
ed
m
o
d
els
ca
n
ef
f
ec
tiv
el
y
ad
d
r
ess
s
ig
n
if
ican
t
ag
r
icu
ltu
r
al
is
s
u
es
in
clu
d
in
g
th
e
p
r
o
m
p
t
a
n
d
p
r
ec
is
e
id
en
t
if
icatio
n
o
f
d
is
ea
s
es.
A
th
o
r
o
u
g
h
ass
ess
m
en
t
o
f
m
o
r
e
th
an
1
6
0
s
tu
d
ies
o
n
DL
-
b
ased
p
lan
t
d
is
ea
s
e
d
etec
tio
n
h
ig
h
lig
h
ted
th
e
n
ee
d
o
f
d
iv
er
s
e,
h
ig
h
-
q
u
ality
d
atasets
in
d
ev
elo
p
i
n
g
r
eliab
le
m
o
d
els
t
h
at
ca
n
d
etec
t
d
is
ea
s
es
ea
r
ly
.
W
ith
th
is
m
eth
o
d
,
f
a
r
m
e
r
s
a
n
d
ag
r
i
cu
lt
u
r
al
e
x
p
e
r
ts
m
a
y
e
m
p
l
o
y
q
u
i
ck
a
n
d
ef
f
i
cie
n
t
m
a
n
a
g
e
m
e
n
t
s
tr
ate
g
i
es.
R
ea
l
-
ti
m
e
d
is
e
ase
d
e
tec
ti
o
n
s
y
s
te
m
s
b
as
ed
o
n
m
o
b
il
e
d
e
v
i
ce
s
h
a
v
e
em
er
g
ed
as
p
r
o
m
is
i
n
g
al
te
r
n
ati
v
es,
p
a
r
t
i
cu
l
a
r
l
y
f
o
r
m
aiz
e.
Kh
a
n
et
a
l.
[
1
7
]
d
em
o
n
s
tr
ated
h
o
w
t
o
in
te
g
r
ate
o
b
ject
r
ec
o
g
n
itio
n
an
d
tr
a
n
s
f
er
lear
n
in
g
m
eth
o
d
s
to
d
ev
elo
p
a
s
y
s
tem
th
at
ef
f
ec
tiv
ely
d
etec
ts
d
is
ea
s
es
in
m
aize
leav
es.
T
h
e
s
e
in
s
tr
u
m
en
ts
en
h
an
ce
d
is
ea
s
e
m
an
ag
e
m
en
t
a
n
d
ex
p
ed
ite
f
ield
d
ec
is
io
n
-
m
ak
in
g
b
y
p
r
o
v
id
i
n
g
f
ar
m
er
s
with
in
s
tan
t
f
ee
d
b
ac
k
.
T
h
e
co
m
b
in
a
tio
n
o
f
AI
an
d
th
e
in
ter
n
et
o
f
th
in
g
s
(
I
o
T
)
h
a
v
e
s
ig
n
if
ican
tly
alter
e
d
lar
g
e
-
s
ca
le
ag
r
icu
ltu
r
al
s
u
r
v
eillan
ce
.
Dr
o
n
es
eq
u
ip
p
ed
with
ad
v
an
ce
d
im
ag
in
g
tech
n
o
lo
g
y
an
d
AI
m
i
g
h
t
s
wif
tly
an
d
p
r
ec
is
ely
id
en
tify
d
is
ea
s
es
o
v
er
v
ast
ag
r
ic
u
ltu
r
al
ar
ea
s
,
o
p
tim
izin
g
r
eso
u
r
ce
u
til
izatio
n
an
d
r
e
d
u
cin
g
cr
o
p
lo
s
s
es.
Su
g
ar
ca
n
e
illn
ess
es
ar
e
b
ein
g
m
an
a
g
ed
with
th
e
u
s
e
o
f
ad
v
a
n
ce
d
AI
tech
n
o
lo
g
y
.
Mu
s
s
a
[
1
8
]
d
escr
ib
ed
a
tech
n
iq
u
e
th
at
em
p
lo
y
s
m
u
ltis
p
ec
tr
al
s
atellite
i
m
ag
er
y
an
d
AI
to
d
etec
t
r
ato
o
n
s
tu
n
tin
g
d
is
ea
s
e
in
s
u
g
ar
ca
n
e.
T
h
is
s
atellite
-
b
ased
m
eth
o
d
f
ac
ilit
ates
th
e
ea
r
ly
d
etec
tio
n
o
f
d
is
ea
s
e
h
o
ts
p
o
ts
o
n
b
ig
p
lan
tatio
n
s
b
y
p
er
m
itti
n
g
tar
g
eted
tr
ea
tm
e
n
t
s
an
d
s
to
p
p
i
n
g
t
h
e
s
p
r
ea
d
o
f
d
is
ea
s
es.
Fu
r
th
er
m
o
r
e,
ad
v
an
ce
m
en
ts
in
AI
f
r
am
ewo
r
k
s
ar
e
m
ak
in
g
it
s
i
m
p
ler
to
id
en
tify
d
is
ea
s
es
in
cr
o
p
s
lik
e
p
o
tato
es
an
d
to
m
ato
es.
R
esear
ch
er
s
ar
e
em
p
lo
y
in
g
ML
an
d
DL
m
o
d
els
to
d
etec
t
ag
r
icu
ltu
r
al
illn
e
s
s
es
in
o
r
d
er
to
s
o
lv
e
s
ig
n
if
i
ca
n
t
is
s
u
es
s
u
ch
as
v
ar
y
in
g
clim
atic
c
o
n
d
itio
n
s
an
d
d
iv
e
r
s
e
d
is
ea
s
e
p
r
ese
n
tatio
n
s
.
T
h
e
co
m
b
in
atio
n
o
f
c
u
ttin
g
-
ed
g
e
AI
tech
n
o
lo
g
y
with
p
r
ac
tical
ap
p
l
icatio
n
s
h
ig
h
lig
h
ts
th
e
p
o
ten
ti
al
o
f
th
ese
to
o
ls
to
tr
an
s
f
o
r
m
m
o
d
er
n
a
g
r
icu
ltu
r
e
.
As
AI
-
p
o
wer
ed
s
y
s
tem
s
ad
v
a
n
ce
,
th
ey
m
ig
h
t
e
n
h
an
ce
f
o
o
d
s
ec
u
r
ity
,
d
is
ea
s
e
p
r
ev
en
tio
n
,
a
n
d
cr
o
p
m
o
n
ito
r
in
g
,
f
ac
ilit
atin
g
wid
esp
r
ea
d
ad
o
p
ti
o
n
o
f
m
o
r
e
r
esil
ien
t
an
d
s
u
s
tai
n
ab
le
ag
r
icu
ltu
r
al
p
r
ac
tices.
T
h
is
s
tu
d
y
[
1
9
]
,
[
2
0
]
p
r
o
p
o
s
es
a
n
o
v
el
lig
h
tweig
h
t
d
ee
p
C
NN
m
o
d
el
to
lear
n
h
ig
h
-
lev
el
h
id
d
en
f
ea
tu
r
e
r
ep
r
esen
tatio
n
s
.
L
o
ca
l
tex
tu
r
e
in
f
o
r
m
atio
n
is
ex
tr
ac
t
ed
f
r
o
m
p
la
n
t
leaf
im
a
g
es
u
s
in
g
d
ee
p
f
ea
t
u
r
es
an
d
co
n
v
en
t
io
n
ally
co
n
s
tr
u
cted
lo
ca
l
b
in
ar
y
p
atter
n
(
L
B
P)
f
ea
tu
r
es.
T
h
r
ee
o
p
e
n
ly
ac
ce
s
s
ib
le
d
atasets
ap
p
le
leaf
,
to
m
ato
leaf
,
an
d
g
r
ap
e
leaf
ar
e
u
s
ed
to
tr
ain
a
n
d
ass
ess
t
h
e
p
r
o
p
o
s
ed
m
o
d
el.
T
h
e
p
r
o
p
o
s
ed
tech
n
iq
u
e
ac
h
iev
es
v
alid
atio
n
ac
cu
r
ac
ies
o
f
9
9
%,
9
6
.
6
%,
an
d
9
8
.
5
%
a
n
d
test
ac
cu
r
ac
ies
o
f
9
8
.
8
%,
9
6
.
5
%,
a
n
d
9
8
.
3
%
ac
r
o
s
s
th
e
th
r
ee
d
atasets
.
Pajan
y
et
a
l.
[
2
1
]
p
r
o
p
o
s
ed
m
o
d
el
en
s
em
b
le
m
eth
o
d
f
o
r
ac
c
u
r
ately
id
en
tif
y
in
g
an
d
class
if
y
in
g
p
la
n
t
d
is
ea
s
es
b
ased
o
n
f
ield
p
ictu
r
es.
T
h
e
s
eg
m
en
t
an
y
th
i
n
g
m
o
d
el
(
SAM)
allo
ws
f
o
r
th
e
r
ec
o
g
n
itio
n
an
d
d
elin
ea
tio
n
o
f
an
y
d
is
ce
r
n
ib
le
item
in
s
id
e
an
im
ag
e.
T
h
e
h
ig
h
lig
h
ted
co
m
p
o
n
en
ts
ar
e
th
en
ex
tr
ac
te
d
f
r
o
m
th
e
o
r
ig
in
al
im
ag
e
u
s
in
g
im
ag
e
p
r
o
ce
s
s
in
g
tech
n
i
q
u
es.
T
h
e
f
u
lly
co
n
v
o
lu
tio
n
al
d
ata
d
escr
ip
tio
n
(
FC
DD)
is
an
ex
p
lain
a
b
le
d
ee
p
one
-
class
class
if
icatio
n
m
o
d
el
f
o
r
an
o
m
aly
d
etec
tio
n
.
I
t
aid
s
in
d
if
f
er
en
tiatin
g
b
etwe
en
b
ac
k
g
r
o
u
n
d
o
b
jects
an
d
ac
tu
al
leaf
th
i
n
g
s
.
Fin
all
y
,
a
ca
te
g
o
r
izatio
n
m
o
d
el
cr
e
ated
b
y
Plan
tv
illag
e
is
u
s
ed
t
o
in
f
er
co
n
clu
s
io
n
s
f
r
o
m
th
e
s
elec
ted
leav
es.
T
aji
et
a
l.
[
2
2
]
r
ec
o
m
m
en
d
s
u
tili
zi
n
g
u
n
m
an
n
e
d
ae
r
ial
v
eh
icle
(
UAV)
-
b
ased
r
em
o
te
s
en
s
in
g
d
ata
in
co
m
b
in
atio
n
with
o
p
tim
al
f
u
zz
y
d
ee
p
n
eu
r
al
n
etwo
r
k
s
-
b
ased
p
lan
t
d
is
e
ase
d
etec
tio
n
an
d
class
if
icatio
n
(
OFDNN
-
PDD
C
)
to
id
en
tif
y
a
n
d
ca
teg
o
r
ize
p
lan
t
illn
ess
es.
T
h
e
OFDNN
-
PDDC
m
eth
o
d
m
ay
b
e
u
s
ed
to
e
f
f
icien
tly
id
en
tif
y
an
d
class
if
y
a
r
an
g
e
o
f
p
lan
t d
i
s
ea
s
es.
T
h
e
h
y
b
r
id
f
r
am
ewo
r
k
em
p
lo
y
ed
in
th
is
s
tu
d
y
is
b
ased
o
n
a
h
y
b
r
id
p
r
ep
r
o
ce
s
s
in
g
tech
n
iq
u
e
th
a
t
in
co
r
p
o
r
ates
an
en
s
em
b
le
f
ea
tu
r
es
en
g
in
ee
r
in
g
p
h
ase
th
at
f
o
cu
s
es
o
n
tex
tu
r
e
f
ea
tu
r
es
to
g
eth
er
with
two
d
is
tin
ct
f
o
r
m
s
o
f
d
ee
p
f
ea
tu
r
e
ex
tr
ac
tio
n
[
2
3
]
.
T
h
e
L
B
P
ch
ar
ac
ter
is
tics
ar
e
in
teg
r
ated
with
th
e
C
NN
q
u
alitie
s
.
T
h
r
ee
d
is
tin
ct
m
eta
-
h
e
u
r
is
tic
alg
o
r
ith
m
s
th
e
b
in
ar
y
d
r
ag
o
n
f
ly
alg
o
r
ith
m
(
B
DA)
,
th
e
an
t
c
o
lo
n
y
o
p
tim
izatio
n
(
AC
O)
tech
n
iq
u
e,
an
d
th
e
m
o
t
h
f
lam
e
o
p
tim
izatio
n
(
MFO)
m
eth
o
d
ar
e
u
s
ed
to
o
p
tim
ize
t
h
e
en
s
em
b
le
f
ea
tu
r
e
v
ec
to
r
.
Mo
d
er
n
ML
m
eth
o
d
s
ar
e
u
s
ed
to
class
if
y
th
e
en
h
an
ce
d
f
ea
tu
r
e
v
ec
to
r
.
3.
M
E
T
H
O
D
AM
S
-
C
o
n
v
Net
en
h
an
ce
s
th
e
s
u
p
er
io
r
f
ea
tu
r
e
ex
tr
ac
ti
o
n
ca
p
ab
ilit
ies
an
d
r
ed
u
ce
d
tr
ain
in
g
r
eq
u
ir
em
e
n
ts
.
I
t
p
er
f
o
r
m
s
b
et
ter
th
an
th
e
b
u
lk
o
f
tr
a
d
itio
n
al
C
NN
m
o
d
els
as
well
as
a
f
ew
o
f
t
r
a
n
s
f
o
r
m
er
m
o
d
els.
T
h
e
AM
S
-
C
o
n
v
Net
i
s
m
ad
e
u
p
o
f
a
f
ea
tu
r
e
e
x
tr
ac
ti
o
n
lay
er
at
th
e
h
ea
d
,
a
f
ea
tu
r
e
class
if
icatio
n
lay
er
at
th
e
en
d
,
an
d
f
o
u
r
AM
S
-
C
o
n
v
Net
b
l
o
ck
s
o
f
v
ar
y
in
g
s
izes
in
th
e
ce
n
te
r
.
T
h
e
in
tr
a
-
lay
er
ar
c
h
itectu
r
e
an
d
s
tack
in
g
tech
n
iq
u
e
o
f
AM
S
-
C
o
n
v
Net
ar
e
d
if
f
er
e
n
t
f
r
o
m
th
o
s
e
o
f
th
e
g
en
e
r
al
ar
ch
it
ec
tu
r
e.
Am
o
n
g
th
e
m
o
d
if
icatio
n
s
ar
e
th
e
f
o
llo
wi
n
g
:
i
n
ea
ch
o
f
th
e
f
o
u
r
s
tag
e
s
,
th
r
ee
,
f
o
u
r
,
s
ix
,
an
d
th
r
ee
b
lo
ck
s
ar
e
s
tack
ed
,
r
esu
ltin
g
in
a
r
atio
o
f
ar
o
u
n
d
1
:1
:2
:1
.
AM
S
-
C
o
n
v
Net
em
p
l
o
y
s
a
b
lo
c
k
s
tack
in
g
r
atio
o
f
1
:1
:3
:1
an
d
m
o
d
if
ies
th
e
s
tack
in
g
to
3
,
3
,
9
,
an
d
3
,
d
r
awin
g
in
s
p
ir
atio
n
f
r
o
m
th
e
t
r
a
n
s
f
o
r
m
er
id
ea
.
AM
S
-
C
o
n
v
Net
p
er
f
o
r
m
s
th
e
p
atch
if
y
p
r
o
ce
s
s
in
p
ar
allel,
s
u
b
s
titu
tin
g
a
4
×
4
co
n
v
o
l
u
tio
n
al
k
er
n
el
s
ize
an
d
a
s
tr
id
e
o
f
4
f
o
r
t
h
e
s
tem
.
Dim
en
s
io
n
ality
ex
p
an
s
io
n
f
o
llo
ws
d
im
en
s
io
n
alit
y
r
ed
u
ctio
n
in
th
e
b
o
ttlen
ec
k
d
esig
n
o
f
C
NN,
an
d
f
ea
tu
r
e
ex
tr
ac
tio
n
f
o
llo
ws.
Usi
n
g
an
in
v
er
ted
b
o
ttlen
e
ck
,
AM
S
-
C
o
n
v
Net
p
er
f
o
r
m
s
f
ea
tu
r
e
ex
tr
ac
tio
n
f
ir
s
t,
f
o
llo
wed
b
y
d
im
en
s
io
n
ality
r
ed
u
ctio
n
an
d
d
im
e
n
s
io
n
ality
in
cr
ea
s
e.
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.
15
,
No
.
3
,
J
u
n
e
20
26
:
2
9
5
6
-
2
9
6
9
2960
A
d
ep
th
-
s
ep
ar
ab
le
co
n
v
o
l
u
tio
n
is
also
em
p
lo
y
ed
,
wh
ich
is
co
m
p
ar
ab
le
to
t
h
e
tr
an
s
f
o
r
m
e
r
'
s
s
elf
-
atten
tio
n
an
d
th
e
ac
tiv
atio
n
f
u
n
ctio
n
is
em
p
lo
y
ed
in
f
r
eq
u
en
tly
.
I
n
th
e
n
o
r
m
alizin
g
lay
er
,
lay
er
n
o
r
m
aliz
atio
n
h
as
tak
e
n
th
e
r
o
le
o
f
b
atch
n
o
r
m
aliza
tio
n
,
a
n
d
th
e
f
r
eq
u
en
cy
o
f
n
o
r
m
aliz
atio
n
ap
p
licatio
n
s
h
as
d
ec
r
ea
s
ed
.
AM
S
-
C
o
n
v
Net
in
co
r
p
o
r
ates a
d
o
wn
s
am
p
lin
g
lay
er
with
lay
er
n
o
r
m
aliza
tio
n
an
d
a
co
n
v
o
lu
tio
n
o
p
er
atio
n
with
a
k
er
n
el
s
ize
o
f
2
an
d
a
s
tr
id
e
o
f
2
at
th
e
s
am
e
tim
e.
AM
S
-
C
o
n
v
Net
em
p
lo
y
s
a
7
×
7
co
n
v
o
l
u
tio
n
k
er
n
el
r
ath
er
th
an
t
h
e
3
×
3
co
n
v
o
lu
tio
n
k
er
n
el.
3
.
1
.
Arc
hite
ct
ure
T
h
e
b
ac
k
g
r
o
u
n
d
o
f
th
e
p
lan
t
d
is
ea
s
e
d
ataset
is
co
m
p
lex
,
an
d
ch
ar
ac
te
r
is
tics
lik
e
tin
y
,
d
is
p
er
s
ed
d
is
ea
s
e
s
p
o
t
p
atch
es
m
ak
e
d
is
ea
s
e
class
if
icatio
n
m
o
r
e
ch
alle
n
g
in
g
.
T
o
ad
d
r
ess
th
e,
th
is
s
tu
d
y
s
u
g
g
ests
a
h
ig
h
-
p
er
f
o
r
m
an
ce
p
lan
t
d
is
ea
s
e
class
if
icatio
n
n
etwo
r
k
th
at
m
a
k
e
s
u
s
e
o
f
an
im
p
r
o
v
ed
AM
S
-
C
o
n
v
Net
a
r
ch
itectu
r
e.
T
h
e
au
g
m
en
te
d
co
n
v
o
lu
tio
n
al
b
lo
ck
atten
tio
n
m
o
d
u
le
is
f
ir
s
t
in
clu
d
ed
af
ter
th
e
s
ec
o
n
d
1
×
1
co
n
v
o
l
u
tio
n
o
f
th
e
AM
S
-
C
o
n
v
Net
b
lo
ck
an
d
at
th
e
co
n
clu
s
io
n
o
f
ea
ch
s
tag
e
in
o
r
d
er
to
im
p
r
o
v
e
th
e
n
etw
o
r
k
'
s
em
p
h
asis
o
n
d
is
ea
s
ed
s
p
o
t
in
f
o
r
m
atio
n
an
d
less
en
th
e
in
f
lu
en
ce
o
f
u
n
r
el
ated
elem
en
ts
.
Seco
n
d
,
to
im
p
r
o
v
e
th
e
n
etwo
r
k
's
ca
p
ac
ity
to
ex
tr
ac
t
f
ea
tu
r
e
in
f
o
r
m
atio
n
r
elate
d
to
illn
ess
es
o
f
v
ar
ied
s
izes.
I
n
th
e
e
n
d
,
th
e
AM
S
-
C
o
n
v
Net
b
lo
ck
m
o
d
el'
s
r
ep
etitio
n
co
u
n
t
is
d
ec
r
ea
s
ed
in
o
r
d
er
to
lo
wer
th
e
m
o
d
el'
s
co
m
p
lex
ity
.
Fig
u
r
e
1
s
h
o
ws
th
e
AM
S
-
C
o
n
v
Net
an
d
F
ig
u
r
e
2
s
h
o
ws th
e
p
r
o
p
o
s
ed
m
o
d
el.
Fig
u
r
e
1
.
AM
S
-
C
o
n
v
Net
Fig
u
r
e
2
.
Pro
p
o
s
ed
m
o
d
el
3
.
2
.
M
ulti
-
do
m
a
in a
t
t
ent
io
n
f
ra
m
ewo
r
k
MD
AF
is
an
atten
tio
n
m
o
d
u
le
co
n
s
is
tin
g
o
f
s
p
atial
a
n
d
ch
an
n
el
m
ec
h
a
n
is
m
s
.
T
h
e
atten
tio
n
is
lev
ied
o
n
ch
an
n
el
in
f
o
r
m
atio
n
with
i
n
th
e
p
o
s
itio
n
tar
g
et
in
f
o
r
m
ati
o
n
;
th
is
is
em
b
ed
d
ed
with
in
t
h
e
C
NN
w
ith
ad
d
ed
co
m
p
u
tatio
n
.
T
h
e
ch
an
n
el
atte
n
tio
n
m
o
d
u
le
p
er
f
o
r
m
s
in
f
o
r
m
atio
n
co
m
b
in
atio
n
th
r
o
u
g
h
av
er
ag
e
m
ax
p
o
o
lin
g
o
p
er
atio
n
s
,
th
en
f
o
r
wa
r
d
s
it
t
o
a
s
h
a
r
ed
n
etwo
r
k
,
wh
ic
h
cr
ea
tes
a
ch
an
n
el
atten
tio
n
m
a
p
b
y
co
m
b
in
in
g
th
e
in
p
u
t
f
ea
t
u
r
e
m
ap
s
an
d
c
o
m
p
r
ess
in
g
th
eir
s
p
atial
d
im
e
n
s
io
n
s
,
h
o
wev
er
lik
e
th
is
th
e
ch
an
n
el
atten
tio
n
m
o
d
el
is
ev
alu
ated
.
I
n
(
1
)
d
en
o
tes
th
e
in
p
u
t
f
ea
tu
r
e
m
ap
,
d
en
o
tes
th
e
ac
tiv
atio
n
f
u
n
ctio
n
,
m
u
ltil
ay
er
p
er
ce
p
tr
o
n
s
(
ML
P)
d
ep
icts
m
u
ltil
ay
er
p
er
ce
p
tr
o
n
,
is
th
e
av
er
ag
e
p
o
o
lin
g
o
p
er
atio
n
,
is
th
e
m
ax
im
u
m
p
o
o
lin
g
o
p
er
atio
n
,
0
∗
,
1
∗
.
an
d
u
p
o
n
co
m
p
r
ess
io
n
o
f
s
p
atial
f
ea
tu
r
e
m
ap
b
y
g
l
o
b
al
m
ax
p
o
o
l a
n
d
g
lo
b
al
a
v
er
ag
e
p
o
o
l.
(
)
=
(
(
(
)
)
+
(
(
)
)
)
(
1
(
(
)
)
+
1
(
(
)
)
(
1
)
T
h
e
s
p
atial
atten
tio
n
m
o
d
u
le
co
n
s
id
er
s
th
e
im
p
o
r
tan
ce
o
f
p
ix
els
at
d
if
f
e
r
en
t
lo
ca
ti
o
n
s
i
n
th
e
s
am
e
ch
an
n
el,
c
o
m
p
r
ess
es
th
e
ch
a
n
n
el,
an
d
p
er
f
o
r
m
s
a
v
er
ag
e
p
o
o
lin
g
an
d
m
ax
im
u
m
p
o
o
li
n
g
in
t
h
e
ch
a
n
n
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
d
a
p
tive
mu
lti
-
s
ca
le
co
n
v
o
lu
ti
o
n
a
l n
etw
o
r
k
fo
r
p
la
n
t le
a
f d
is
ea
s
e
d
etec
tio
n
a
n
d
… (
Teja
s
h
w
in
i C.
Ga
d
a
g
)
2961
d
im
en
s
io
n
to
co
n
ce
n
tr
ate
th
e
in
f
o
r
m
atio
n
at
th
e
s
p
atial
lo
ca
tio
n
s
o
f
im
p
o
r
tan
t
f
ea
tu
r
es.
T
h
e
s
p
atial
atten
tio
n
m
o
d
u
le
is
ca
lcu
lated
as
(
2
)
.
(
)
=
(
ℎ
7
∗
7
(
[
(
)
,
(
)
]
)
)
(
ℎ
7
∗
7
(
[
,
]
)
)
)
(
2
)
I
n
(
2
)
d
en
o
tes
th
e
f
ea
tu
r
e
m
a
p
,
as
a
r
esu
lt
o
f
th
e
s
p
atial
atten
tio
n
m
o
d
u
le,
is
th
e
in
p
u
t
f
e
atu
r
e
m
ap
,
is
th
e
s
ig
m
o
id
f
u
n
ctio
n
,
ℎ
7
∗
7
d
ep
icts
th
e
co
n
v
o
l
u
tio
n
o
p
er
atio
n
an
d
co
n
v
o
lu
tio
n
al
k
e
r
n
el
o
f
s
iz
e
7
,
d
ep
icts
th
e
p
o
o
lin
g
o
p
e
r
atio
n
an
d
d
ep
icts
th
e
m
ax
im
u
m
p
o
o
lin
g
o
p
e
r
atio
n
.
T
h
e
MD
AF
ch
an
n
el
atten
tio
n
m
o
d
u
le
a
n
d
th
e
s
p
a
tial
atten
tio
n
m
o
d
u
le
ar
e
s
er
ially
co
n
n
ec
te
d
.
C
h
an
n
el
atte
n
tio
n
is
ca
r
r
ie
d
o
u
t
p
r
io
r
t
o
th
e
ap
p
licatio
n
o
f
s
p
atial
atten
tio
n
.
T
h
e
latter
'
s
in
p
u
t
co
m
es
f
r
o
m
th
e
p
r
ev
io
u
s
ly
alter
ed
f
e
atu
r
es,
wh
ich
will
af
f
ec
t
h
o
w
s
p
atial
f
ea
tu
r
es
a
r
e
lear
n
ed
a
n
d
m
ig
h
t
ca
u
s
e
th
e
atten
tio
n
m
o
d
u
le
t
o
f
u
n
ctio
n
er
r
atica
lly
.
ML
P
ar
e
u
s
ed
in
t
an
d
em
to
r
e
d
u
ce
th
e
d
im
en
s
io
n
ality
o
f
ch
a
n
n
el
atten
tio
n
,
wh
ich
r
esu
lts
in
a
lo
s
s
o
f
f
ea
tu
r
es
an
d
an
in
c
r
ea
s
e
in
th
e
n
u
m
b
er
o
f
p
ar
am
eter
s
.
B
y
s
witch
in
g
f
r
o
m
a
s
er
ial
co
n
n
ec
tio
n
to
a
p
a
r
allel
co
n
n
ec
tio
n
,
th
is
s
tu
d
y
ad
d
r
e
s
s
es
th
e
d
if
f
icu
lties
f
o
u
n
d
a
n
d
en
h
an
ce
s
MD
AF
.
T
h
is
e
n
s
u
r
es
th
at
ch
an
n
el
atten
tio
n
an
d
s
p
atial
atten
tio
n
in
p
u
ts
co
m
e
f
r
o
m
th
e
s
am
e
s
o
u
r
ce
,
c
o
n
tin
u
e
to
b
e
n
o
n
-
in
ter
f
er
in
g
,
a
n
d
p
r
o
v
id
e
b
etter
f
ea
tu
r
e
weig
h
t
in
f
o
r
m
at
io
n
.
T
o
in
cr
e
ase
th
e
m
o
d
el'
s
cr
o
s
s
-
ch
an
n
el
in
te
r
ac
tio
n
ca
p
ab
ilit
ies,
s
im
p
lify
th
e
alg
o
r
ith
m
,
im
p
r
o
v
e
in
f
o
r
m
ati
o
n
s
h
ar
in
g
ac
r
o
s
s
ch
an
n
els,
a
n
d
r
ed
u
ce
f
ea
t
u
r
e
d
eg
r
ad
atio
n
,
a
o
n
e
-
d
im
en
s
io
n
al
co
n
v
o
l
u
tio
n
o
p
er
atio
n
s
im
u
lta
n
eo
u
s
ly
s
er
v
es a
s
th
e
p
er
c
ep
tr
o
n
lay
er
.
T
h
e
f
e
at
u
r
e
e
x
t
r
ac
ti
o
n
p
r
o
ce
d
u
r
e
s
u
cc
ess
f
u
ll
y
r
es
o
l
v
es
t
h
e
i
n
te
r
f
e
r
e
n
c
e
p
r
o
b
le
m
s
b
r
o
u
g
h
t
o
n
b
y
t
h
e
n
et
wo
r
k
'
s
i
n
t
r
i
ca
t
e
n
at
u
r
al
en
v
i
r
o
n
m
e
n
t
b
y
all
o
w
in
g
t
h
e
n
etw
o
r
k
t
o
c
o
n
c
e
n
tr
at
e
o
n
l
esi
o
n
ch
ar
ac
t
er
is
tics
v
i
a
t
h
e
atte
n
t
io
n
m
ec
h
a
n
is
m
.
A
n
i
m
p
r
o
v
e
d
M
DAF
att
en
ti
o
n
m
o
d
u
le
th
at
is
i
n
cl
u
d
e
d
i
n
t
o
th
e
A
MS
-
C
o
n
v
Ne
t
m
o
d
el
is
s
h
o
wn
in
t
h
is
w
o
r
k
.
F
o
c
u
s
i
n
g
o
n
t
h
e
e
x
t
r
ac
ti
o
n
o
f
v
is
u
al
d
ata
,
all
o
ca
t
in
g
w
ei
g
h
ts
b
etw
ee
n
s
ic
k
p
at
ch
es
a
n
d
b
a
ck
g
r
o
u
n
d
i
n
f
o
r
m
ati
o
n
o
p
ti
m
al
ly
,
a
n
d
e
n
h
a
n
c
i
n
g
a
d
a
p
ta
tio
n
t
o
a
v
a
r
ie
ty
o
f
c
o
n
te
x
ts
ar
e
al
l
c
r
u
ci
al
f
o
r
im
p
r
o
v
i
n
g
n
e
tw
o
r
k
i
d
e
n
ti
f
i
ca
t
i
o
n
ac
c
u
r
a
cy
.
T
h
e
n
u
m
b
e
r
o
f
m
o
d
el
p
a
r
a
m
et
er
s
s
t
ay
s
r
o
u
g
h
l
y
co
n
s
t
an
t.
3
.
3
.
Ada
ptiv
e
s
ca
le
f
us
io
n m
o
du
le
T
h
is
m
o
d
u
le
in
cr
ea
s
es
th
e
n
etwo
r
k
wid
th
u
p
o
n
r
ep
lacin
g
th
e
s
p
ar
s
e
an
d
d
en
s
e
s
tr
u
ctu
r
es
th
at
en
h
an
ce
s
n
etwo
r
k
p
er
f
o
r
m
a
n
ce
b
y
v
o
id
in
g
c
o
m
p
u
tatio
n
al
tim
e.
T
h
e
in
p
u
t
f
u
r
th
e
r
ex
tr
a
cts
v
ar
io
u
s
f
ea
tu
r
es
th
r
o
u
g
h
v
a
r
io
u
s
s
izes
o
f
co
n
v
o
lu
tio
n
al
k
e
r
n
els
in
th
e
a
d
jace
n
t
b
r
an
c
h
es;
th
e
f
o
u
r
b
r
an
ch
es
ar
e
th
en
s
p
liced
in
ch
an
n
el
d
im
en
s
io
n
an
d
is
f
u
r
t
h
er
f
u
s
ed
with
th
e
f
ea
tu
r
e
m
a
tr
ix
is
g
iv
e
n
as
o
u
tp
u
t.
T
h
e
m
o
d
el'
s
s
ize
m
ay
b
e
d
ec
r
ea
s
ed
an
d
th
e
co
m
p
u
tatio
n
al
co
m
p
lex
ity
d
ec
r
ea
s
ed
v
ia
d
ep
th
wis
e
s
ep
ar
ab
le
co
n
v
o
lu
tio
n
,
wh
ich
in
clu
d
es
d
ep
th
wis
e
an
d
p
o
in
t
-
wis
e
co
n
v
o
lu
tio
n
.
Ass
u
m
e
th
at
th
e
in
p
u
t
f
ea
tu
r
e
m
atr
ix
'
s
wid
th
an
d
h
eig
h
t
co
r
r
esp
o
n
d
to
th
e
co
n
v
o
lu
tio
n
k
er
n
el'
s
s
ize,
th
at
th
e
in
p
u
t
f
e
atu
r
e
m
atr
ix
'
s
n
u
m
b
er
o
f
c
h
an
n
els
is
,
an
d
th
at
th
e
o
u
tp
u
t
m
atr
ix
'
s
n
u
m
b
er
o
f
ch
an
n
els is
.
I
n
(
3
)
d
en
o
tes
th
e
am
o
u
n
t
o
f
c
o
n
v
o
lu
ti
o
n
al
co
m
p
u
tatio
n
,
(
4
)
is
a
s
ep
ar
ab
le
co
n
v
o
lu
tio
n
c
o
m
p
u
tatio
n
.
in
th
e
(
5
)
r
ep
r
esen
ts
s
ep
ar
ate
co
n
v
o
l
u
tio
n
t
h
at
s
av
es
m
o
r
e
p
ar
am
eter
s
in
co
m
p
ar
is
o
n
wit
h
th
e
co
n
v
en
tio
n
al
co
n
v
o
l
u
tio
n
al
co
m
p
u
tatio
n
.
T
h
e
ASFM
s
tr
u
ctu
r
e
im
p
lem
en
ted
in
th
is
s
tu
d
y
is
f
o
u
n
d
ed
o
n
d
e
p
th
wis
e
s
ep
ar
ab
le
co
n
v
o
lu
tio
n
an
d
t
h
e
I
n
ce
p
tio
n
ar
ch
itectu
r
e
.
I
n
iti
ally
,
th
e
1
×1
an
d
3
×3
o
r
d
in
a
r
y
co
n
v
o
lu
tio
n
s
in
I
n
ce
p
tio
n
ar
e
s
u
b
s
titu
ted
with
a
3
×3
d
e
p
th
wis
e
co
n
v
o
lu
tio
n
an
d
a
1
×
1
p
o
in
t
-
wis
e
co
n
v
o
lu
tio
n
.
Su
b
s
eq
u
en
tly
,
a
5
×5
co
n
v
o
lu
tio
n
is
r
ep
lace
d
b
y
two
3
×3
co
n
v
o
l
u
tio
n
s
.
L
astl
y
,
m
u
ltip
le
s
m
all
-
k
er
n
el
co
n
v
o
lu
tio
n
o
p
er
atio
n
s
ca
n
in
tr
o
d
u
ce
in
c
r
ea
s
ed
n
o
n
lin
ea
r
ities
wh
ile
m
ain
tain
in
g
th
e
s
am
e
r
ec
ep
tiv
e
f
ield
with
r
ed
u
ce
d
co
m
p
u
tatio
n
al
r
eq
u
ir
em
e
n
ts
.
T
h
e
m
o
d
el
en
c
o
u
n
ter
s
ch
allen
g
es
in
ex
tr
ac
ti
n
g
lesi
o
n
ch
ar
ac
ter
is
tics
with
i
n
a
co
m
p
lex
n
at
u
r
al
en
v
ir
o
n
m
en
t,
wh
ic
h
af
f
ec
ts
r
ec
o
g
n
itio
n
ac
cu
r
ac
y
.
E
n
h
an
cin
g
th
e
p
r
o
ce
s
s
is
ch
allen
g
in
g
;
h
o
wev
e
r
,
it
is
eq
u
ally
d
if
f
icu
lt
to
im
p
r
o
v
e
t
h
e
ex
tr
ac
tio
n
o
f
d
is
ea
s
e
in
f
o
r
m
atio
n
u
s
in
g
o
n
l
y
a
s
in
g
le
co
n
v
o
lu
ti
o
n
al
k
er
n
el
s
ize
is
u
tili
ze
d
d
u
e
to
th
e
v
ar
iab
ilit
y
in
s
ize
a
n
d
s
h
ap
e
o
f
t
h
e
d
is
ea
s
e
s
p
o
ts
.
T
h
is
s
tu
d
y
r
ep
r
esen
ts
a
ASFM
m
o
d
u
le
d
esig
n
e
d
to
ex
tr
ac
t
d
e
ep
er
ch
ar
ac
ter
is
tics
o
f
illn
ess
p
ictu
r
es.
T
h
e
p
er
f
o
r
m
an
ce
o
f
th
e
AM
S
-
C
o
n
v
Net
n
etwo
r
k
ca
n
b
e
im
p
r
o
v
e
d
th
r
o
u
g
h
th
e
in
te
g
r
atio
n
o
f
th
e
ASFM
m
o
d
u
le,
wh
ich
f
ac
ilit
ates
th
e
ex
tr
ac
tio
n
o
f
d
iv
er
s
e
s
ca
les
o
f
illn
es
s
s
p
o
t
in
f
o
r
m
atio
n
.
Fig
u
r
e
3
s
h
o
ws
th
e
h
ier
ar
ch
ical
f
ea
tu
r
e
ag
g
r
eg
atio
n
m
o
d
u
le
(
HFAM)
m
o
d
el.
∗
∗
∗
∗
∗
(
3
)
∗
∗
∗
∗
+
∗
∗
∗
(
4
)
(
∗
∗
∗
∗
∗
)
/
(
∗
∗
∗
∗
+
∗
∗
∗
)
=
1
+
1
∗
=
1
+
1
9
(
5
)
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.
15
,
No
.
3
,
J
u
n
e
20
26
:
2
9
5
6
-
2
9
6
9
2962
3
.
4
.
P
re
-
t
ra
ined k
no
wledg
e
a
da
pta
t
io
n
T
h
e
m
o
d
el
is
v
u
ln
er
ab
le
to
is
s
u
es
l
ik
e
o
v
er
f
itti
n
g
an
d
p
o
o
r
g
en
er
aliza
tio
n
ab
ilit
y
in
t
h
e
f
ield
o
f
d
is
ea
s
e
r
ec
o
g
n
itio
n
s
in
ce
it
is
ch
allen
g
in
g
to
g
at
h
er
f
u
r
th
er
citr
u
s
d
is
ea
s
e
leaf
d
ata.
A
te
ch
n
iq
u
e
f
o
r
m
o
v
in
g
p
r
e
-
tr
ain
in
g
weig
h
ts
f
r
o
m
th
e
s
o
u
r
ce
d
o
m
ain
t
o
th
e
tar
g
et
d
o
m
ain
is
ca
lled
PTKA
,
a
n
d
it
ca
n
h
elp
with
s
m
all
s
am
p
le
is
s
u
es
an
d
en
h
an
ce
n
e
two
r
k
p
er
f
o
r
m
an
ce
.
Usu
ally
,
a
s
izab
le
d
ataset
is
u
tili
ze
d
as
th
e
s
o
u
r
ce
d
o
m
ain
to
f
ir
s
t
tr
ain
th
e
m
o
d
el.
A
d
at
a
co
llectio
n
th
at
y
o
u
wis
h
to
a
p
p
ly
t
o
a
lim
ited
n
u
m
b
er
o
f
s
am
p
les
is
k
n
o
wn
as
th
e
tar
g
et
d
o
m
ain
.
I
m
a
g
eNe
t
is
th
e
b
ig
g
est
im
ag
e
r
ec
o
g
n
i
tio
n
d
atab
ase
a
n
d
is
f
r
eq
u
en
t
ly
u
s
ed
f
o
r
p
ictu
r
e
ca
teg
o
r
izatio
n
,
d
etec
tio
n
,
an
d
p
lace
m
en
t.
T
o
p
r
e
-
tr
ain
th
e
m
o
d
el
in
th
is
s
tu
d
y
,
th
e
I
m
ag
eN
et
d
ata
s
et
wa
s
u
s
ed
as
th
e
s
o
u
r
ce
d
o
m
ain
.
T
h
e
n
,
a
m
o
n
g
f
r
ee
zin
g
a
n
d
f
in
e
-
t
u
n
in
g
tech
n
i
q
u
es,
t
h
e
PTKA
a
p
p
r
o
ac
h
a
p
p
r
o
p
r
iate
f
o
r
d
iag
n
o
s
in
g
p
lan
t
d
is
ea
s
e
d
ataset
was
u
s
ed
.
Fre
ez
in
g
is
t
h
e
p
r
o
ce
s
s
o
f
s
ettin
g
th
e
weig
h
ts
in
th
e
s
o
u
r
ce
d
o
m
ain
'
s
n
etwo
r
k
m
o
d
el,
f
r
e
ez
in
g
all
o
f
th
e
n
etwo
r
k
'
s
co
n
v
o
lu
tio
n
al
lay
er
s
,
an
d
o
n
ly
m
o
d
if
y
in
g
th
e
f
u
lly
co
n
n
ec
ted
o
n
es.
Fin
e
-
tu
n
in
g
is
th
e
p
r
o
ce
s
s
o
f
k
ee
p
in
g
s
o
m
e
o
f
th
e
n
etwo
r
k
'
s
co
n
v
o
lu
tio
n
al
an
d
f
u
lly
co
n
n
ec
ted
la
y
er
s
f
o
r
ad
d
itio
n
al
tr
ain
in
g
an
d
f
in
e
-
tu
n
in
g
wh
ile
f
r
ee
zin
g
s
o
m
e
o
f
t
h
e
n
etw
o
r
k
'
s
co
n
v
o
lu
tio
n
al
lay
er
s
.
Fig
u
r
e
4
s
h
o
ws th
e
PTKA
m
ec
h
an
is
m
.
Fig
u
r
e
3
.
HFAM
m
o
d
el
Fig
u
r
e
4
.
PTKA
m
ec
h
a
n
is
m
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
d
a
p
tive
mu
lti
-
s
ca
le
co
n
v
o
lu
ti
o
n
a
l n
etw
o
r
k
fo
r
p
la
n
t le
a
f d
is
ea
s
e
d
etec
tio
n
a
n
d
… (
Teja
s
h
w
in
i C.
Ga
d
a
g
)
2963
4.
RE
SU
L
T
S
AND
D
I
SCU
SS
I
O
N
T
h
e
p
er
f
o
r
m
a
n
ce
e
v
alu
atio
n
r
ev
ea
ls
th
at
m
o
d
els
ac
h
iev
e
v
ar
y
in
g
lev
els
o
f
ef
f
ec
tiv
e
n
e
s
s
ac
r
o
s
s
d
if
f
er
en
t
d
atasets
,
with
m
etr
ics
s
u
ch
as
p
r
ec
is
io
n
,
r
ec
all,
F1
-
s
co
r
e,
an
d
ac
cu
r
ac
y
r
ef
lectin
g
th
e
co
m
p
lex
ity
o
f
th
e
task
s
.
Data
s
ets
lik
e
s
u
g
ar
ca
n
e
leaf
d
is
ea
s
e
s
h
o
w
r
elativ
ely
lo
wer
p
e
r
f
o
r
m
an
ce
,
in
d
i
ca
tin
g
ch
allen
g
es
i
n
ac
cu
r
ate
d
etec
tio
n
,
wh
ile
b
et
ter
r
esu
lts
ar
e
o
b
s
er
v
ed
f
o
r
co
r
n
,
p
o
tato
,
an
d
g
r
o
u
n
d
n
u
t
d
atasets
.
Ad
v
an
ce
d
ar
ch
itectu
r
es
lik
e
Mo
b
ileNetV2
an
d
E
f
f
icien
tNetB
5
d
em
o
n
s
tr
ate
s
tr
o
n
g
p
er
f
o
r
m
an
ce
,
s
h
o
wca
s
in
g
th
e
b
en
ef
its
o
f
m
o
d
er
n
d
esig
n
s
.
4
.
1
.
Da
t
a
s
et
det
a
ils
T
h
e
d
etails o
f
th
e
d
atasets
u
s
ed
in
th
is
s
tu
d
y
ar
e
p
r
esen
ted
as
f
o
llo
ws:
i)
Gr
o
u
n
d
n
u
t
d
ataset
[
2
3
]
:
t
h
e
d
ataset
u
s
ed
f
o
r
th
is
r
esear
ch
i
n
clu
d
es
d
ig
ital
im
ag
es
o
f
g
r
o
u
n
d
n
u
t
leav
es
ca
p
tu
r
ed
u
n
d
er
v
ar
y
i
n
g
wea
t
h
er
co
n
d
itio
n
s
an
d
d
iv
er
s
e
lig
h
tin
g
en
v
ir
o
n
m
en
ts
in
th
e
k
o
p
p
al
ar
ea
o
f
Kar
n
atak
a
,
I
n
d
ia.
I
t
co
n
s
is
ts
o
f
a
t
o
tal
o
f
1
0
,
3
6
1
i
m
ag
es,
wh
ich
h
av
e
b
ee
n
d
iv
id
ed
in
to
s
ep
ar
ate
tr
ain
in
g
an
d
test
s
ets.
T
h
e
test
d
ataset,
o
n
th
e
o
t
h
er
h
an
d
,
co
n
s
is
ts
o
f
2
,
4
5
1
p
h
o
to
s
,
with
ea
c
h
cl
ass
in
clu
d
in
g
,
d
ep
en
d
i
n
g
o
n
th
e
ca
teg
o
r
y
,
4
0
9
o
r
4
0
5
im
a
g
es.
ii)
Su
g
ar
ca
n
e
leaf
d
is
ea
s
e
d
ataset
[
2
4
]
:
t
h
e
s
u
g
a
r
ca
n
e
leaf
d
is
ea
s
e
d
ataset
co
m
p
r
is
es
2
,
5
6
7
im
ag
es
d
is
tr
ib
u
ted
ac
r
o
s
s
f
iv
e
d
is
tin
ct
class
es:
h
ea
lth
y
,
m
o
s
aic
,
r
ed
r
o
t
,
r
u
s
t
,
an
d
y
ello
w
.
T
h
is
d
ataset
is
d
esig
n
ed
to
ca
p
t
u
r
e
a
wid
e
v
ar
iety
o
f
v
is
u
al
f
ea
tu
r
es
ass
o
ciate
d
with
th
ese
co
n
d
itio
n
s
,
o
f
f
e
r
in
g
a
co
m
p
r
eh
e
n
s
iv
e
r
eso
u
r
ce
f
o
r
t
h
e
d
e
v
elo
p
m
e
n
t
o
f
ML
m
o
d
els
f
o
r
d
is
ea
s
e
class
if
icatio
n
.
T
h
e
in
clu
s
io
n
o
f
d
iv
er
s
e
d
is
ea
s
e
ty
p
es
en
s
u
r
es
r
o
b
u
s
t
tr
ai
n
in
g
an
d
test
in
g
ca
p
ab
ilit
ies
f
o
r
al
g
o
r
ith
m
s
tar
g
e
tin
g
s
u
g
ar
ca
n
e
leaf
h
ea
lth
.
iii)
C
o
r
n
(
o
r
m
aize
)
leaf
d
is
ea
s
e
d
ataset
[
2
5
]
:
t
h
e
c
o
r
n
(
o
r
m
aize
)
leaf
d
is
ea
s
e
d
ataset
co
n
tain
s
4
,
1
8
8
im
ag
es
ca
teg
o
r
ized
in
to
f
o
u
r
g
r
o
u
p
s
:
co
m
m
o
n
r
u
s
t
,
g
r
ay
leaf
s
p
o
t
,
b
lig
h
t
,
an
d
h
ea
lth
y
.
T
h
is
d
ata
s
et
p
r
o
v
id
es
a
r
ich
an
d
d
iv
er
s
e
s
et
o
f
v
is
u
al
p
atter
n
s
f
o
r
d
is
ea
s
e
an
d
h
ea
lth
y
leaf
id
e
n
tific
atio
n
.
I
t p
lay
s
a
c
r
itical
r
o
le
in
ad
v
an
cin
g
p
r
ec
is
io
n
ag
r
icu
ltu
r
e,
en
ab
lin
g
th
e
d
e
v
elo
p
m
e
n
t
o
f
ac
cu
r
ate
an
d
ef
f
icien
t
m
o
d
els
f
o
r
ea
r
ly
d
is
ea
s
e
d
etec
tio
n
an
d
ef
f
ec
tiv
e
cr
o
p
m
a
n
ag
em
e
n
t in
m
aize
f
a
r
m
in
g
.
iv
)
Po
tato
leaf
d
is
ea
s
e
d
ataset
[
2
6
]
:
th
e
p
o
tato
leaf
d
is
ea
s
e
d
ataset
co
n
s
is
ts
o
f
4
,
0
6
2
im
ag
es
class
if
ied
in
to
th
r
ee
ca
teg
o
r
ies:
ea
r
ly
b
lig
h
t
,
late
b
lig
h
t
,
an
d
h
ea
lth
y
.
I
t
s
er
v
es
as
a
v
ital
to
o
l
f
o
r
s
tu
d
y
in
g
th
e
v
is
u
al
d
is
tin
ctio
n
s
b
etwe
en
d
if
f
er
en
t
ty
p
es
o
f
b
lig
h
ts
an
d
h
ea
lth
y
p
o
tato
leav
es.
T
h
is
d
ataset
is
in
s
tr
u
m
en
tal
in
f
ac
ilit
atin
g
th
e
d
ev
elo
p
m
en
t
o
f
au
t
o
m
ated
d
is
ea
s
e
d
etec
tio
n
s
y
s
tem
s
,
h
elp
in
g
to
r
ed
u
ce
cr
o
p
l
o
s
s
an
d
im
p
r
o
v
e
p
r
o
d
u
ctiv
ity
in
p
o
tato
cu
ltiv
atio
n
.
4
.
2
.
M
et
rics c
o
m
pa
riso
n
4
.
2
.
1
.
P
re
cisi
o
n
Pre
cisi
o
n
m
ea
s
u
r
es
th
e
p
r
o
p
o
r
tio
n
o
f
c
o
r
r
ec
tly
i
d
en
tifie
d
p
o
s
itiv
e
in
s
tan
ce
s
am
o
n
g
al
l
in
s
tan
ce
s
p
r
ed
icted
as
p
o
s
itiv
e
b
y
th
e
m
o
d
el.
I
t
f
o
cu
s
es
o
n
th
e
ac
c
u
r
ac
y
o
f
p
o
s
itiv
e
p
r
e
d
ictio
n
s
an
d
is
p
ar
ticu
lar
ly
im
p
o
r
tan
t
in
s
ce
n
ar
i
o
s
wh
er
e
f
alse
p
o
s
itiv
es
ca
r
r
y
s
ig
n
if
ica
n
t
co
n
s
eq
u
en
ce
s
.
Fo
r
ex
a
m
p
le
,
in
s
p
a
m
d
etec
tio
n
,
a
h
ig
h
p
r
ec
is
io
n
m
ea
n
s
th
e
m
o
d
el
ef
f
ec
tiv
ely
a
v
o
id
s
lab
eli
n
g
leg
itima
te
em
ails
as
s
p
am
.
Pre
cisi
o
n
p
r
o
v
id
es
in
s
ig
h
t
in
to
th
e
r
eliab
ilit
y
o
f
th
e
m
o
d
el'
s
p
o
s
itiv
e
p
r
ed
ictio
n
s
,
in
d
icatin
g
h
o
w
well
th
e
m
o
d
el
d
is
cr
im
in
ates
b
etwe
en
r
elev
an
t a
n
d
ir
r
ele
v
a
n
t p
o
s
itiv
e
class
if
icatio
n
s
.
4
.
2
.
2
.
Rec
a
ll
R
ec
all,
also
k
n
o
wn
as
s
en
s
itiv
ity
o
r
th
e
tr
u
e
p
o
s
itiv
e
r
ate,
ev
alu
ates
th
e
m
o
d
el’
s
ab
ilit
y
to
co
r
r
ec
tl
y
id
en
tify
all
ac
t
u
al
p
o
s
itiv
e
in
s
tan
ce
s
in
th
e
d
ataset.
I
t
m
ea
s
u
r
es
h
o
w
well
th
e
m
o
d
el
ca
p
tu
r
es
p
o
s
itiv
e
ca
s
es
an
d
is
cr
u
cial
in
ap
p
licatio
n
s
wh
er
e
m
is
s
in
g
a
p
o
s
itiv
e
in
s
tan
ce
ca
n
h
av
e
s
ev
er
e
co
n
s
eq
u
en
ce
s
,
s
u
ch
as
in
m
ed
ical
d
iag
n
o
s
es
o
r
f
r
au
d
d
etec
tio
n
.
A
h
ig
h
r
ec
all
in
d
ica
tes
th
e
m
o
d
el
s
u
cc
ess
f
u
lly
id
en
tifie
s
m
o
s
t
o
f
th
e
tr
u
e
p
o
s
itiv
es,
en
s
u
r
in
g
th
at
cr
itical
ca
s
es a
r
e
n
o
t o
v
er
lo
o
k
ed
,
ev
en
if
it in
cl
u
d
es so
m
e
ir
r
el
ev
an
t p
o
s
itiv
es.
4
.
2
.
3
.
F1
-
s
co
re
F1
-
s
co
r
e
is
th
e
h
a
r
m
o
n
ic
m
ea
n
o
f
p
r
ec
is
io
n
a
n
d
r
ec
all,
p
r
o
v
id
in
g
a
b
ala
n
ce
d
ev
alu
atio
n
o
f
a
m
o
d
el'
s
p
er
f
o
r
m
an
ce
wh
e
n
d
ea
lin
g
wi
th
im
b
alan
ce
d
d
atasets
.
I
t
is
p
ar
ticu
lar
ly
u
s
ef
u
l
wh
en
b
o
t
h
f
alse
p
o
s
itiv
es
an
d
f
alse
n
eg
ativ
es
ca
r
r
y
s
ig
n
i
f
ican
t
co
n
s
eq
u
en
ce
s
,
a
n
d
a
b
ala
n
c
e
b
etwe
en
p
r
ec
is
io
n
a
n
d
r
ec
all
is
n
ee
d
e
d
.
A
h
ig
h
F1
-
s
co
r
e
in
d
icate
s
th
at
th
e
m
o
d
el
m
ain
tain
s
a
g
o
o
d
tr
ad
e
-
o
f
f
,
ca
p
tu
r
in
g
m
o
s
t
p
o
s
itiv
e
ca
s
e
s
wh
ile
m
in
im
izin
g
in
co
r
r
ec
t
p
r
ed
ictio
n
s
.
T
h
is
m
etr
ic
is
id
ea
l
f
o
r
ap
p
licatio
n
s
lik
e
m
ed
ical
d
iag
n
o
s
es
o
r
s
p
a
m
d
etec
tio
n
,
wh
e
r
e
b
o
th
ty
p
es o
f
er
r
o
r
s
m
u
s
t b
e
c
ar
ef
u
lly
m
a
n
ag
ed
.
4
.
2
.
4
.
Acc
ura
cy
Acc
u
r
ac
y
is
th
e
p
r
o
p
o
r
tio
n
o
f
co
r
r
ec
tly
p
r
ed
icted
in
s
tan
ce
s
(
b
o
th
p
o
s
itiv
e
an
d
n
eg
ativ
e)
o
u
t
o
f
th
e
to
tal
in
s
tan
ce
s
in
th
e
d
ataset
.
I
t
is
a
s
tr
aig
h
tf
o
r
war
d
a
n
d
wid
ely
u
s
ed
m
et
r
ic
th
at
m
e
asu
r
es
th
e
o
v
er
all
ef
f
ec
tiv
en
ess
o
f
a
m
o
d
el.
Ho
wev
er
,
it
ca
n
b
e
m
is
lead
in
g
in
im
b
alan
ce
d
d
atasets
wh
er
e
o
n
e
class
s
ig
n
if
ican
tly
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.
15
,
No
.
3
,
J
u
n
e
20
26
:
2
9
5
6
-
2
9
6
9
2964
o
u
tweig
h
s
th
e
o
th
er
.
Fo
r
ex
a
m
p
le,
in
a
d
ataset
with
9
5
%
n
eg
ativ
es,
a
m
o
d
el
p
r
e
d
ictin
g
a
ll
n
eg
ativ
es
ac
h
iev
es
9
5
%
ac
cu
r
ac
y
b
u
t
f
ails
to
id
en
tify
p
o
s
itiv
es.
T
h
u
s
,
wh
ile
ac
c
u
r
ac
y
is
a
h
elp
f
u
l
m
etr
ic
f
o
r
b
alan
ce
d
d
atasets
,
it
is
o
f
ten
s
u
p
p
lem
e
n
ted
b
y
o
th
e
r
m
etr
ics
lik
e
p
r
ec
is
io
n
,
r
ec
all
,
an
d
F1
-
s
co
r
e
to
p
r
o
v
id
e
a
m
o
r
e
co
m
p
r
eh
e
n
s
iv
e
ev
alu
atio
n
o
f
m
o
d
el
p
er
f
o
r
m
a
n
ce
.
4
.
3
.
G
ro
un
dn
ut
da
t
a
s
et
re
s
u
lt
s
T
h
e
T
a
b
l
e
1
a
n
d
F
i
g
u
r
e
5
i
l
l
u
s
tr
a
t
e
t
h
e
p
r
e
c
is
i
o
n
r
a
t
es
(
%
)
o
f
v
a
r
i
o
u
s
m
o
d
e
l
s
i
n
a
c
o
m
p
a
r
a
tiv
e
a
n
a
l
y
s
is
.
I
t
s
h
o
w
s
t
h
at
t
r
a
d
i
ti
o
n
a
l
m
o
d
e
l
s
li
k
e
V
G
G
1
6
e
x
h
i
b
i
t
r
el
a
t
iv
e
l
y
l
o
w
e
r
p
r
e
c
i
s
i
o
n
(
8
5
.
8
6
3
%
)
,
w
h
i
l
e
a
d
v
a
n
c
ed
a
r
c
h
i
t
e
c
t
u
r
es
l
i
k
e
D
e
n
s
e
N
et
1
6
1
,
E
f
f
i
c
i
e
n
t
N
e
tB
5
,
a
n
d
M
o
b
i
le
N
e
t
V
2
d
e
m
o
n
s
t
r
at
e
i
m
p
r
o
v
e
d
p
e
r
f
o
r
m
a
n
c
e
,
w
i
t
h
M
o
b
i
l
e
N
et
V
2
a
c
h
i
e
v
i
n
g
9
3
.
8
5
2
%
.
E
f
f
i
ci
e
n
t
N
et
B
5
s
h
o
ws
s
l
ig
h
t
l
y
l
es
s
p
r
e
c
is
i
o
n
(
9
3
.
1
5
5
%
)
t
h
a
n
M
o
b
i
le
N
e
tV
2
.
R
e
m
a
r
k
a
b
l
y
,
t
h
e
c
u
s
t
o
m
o
r
e
x
p
e
r
i
m
e
n
t
a
l
m
o
d
el
s
l
a
b
e
le
d
E
S
an
d
P
S
a
c
h
i
e
v
e
s
i
g
n
i
f
ic
a
n
t
p
r
e
cis
i
o
n
i
m
p
r
o
v
e
m
e
n
ts
,
w
i
t
h
E
S
r
ea
c
h
i
n
g
9
7
.
3
6
5
%
a
n
d
P
S
a
c
h
i
e
v
i
n
g
t
h
e
h
i
g
h
e
s
t
p
r
ec
i
s
i
o
n
a
t
9
8
.
3
4
%
.
T
h
is
a
n
al
y
s
is
s
u
g
g
e
s
ts
t
h
a
t
w
h
i
le
m
o
d
e
r
n
D
L
a
r
c
h
i
t
e
c
t
u
r
e
s
e
n
h
a
n
c
e
p
e
r
f
o
r
m
a
n
c
e
,
c
u
s
t
o
m
m
o
d
e
l
s
l
i
k
e
PS
a
n
d
E
S
o
u
t
p
e
r
f
o
r
m
s
t
a
n
d
a
r
d
p
r
e
-
t
r
a
i
n
e
d
m
o
d
e
l
s
,
i
n
d
i
c
at
i
n
g
p
o
t
e
n
t
i
al
o
p
t
i
m
i
z
at
i
o
n
s
ta
i
l
o
r
e
d
t
o
t
h
e
s
p
e
cif
i
c
d
a
t
a
s
e
t
o
r
t
as
k
.
T
ab
le
1
.
Me
tr
ic
co
m
p
ar
is
o
n
o
n
g
r
o
u
n
d
n
u
t d
ataset
M
o
d
e
l
P
r
e
c
i
s
i
o
n
(
%)
R
e
c
a
l
l
(
%)
F1
-
s
c
o
r
e
(
%)
A
c
c
u
r
a
c
y
(
%)
D
e
n
seN
e
t
1
6
1
9
2
.
1
4
5
9
1
.
8
4
9
1
.
8
3
9
1
.
8
4
Ef
f
i
c
i
e
n
t
N
e
t
9
3
.
1
5
5
9
4
.
0
8
4
9
4
.
0
6
4
9
4
.
0
8
4
V
G
G
1
6
8
5
.
8
6
3
8
5
.
5
9
7
8
5
.
4
2
2
8
5
.
5
9
7
M
o
b
i
l
e
N
e
t
V
2
9
3
.
8
5
2
9
3
.
7
9
8
9
5
.
5
8
8
9
3
.
7
9
8
I
n
c
e
p
t
i
o
n
R
e
sN
e
t
V
2
8
8
.
4
7
1
8
8
.
2
9
8
8
.
2
5
5
8
8
.
2
9
ES
9
7
.
3
6
5
9
7
.
7
5
6
9
7
.
2
2
5
9
7
.
2
2
5
PS
9
8
.
3
4
9
8
.
7
6
9
8
.
9
8
9
8
.
4
3
Fig
u
r
e
5
.
Pre
cisi
o
n
m
et
r
ic
co
m
p
ar
is
o
n
o
n
g
r
o
u
n
d
n
u
t
d
ataset
T
h
e
Fig
u
r
e
6
p
r
esen
ts
th
e
r
ec
all
(
%)
p
er
f
o
r
m
an
ce
o
f
v
ar
io
u
s
DL
m
o
d
els.
Am
o
n
g
th
e
lis
t
ed
m
o
d
els,
VGG1
6
d
em
o
n
s
tr
ates
th
e
lo
west
r
ec
all
at
8
5
.
5
9
7
%,
wh
i
le
n
ewe
r
ar
ch
itectu
r
es
lik
e
I
n
ce
p
tio
n
R
esNetV2
(
8
8
.
2
9
%)
an
d
Den
s
eNe
t1
6
1
(
9
1
.
8
4
%)
s
h
o
w
in
cr
em
e
n
tal
i
m
p
r
o
v
e
m
en
ts
.
Mo
b
ileNetV2
an
d
E
f
f
icie
n
tNetB
5
ac
h
iev
e
n
o
tab
le
r
ec
all
v
alu
e
s
o
f
9
3
.
7
9
8
%
an
d
9
4
.
0
8
4
%,
r
esp
ec
tiv
ely
,
i
n
d
icatin
g
t
h
eir
ef
f
ec
tiv
e
n
ess
in
id
en
tify
in
g
tr
u
e
p
o
s
itiv
es.
T
h
e
cu
s
to
m
m
o
d
els
E
S
an
d
PS
s
ig
n
if
ican
tly
o
u
tp
e
r
f
o
r
m
th
e
s
tan
d
ar
d
ar
ch
itectu
r
es
,
with
E
S
ac
h
iev
in
g
an
im
p
r
ess
iv
e
9
7
.
7
5
6
%
r
ec
all
an
d
PS
att
ain
in
g
th
e
h
ig
h
est
r
ec
all
o
f
9
8
.
7
6
%.
T
h
is
an
aly
s
is
h
ig
h
lig
h
ts
th
e
ev
o
lu
tio
n
o
f
m
o
d
el
p
er
f
o
r
m
an
ce
an
d
u
n
d
er
s
co
r
es
th
e
s
u
p
er
io
r
r
ec
al
l
ac
h
iev
ed
b
y
th
e
s
p
ec
ialized
m
o
d
els,
s
u
g
g
esti
n
g
th
eir
ad
a
p
tatio
n
to
s
p
ec
if
ic
d
ataset
ch
ar
ac
ter
is
tics
o
r
task
r
eq
u
ir
em
en
ts
.
T
h
e
Fig
u
r
e
7
s
h
o
wca
s
es
th
e
F1
-
s
co
r
e
p
er
f
o
r
m
an
ce
o
f
v
a
r
io
u
s
DL
m
o
d
els.
T
h
e
F1
-
s
co
r
e,
wh
ic
h
b
alan
ce
s
p
r
ec
is
io
n
a
n
d
r
ec
all,
in
d
icate
s
th
at
VGG1
6
h
as
th
e
lo
west
p
er
f
o
r
m
an
ce
at
8
5
.
4
2
2
%.
I
m
p
r
o
v
e
m
en
ts
ar
e
o
b
s
er
v
ed
with
I
n
ce
p
tio
n
R
esNet
V2
(
8
8
.
2
5
5
%)
an
d
D
en
s
eNe
t1
6
1
(
9
1
.
8
3
%),
wh
ile
E
f
f
icien
tNetB
5
a
n
d
Mo
b
ileNetV2
d
em
o
n
s
tr
ate
s
tr
o
n
g
p
er
f
o
r
m
an
ce
at
9
4
.
0
6
4
%
an
d
9
5
.
5
8
8
%,
r
esp
ec
tiv
ely
.
H
o
wev
er
,
th
e
cu
s
to
m
m
o
d
els
E
S
an
d
PS
s
ig
n
if
ican
tl
y
o
u
tp
er
f
o
r
m
th
ese
s
tan
d
ar
d
a
r
ch
itectu
r
es,
with
E
S
ac
h
iev
in
g
9
7
.
2
2
5
%
an
d
PS
r
ea
ch
in
g
an
e
x
ce
p
tio
n
al
9
8
.
9
8
%.
T
h
is
an
aly
s
is
u
n
d
er
s
co
r
e
s
th
e
s
u
p
er
io
r
b
alan
ce
ac
h
ie
v
ed
b
y
th
e
cu
s
to
m
m
o
d
els,
esp
ec
ially
PS
,
s
u
g
g
esti
n
g
th
eir
en
h
an
ce
d
o
p
tim
izati
o
n
an
d
s
u
itab
ilit
y
f
o
r
t
h
e
s
p
ec
if
ic
d
ataset
o
r
task
co
m
p
ar
ed
to
th
e
tr
a
d
itio
n
al
ar
c
h
itectu
r
es.
75
80
85
90
95
1
0
0
P
re
c
isio
n
v
a
lu
e
M
e
th
o
d
o
lo
g
y
P
re
c
isio
n
(%
)
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
d
a
p
tive
mu
lti
-
s
ca
le
co
n
v
o
lu
ti
o
n
a
l n
etw
o
r
k
fo
r
p
la
n
t le
a
f d
is
ea
s
e
d
etec
tio
n
a
n
d
… (
Teja
s
h
w
in
i C.
Ga
d
a
g
)
2965
Fig
u
r
e
6
.
R
ec
all
m
etr
ic
co
m
p
a
r
is
o
n
o
n
g
r
o
u
n
d
n
u
t d
ataset
Fig
u
r
e
7
.
F1
-
s
co
r
e
m
etr
ic
co
m
p
ar
is
o
n
o
n
g
r
o
u
n
d
n
u
t
d
ataset
T
h
e
F
ig
u
r
e
8
s
u
m
m
a
r
izes
th
e
ac
cu
r
ac
y
(
%)
o
f
d
if
f
e
r
en
t
DL
m
o
d
els.
VGG1
6
ex
h
ib
its
th
e
lo
west
ac
cu
r
ac
y
at
8
5
.
5
9
7
%,
i
n
d
ic
atin
g
r
o
o
m
f
o
r
im
p
r
o
v
em
en
t
in
its
p
er
f
o
r
m
an
ce
.
I
n
ce
p
t
io
n
R
esNetV2
an
d
Den
s
eNe
t1
6
1
d
em
o
n
s
tr
ate
b
etter
r
esu
lts
,
with
ac
cu
r
ac
y
v
alu
es
o
f
8
8
.
2
9
%
a
n
d
9
1
.
8
4
%,
r
esp
ec
tiv
ely
.
Mo
b
ileNetV2
an
d
E
f
f
icien
tN
etB
5
s
h
o
w
ev
en
g
r
ea
ter
ac
cu
r
ac
y
at
9
3
.
7
9
8
%
an
d
9
4
.
0
8
4
%,
s
h
o
wca
s
in
g
th
e
im
p
ac
t
o
f
ad
v
a
n
ce
m
en
ts
in
ar
ch
itectu
r
e
d
esig
n
.
T
h
e
cu
s
t
o
m
m
o
d
el
PS
s
ig
n
if
ican
tly
s
u
r
p
ass
es
all
o
th
er
m
o
d
els,
ac
h
iev
in
g
a
n
ex
ce
p
tio
n
al
ac
cu
r
ac
y
o
f
9
8
.
4
3
%,
h
ig
h
l
ig
h
tin
g
its
s
u
p
er
io
r
ad
a
p
tab
ilit
y
an
d
o
p
tim
izatio
n
f
o
r
t
h
e
s
p
ec
if
ic
d
ataset
o
r
task
.
T
h
is
an
aly
s
is
u
n
d
er
s
co
r
es
t
h
e
ev
o
lu
tio
n
ar
y
p
er
f
o
r
m
an
ce
i
m
p
r
o
v
e
m
en
ts
,
with
PS
em
er
g
in
g
as th
e
m
o
s
t e
f
f
ec
tiv
e
m
o
d
el.
Fig
u
r
e
8
.
Acc
u
r
ac
y
m
etr
ic
c
o
m
p
ar
is
o
n
o
n
g
r
o
u
n
d
n
u
t
d
ataset
75
80
85
90
95
1
0
0
R
e
c
a
l
l
v
a
l
u
e
M
e
t
h
o
d
o
l
o
g
y
Recall
(%
)
75
80
85
90
95
1
0
0
F1
-
sco
r
e
v
a
l
u
e
M
e
t
h
o
d
o
l
o
g
y
F1
-
sc
o
re
(%
)
75
80
85
90
95
1
0
0
A
c
c
u
r
a
c
y
v
a
l
u
e
M
e
t
h
o
d
o
l
o
g
y
A
c
c
u
r
a
c
y
(
%)
Evaluation Warning : The document was created with Spire.PDF for Python.