I
nte
rna
t
io
na
l J
o
urna
l o
f
I
nfo
rm
a
t
ics a
nd
Co
m
m
un
ica
t
io
n T
ec
hn
o
lo
g
y
(
I
J
-
I
CT
)
Vo
l.
15
,
No
.
3
,
Sep
tem
b
er
20
26
,
p
p
.
1
3
6
4
~
1
3
7
5
I
SS
N:
2252
-
8
7
7
6
,
DOI
:
1
0
.
1
1
5
9
1
/iji
ct
.
v15
i
3
.
pp
1
3
6
4
-
1
3
7
5
1364
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
ict.
ia
esco
r
e.
co
m
Explain
a
ble AI
-
b
a
sed hy
brid
G
NN
-
M
LP mo
del
for s
trawberry
fruit
disea
se de
tec
tion usin
g
hypers
pectral ima
g
ing
Va
rsh
a
K
ira
n B
ho
s
a
le
1,
2
,
Ch
in
-
Sh
iu
h Shi
e
h
1
1
D
e
p
a
r
t
me
n
t
o
f
El
e
c
t
r
o
n
i
c
E
n
g
i
n
e
e
r
i
n
g
,
N
a
t
i
o
n
a
l
K
a
o
h
si
u
n
g
U
n
i
v
e
r
si
t
y
o
f
S
c
i
e
n
c
e
a
n
d
T
e
c
h
n
o
l
o
g
y
,
K
a
o
h
si
u
n
g
C
i
t
y
,
Ta
i
w
a
n
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
,
D
n
y
a
n
s
h
r
e
e
I
n
st
i
t
u
t
e
o
f
En
g
i
n
e
e
r
i
n
g
a
n
d
Te
c
h
n
o
l
o
g
y
,
S
a
t
a
r
a
,
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
6
,
2
0
2
5
R
ev
is
ed
Ma
y
2
5
,
2
0
2
6
Acc
ep
ted
J
u
l 5
,
2
0
2
6
S
traw
b
e
rries
a
re
se
v
e
re
ly
a
ffe
c
t
e
d
b
y
t
h
e
m
a
in
fu
n
g
a
l
d
ise
a
se
s
su
c
h
a
s
a
n
th
ra
c
n
o
se
fr
u
it
r
o
t,
g
re
y
m
o
u
ld
,
a
n
d
p
o
wd
e
ry
m
il
d
e
w,
d
irec
tl
y
re
d
u
c
in
g
c
o
m
m
e
rc
ial
y
ield
a
n
d
p
o
st
-
h
a
rv
e
st
q
u
a
li
t
y
.
In
th
is
p
a
p
e
r,
we
p
ro
p
o
se
a
n
e
n
h
a
n
c
e
d
h
y
b
ri
d
d
e
e
p
-
lea
rn
in
g
m
e
th
o
d
b
y
c
o
m
b
i
n
in
g
g
ra
p
h
n
e
u
ra
l
n
e
two
r
k
s
(G
NN
s)
a
n
d
m
u
lt
i
-
lay
e
r
p
e
rc
e
p
tr
o
n
s
(
M
LP
s)
fo
r
e
ffe
c
ti
v
e
stra
wb
e
rry
d
ise
a
s
e
d
e
tec
ti
o
n
in
re
a
l
e
n
v
ir
o
n
m
e
n
ts
o
f
fiel
d
s.
T
h
is
d
a
tas
e
t
c
o
n
tai
n
s
ima
g
e
s
a
c
q
u
ired
fr
o
m
b
o
t
h
p
u
b
li
c
d
o
m
a
in
re
p
o
sit
o
ries
a
s
we
ll
a
s
fa
rm
o
p
e
ra
ti
o
n
a
l
se
tt
in
g
s,
e
n
c
o
m
p
a
ss
in
g
th
re
e
c
l
a
ss
e
s
o
f
d
ise
a
se
s
with
v
a
ry
i
n
g
li
g
h
ti
n
g
c
o
n
d
i
ti
o
n
s,
b
a
c
k
g
ro
u
n
d
n
o
ise
,
a
n
d
o
c
c
l
u
sio
n
.
A
str
u
c
tu
re
d
p
re
-
p
ro
c
e
ss
in
g
wo
rk
fl
o
w
c
o
m
p
risi
n
g
c
o
n
tras
t
e
n
h
a
n
c
e
m
e
n
t,
d
e
n
o
isin
g
,
a
n
d
sy
n
t
h
e
ti
c
h
y
p
e
rs
p
e
c
tral
sim
u
latio
n
is
u
se
d
t
o
e
n
h
a
n
c
e
su
b
tl
e
les
io
n
fe
a
tu
re
s
a
n
d
sta
b
il
ize
th
e
su
b
se
q
u
e
n
t
fe
a
tu
re
e
x
trac
ti
o
n
.
T
h
e
h
y
b
ri
d
GN
N
–
M
L
P
fra
m
e
wo
rk
p
o
ss
e
ss
e
s
th
e
m
e
rit
s
o
f
sp
a
ti
a
ll
y
l
o
c
a
l
les
io
n
to
p
o
lo
g
y
a
n
d
two
-
p
o
i
n
t
c
o
n
tex
t
u
a
l
i
n
fo
rm
a
ti
o
n
,
wh
ich
c
a
n
imp
r
o
v
e
d
ise
a
se
c
las
sifica
ti
o
n
c
o
m
p
a
re
d
with
c
o
m
m
o
n
CNN
-
b
a
se
d
stru
c
tu
re
s.
5
-
f
o
ld
c
ro
ss
-
v
a
li
d
a
ti
o
n
sh
o
ws
t
h
a
t
t
h
e
m
o
d
e
l
h
o
l
d
s
9
3
.
5
9
%
a
c
c
u
ra
c
y
a
n
d
a
m
a
c
r
o
F
1
-
sc
o
re
o
f
9
0
.
3
9
%
,
s
h
o
wi
n
g
g
o
o
d
g
e
n
e
ra
li
z
a
ti
o
n
e
v
e
n
wit
h
h
e
tero
g
e
n
e
o
u
s
i
n
p
u
t
re
g
ime
s.
Tran
sp
a
re
n
t
m
o
d
e
ls
u
se
l
o
c
a
l
i
n
terp
re
tab
le
m
o
d
e
l
-
a
g
n
o
stic
e
x
p
lan
a
ti
o
n
s
(LI
M
E)
a
n
d
g
ra
d
ien
t
-
we
ig
h
ted
c
las
s
a
c
ti
v
a
ti
o
n
m
a
p
p
i
n
g
(G
ra
d
-
CAM)
in
c
o
m
b
i
n
a
ti
o
n
,
h
ig
h
li
g
h
ti
n
g
d
ise
a
se
-
re
lev
a
n
t
a
re
a
s
th
a
t
g
iv
e
a
n
in
terp
re
tab
le
ra
ti
o
n
a
le
fo
r
e
a
c
h
p
re
d
ictio
n
.
To
c
o
n
c
lu
d
e
,
t
h
e
d
e
v
e
lo
p
e
d
sy
ste
m
o
ffe
rs
a
n
a
c
c
u
ra
te
a
n
d
e
x
p
lain
a
b
le
so
l
u
ti
o
n
th
a
t
h
a
s
a
c
o
m
p
u
tati
o
n
a
ll
y
e
fficie
n
t
c
o
m
m
it
m
e
n
t
fo
r
re
a
l
-
ti
m
e
m
o
n
it
o
ri
n
g
o
f
d
ise
a
se
in
sm
a
rt
a
g
r
icu
lt
u
re
se
tt
i
n
g
s,
p
a
rti
c
u
larly
o
n
l
o
w
-
c
o
st
h
a
r
d
wa
re
a
ss
e
ts.
K
ey
w
o
r
d
s
:
Dee
p
l
ea
r
n
in
g
E
x
p
lain
ab
le
AI
Gr
ap
h
n
e
u
r
al
n
etwo
r
k
Hy
p
er
s
p
ec
tr
al
im
ag
in
g
I
m
ag
e
class
if
icatio
n
MLP
Sm
ar
t f
ar
m
in
g
Stra
wb
er
r
y
d
is
ea
s
es
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
:
Var
s
h
a
Kir
an
B
h
o
s
ale
Dep
ar
tm
en
t o
f
C
o
m
p
u
ter
Scie
n
ce
an
d
E
n
g
in
ee
r
in
g
,
D
n
y
an
s
h
r
ee
I
n
s
titu
te
o
f
E
n
g
in
ee
r
i
n
g
an
d
T
ec
h
n
o
lo
g
y
A/P So
n
v
ad
i
-
Gajv
ad
i,
Sajjan
g
ad
R
o
ad
,
Satar
a
4
1
5
0
1
3
,
Satar
a,
I
n
d
ia
E
m
ail:
v
k
b
h
o
s
ale2
1
@
g
m
ail.
co
m
1.
I
NT
RO
D
UCT
I
O
N
Stra
wb
er
r
y
(
F
r
a
g
a
r
ia
×
a
n
a
n
a
s
s
a
L
.
)
is
am
o
n
g
th
e
m
o
s
t
ec
o
n
o
m
ically
im
p
o
r
tan
t
h
o
r
ticu
l
tu
r
al
cr
o
p
s
b
ec
au
s
e
o
f
its
h
ig
h
n
u
tr
ien
t
co
n
ten
t,
attr
ac
tiv
e
f
lav
o
r
,
a
n
d
s
tr
o
n
g
g
lo
b
al
m
ar
k
et
d
e
m
an
d
.
Nev
e
r
th
eless
,
f
u
n
g
al
p
ath
o
g
en
s
s
tr
o
n
g
ly
lim
it
ab
s
cisi
c
ac
id
(
AB
A)
p
r
o
d
u
ctio
n
a
n
d
B
o
tr
ytis
cin
erea
(
th
e
ca
u
s
al
ag
en
t
o
f
g
r
e
y
mo
u
ld
)
ac
co
u
n
ts
f
o
r
u
p
to
5
0
%
o
f
p
r
e
-
an
d
p
o
s
th
ar
v
est
lo
s
s
e
s
wo
r
ld
wid
e
u
n
d
er
f
av
o
u
r
ab
le
e
n
v
ir
o
n
m
en
tal
co
n
d
itio
n
s
[
1
]
.
T
h
is
lo
s
s
is
ec
o
n
o
m
ically
d
e
v
astatin
g
f
o
r
f
ar
m
er
s
,
esp
ec
ially
s
m
allh
o
ld
er
s
wh
o
d
o
n
o
t
h
av
e
ad
v
an
ce
d
cr
o
p
-
m
o
n
ito
r
in
g
s
y
s
tem
s
in
p
lace
.
Acc
o
r
d
in
g
ly
,
s
ca
lab
le
,
ea
r
ly
-
s
tag
e
,
a
n
d
ac
cu
r
ate
d
is
ea
s
e
d
etec
tio
n
m
eth
o
d
s
ar
e
in
c
r
ea
s
in
g
ly
im
p
er
ativ
e
in
m
o
d
e
r
n
p
r
ec
is
io
n
ag
r
icu
ltu
r
e.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J I
n
f
&
C
o
m
m
u
n
T
ec
h
n
o
l
I
SS
N:
2252
-
8
7
7
6
E
xp
la
in
a
b
le
A
I
-
b
a
s
ed
h
yb
r
id
GN
N
-
ML
P
mo
d
el
fo
r
s
tr
a
w
b
err
y
fr
u
it
d
is
ea
s
e
… (
V
a
r
s
h
a
K
i
r
a
n
B
h
o
s
a
le
)
1365
C
o
m
b
in
in
g
h
y
p
er
s
p
ec
tr
al
im
a
g
in
g
with
m
ac
h
i
n
e
lear
n
in
g
ap
p
r
o
ac
h
es o
f
f
e
r
s
a
n
o
v
el
m
eth
o
d
f
o
r
ea
r
ly
d
etec
tio
n
o
f
s
tr
awb
er
r
y
d
is
ea
s
es
b
y
id
en
tify
in
g
s
p
ec
tr
al
alter
atio
n
s
o
cc
u
r
r
in
g
b
ef
o
r
e
v
is
ib
le
s
y
m
p
to
m
s
[
2
]
.
Dee
p
co
n
v
o
lu
tio
n
al
n
eu
r
al
n
et
wo
r
k
s
(
C
NNs)
h
av
e
also
b
ee
n
ap
p
lied
to
m
u
ltip
le
p
lan
t
-
d
is
e
ase
clas
s
es
an
d
can
ac
h
iev
e
s
tate
-
of
-
th
e
-
ar
t r
esu
lts
with
co
n
v
en
tio
n
al
im
ag
e
-
b
ase
d
d
iag
n
o
s
is
[
3
]
.
As a
r
esu
lt,
ar
tific
ial
in
tellig
en
ce
(
AI
)
an
d
,
r
elate
d
ly
,
d
ee
p
lea
r
n
in
g
ar
e
b
ei
n
g
u
s
ed
m
o
r
e
wid
ely
f
o
r
cr
o
p
-
h
ea
lth
m
o
n
ito
r
i
n
g
a
n
d
co
n
tr
ib
u
tin
g
to
ag
r
icu
ltu
r
al
d
ec
is
io
n
-
s
u
p
p
o
r
t
s
y
s
tem
s
.
Fer
en
tin
o
s
[
4
]
illu
s
tr
ated
th
at
d
ee
p
lear
n
in
g
ca
n
b
e
ex
tr
em
ely
u
s
ef
u
l
i
n
id
en
tify
in
g
p
lan
t
d
is
ea
s
es
f
o
r
m
an
y
d
if
f
e
r
en
t
cr
o
p
s
wh
ile
s
h
o
win
g
th
e
lim
itatio
n
s
a
n
d
ad
v
an
tag
es
o
f
tr
ad
itio
n
al
tech
n
iq
u
es u
s
ed
in
p
lan
t
-
d
is
ea
s
e
id
en
tific
atio
n
.
AI
h
as
also
b
ee
n
in
teg
r
ated
in
to
au
to
m
ated
p
lan
t
-
d
is
ea
s
e
d
etec
tio
n
p
ip
elin
es
[
5
]
an
d
d
ep
lo
y
ed
in
r
ea
l
-
tim
e
m
o
b
ile
s
y
s
tem
s
to
m
o
n
ito
r
s
tr
awb
er
r
y
d
is
ea
s
es
in
o
p
en
-
f
ield
co
n
d
itio
n
s
[
6
]
.
Alth
o
u
g
h
t
h
ese
ad
v
an
ce
s
allo
w
th
e
i
d
en
tific
atio
n
a
n
d
a
n
aly
s
is
o
f
d
is
ea
s
e
n
et
wo
r
k
s
in
estab
lis
h
ed
an
d
r
ela
p
s
ed
h
ae
m
ato
lo
g
ical
m
alig
n
an
cies,
d
is
ea
s
e
d
etec
ti
o
n
at
t
h
o
s
e
ea
r
ly
s
tag
es
r
em
ain
s
ch
allen
g
in
g
,
p
a
r
ticu
lar
ly
u
n
d
e
r
u
n
co
n
tr
o
lled
im
ag
in
g
c
o
n
d
itio
n
s
.
C
o
n
v
e
n
tio
n
al
d
i
g
ital
im
ag
e
-
p
r
o
ce
s
s
in
g
tech
n
iq
u
es
g
en
e
r
ally
d
o
n
o
t
p
er
f
o
r
m
well
d
u
e
t
o
v
ar
y
in
g
illu
m
in
atio
n
,
b
ac
k
g
r
o
u
n
d
cl
u
tter
,
an
d
s
en
s
o
r
n
o
is
e
[
7
]
,
an
d
tr
a
d
itio
n
al
co
m
p
u
ter
-
v
is
io
n
p
ip
elin
es
m
ay
b
e
less
r
o
b
u
s
t f
o
r
f
ield
-
s
ca
le
d
ep
lo
y
m
en
t
[
8
]
.
Deep
-
lear
n
in
g
a
r
ch
itectu
r
es
th
at
en
ab
le
h
ier
ar
c
h
ical
tr
ain
in
g
o
f
th
e
n
etwo
r
k
h
a
v
e
b
ee
n
p
r
o
p
o
s
ed
to
o
v
er
co
m
e
th
ese
lim
itatio
n
s
.
Fo
r
in
ter
p
r
eta
b
le
d
is
ea
s
e
d
iag
n
o
s
is
in
cr
o
p
s
lik
e
s
o
y
b
ea
n
,
h
y
b
r
id
C
NN
–
g
r
ap
h
n
eu
r
al
n
etwo
r
k
(
GNN)
m
o
d
els
h
av
e
b
ee
n
s
tu
d
ied
[
9
]
,
an
d
tr
an
s
f
o
r
m
er
-
b
ased
v
is
io
n
m
o
d
els
o
f
f
er
a
s
tr
o
n
g
f
ea
tu
r
e
-
ex
tr
ac
tio
n
ca
p
ab
ilit
y
f
o
r
d
ietar
y
im
a
g
es
u
s
ed
in
a
g
r
icu
l
tu
r
al
a
p
p
licatio
n
s
[
1
0
]
.
I
n
th
e
c
o
n
tex
t
o
f
s
tr
awb
er
r
y
r
esear
ch
,
m
o
r
e
d
etec
t
-
o
n
ly
YOL
Ov
8
-
b
ased
p
est
an
d
d
is
ea
s
e
d
etec
t
io
n
m
eth
o
d
s
h
av
e
b
ee
n
estab
lis
h
ed
.
He
et
a
l.
[
1
1
]
p
r
o
p
o
s
ed
th
e
KT
D
-
YOL
Ov
8
m
o
d
el,
wh
er
e
KT
D
s
tan
d
s
f
o
r
Ker
n
elW
ar
eh
o
u
s
e
co
n
v
o
l
u
tio
n
+
tr
ip
let
atten
tio
n
+
m
u
lti
-
tem
p
late
b
r
an
ch
b
l
o
ck
s
h
ar
in
g
h
ea
d
.
C
lass
ical
C
NN
-
b
ased
s
y
s
tem
s
also
r
em
ain
u
s
ef
u
l f
o
r
s
tr
awb
e
r
r
y
d
is
ea
s
e
d
etec
tio
n
u
n
d
er
s
em
i
-
co
n
tr
o
lled
an
d
f
ield
co
n
d
itio
n
s
[
1
2
]
.
Fu
r
th
er
m
o
r
e
,
r
at
h
er
t
h
an
s
im
p
le
ca
teg
o
r
i
s
atio
n
(
f
r
o
m
wh
i
ch
C
NN
-
b
ased
n
etwo
r
k
s
in
f
ac
t
d
o
n
o
t
o
f
f
er
a
co
m
p
etitiv
e
ad
v
a
n
tag
e
)
,
m
o
r
e
s
o
p
h
is
ticated
d
ee
p
n
etwo
r
k
ar
c
h
itectu
r
es
h
av
e
b
ee
n
ap
p
lied
to
p
r
e
v
en
t
d
eg
r
ad
atio
n
o
f
d
etec
to
r
p
er
f
o
r
m
an
ce
with
r
esp
ec
t
t
o
d
o
m
ain
-
s
p
ec
if
ic
d
is
to
r
tio
n
s
s
u
ch
as
o
cc
lu
s
io
n
an
d
b
ac
k
g
r
o
u
n
d
clu
tter
[
1
3
]
.
Fo
r
s
tr
awb
er
r
y
leaf
d
is
ea
s
e
id
e
n
tific
atio
n
,
in
ter
p
r
etab
le
an
d
lig
h
tweig
h
t
d
ee
p
lear
n
in
g
p
ip
elin
es h
a
v
e
b
ee
n
d
esig
n
ed
to
en
h
an
ce
p
r
o
ce
s
s
in
g
p
er
f
o
r
m
an
ce
o
n
e
d
g
e
-
c
o
m
p
u
t
in
g
d
ev
ices a
s
well
[
1
4
]
.
M
o
r
e
g
en
er
al
ly
,
th
e
ev
o
lu
tio
n
o
f
p
lan
t
p
h
en
o
ty
p
in
g
,
s
u
ch
as
3
D
r
ec
o
n
s
tr
u
ctio
n
,
tr
ait
m
o
d
ellin
g
,
an
d
s
p
atial
f
ea
tu
r
e
u
n
d
er
s
tan
d
in
g
,
h
ig
h
lig
h
ts
th
at
m
o
r
e
ad
v
a
n
ce
d
d
ata
r
ep
r
esen
tatio
n
s
ar
e
b
ec
o
m
in
g
m
o
r
e
p
r
o
m
in
e
n
t
in
th
e
f
i
eld
o
f
c
r
o
p
h
ea
lth
[
1
5
]
.
Similar
ly
,
s
en
s
o
r
f
u
s
io
n
an
d
AI
-
en
ab
le
d
an
aly
tics
h
av
e
m
ad
e
tr
ee
-
cr
o
p
s
m
ar
t
s
p
r
ay
e
r
s
f
ar
s
m
ar
ter
,
with
h
ig
h
-
lev
el
d
e
m
o
n
s
tr
atio
n
s
s
h
o
win
g
h
o
w
th
ese
s
m
ar
t
m
o
d
els
p
r
o
v
id
e
a
p
r
ac
tical
s
o
lu
tio
n
s
f
o
r
d
is
ea
s
e
m
an
ag
em
e
n
t a
t th
e
f
ar
m
lev
el
[
1
6
]
.
E
x
p
lai
n
ab
le
ar
tific
ial
in
tellig
en
ce
(
XAI
)
h
as
e
m
er
g
e
d
to
clar
if
y
h
o
w
m
o
d
el
in
p
u
ts
in
f
lu
en
ce
p
r
ed
ictio
n
s
.
L
o
ca
l
in
ter
p
r
etab
l
e
m
o
d
el
-
ag
n
o
s
tic
ex
p
lan
atio
n
s
(
L
I
ME
)
p
r
o
v
i
d
e
in
s
tan
ce
-
le
v
el
ex
p
lan
atio
n
s
b
y
id
en
tify
in
g
im
a
g
e
r
e
g
io
n
s
t
h
at
in
f
lu
en
ce
a
class
if
ier
's
o
u
tp
u
t
[
1
7
]
.
I
n
a
g
r
icu
ltu
r
al
d
is
ea
s
e
d
etec
tio
n
,
m
ac
h
in
e
-
lear
n
i
n
g
m
o
d
els
h
a
v
e
also
b
ee
n
ev
alu
ated
o
n
cu
s
t
omi
s
ed
d
atasets
to
ex
am
in
e
th
eir
p
r
ac
tical
p
er
f
o
r
m
an
ce
a
n
d
d
e
p
lo
y
m
en
t
lim
itatio
n
s
[
1
8
]
.
C
o
m
p
r
eh
en
s
iv
e
r
ev
iews
o
f
d
ee
p
lear
n
in
g
in
ag
r
icu
ltu
r
e
co
n
tin
u
e
to
em
p
h
asi
s
e
th
e
n
ee
d
f
o
r
s
ca
lab
le,
r
o
b
u
s
t,
an
d
in
ter
p
r
etab
le
ap
p
r
o
ac
h
es
to
d
is
ea
s
e
d
etec
tio
n
an
d
cr
o
p
m
o
n
ito
r
in
g
[
1
9
]
.
On
g
o
in
g
ef
f
o
r
ts
to
cr
ea
te
m
o
d
els
th
at
o
p
er
ate
e
f
f
icien
tly
wit
h
lim
ited
r
eso
u
r
ce
s
ar
e
o
u
tlin
ed
in
[
2
0
]
th
r
o
u
g
h
co
m
p
a
r
ativ
e
s
tu
d
ies
o
f
s
tr
ateg
ies
f
o
r
f
in
e
-
tu
n
in
g
f
o
r
p
lan
t
-
d
is
ea
s
e
r
ec
o
g
n
itio
n
an
d
in
[
2
1
]
th
r
o
u
g
h
th
e
d
esig
n
of
lig
h
tweig
h
t
atten
ti
o
n
-
b
ased
C
NNs.
XAI
is
also
b
e
co
m
i
n
g
in
cr
ea
s
in
g
ly
s
ig
n
i
f
ican
t
in
p
r
ec
is
io
n
ag
r
icu
ltu
r
e,
as
f
ar
m
er
s
an
d
ag
r
icu
ltu
r
al
in
d
u
s
tr
y
p
lay
e
r
s
r
eq
u
ir
e
tr
a
n
s
p
ar
en
t
au
t
o
m
at
ed
d
ec
is
io
n
-
s
u
p
p
o
r
t
s
y
s
tem
s
[
2
2
]
.
Nev
er
th
eless
,
s
t
an
d
ar
d
d
ata
-
d
r
iv
e
n
tech
n
iq
u
es
co
n
tin
u
e
to
p
e
r
f
o
r
m
well
as
b
aselin
es
f
o
r
p
lan
t
-
d
is
ea
s
e
class
if
icat
io
n
[
2
3
]
,
wh
ils
t
m
o
r
e
in
-
d
ep
th
s
tu
d
y
o
f
d
ee
p
-
lear
n
i
n
g
m
eth
o
d
s
f
o
r
p
lan
t
p
h
e
n
o
ty
p
i
n
g
u
n
co
v
e
r
s
th
e
p
o
te
n
tial o
f
ad
v
a
n
ce
d
f
ea
tu
r
e
lear
n
in
g
with
in
a
g
r
icu
ltu
r
al
d
o
m
ain
s
[
2
4
]
.
I
n
ad
d
itio
n
to
p
lan
t
an
d
d
is
ea
s
e
clas
s
if
icatio
n
,
d
ee
p
-
lear
n
i
n
g
tech
n
iq
u
es
h
av
e
b
ee
n
s
u
g
g
ested
f
o
r
p
lan
t
-
d
is
ea
s
e
s
ev
er
ity
esti
m
atio
n
[
2
5
]
o
r
d
etec
tin
g
b
io
ti
c
s
tr
ess
in
d
if
f
er
en
t
g
r
o
wth
co
n
d
itio
n
s
[
2
6
]
.
Mo
r
e
co
m
p
r
eh
en
s
iv
e
r
e
v
iews o
f
d
ee
p
lea
r
n
in
g
ap
p
licatio
n
s
i
n
ag
r
icu
ltu
r
e
[
2
0
]
co
n
tin
u
e
to
em
p
h
asi
s
e
th
e
n
ee
d
f
o
r
g
en
er
aliza
b
le,
s
ca
lab
le
,
an
d
r
o
b
u
s
t
AI
s
y
s
tem
s
th
at
ar
e
ca
p
a
b
le
o
f
d
ep
l
o
y
m
e
n
t
in
to
r
ea
l
f
ield
en
v
ir
o
n
m
en
ts
.
T
o
ad
d
r
ess
th
ese
ch
allen
g
es,
th
is
p
a
p
er
p
r
esen
ts
a
GNN
–
ML
P
h
y
b
r
id
ar
ch
itectu
r
e
f
o
r
class
if
icatio
n
o
f
d
if
f
er
en
t
ty
p
es
o
f
s
tr
awb
er
r
y
d
is
ea
s
es
b
ased
o
n
f
ar
m
-
r
ea
lis
tic
im
ag
es.
T
h
e
m
o
d
el
in
co
r
p
o
r
ates
th
e
f
o
llo
win
g
elem
en
ts
:
i)
co
n
tr
ast
n
o
r
m
al
i
s
atio
n
to
m
itig
ate
illu
m
in
at
io
n
v
ar
ia
b
ilit
y
;
ii)
Gau
s
s
ian
d
en
o
is
in
g
to
f
ilter
o
u
t
b
ac
k
g
r
o
u
n
d
n
o
is
e;
iii)
s
y
n
th
etic
h
y
p
er
s
p
ec
tr
al
s
im
u
la
tio
n
f
o
r
en
r
ich
in
g
d
is
ea
s
e
-
in
d
icativ
e
in
f
o
r
m
atio
n
;
iv
)
g
r
ap
h
-
b
ased
r
ea
s
o
n
in
g
to
m
o
d
el
r
elatio
n
s
b
etwe
en
lesi
o
n
to
p
o
lo
g
y
an
d
s
tr
u
ctu
r
e;
v
)
ML
P
-
b
ased
s
em
an
tic
f
u
s
io
n
o
f
v
is
u
al
atten
tio
n
m
ap
s
;
an
d
v
i)
XAI
m
o
d
u
les
,
s
p
ec
if
ical
ly
L
I
ME
an
d
Gr
ad
ien
t
-
weig
h
te
d
C
lass
Activ
atio
n
Ma
p
p
in
g
(
Gr
a
d
-
C
AM
)
,
ca
p
ab
le
o
f
g
en
e
r
atin
g
h
u
m
an
-
in
ter
p
r
etab
le
b
io
lo
g
ical
d
ec
is
io
n
m
ap
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
7
7
6
I
n
t J
I
n
f
&
C
o
m
m
u
n
T
ec
h
n
o
l
,
Vo
l.
15
,
No
.
3
,
Sep
tem
b
er
20
26
:
1
3
6
4
-
1
3
7
5
1366
T
o
th
is
en
d
,
we
p
r
o
p
o
s
e
a
k
n
o
wled
g
e
-
b
ased
d
is
ea
s
e
-
d
iag
n
o
s
is
p
ip
elin
e
th
at
is
co
m
p
ac
t
an
d
u
s
ab
le
ac
r
o
s
s
h
eter
o
g
en
e
o
u
s
ag
r
i
cu
ltu
r
al
en
v
ir
o
n
m
e
n
ts
an
d
e
d
g
e
-
c
o
m
p
u
tin
g
p
latf
o
r
m
s
wh
ile
en
s
u
r
in
g
c
o
m
p
u
tatio
n
al
ef
f
icien
cy
;
th
u
s
,
p
r
esen
tin
g
an
en
d
-
to
-
en
d
a
u
to
m
atic
p
e
s
t
p
ath
o
g
en
d
etec
tio
n
s
o
lu
tio
n
t
h
at
is
r
o
b
u
s
t,
ce
r
tain
,
an
d
in
ter
p
r
eta
b
le.
T
h
e
r
est
o
f
th
e
m
an
u
s
cr
ip
t
is
o
r
g
an
i
s
ed
as
f
o
llo
ws:
b
ac
k
g
r
o
u
n
d
liter
atu
r
e
o
n
p
lan
t
d
is
ea
s
e
d
etec
tio
n
,
h
y
p
er
s
p
e
ctr
al
m
eth
o
d
s
,
an
d
g
r
ap
h
-
b
ased
m
o
d
elin
g
a
r
e
r
ev
ie
wed
in
Sectio
n
2
,
m
eth
o
d
s
d
escr
ip
tio
n
is
in
Sectio
n
3
,
th
e
ex
p
e
r
im
en
tal
r
esu
lts
in
Sectio
n
4
,
f
o
llo
wed
b
y
d
i
s
cu
s
s
io
n
in
Sectio
n
5
,
an
d
t
h
e
co
n
clu
s
io
n
in
Sectio
n
6
.
2.
RE
L
AT
E
D
WO
RK
E
ar
ly
p
lan
t
-
d
is
ea
s
e
d
etec
tio
n
m
eth
o
d
s
u
s
ed
h
an
d
cr
af
ted
c
o
lo
u
r
,
te
x
tu
r
e,
a
n
d
s
h
a
p
e
f
ea
tu
r
es
with
class
if
ier
s
s
u
ch
as
s
u
p
p
o
r
t
v
ec
to
r
m
ac
h
in
es
(
SVMs),
r
an
d
o
m
f
o
r
ests
(
R
F
s
)
,
an
d
k
-
n
ea
r
est
n
eig
h
b
o
u
r
s
(
k
-
NN)
[
8
]
.
T
h
eir
p
er
f
o
r
m
a
n
ce
o
f
ten
d
ec
l
in
es
in
r
ea
l
-
wo
r
ld
co
n
d
iti
o
n
s
b
ec
au
s
e
o
f
illu
m
in
atio
n
ch
an
g
es,
b
ac
k
g
r
o
u
n
d
clu
tter
,
an
d
s
en
s
o
r
n
o
is
e
[7
]
,
[
1
9
]
.
T
h
is
is
s
im
ilar
to
h
ier
ar
ch
ical
f
ea
tu
r
e
ex
tr
ac
tio
n
,
wh
ich
y
o
u
ca
n
tak
e
ad
v
an
tag
e
o
f
with
d
ee
p
lear
n
in
g
a
n
d
co
n
v
o
lu
tio
n
al
n
e
u
r
a
l
n
etwo
r
k
s
(
C
NNs)
ca
n
tak
e
ad
v
a
n
tag
e
o
f
with
r
esp
ec
t
to
m
u
lti
-
cr
o
p
d
is
ea
s
e
class
if
icatio
n
;
Fer
en
tin
o
s
[
4
]
h
ig
h
lig
h
ted
th
at
th
is
task
co
u
l
d
tr
iv
ially
b
e
s
o
lv
ed
by
one
-
vs
-
all
-
b
ased
class
if
ier
en
s
em
b
les
(
r
ev
is
itin
g
th
e
ea
r
ly
1
9
9
0
s
)
.
At
th
e
s
am
e
tim
e,
L
iu
an
d
W
an
g
[
5
]
p
r
esen
ted
a
r
ev
iew
o
n
d
ee
p
-
le
ar
n
in
g
me
th
o
d
s
f
o
r
v
ar
io
u
s
p
l
an
t
-
d
is
ea
s
e
an
d
p
est
d
etec
tio
n
u
s
e
c
ases
.
Ma
h
m
u
d
et
a
l.
[
6
]
p
r
o
p
o
s
ed
a
m
o
b
ile
m
ac
h
in
e
-
v
is
io
n
s
y
s
tem
to
i
d
en
tify
s
tr
awb
er
r
y
p
o
wd
er
y
m
ild
ew
in
r
ea
l
tim
e.
L
o
n
g
-
r
an
g
e
v
is
u
al
f
ea
tu
r
e
m
o
d
ellin
g
is
also
s
u
p
p
o
r
ted
b
y
tr
an
s
f
o
r
m
er
-
b
ased
m
o
d
els
[
1
0
]
.
B
u
t
g
r
id
-
b
ased
C
NNs
an
d
tr
an
s
f
o
r
m
er
s
m
ay
n
o
t
ex
p
licitly
m
o
d
el
ir
r
eg
u
lar
r
elatio
n
s
h
ip
s
b
etwe
en
lesi
o
n
s
.
T
o
o
v
er
co
m
e
th
is
lim
itatio
n
,
GNNs
r
ep
r
esen
t
lesi
o
n
r
eg
i
o
n
s
as
g
r
a
p
h
a
n
d
t
h
eir
s
p
atial
r
elatio
n
s
h
ip
as
ed
g
es
[
9
]
,
[
2
7
]
.
T
h
er
e
f
o
r
e,
h
y
b
r
id
C
NN
–
GNN
s
tr
u
ctu
r
es
ca
n
in
teg
r
ate
lo
ca
l
tex
tu
r
e
f
ea
tu
r
es
with
r
elatio
n
al
in
f
o
r
m
atio
n
an
d
en
h
an
ce
th
eir
r
o
b
u
s
tn
ess
u
n
d
er
h
ete
r
o
g
en
eo
u
s
f
ield
c
o
n
d
itio
n
s
o
f
th
e
f
ield
[
9
]
.
Hy
p
er
s
p
ec
tr
al
a
n
d
m
u
ltis
p
e
ctr
al
im
ag
in
g
ar
e
u
s
ed
b
ec
a
u
s
e
t
h
ey
allo
w
an
ea
r
ly
d
iag
n
o
s
is
o
f
d
is
ea
s
e
s
b
y
v
is
u
ali
s
in
g
th
e
f
ir
s
t
b
io
ch
em
ical
ch
an
g
es
b
ef
o
r
e
v
is
ib
le
s
y
m
p
to
m
s
b
ec
o
m
e
o
b
v
io
u
s
.
J
ian
g
et
a
l.
[
2
]
,
u
s
in
g
m
ac
h
in
e
lear
n
i
n
g
a
n
d
s
p
ec
tr
al
f
in
g
er
p
r
in
t
f
ea
t
u
r
es,
s
h
o
wed
th
at
h
y
p
e
r
s
p
ec
tr
al
im
ag
in
g
is
a
n
ef
f
ec
tiv
e
t
o
o
l
f
o
r
ea
r
ly
id
e
n
tific
atio
n
o
f
s
tr
awb
er
r
y
lea
f
d
is
ea
s
es.
Du
e
to
t
h
e
c
o
s
t
-
p
r
o
h
ib
itiv
e
n
atu
r
e
a
n
d
s
en
s
itiv
ity
o
f
d
ed
icate
d
hype
r
s
p
ec
tr
al
eq
u
ip
m
en
t
to
a
cq
u
is
itio
n
co
n
d
itio
n
s
,
co
m
p
u
t
atio
n
al
s
p
ec
tr
al
s
u
r
r
o
g
ate
m
eth
o
d
s
d
er
iv
ed
f
r
o
m
R
GB
im
ag
es
m
ay
p
r
o
v
id
e
a
p
r
ac
tical
alter
n
ativ
e.
An
o
t
h
er
p
r
o
b
lem
is
th
e
d
iv
er
s
ity
o
f
th
e
d
atasets
:
lab
-
s
ty
le
d
atasets
ca
n
d
r
iv
e
m
o
d
el
d
ev
elo
p
m
en
t
b
u
t
ar
e
n
o
t
n
ec
ess
ar
ily
r
ep
r
esen
tativ
e
o
f
f
ield
co
n
d
itio
n
s
,
wh
ich
ar
e
o
f
ten
ch
a
r
ac
ter
i
s
ed
b
y
im
b
ala
n
ce
,
o
cc
lu
s
io
n
,
an
d
b
ac
k
g
r
o
u
n
d
clu
tter
[
7
]
,
[
1
9
]
,
[
2
3
]
.
E
x
p
lain
ab
ilit
y
m
eth
o
d
s
s
u
ch
as
L
I
ME
[
1
7
]
,
SHap
le
y
ad
d
itiv
e
e
x
Plan
atio
n
s
(
SHAP),
an
d
Gr
ad
-
C
AM
ca
n
also
g
en
er
ate
v
is
u
al
ev
id
en
ce
o
f
m
o
d
el
d
ec
is
io
n
s
;
h
o
wev
er
,
in
s
tr
awb
er
r
y
d
is
ea
s
e
r
esear
ch
,
ex
p
lain
ab
ilit
y
h
as
n
o
t
b
ee
n
f
u
lly
em
p
lo
y
ed
[
2
2
]
.
R
ec
en
t
d
ev
elo
p
m
en
ts
in
th
e
i
n
ter
n
et
o
f
th
i
n
g
s
(
I
o
T
)
,
u
n
cr
e
w
ed
ae
r
ial
v
eh
icle
(
UAV)
im
ag
in
g
,
an
d
ed
g
e
AI
h
a
v
e
ex
p
an
d
e
d
t
h
e
p
o
s
s
ib
ilit
ie
s
f
o
r
s
ca
lab
le
cr
o
p
s
u
r
v
eillan
ce
.
Mo
h
an
t
y
et
a
l.
[
2
3
]
d
e
m
o
n
s
tr
ated
th
e
ef
f
ec
tiv
en
ess
o
f
d
ee
p
lear
n
in
g
f
o
r
lar
g
e
-
s
ca
le
im
ag
e
-
b
ased
p
lan
t
-
d
is
ea
s
e
d
etec
tio
n
,
wh
il
e
Sin
g
h
et
a
l.
[
2
4
]
em
p
h
asi
s
ed
its
im
p
o
r
tan
ce
f
o
r
h
ig
h
-
th
r
o
u
g
h
p
u
t
p
lan
t
p
h
en
o
ty
p
in
g
.
L
ig
h
tweig
h
t
d
is
ea
s
e
-
s
ev
er
ity
esti
m
atio
n
m
eth
o
d
s
h
a
v
e
also
b
ee
n
d
ev
e
lo
p
ed
f
o
r
p
r
ac
tical
f
ield
ap
p
li
ca
tio
n
s
[
2
5
]
.
Stra
wb
er
r
y
-
f
o
cu
s
ed
s
tu
d
ies
in
clu
d
e
im
p
r
o
v
e
d
YOL
Ov
8
m
o
d
els
[
2
8
]
,
KT
D
-
YOL
Ov
8
d
is
ea
s
e
d
etec
tio
n
[
1
1
]
,
C
NN
-
b
ased
d
et
ec
tio
n
s
y
s
tem
s
[
4
]
,
[
6
]
,
[
1
2
]
,
an
d
lig
h
tweig
h
t
i
n
ter
p
r
etab
le
d
is
ea
s
e
-
r
ec
o
g
n
itio
n
ap
p
r
o
ac
h
es
[
1
4
]
.
Oth
er
r
elat
ed
wo
r
k
in
clu
d
es
g
r
ap
h
-
b
ased
p
la
n
t
s
p
ec
ies
class
if
icatio
n
[
2
8
]
,
YOL
Ov
5
-
b
a
s
ed
h
o
r
ticu
ltu
r
al
p
h
e
n
o
ty
p
in
g
[
2
8
]
,
d
u
al
-
b
r
a
n
c
h
cr
o
p
-
d
is
ea
s
e
r
ec
o
g
n
itio
n
[
2
8
]
,
s
atellite
tim
e
-
s
er
ie
s
atten
tio
n
m
o
d
els
[
2
8
]
,
s
p
ec
tr
al
-
s
p
atia
l
S
iam
ese
C
NN
s
[
2
8
]
,
an
d
lig
h
tweig
h
t Y
OL
Ov
5
s
tr
awb
er
r
y
d
etec
t
o
r
s
[
2
8
]
.
C
o
llectiv
ely
,
p
r
io
r
liter
atu
r
e
r
ev
ea
ls
p
er
s
is
ten
t
lim
itatio
n
s
:
i
)
in
s
u
f
f
icien
t
i
n
ter
p
r
et
ab
ilit
y
in
h
ig
h
-
p
e
r
f
o
r
m
in
g
m
o
d
els;
ii)
r
ed
u
ce
d
r
o
b
u
s
tn
ess
u
n
d
er
u
n
co
n
tr
o
lled
f
ield
c
o
n
d
itio
n
s
;
i
ii)
in
ad
eq
u
ate
m
o
d
ellin
g
o
f
r
elatio
n
al
lesi
o
n
s
tr
u
ctu
r
e;
iv
)
lack
o
f
u
n
if
ied
s
p
ec
tr
al
-
en
h
an
ce
d
an
d
g
r
ap
h
-
b
ased
ex
p
lain
ab
le
p
ip
elin
es;
v)
d
o
m
ain
-
s
h
if
t
-
d
r
iv
en
g
en
e
r
ali
s
atio
n
f
ailu
r
es.
T
h
ese
g
ap
s
m
o
tiv
ate
in
t
eg
r
atin
g
s
y
n
th
etic
h
y
p
er
s
p
ec
tr
al
en
h
an
ce
m
e
n
t,
GNN
-
b
ased
r
elatio
n
al
m
o
d
ell
in
g
,
an
d
XAI
in
to
a
u
n
if
ied
,
r
o
b
u
s
t
s
tr
awb
er
r
y
-
d
is
ea
s
e
d
etec
tio
n
f
r
am
ewo
r
k
.
3.
M
E
T
H
O
D
I
n
th
is
wo
r
k
,
we
p
r
o
p
o
s
e
a
h
y
b
r
id
d
ee
p
lear
n
in
g
m
eth
o
d
th
at
co
m
b
in
es
GNNs
an
d
Mu
lti
-
L
ay
er
Per
ce
p
tr
o
n
s
(
ML
Ps
)
f
o
r
au
t
o
m
ated
d
etec
ti
o
n
o
f
s
tr
awb
e
r
r
y
f
r
u
it
d
is
ea
s
es.
T
h
e
m
o
d
el
ai
m
s
ex
p
licitly
at
th
r
ee
co
m
m
o
n
l
y
o
cc
u
r
r
in
g
a
n
d
ec
o
n
o
m
ically
im
p
o
r
tan
t
d
is
ea
s
es
:
an
th
r
ac
n
o
s
e
f
r
u
it
r
o
t,
g
r
ey
mo
u
ld
,
an
d
p
o
wd
er
y
m
ild
ew
,
all
o
f
wh
ic
h
v
is
u
ally
r
esem
b
le
ea
ch
o
th
er
to
s
o
m
e
ex
ten
t
u
n
d
er
u
n
c
o
n
tr
o
lle
d
f
ar
m
c
o
n
d
itio
n
s
.
T
h
eir
m
o
r
p
h
o
lo
g
y
is
af
f
ec
ted
b
y
en
v
ir
o
n
m
e
n
tal
co
n
d
itio
n
s
,
lig
h
t
q
u
ality
,
ir
r
ig
atio
n
lev
els
,
as
well
as
cu
ltiv
ar
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J I
n
f
&
C
o
m
m
u
n
T
ec
h
n
o
l
I
SS
N:
2252
-
8
7
7
6
E
xp
la
in
a
b
le
A
I
-
b
a
s
ed
h
yb
r
id
GN
N
-
ML
P
mo
d
el
fo
r
s
tr
a
w
b
err
y
fr
u
it
d
is
ea
s
e
… (
V
a
r
s
h
a
K
i
r
a
n
B
h
o
s
a
le
)
1367
an
d
f
r
u
it
r
ip
e
n
in
g
s
tag
e.
T
h
e
r
e
b
y
,
m
a
n
u
al
c
h
ec
k
in
g
is
u
n
r
eliab
le,
illo
g
ical
,
a
n
d
tim
e
-
c
o
n
s
u
m
in
g
,
esp
ec
ially
o
n
lar
g
e
-
s
ca
le
p
r
o
d
u
ctio
n
s
y
s
tem
s
[
1
]
,
[
1
9
]
.
T
h
e
id
ea
o
f
th
e
p
r
o
p
o
s
ed
m
et
h
o
d
is
t
o
tak
e
ad
v
an
ta
g
e
o
f
b
o
t
h
th
e
s
p
atial
s
tr
u
ctu
r
al
lear
n
i
n
g
ab
ilit
y
o
f
GNNs
an
d
th
e
s
em
an
tic
ab
s
tr
ac
tio
n
ab
ilit
y
of
MLP
s
.
GNNs
ca
p
tu
r
e
s
p
atial
lesi
o
n
t
o
p
o
lo
g
y
,
ad
jace
n
cy
p
atter
n
s
,
an
d
c
r
o
s
s
-
r
eg
io
n
co
n
tex
tu
al
r
elatio
n
s
h
ip
s
,
wh
ile
ML
Ps
u
s
e
p
o
o
led
f
ea
tu
r
es
f
o
r
p
o
wer
f
u
l
g
l
o
b
al
d
is
cr
im
in
atio
n
.
T
h
is
d
u
al
-
lea
r
n
in
g
d
ef
i
n
itio
n
is
well
ad
ap
ted
f
o
r
th
e
a
g
r
icu
ltu
r
al
im
ag
e
s
wh
er
e
s
y
m
p
to
m
s
r
an
d
o
m
l
y
h
ap
p
en
o
n
th
e
f
r
u
its
with
o
cc
lu
s
io
n
o
r
im
ag
e
n
o
is
e
[
4
]
,
[
7
]
.
3
.
1
.
Da
t
a
s
et
a
cquis
it
io
n a
nd
prepro
ce
s
s
i
ng
An
ad
ap
tab
le
d
atab
ase
was c
r
ea
ted
f
o
r
t
h
e
tr
ain
in
g
an
d
v
alid
atio
n
o
f
th
e
h
y
b
r
id
m
o
d
el:
-
Pu
b
lic
-
r
elea
s
e
s
am
p
les
(
Plan
tVillag
e,
Kag
g
le)
with
clea
n
f
ield
c
o
n
d
itio
n
im
ag
e
s
p
r
o
m
p
tin
g
well
-
co
n
s
tr
u
cted
an
n
o
tatio
n
s
[
8
]
,
[
9
]
.
-
Natu
r
al
f
ar
m
im
ag
es
c
o
llected
u
n
d
er
ex
tr
em
el
y
v
a
r
iab
le
e
n
v
ir
o
n
m
en
tal
c
o
n
d
itio
n
s
,
in
cl
u
d
in
g
u
n
e
v
e
n
illu
m
in
atio
n
,
ca
s
t sh
ad
o
ws,
d
ir
t,
cr
o
p
r
esid
u
es,
leaf
o
cc
lu
s
io
n
s
,
an
d
f
r
u
it a
t d
if
f
er
e
n
t a
n
g
les
[
1
0
]
,
[
1
1
]
.
-
L
ab
els
:
t
r
ain
ed
p
lan
t
p
ath
o
l
o
g
is
ts
cu
r
ated
lab
els
to
g
u
ar
a
n
tee
d
iag
n
o
s
tic
ac
cu
r
ac
y
an
d
to
elim
in
ate
am
b
ig
u
ities
in
s
am
p
les
[
1
2
]
.
Usi
n
g
a
c
o
m
b
in
atio
n
o
f
c
o
n
tr
o
lled
an
d
r
ea
l
-
f
ield
im
ag
es,
a
s
it
in
cr
ea
s
es
r
o
b
u
s
tn
ess
,
d
ec
r
ea
s
es
o
v
er
f
itti
n
g
o
n
clea
n
b
ac
k
g
r
o
u
n
d
s
,
an
d
at
last
it
en
s
u
r
es
th
at
th
e
m
o
d
e
l
can
ac
tu
ally
b
e
u
s
ed
in
f
ar
m
s
.
3
.
1
.
1
.
I
m
a
g
e
re
s
izing
All
th
e
lesi
o
n
s
ar
e
b
ein
g
r
esized
to
2
5
6
×
2
5
6
in
d
iv
id
u
al
p
i
x
els,
s
o
th
eir
wid
th
an
d
h
eig
h
t
wer
e
s
et
t
o
ac
h
iev
e
th
e
co
m
m
o
n
ly
f
o
u
n
d
r
eso
lu
tio
n
lim
its
,
wh
ich
allo
w
p
r
eser
v
in
g
b
asic
lesi
o
n
p
r
o
p
e
r
ties
an
d
r
ed
u
cin
g
ex
ce
s
s
iv
e
co
m
p
u
tatio
n
c
o
s
ts
.
T
h
is
s
tan
d
ar
d
is
at
io
n
:
-
Facilitate
s
ef
f
icien
t b
atch
tr
ain
in
g
.
-
R
ed
u
ce
s
GPU/
C
PU m
em
o
r
y
lo
ad
.
-
E
n
s
u
r
es
th
at
th
e
f
ix
e
d
in
p
u
t la
y
er
s
o
f
th
e
n
eu
r
al
n
etwo
r
k
a
r
e
co
m
p
atib
le
.
-
R
ed
u
ce
s
in
ter
p
o
latio
n
er
r
o
r
s
o
n
s
m
all
lesi
o
n
s
(
wh
ich
h
av
e
cl
in
ical
s
ig
n
if
ican
ce
)
[
1
4
]
,
[
1
5
]
.
3
.
1
.
2
.
Co
ntr
a
s
t
enha
ncem
ent
C
o
n
tr
ast
-
lim
ited
ad
ap
tiv
e
h
is
to
g
r
am
eq
u
alis
atio
n
(
C
L
AHE
)
,
is
a
m
eth
o
d
f
o
r
co
n
tr
o
llin
g
n
o
n
-
u
n
if
o
r
m
illu
m
in
atio
n
m
ec
h
a
n
is
m
s
th
at
ca
n
b
e
p
r
esen
t
in
o
p
en
-
f
ield
f
ar
m
en
v
ir
o
n
m
en
ts
b
ec
au
s
e
o
f
d
ir
ec
t
s
u
n
lig
h
t,
s
h
ad
o
win
g
,
a
n
d
a
c
o
m
b
in
atio
n
o
f
d
if
f
u
s
e
lig
h
t
[
1
6
]
,
[
1
9
]
.
C
L
AHE
:
-
Am
p
lifie
s
f
in
e
-
g
r
ain
e
d
d
am
a
g
es f
r
eq
u
en
tly
n
o
t v
is
ib
le
in
t
h
e
r
aw
r
ed
-
g
r
ee
n
-
b
lu
e
(
R
GB
)
im
ag
es
.
-
R
ev
ea
ls
ea
r
ly
-
s
tag
e
in
f
ec
tio
n
p
atter
n
s
.
-
E
n
h
an
ce
s
d
is
cr
im
in
atio
n
o
f
p
a
th
o
lo
g
ical
tis
s
u
e
f
r
o
m
n
o
r
m
al
s
k
in
.
-
Gu
ar
d
s
ag
ain
s
t o
v
er
-
im
p
ac
ted
n
o
is
e
(
b
y
c
o
n
tr
ast lim
itin
g
)
.
-
T
h
is
en
h
an
ce
s
th
e
d
e
f
in
itio
n
o
f
lesi
o
n
m
ar
g
in
s
,
f
u
n
g
al
p
atter
n
s
,
an
d
d
is
co
lo
u
r
ed
a
r
ea
s
.
3
.
1
.
3
.
Deno
is
ing
a
nd
s
y
nthet
i
c
hy
persp
ec
t
ra
l sim
ula
t
io
n
:
A
Gau
s
s
ian
f
ilter
is
ap
p
lied
to
t
h
e
ca
lib
r
atio
n
im
a
g
e,
wh
ich
h
elp
s
ef
f
ec
tiv
ely
s
u
p
p
r
ess
s
en
s
o
r
n
o
is
e,
m
o
tio
n
n
o
is
e
,
an
d
lo
w
lig
h
t
n
o
is
e
p
r
esen
t
in
th
e
f
ar
m
im
ag
es
tak
en
f
r
o
m
s
m
ar
tp
h
o
n
es
[
1
9
]
.
Den
o
is
in
g
is
f
o
llo
wed
b
y
a
s
y
n
th
etic
h
y
p
er
s
p
ec
tr
al
tr
an
s
f
o
r
m
,
wh
ich
ex
p
an
d
s
th
e
n
u
m
b
er
o
f
ch
an
n
els
with
r
ich
er
in
f
o
r
m
atio
n
.
I
n
s
tead
o
f
u
s
in
g
(
s
till
p
r
icey
)
h
y
p
e
r
s
p
ec
tr
al
c
am
er
as,
th
e
ap
p
r
o
ac
h
c
o
m
p
u
t
atio
n
ally
g
en
er
ates
ad
d
itio
n
al
f
r
e
q
u
en
c
y
b
a
n
d
s
b
y
:
-
PC
A
-
b
ased
s
p
ec
tr
al
p
r
o
jectio
n
s
.
-
Gr
ad
ien
t a
n
d
te
x
tu
r
e
f
ilter
s
.
-
Veg
etatio
n
s
tr
ess
in
d
ices
.
-
C
o
lo
u
r
–
f
r
e
q
u
en
c
y
d
ec
o
m
p
o
s
it
io
n
.
T
h
ese
s
p
ec
tr
al
s
u
r
r
o
g
ates
ca
p
t
u
r
e
in
f
o
r
m
atio
n
a
b
o
u
t
th
e
ea
r
l
y
b
io
c
h
em
ical
c
h
an
g
es
i
n
d
u
ce
d
b
y
p
at
h
o
g
e
n
s
-
a
p
h
y
s
io
lo
g
ical
r
esp
o
n
s
e
wh
ich
is
o
f
ten
in
v
is
ib
le
in
R
GB
d
ata
[
2
0
]
,
[
2
4
]
.
3
.
1
.
4
.
Da
t
a
pa
rt
it
io
nin
g
a
nd
v
a
lid
a
t
io
n
T
h
e
r
aw
d
ata
was
s
p
lit
in
to
3
s
ets
:
tr
ain
(
7
0
%
o
f
th
e
d
ata)
,
v
alid
atio
n
s
et
(
1
5
%)
a
n
d
a
test
s
et
(
1
5
%).
T
o
d
ec
r
ea
s
e
t
h
e
lik
elih
o
o
d
o
f
a
o
n
e
-
class
s
p
lit,
we
p
er
f
o
r
m
ed
s
tr
atif
ied
s
p
litt
in
g
s
u
c
h
th
at
ea
ch
s
u
b
s
et
co
n
tain
ed
all
class
es.
To
ac
h
iev
e
g
r
ea
ter
s
tat
is
tical
r
o
b
u
s
tn
ess
an
d
p
r
ev
e
n
t
o
v
e
r
-
p
ar
am
eter
i
s
atio
n
,
5
-
f
o
ld
cr
o
s
s
-
v
alid
atio
n
[
1
9
]
,
[
2
5
]
was a
ls
o
ap
p
lied
d
u
r
in
g
m
o
d
el
tr
ain
in
g
an
d
v
alid
atio
n
.
I
n
all
p
ar
titi
o
n
s
,
8
0
%
o
f
th
e
d
ata
was
u
s
ed
f
o
r
tr
ain
in
g
an
d
2
0
%
f
o
r
v
alid
atio
n
in
a
r
o
tated
m
an
n
er
.
T
h
ese
ac
c
u
r
ac
y
m
etr
ics
wer
e
av
er
ag
ed
o
v
er
th
e
f
o
l
d
s
to
p
r
o
v
id
e
g
en
e
r
aliza
b
le
an
d
s
tab
le
esti
m
ates o
f
ac
cu
r
ac
y
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
7
7
6
I
n
t J
I
n
f
&
C
o
m
m
u
n
T
ec
h
n
o
l
,
Vo
l.
15
,
No
.
3
,
Sep
tem
b
er
20
26
:
1
3
6
4
-
1
3
7
5
1368
T
h
e
d
if
f
er
en
ces
in
th
is
p
r
ep
r
o
ce
s
s
in
g
p
ip
elin
e
an
d
th
e
d
is
tr
ib
u
tio
n
u
s
ed
b
ef
o
r
e
in
jectio
n
al
l
h
elp
ed
to
g
iv
e
th
e
GNN
-
ML
P
m
o
d
el
d
o
wn
s
tr
ea
m
“c
lean
ed
”,
c
o
n
tr
ast
-
en
h
an
ce
d
,
an
d
in
f
o
r
m
ativ
e
f
ea
tu
r
es
,
wh
ich
wo
u
l
d
in
tu
r
n
aid
ed
its
lear
n
in
g
o
f
f
i
n
e
-
s
ca
le
d
is
ea
s
e
p
atter
n
s
f
o
r
s
tr
awb
er
r
y
f
r
u
its
.
3
.
2
.
M
o
del
des
ig
n a
nd
a
rc
hite
ct
ure
T
h
e
m
o
d
el
is
a
f
u
s
io
n
m
o
d
el
b
ased
o
n
th
e
in
teg
r
atio
n
o
f
GNN
s
[
8
]
an
d
a
n
MLP
,
in
wh
ich
t
h
e
s
p
ec
ies
ar
e
ca
teg
o
r
i
s
ed
f
r
o
m
lear
n
ed
s
p
atial
an
d
s
em
a
n
tic
em
b
e
d
d
in
g
s
s
ep
ar
ately
f
o
r
m
ap
p
in
g
.
O
u
r
m
ain
u
n
d
er
ly
in
g
d
esig
n
p
h
ilo
s
o
p
h
y
is
to
en
a
b
le
r
o
b
u
s
t
s
p
atio
-
s
em
an
tic
lear
n
in
g
,
s
o
th
at
th
e
m
o
d
el
will
n
o
t
o
n
ly
r
ea
s
o
n
a
b
o
u
t
th
e
v
alu
es
o
f
th
e
p
ix
els
th
em
s
elv
es,
b
u
t
also
ab
o
u
t
th
e
s
tr
u
ctu
r
al
lay
o
u
ts
an
d
c
o
n
tex
tu
al
d
is
tr
ib
u
tio
n
s
o
f
th
e
d
is
ea
s
e
s
y
m
p
to
m
s
s
p
r
ea
d
o
u
t
o
v
er
th
e
f
r
u
it su
r
f
ac
e.
An
o
v
er
v
iew
o
f
th
e
p
r
o
p
o
s
ed
ar
ch
itectu
r
e
is
illu
s
tr
ated
in
Fig
u
r
e
1
;
t
h
e
o
v
e
r
all
p
r
o
ce
d
u
r
e
in
clu
d
es:
-
I
n
p
u
t r
aw
im
ag
es o
f
i
n
f
ec
ted
s
am
p
les o
f
s
tr
awb
er
r
ies with
le
s
io
n
s
,
m
o
u
ld
p
atc
h
es
,
o
r
n
ec
r
o
tic
zo
n
es.
-
Patch
ex
tr
ac
tio
n
an
d
b
u
ild
in
g
th
e
g
r
ap
h
.
T
h
e
i
m
ag
e
is
d
iv
id
ed
in
to
s
u
p
er
p
ix
els
;
ea
ch
s
u
p
er
p
ix
e
l
co
r
r
esp
o
n
d
s
to
a
g
r
ap
h
n
o
d
e,
an
d
its
n
eig
h
b
o
u
r
in
g
p
atch
e
s
c
o
r
r
esp
o
n
d
to
e
d
g
es,
th
u
s
r
esu
ltin
g
in
a
R
eg
io
n
Ad
jace
n
cy
Gr
a
p
h
(
R
A
G)
,
wh
ich
in
d
icate
s
th
e
lo
ca
l s
p
atial
d
is
tr
ib
u
tio
n
o
f
lesi
o
n
s
[
2
7
]
,
[
2
8
]
.
-
T
h
is
lay
er
,
t
h
e
GNN
m
o
d
u
le,
is
wh
er
e
n
eig
h
b
o
u
r
in
g
n
o
d
es
s
h
ar
e
in
f
o
r
m
atio
n
to
m
o
d
el
p
att
er
n
s
o
f
lesi
o
n
ex
p
an
s
io
n
,
s
p
atial
s
tr
u
ctu
r
e
o
f
d
is
o
r
d
er
ed
m
o
r
p
h
o
lo
g
y
,
a
n
d
ir
r
eg
u
lar
ities
in
b
o
r
d
er
s
[
2
7
]
-
T
h
e
ML
P m
o
d
u
le
m
ap
s
th
e
g
lo
b
ally
ag
g
r
eg
ated
GNN
em
b
e
d
d
in
g
t
o
class
l
ab
els
[
2
5
]
,
[
2
6
]
.
-
T
h
e
f
in
al
p
r
ed
ictio
n
,
co
m
p
r
is
in
g
th
e
p
r
ed
icted
d
is
ea
s
e
class
an
d
its
co
n
f
id
e
n
ce
s
co
r
e.
Ov
er
all,
th
is
h
y
b
r
id
s
tr
u
ctu
r
e
in
teg
r
ates
s
p
atial
r
ea
s
o
n
in
g
an
d
s
em
an
tic
class
if
icatio
n
in
to
a
s
in
g
le
lig
h
tweig
h
t a
r
ch
itectu
r
e
s
u
itab
le
f
o
r
s
m
ar
t
-
f
a
r
m
in
g
d
ep
lo
y
m
en
t.
Fig
u
r
e
1
.
B
lo
ck
d
iag
r
am
o
f
th
e
p
r
o
p
o
s
ed
h
y
b
r
id
GNN
–
ML
P
m
o
d
el
f
o
r
s
tr
awb
er
r
y
d
is
ea
s
e
class
if
icatio
n
3
.
2
.
1
.
G
NN
mo
du
le
T
h
e
GNN
m
o
d
u
le
is
tailo
r
e
d
t
o
ca
p
tu
r
e
an
d
r
etain
s
p
atial
p
a
tter
n
s
o
f
d
is
ea
s
e
-
af
f
ec
ted
r
eg
i
o
n
s
.
T
o
d
o
th
is
,
ea
ch
in
p
u
t
im
ag
e
is
f
ir
s
t
d
iv
id
ed
in
to
lo
ca
l
p
atch
es
o
r
s
u
p
er
p
ix
els,
wh
ich
r
ep
r
esen
t
th
e
n
o
d
es
o
f
a
g
r
ap
h
.
T
h
e
co
n
n
ec
tio
n
s
b
etwe
en
n
o
d
es
ar
e
co
n
tr
o
lled
b
y
d
is
tan
ce
(
g
eo
m
etr
ic
cl
o
s
en
ess
)
,
co
lo
r
s
im
il
ar
ity
,
an
d
tex
tu
r
e
co
h
er
en
ce
.
T
h
is
r
esu
lts
in
a
g
r
ap
h
-
lik
e
m
o
d
el
o
f
th
e
s
u
r
f
ac
e
o
f
ea
c
h
f
r
u
it,
with
n
o
d
es
(
i.e
.
,
clu
s
ter
s
)
r
ep
r
esen
tin
g
s
ets
o
f
ad
jace
n
t
p
ix
els
ass
ig
n
ed
to
f
ea
tu
r
e
v
ec
to
r
s
d
escr
ib
in
g
lo
ca
l
n
eig
h
b
o
u
r
h
o
o
d
r
e
g
io
n
s
an
d
ed
g
es in
d
icatin
g
s
p
atial
r
elatio
n
s
h
ip
s
.
Gr
ap
h
co
n
v
o
l
u
tio
n
al
lay
e
r
s
(
GC
N
s
)
ar
e
u
s
ed
to
p
r
o
p
ag
ate
th
ese
f
ea
tu
r
es
iter
ativ
ely
,
t
h
r
o
u
g
h
a
k
in
d
o
f
m
ess
ag
e
-
p
ass
in
g
-
b
ased
s
tr
ateg
y
:
f
o
r
ea
ch
n
o
d
e
,
th
e
GC
N
ag
g
r
eg
ate
s
in
f
o
r
m
atio
n
f
r
o
m
i
ts
n
eig
h
b
o
u
r
s
.
T
h
is
p
r
o
ce
s
s
h
elp
s
teac
h
th
e
m
o
d
e
l
h
o
w
s
y
m
p
to
m
s
m
ay
s
p
r
ea
d
ac
r
o
s
s
th
e
s
u
r
f
ac
e
o
f
f
r
u
it
(
t
h
u
s
lear
n
in
g
ab
o
u
t
in
ter
-
p
atch
co
r
r
elatio
n
s
,
lesi
o
n
b
o
r
d
e
r
s
,
as
well
a
s
co
-
o
cc
u
r
r
in
g
d
is
ea
s
e)
.
GNNs
ar
e
p
ar
ticu
lar
ly
well
-
s
u
ited
f
o
r
m
o
d
elin
g
th
e
n
o
n
-
E
u
clid
ea
n
s
p
atial
d
ep
en
d
en
cies
t
h
at
co
n
v
en
tio
n
al
C
NNs
o
f
t
e
n
d
o
n
o
t
ca
p
tu
r
e.
Ma
th
em
atica
lly
,
th
e
f
o
r
war
d
p
ass
o
f
ea
ch
g
r
ap
h
lay
er
ca
n
b
e
f
o
r
m
u
lated
as:
ℎ
(
+
1
)
=
(
∑
1
∈
(
)
(
)
ℎ
(
)
)
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J I
n
f
&
C
o
m
m
u
n
T
ec
h
n
o
l
I
SS
N:
2252
-
8
7
7
6
E
xp
la
in
a
b
le
A
I
-
b
a
s
ed
h
yb
r
id
GN
N
-
ML
P
mo
d
el
fo
r
s
tr
a
w
b
err
y
fr
u
it
d
is
ea
s
e
… (
V
a
r
s
h
a
K
i
r
a
n
B
h
o
s
a
le
)
1369
w
h
er
e
h
i
(
l
)
is
th
e
f
ea
tu
r
e
v
ec
to
r
o
f
n
o
d
e
i
at
lay
er
1
,
is
th
e
s
et
o
f
n
eig
h
b
o
u
r
in
g
n
o
d
es,
(
1
)
is
th
e
weig
h
t
m
atr
ix
,
is
th
e
n
o
r
m
ali
s
atio
n
c
o
n
s
tan
t,
an
d
σ
is
th
e
n
o
n
-
lin
ea
r
ac
tiv
atio
n
f
u
n
cti
o
n
.
3
.
2
.
2
.
M
L
P
m
o
du
le
Af
ter
th
e
s
p
atial
f
ea
tu
r
es
in
th
e
en
r
ich
ed
n
o
d
e
em
b
ed
d
i
n
g
s
ar
e
ag
g
r
eg
ate
d
in
to
a
f
ix
ed
-
s
ize
v
ec
to
r
,
th
ey
p
ass
th
r
o
u
g
h
to
th
e
ML
P
class
if
ier
.
T
h
is
m
o
d
u
le
s
er
v
es
as
th
e
s
em
an
tic
d
ec
is
io
n
lay
er
,
m
ap
p
in
g
th
e
h
ig
h
-
d
im
e
n
s
io
n
al
g
r
ap
h
f
ea
tu
r
es
b
ac
k
in
to
class
lab
el
s
.
T
h
e
ML
P
is
co
m
p
o
s
ed
o
f
s
ev
er
al
f
u
lly
co
n
n
ec
ted
lay
er
s
f
o
llo
wed
b
y
th
e
ap
p
licatio
n
o
f
R
eL
U
ac
tiv
atio
n
,
d
r
o
p
o
u
t (
to
a
v
o
id
o
v
er
f
itti
n
g
)
,
an
d
b
atc
h
n
o
r
m
ali
s
atio
n
(
to
allo
w
co
n
v
er
g
e
n
ce
)
.
I
ts
p
u
r
p
o
s
e
is
to
ca
p
t
u
r
e
g
lo
b
al,
ab
s
tr
ac
t
r
e
p
r
esen
tatio
n
s
,
e.
g
.
,
th
e
d
is
tr
ib
u
ti
o
n
o
f
d
is
ea
s
e
s
y
m
p
to
m
s
o
v
er
t
h
e
e
n
tire
f
r
u
it
wh
ile
m
a
in
tain
in
g
th
e
s
p
atial
in
f
o
r
m
ati
o
n
lear
n
ed
b
y
th
e
GNN.
T
h
e
last
s
o
f
tm
ax
lay
er
p
r
o
v
id
es c
lass
p
r
o
b
ab
ilit
ies f
o
r
th
e
th
r
ee
d
is
ea
s
es.
3
.
2
.
3
.
T
ra
ini
ng
env
iro
nm
ent
a
nd
co
nfig
ura
t
io
n
W
e
tr
ain
ed
it
u
s
in
g
th
e
Py
T
o
r
ch
f
r
a
m
ewo
r
k
,
wh
ich
is
m
o
d
u
lar
an
d
allo
ws
cu
s
to
m
GNN
l
ay
er
s
to
b
e
ad
d
ed
ea
s
ily
,
as
well
a
s
d
y
n
am
ic
g
r
ap
h
s
to
b
e
h
an
d
le
d
ef
f
icien
tly
.
T
h
e
m
o
d
el
was
th
u
s
ap
p
lied
b
y
b
ein
g
d
ep
lo
y
e
d
an
d
tr
ain
e
d
o
n
t
h
e
f
u
ll
d
ataset,
ex
ce
p
t
h
er
e
it
was
r
u
n
o
n
a
C
PU
-
o
n
ly
m
ac
h
in
e
in
o
r
d
er
to
s
im
u
late
wh
at
wo
u
ld
o
cc
u
r
u
p
o
n
f
u
r
th
er
on
-
s
ite
d
ep
lo
y
m
en
t
o
f
t
h
is
m
o
d
el
at
f
ar
m
lev
el,
eith
e
r
wh
er
e
GPUs
ar
e
n
o
t
r
ea
d
ily
av
ailab
le
o
r
alter
n
ativ
e
ly
in
s
h
o
r
t su
p
p
ly
.
Key
tr
ain
in
g
p
ar
am
eter
s
in
clu
d
ed
:
-
L
o
s
s
f
u
n
ctio
n
: c
ateg
o
r
ical
cr
o
s
s
-
en
tr
o
p
y
-
Op
tim
i
s
er
: A
d
am
with
an
in
itial lea
r
n
in
g
r
ate
o
f
0
.
0
0
0
1
-
B
atch
s
ize:
3
2
-
E
p
o
ch
s
: 5
0
(
with
ea
r
ly
s
to
p
p
in
g
b
ased
o
n
v
alid
atio
n
lo
s
s
)
-
Dr
o
p
o
u
t
r
ate:
0
.
3
T
h
ese
p
ar
am
eter
s
wer
e
f
in
e
-
tu
n
ed
th
r
o
u
g
h
ex
p
er
im
en
tat
io
n
o
n
th
e
v
alid
atio
n
s
et
u
s
in
g
a
g
r
id
s
ea
r
ch
p
r
o
ce
d
u
r
e.
T
h
is
h
y
b
r
i
d
ar
ch
ite
ctu
r
e
allo
ws
f
o
r
tr
ain
i
n
g
o
f
s
p
atial
s
tr
u
ctu
r
e
an
d
g
l
o
b
al
s
em
an
tics
co
n
cu
r
r
e
n
tly
,
with
b
ette
r
p
er
f
o
r
m
an
ce
th
an
t
r
ad
itio
n
al
im
a
g
e
class
if
ier
s
in
co
m
p
lex
f
ield
e
n
v
ir
o
n
m
en
ts
w
h
er
e
v
is
u
al
d
is
ea
s
e
s
ig
n
als ca
n
b
e
f
ain
t,
s
u
b
tle,
o
v
er
lap
p
in
g
,
o
r
et
h
er
ea
l.
3
.
3
.
XAI int
eg
ra
t
i
o
n
I
n
ter
p
r
etab
ilit
y
is
th
e
p
r
im
ar
y
lim
itatio
n
o
f
d
ee
p
lear
n
in
g
m
eth
o
d
s
,
esp
ec
ial
ly
in
s
en
s
itiv
e
d
o
m
ain
s
lik
e
ag
r
icu
ltu
r
e
.
Alth
o
u
g
h
th
e
s
e
m
o
d
els
r
ea
ch
r
e
m
ar
k
a
b
le
a
cc
u
r
ac
y
,
t
h
ey
ar
e
ty
p
ically
s
ee
n
as
“b
lack
b
o
x
es”,
as
th
ey
d
o
n
o
t
o
f
f
e
r
an
in
tu
it
iv
e
r
ea
s
o
n
in
g
r
eg
ar
d
i
n
g
wh
y
a
ce
r
tain
ca
teg
o
r
y
was
id
en
tifie
d
.
T
h
is
lack
o
f
tr
an
s
p
ar
en
cy
ca
n
also
er
o
d
e
u
s
er
co
n
f
id
e
n
ce
an
d
p
r
e
v
e
n
t
n
o
n
-
tech
n
o
c
r
atic
ad
o
p
te
r
s
s
u
ch
as
f
ar
m
er
s
,
ag
r
o
n
o
m
is
ts
,
an
d
ag
r
ic
u
ltu
r
al
ad
v
is
o
r
s
f
r
o
m
u
s
in
g
th
ese
tech
n
iq
u
es.
T
o
allev
iate
th
is
p
r
o
b
lem
,
in
th
is
wo
r
k
,
we
h
av
e
in
teg
r
at
ed
XAI
m
eth
o
d
s
in
to
o
u
r
cla
s
s
if
i
ca
tio
n
p
ip
elin
e.
T
h
e
s
e
m
eth
o
d
s
aim
to
m
ak
e
th
e
m
o
d
el’
s
d
ec
is
io
n
-
m
ak
i
n
g
p
r
o
ce
s
s
tr
a
n
s
p
ar
e
n
t
an
d
in
ter
p
r
etab
le,
h
elp
in
g
e
n
d
-
u
s
er
s
v
alid
ate
m
o
d
el
p
r
ed
ictio
n
s
a
n
d
h
a
v
e
co
n
f
i
d
en
ce
in
th
e
s
y
s
tem
’
s
tr
u
s
two
r
th
in
ess
.
3
.
3
.
1
.
L
I
M
E
LIME
[
1
7
]
is
a
m
o
d
el
-
a
g
n
o
s
ti
c
ex
p
lan
atio
n
tech
n
i
q
u
e
th
at
e
x
p
lain
s
in
d
iv
id
u
al
p
r
ed
ictio
n
s
f
r
o
m
th
e
m
o
d
el
b
y
lear
n
in
g
a
lo
ca
lly
i
n
ter
p
r
etab
le
m
o
d
el
ar
o
u
n
d
th
e
p
r
ed
ictio
n
.
I
t
wo
r
k
s
b
y
g
en
er
atin
g
a
b
u
n
ch
o
f
p
er
tu
r
b
e
d
s
am
p
les
f
r
o
m
t
h
e
o
r
ig
in
al
in
p
u
t
(
e.
g
,
b
y
b
l
o
ck
in
g
o
u
t
p
ar
ts
o
f
th
e
im
ag
e
)
,
an
d
t
h
en
s
ee
in
g
h
o
w
t
h
e
m
o
d
el
o
u
tp
u
t c
h
a
n
g
es.
I
n
th
is
s
tu
d
y
,
L
I
ME
[
1
7
]
w
as
em
p
lo
y
ed
to
r
ev
ea
l
th
e
m
o
s
t
im
p
ac
tf
u
l
r
eg
io
n
s
o
r
p
i
x
els
o
f
th
e
s
tr
awb
er
r
y
f
r
u
it
im
a
g
e
o
n
th
e
m
o
d
el
p
r
ed
ictio
n
.
T
h
e
r
esu
ltin
g
o
u
tp
u
t
is
a
b
in
a
r
y
s
alien
cy
m
ap
th
at
h
ig
h
lig
h
ts
th
e
to
p
-
k
i
m
p
o
r
ta
n
t
r
eg
io
n
s
th
at
co
n
tr
ib
u
te
t
o
th
e
p
r
ed
ict
ed
d
is
ea
s
e
class
p
o
s
itiv
ely
o
r
n
eg
ativ
ely
.
B
o
th
r
esear
ch
er
s
an
d
p
r
a
ctitio
n
er
s
c
an
u
s
e
th
is
to
v
e
r
if
y
w
h
eth
er
th
e
m
o
d
el
is
atten
d
in
g
to
b
io
l
o
g
ically
m
ea
n
in
g
f
u
l
f
ea
tu
r
es o
f
in
ter
est (
e.
g
.
,
lesi
o
n
b
o
u
n
d
ar
ies,
f
u
n
g
al
te
x
tu
r
es)
in
s
tead
o
f
ir
r
elev
a
n
t b
ac
k
g
r
o
u
n
d
n
o
is
e.
3
.
3
.
2
.
G
ra
d
-
CAM
Alth
o
u
g
h
L
I
ME
p
r
o
v
id
es
in
s
tan
ce
-
s
p
ec
if
ic
lo
ca
l
ex
p
lan
atio
n
s
,
Gr
ad
-
C
AM
p
r
o
d
u
ce
s
g
lo
b
al
v
is
u
al
r
ea
s
o
n
in
g
b
y
g
en
e
r
atin
g
h
ea
t
m
ap
s
th
at
s
h
o
w
th
e
m
o
s
t
d
is
c
r
im
in
ativ
e
r
e
g
io
n
s
u
s
ed
d
u
r
in
g
class
if
icatio
n
b
y
a
m
o
d
el.
G
r
ad
-
C
AM
[
1
8
]
co
m
p
u
tes
th
e
g
r
ad
ien
t
o
f
th
e
o
u
tp
u
t
class
s
co
r
e
with
r
esp
ec
t
to
th
e
f
ea
tu
r
e
m
a
p
s
o
f
th
e
last
co
n
v
o
lu
tio
n
al
lay
e
r
.
T
h
e
g
r
ad
ie
n
ts
ar
e
f
u
r
th
er
p
o
o
le
d
an
d
e
m
p
lo
y
e
d
to
weig
h
th
e
f
ea
tu
r
e
m
ap
s
,
wh
ich
ar
e
u
p
s
am
p
led
an
d
s
u
p
er
im
p
o
s
ed
with
th
e
o
r
ig
in
al
in
p
u
t
im
ag
e
to
g
en
er
ate
a
class
-
s
p
ec
if
ic
lo
ca
li
s
atio
n
m
ap
.
I
n
o
u
r
im
p
lem
e
n
tatio
n
,
Gr
ad
-
C
AM
[
1
8
]
was p
ar
ticu
lar
ly
ef
f
ec
tiv
e
in
:
-
L
o
ca
li
s
in
g
d
is
ea
s
e
s
p
o
ts
,
ev
en
in
th
e
p
r
esen
ce
o
f
o
cc
lu
s
io
n
o
r
n
o
is
e.
-
Hig
h
lig
h
tin
g
s
y
m
p
to
m
-
s
p
ec
if
i
c
tex
tu
r
es a
n
d
c
o
lo
u
r
ch
an
g
es.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
7
7
6
I
n
t J
I
n
f
&
C
o
m
m
u
n
T
ec
h
n
o
l
,
Vo
l.
15
,
No
.
3
,
Sep
tem
b
er
20
26
:
1
3
6
4
-
1
3
7
5
1370
-
Pro
v
id
in
g
ac
tio
n
a
b
le
in
s
ig
h
ts
f
o
r
d
o
m
ain
ex
p
e
r
ts
,
s
u
ch
as
wh
er
e
to
f
u
r
th
er
in
s
p
ec
t
th
e
f
r
u
it
f
o
r
ea
r
ly
-
s
tag
e
s
y
m
p
to
m
s
.
3
.
3
.
3
.
B
enef
it
s
o
f
inte
g
ra
t
ing
XAI in
a
g
ricult
ura
l dia
g
no
s
i
s
T
h
e
d
u
al
in
teg
r
atio
n
o
f
L
I
ME
[
1
7
]
an
d
Gr
ad
-
C
AM
en
s
u
r
es
th
at
b
o
th
lo
ca
l
in
ter
p
r
etab
ilit
y
(
v
ia
L
I
ME
)
an
d
class
-
d
is
cr
im
in
ativ
e
v
is
u
al
lo
ca
li
s
atio
n
(
v
ia
Gr
ad
-
C
AM
)
ar
e
ad
d
r
ess
ed
.
T
h
ese
to
o
ls
p
lay
s
ev
er
al
im
p
o
r
tan
t
r
o
les:
-
I
n
cr
ea
s
e
en
d
-
u
s
er
tr
u
s
t b
y
v
is
u
ali
s
in
g
th
e
r
atio
n
al
e
b
e
h
in
d
m
o
d
el
d
ec
is
io
n
s
.
-
E
n
ab
le
non
-
tec
h
n
ical
s
tak
eh
o
ld
er
s
to
ad
ju
s
t
in
te
r
v
en
tio
n
s
b
a
s
ed
o
n
i
n
f
o
r
m
ed
s
elec
t
iv
e
m
ea
s
u
r
es
(
e.
g
.
,
a
p
p
ly
in
g
p
esti
cid
es o
n
ly
in
lo
ca
li
s
ed
ar
ea
s
af
f
ec
ted
b
y
t
h
e
p
est)
.
-
E
n
h
an
ce
d
eb
u
g
g
in
g
a
n
d
m
o
d
el
tu
n
in
g
by
i
d
en
tific
atio
n
o
f
m
is
class
if
ied
r
eg
io
n
s
o
r
b
ias
in
t
h
e
tr
ain
in
g
d
ata
.
-
Facilitatin
g
tech
n
o
lo
g
y
a
d
o
p
tio
n
in
a
g
r
icu
ltu
r
al
c
o
m
m
u
n
ities
wh
er
e
p
r
a
g
m
atic
u
s
ab
ilit
y
an
d
in
ter
p
r
etab
ilit
y
ar
e
o
f
ten
m
u
ch
m
o
r
e
im
p
o
r
tan
t th
a
n
r
aw
p
e
r
f
o
r
m
an
ce
s
tatis
tics
.
W
ith
th
e
in
co
r
p
o
r
atio
n
o
f
XA
I
m
o
d
u
les
in
o
u
r
p
i
p
elin
e
,
ev
e
n
af
ter
tr
ai
n
in
g
,
t
h
e
en
tir
e
s
y
s
t
em
s
h
o
u
ld
w
o
r
k
as
a
tr
ad
itio
n
al
s
o
lu
tio
n
.
Sti
ll,
n
o
w
it
p
r
o
v
id
es
in
ter
p
r
eta
b
ilit
y
to
th
e
o
u
tp
u
t
,
en
a
b
lin
g
th
e
f
ar
m
er
to
tr
u
s
t
th
e
m
o
d
el
an
d
m
ak
in
g
p
r
ec
is
io
n
a
g
r
icu
ltu
r
e
f
r
ien
d
lier
.
3
.
4
.
Sy
s
t
e
m
f
lo
w
a
nd
ev
a
lua
t
io
n
An
o
v
er
v
iew
o
f
th
e
wh
o
le
s
y
s
t
em
p
ip
elin
e
is
d
ep
icted
in
F
ig
u
r
e
2
,
h
ig
h
lig
h
tin
g
g
en
e
r
al
wo
r
k
f
lo
w
s
tep
s
s
u
ch
as
d
ata
ac
q
u
is
itio
n
,
m
o
d
el
b
u
ild
i
n
g
,
an
d
in
ter
p
r
etin
g
r
esu
lts
.
T
h
e
p
ip
elin
e
also
ep
ito
m
izes
a
d
esira
b
le
m
o
d
u
lar
,
e
x
p
lain
ab
l
e,
co
n
ten
t
-
r
ich
f
r
am
ewo
r
k
with
ea
s
y
d
ep
lo
y
m
en
t,
with
le
s
s
f
o
cu
s
o
n
d
ataset
b
en
ch
m
ar
k
s
an
d
m
o
r
e
o
n
r
ea
l
-
wo
r
ld
ag
r
ic
u
ltu
r
al
ap
p
licatio
n
s
.
Fig
u
r
e
2
.
Mo
d
el
tr
ain
in
g
an
d
e
v
alu
atio
n
p
r
o
ce
s
s
I
t
s
tar
ts
f
r
o
m
r
aw
im
ag
es
o
f
n
o
n
-
d
is
ea
s
ed
an
d
d
is
ea
s
ed
s
tr
a
wb
er
r
y
s
am
p
les
,
c
o
l
le
ct
ed
u
n
d
er
d
if
f
e
r
en
t
f
ield
co
n
d
itio
n
s
(
e.
g
.
,
lig
h
tin
g
,
r
eso
lu
tio
n
,
a
n
g
le
,
etc
)
.
T
h
is
k
ee
p
s
all
o
u
r
d
ata
r
ea
li
s
tic
an
d
n
o
t
o
v
e
r
-
s
u
p
p
r
ess
ed
f
r
o
m
wh
er
e
we
c
o
llected
it,
as we
ll a
s
g
ettin
g
th
e
r
ig
h
t a
ttrib
u
te
d
iv
er
s
ity
.
Fo
r
th
e
d
ata
co
llectio
n
s
tep
,
h
er
e
we
u
s
e
a
p
r
o
ce
s
s
in
g
p
h
ase
with
d
ataset
s
s
u
ch
as
C
L
AHE
[
2
]
m
eth
o
d
s
an
d
h
y
p
e
r
s
p
ec
tr
al
im
ag
in
g
s
im
u
latio
n
.
T
h
ese
tr
an
s
f
o
r
m
atio
n
s
ar
e
cr
itical
f
o
r
th
e
u
n
d
er
ly
in
g
o
b
jectiv
e
o
f
in
d
u
cin
g
d
is
ea
s
e
ab
n
o
r
m
alities
an
d
id
en
tif
y
in
g
laten
t
v
ar
iab
les
th
at
elu
d
e
d
etec
tio
n
in
R
G
B
in
p
u
ts
.
Usi
n
g
s
tr
atif
ied
s
am
p
lin
g
,
th
e
d
ata
is
d
iv
id
ed
in
to
7
0
%
tr
ain
in
g
,
1
5
%
v
alid
atio
n
,
an
d
1
5
%
test
s
ets,
tak
in
g
ca
r
e
to
p
r
eser
v
e
th
e
cla
s
s
d
is
tr
ib
u
tio
n
in
ea
c
h
o
f
t
h
ese
s
u
b
s
ets.
Fin
ally
,
th
e
p
er
f
o
r
m
an
ce
esti
m
ates
ar
e
m
ad
e
with
5
-
f
o
ld
cr
o
s
s
-
v
alid
a
tio
n
[
2
7
]
as
p
ar
t
o
f
m
o
d
el
d
e
v
elo
p
m
en
t
to
im
p
r
o
v
e
r
eliab
ility
an
d
m
itig
ate
th
e
r
is
k
o
f
o
v
er
f
itti
n
g
.
Fo
r
e
v
er
y
f
o
ld
,
th
e
c
o
m
p
lete
d
e
v
elo
p
m
en
t
d
a
ta
a
r
e
cy
cle
d
th
r
o
u
g
h
s
p
lits
p
r
o
v
i
d
in
g
s
ep
ar
ate
tr
ain
in
g
an
d
v
alid
ati
o
n
p
a
r
titi
o
n
s
,
f
o
llo
we
d
b
y
f
in
al
ev
alu
atio
n
o
n
th
e
u
n
s
ee
n
te
s
t
s
et.
T
h
ese
g
r
ap
h
-
b
ased
im
ag
e
r
e
p
r
esen
tatio
n
s
a
r
e
th
en
u
s
ed
to
tr
ain
th
e
GNN
–
ML
P h
y
b
r
i
d
ar
ch
itectu
r
e
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J I
n
f
&
C
o
m
m
u
n
T
ec
h
n
o
l
I
SS
N:
2252
-
8
7
7
6
E
xp
la
in
a
b
le
A
I
-
b
a
s
ed
h
yb
r
id
GN
N
-
ML
P
mo
d
el
fo
r
s
tr
a
w
b
err
y
fr
u
it
d
is
ea
s
e
… (
V
a
r
s
h
a
K
i
r
a
n
B
h
o
s
a
le
)
1371
Af
ter
war
d
,
it
p
r
o
ce
e
d
s
to
ev
al
u
atio
n
,
w
h
er
e
th
e
m
o
d
el
is
e
v
alu
ated
b
y
r
u
n
n
i
n
g
c
r
o
s
s
-
v
alid
atio
n
a
n
d
ca
lcu
latin
g
co
m
m
o
n
class
if
icatio
n
m
etr
ics.
T
h
e
s
y
s
tem
co
m
p
u
tes:
-
Acc
u
r
ac
y
:
th
e
n
u
m
b
e
r
o
f
c
o
r
r
e
ctly
p
r
ed
icted
o
b
s
er
v
atio
n
s
o
v
er
th
e
to
tal
o
b
s
er
v
atio
n
s
.
-
Pre
cisi
o
n
: tr
u
e
p
o
s
itiv
es /
(
tr
u
e
p
o
s
itiv
es +
f
alse p
o
s
itiv
es).
-
R
ec
all
(
s
en
s
iti
v
ity
)
:
tr
u
e
p
o
s
itiv
e/ac
tu
al
p
o
s
itiv
e
f
o
r
a
ce
r
tain
class
-
F1
-
s
co
r
e:
it
is
th
e
Har
m
o
n
ic
m
ea
n
o
f
Pre
cisi
o
n
an
d
r
ec
a
ll,
g
iv
in
g
a
s
in
g
le
weig
h
te
d
m
etr
ic
f
o
r
a
n
im
b
alan
ce
d
d
ataset.
T
h
en
,
i
n
ter
p
r
etiv
e
ex
to
o
ls
(
e.
g
.
,
Gr
ad
-
C
AM
an
d
L
I
ME
[
1
7
]
)
ca
n
ex
p
o
u
n
d
m
o
d
el
d
ec
is
io
n
s
in
th
e
co
n
tex
t
o
f
f
ewe
r
p
atien
ce
.
T
h
ese
m
eth
o
d
s
g
en
er
ate
atten
tio
n
o
v
er
lay
s
an
d
s
alien
cy
m
ap
s
wh
ich
p
lo
t
ar
ea
s
o
v
er
t
h
e
im
ag
e
th
at
h
av
e
b
ee
n
in
f
lu
e
n
tial
in
t
h
e
f
in
al
class
if
icatio
n
d
ec
is
io
n
.
T
h
is
will
a
llo
w
f
o
r
im
m
ed
iate
d
ec
is
io
n
-
ma
k
in
g
in
th
e
f
ield
,
wh
ich
in
s
p
ir
es
tr
u
s
t.
E
v
en
tu
al
ly
,
th
e
f
in
al
o
u
tp
u
t
r
etu
r
n
s
y
o
u
th
e
m
o
d
el
o
u
tp
u
ts
as f
o
llo
ws:
-
Pre
d
icted
d
is
ea
s
e
class
lab
els (
class
lab
els ar
e
an
th
r
ac
n
o
s
e,
g
r
ey
m
o
u
ld
,
an
d
p
o
wd
er
y
m
ild
ew
)
.
-
C
o
n
f
id
en
ce
s
co
r
es
.
-
Fo
r
ea
ch
p
r
ed
ictio
n
in
s
tan
ce
,
v
is
u
al
ex
p
lan
atio
n
m
ap
s
.
T
h
is
p
ip
elin
e
p
r
o
v
id
es
a
well
-
f
o
u
n
d
ed
,
ad
v
an
ce
d
,
tr
an
s
p
ar
en
t
,
an
d
s
ca
lab
le
s
o
lu
tio
n
f
o
r
th
e
ea
r
ly
d
etec
tio
n
o
f
s
tr
awb
er
r
y
d
is
ea
s
es,
en
ab
lin
g
th
e
d
ev
elo
p
m
en
t
o
f
s
m
ar
t
f
ar
m
in
g
s
y
s
tem
s
an
d
p
r
ec
is
io
n
ag
r
icu
ltu
r
e
a
p
p
licatio
n
s
.
4.
RE
SU
L
T
S
4
.
1
.
O
v
er
a
ll per
f
o
r
m
a
nce
T
h
e
p
r
o
p
o
s
ed
h
y
b
r
id
GNN
–
ML
P
m
o
d
el
o
b
tain
ed
a
n
ac
c
u
r
ac
y
an
d
m
ac
r
o
-
F1
o
f
9
3
.
5
9
a
n
d
9
0
.
3
9
%.
T
h
e
co
n
f
u
s
io
n
m
atr
ix
s
h
o
wn
in
Fig
u
r
e
3
d
em
o
n
s
tr
ates
a
b
alan
ce
d
class
if
icatio
n
a
cr
o
s
s
all
th
r
ee
d
is
ea
s
e
class
es,
wi
th
o
n
ly
s
lig
h
t m
is
class
if
icatio
n
.
Fig
u
r
e
3
.
C
o
n
f
u
s
io
n
m
atr
i
x
illu
s
tr
atin
g
class
if
icatio
n
o
u
tco
m
es f
o
r
an
th
r
ac
n
o
s
e
f
r
u
it r
o
t,
g
r
ey
m
o
u
l
d
,
an
d
p
o
wd
er
y
m
ild
ew
4
.
2
.
Cla
s
s
-
wis
e
a
nd
cr
o
s
s
-
v
a
lid
a
t
io
n m
et
rics
Per
f
o
r
m
an
ce
o
f
th
e
class
-
lev
el
p
r
ec
is
io
n
,
r
ec
all
,
an
d
F1
-
s
co
r
e
is
s
u
m
m
ar
ized
in
T
ab
le
1
.
Gr
ey
mo
u
ld
r
ec
eiv
ed
t
h
e
h
i
g
h
est
r
atin
g
s
,
f
o
llo
wed
b
y
a
n
t
h
r
ac
n
o
s
e
an
d
p
o
wd
er
y
m
il
d
ew.
R
ea
s
o
n
ab
l
y
s
tab
le
5
-
f
o
ld
cr
o
s
s
-
v
a
l
i
d
a
t
i
o
n
p
e
r
f
o
r
m
a
n
c
e
w
as
a
ch
i
e
v
e
d
,
d
e
m
o
n
s
t
r
a
t
i
n
g
c
o
n
s
i
s
t
en
t
b
e
h
a
v
i
o
u
r
a
c
r
o
s
s
d
i
f
f
e
r
e
n
t
d
a
t
a
s
p
li
ts
.
T
ab
le
1.
C
lass
if
icatio
n
m
etr
ics s
u
m
m
ar
y
C
l
a
s
s
P
r
e
c
i
s
i
o
n
R
e
c
a
l
l
F1
-
s
c
o
r
e
S
u
p
p
o
r
t
A
n
t
h
r
a
c
n
o
se
f
r
u
i
t
r
o
t
0
.
8
5
7
1
0
.
9
0
0
0
0
.
8
7
8
0
60
G
r
e
y
m
o
u
ld
0
.
9
6
3
4
0
.
9
6
0
5
0
.
9
6
1
9
3
2
9
P
o
w
d
e
r
y
m
i
l
d
e
w
0
.
8
8
3
1
0
.
8
6
0
8
0
.
8
7
1
8
79
A
c
c
u
r
a
c
y
0
.
9
3
5
9
4
6
8
M
a
c
r
o
a
v
e
r
a
g
e
0
.
9
0
1
2
0
.
9
0
7
1
0
.
9
0
3
9
4
6
8
W
e
i
g
h
t
e
d
a
v
e
r
a
g
e
0
.
9
3
6
2
0
.
9
3
5
9
0
.
9
3
6
0
4
6
8
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
7
7
6
I
n
t J
I
n
f
&
C
o
m
m
u
n
T
ec
h
n
o
l
,
Vo
l.
15
,
No
.
3
,
Sep
tem
b
er
20
26
:
1
3
6
4
-
1
3
7
5
1372
4
.
3
.
Co
m
pa
ra
t
iv
e
perf
o
r
m
a
nce
a
g
a
ins
t
ba
s
eline
m
o
dels
C
o
m
p
ar
is
o
n
s
wer
e
m
a
d
e
with
SVM,
R
F,
s
m
aller
C
NNs,
YOL
O
d
etec
to
r
s
,
an
d
p
r
ev
io
u
s
ly
r
ep
o
r
ted
GNN/h
y
b
r
id
m
eth
o
d
s
.
T
h
e
r
esu
lts
co
llected
in
T
ab
le
2
an
d
T
ab
le
3
s
h
o
w
th
at
th
e
p
r
o
p
o
s
ed
GNN
–
ML
P
m
o
d
el
p
er
f
o
r
m
s
m
u
ch
b
etter
th
an
all
th
e
b
aselin
es
in
ter
m
s
o
f
ac
cu
r
ac
y
,
p
r
ec
is
io
n
,
r
e
c
all,
an
d
F1
-
s
co
r
e.
T
h
e
p
er
f
o
r
m
an
ce
c
o
m
p
ar
is
o
n
s
a
r
e
g
r
ap
h
ically
s
h
o
wn
in
Fig
u
r
e
4
.
Fig
u
r
e
4
s
h
o
ws
r
ep
r
esen
t
ativ
e
in
p
u
t
im
ag
es
an
d
t
h
e
c
o
r
r
esp
o
n
d
in
g
d
is
ea
s
e
class
if
icatio
n
r
esu
lts
p
r
o
d
u
ce
d
b
y
t
h
e
p
r
o
p
o
s
ed
h
y
b
r
id
GNN
–
ML
P
m
o
d
el.
Fig
u
r
e
4
(
a
)
p
r
esen
ts
th
e
o
r
ig
i
n
al
d
is
ea
s
ed
s
tr
awb
er
r
y
f
r
u
it
i
m
ag
es,
wh
ile
Fig
u
r
e
4
(
b
)
illu
s
tr
ates
th
e
p
r
ed
icte
d
d
is
ea
s
e
class
e
s
with
th
eir
co
r
r
esp
o
n
d
in
g
p
r
ed
ictio
n
ac
cu
r
ac
i
es.
(
a)
(
b
)
Fig
u
r
e
4
.
Stra
wb
er
r
y
d
is
ea
s
e
c
lass
if
icatio
n
:
(
a)
d
is
ea
s
ed
s
tr
a
wb
er
r
y
f
r
u
it im
ag
es b
e
f
o
r
e
cla
s
s
if
icatio
n
an
d
(
b
)
th
e
co
r
r
esp
o
n
d
in
g
class
if
icatio
n
r
esu
lts
g
en
er
ated
b
y
th
e
p
r
o
p
o
s
ed
h
y
b
r
id
GNN
–
ML
P m
o
d
el
T
ab
le
2
.
C
o
m
p
a
r
ativ
e
p
er
f
o
r
m
an
ce
o
f
p
r
o
p
o
s
ed
m
o
d
el
v
s
.
p
r
ev
io
u
s
s
tu
d
ies
T
ab
le
3
.
C
o
m
p
a
r
ativ
e
p
er
f
o
r
m
an
ce
o
f
m
o
d
els
M
o
d
e
l
A
c
c
u
r
a
c
y
(
%)
P
r
e
c
i
s
i
o
n
(
%)
R
e
c
a
l
l
(
%)
F1
-
sc
o
r
e
(
%)
S
V
M
8
7
.
0
2
8
5
.
6
3
8
4
.
7
2
8
5
.
1
0
RF
8
8
.
4
5
8
6
.
2
1
8
5
.
3
0
8
5
.
7
5
C
N
N
9
1
.
3
0
8
9
.
1
5
8
8
.
7
2
8
8
.
9
3
GNN
–
M
LP (
P
r
o
p
o
se
d
)
9
3
.
5
9
9
0
.
1
2
9
0
.
7
1
9
0
.
3
9
R
e
f
e
r
e
n
c
e
s
M
o
d
e
l
t
y
p
e
D
a
t
a
s
e
t
t
y
p
e
A
c
c
u
r
a
c
y
(
%)
F1
-
sc
o
r
e
(
%)
N
o
t
e
s
(
c
o
r
r
e
c
t
e
d
a
u
t
h
o
r
a
t
t
r
i
b
u
t
i
o
n
)
K
a
r
k
i
e
t
a
l
.
[
3
]
C
N
N
C
o
n
t
r
o
l
l
e
d
l
a
b
90
–
92
~
8
9
S
e
n
s
i
t
i
v
e
t
o
l
i
g
h
t
i
n
g
;
l
i
mi
t
e
d
r
o
b
u
st
n
e
ss i
n
f
i
e
l
d
c
o
n
d
i
t
i
o
n
s
Li
u
e
t
a
l
.
[
5
]
Ex
p
l
a
i
n
a
b
l
e
C
N
N
M
i
x
e
d
~
9
1
~
9
0
G
o
o
d
t
e
x
t
u
r
e
e
x
t
r
a
c
t
i
o
n
;
w
e
a
k
s
p
a
t
i
a
l
–
st
r
u
c
t
u
r
a
l
p
a
t
t
e
r
n
l
e
a
r
n
i
n
g
M
a
h
m
u
d
e
t
a
l
.
[
6
]
M
o
b
i
l
e
C
N
N
F
i
e
l
d
~
9
0
~
8
8
P
e
r
f
o
r
ms
p
o
o
r
l
y
w
i
t
h
o
v
e
r
l
a
p
p
i
n
g
o
r
c
l
u
t
t
e
r
e
d
s
y
m
p
t
o
m re
g
i
o
n
s
X
i
e
e
t
a
l
.
[
2
8
]
Y
O
LO
v
8
V
a
r
i
a
n
t
F
i
e
l
d
+
g
r
e
e
n
h
o
u
se
92
–
94
~
9
2
S
t
r
o
n
g
d
e
t
e
c
t
i
o
n
c
a
p
a
b
i
l
i
t
y
;
w
e
a
k
e
r
f
i
n
e
-
g
r
a
i
n
e
d
c
l
a
ssi
f
i
c
a
t
i
o
n
He
e
t
a
l
.
[
1
1
]
K
TD
-
Y
O
LO
v
8
F
i
e
l
d
92
91
M
i
s
c
l
a
ssi
f
i
e
s
v
i
s
u
a
l
l
y
s
i
m
i
l
a
r
d
i
se
a
se
c
l
a
ss
e
s
K
i
m
e
t
a
l
.
[
1
3
]
D
e
e
p
C
N
N
C
o
n
t
r
o
l
l
e
d
+
semi
-
f
i
e
l
d
90
–
93
~
8
9
S
e
n
s
i
t
i
v
e
t
o
n
o
i
s
e
,
o
c
c
l
u
si
o
n
,
a
n
d
b
a
c
k
g
r
o
u
n
d
a
r
t
i
f
a
c
t
s
O
c
h
o
a
-
O
r
n
e
l
a
s
e
t
a
l
.
[
1
4
]
Li
g
h
t
w
e
i
g
h
t
d
e
e
p
l
e
a
r
n
i
n
g
F
i
e
l
d
87
–
90
~
8
6
F
a
st
i
n
f
e
r
e
n
c
e
;
r
e
d
u
c
e
d
a
c
c
u
r
a
c
y
i
n
n
o
i
s
y
o
u
t
d
o
o
r
c
o
n
d
i
t
i
o
n
s
H
a
r
a
n
d
i
e
t
a
l
.
[
1
5
]
GNN
(
P
h
e
n
o
t
y
p
i
n
g
)
C
o
n
t
r
o
l
l
e
d
~
9
1
~
9
0
S
t
r
o
n
g
g
r
a
p
h
-
b
a
se
d
r
e
a
so
n
i
n
g
;
l
i
mi
t
e
d
v
a
l
i
d
a
t
i
o
n
o
n
f
r
u
i
t
c
r
o
p
s
P
a
r
t
e
l
e
t
a
l
.
[
1
6
]
H
y
b
r
i
d
G
r
a
p
h
–
C
N
N
C
o
n
t
r
o
l
l
e
d
~
9
2
~
9
1
Li
mi
t
e
d
g
e
n
e
r
a
l
i
z
a
t
i
o
n
t
o
r
e
a
l
-
f
i
e
l
d
st
r
a
w
b
e
r
r
y
i
ma
g
e
r
y
P
r
o
p
o
se
d
m
o
d
e
l
H
y
b
r
i
d
G
N
N
+
M
LP
F
i
e
l
d
+
o
p
e
n
-
so
u
r
c
e
9
3
.
5
9
9
0
.
3
9
B
e
st
g
e
n
e
r
a
l
i
z
a
t
i
o
n
;
r
o
b
u
st
u
n
d
e
r
n
o
i
s
e
,
o
c
c
l
u
si
o
n
,
a
n
d
o
v
e
r
l
a
p
p
i
n
g
s
y
m
p
t
o
ms
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J I
n
f
&
C
o
m
m
u
n
T
ec
h
n
o
l
I
SS
N:
2252
-
8
7
7
6
E
xp
la
in
a
b
le
A
I
-
b
a
s
ed
h
yb
r
id
GN
N
-
ML
P
mo
d
el
fo
r
s
tr
a
w
b
err
y
fr
u
it
d
is
ea
s
e
… (
V
a
r
s
h
a
K
i
r
a
n
B
h
o
s
a
le
)
1373
4
.
4
.
E
x
pla
ina
bil
it
y
v
is
ua
lis
a
t
io
ns
(
XAI)
E
x
p
lain
ab
ilit
y
to
o
ls
wer
e
u
s
e
d
to
v
er
if
y
th
at
th
e
m
o
d
el
f
o
cu
s
was
s
o
u
n
d
.
L
I
ME
an
d
G
r
ad
-
C
AM
v
is
u
ali
s
atio
n
s
ar
e
p
lo
tte
d
to
g
eth
er
in
Fig
u
r
e
5
to
d
e
m
o
n
s
tr
ate
th
at
th
e
m
o
d
el
c
o
n
s
is
ten
tly
id
en
tifie
s
b
io
lo
g
ically
m
ea
n
in
g
f
u
l
le
s
io
n
r
eg
io
n
s
co
n
s
is
ten
tly
.
Su
ch
r
esu
lts
v
alid
ate
th
at
th
e
m
o
d
e
l
p
r
ed
ictio
n
s
ar
e
in
ac
co
r
d
an
ce
with
o
b
s
er
v
ab
le
o
u
tb
r
ea
k
d
is
tr
ib
u
tio
n
s
ev
e
n
u
n
d
er
d
if
f
icu
lt
f
ield
c
o
n
d
itio
n
s
.
Fig
u
r
e
5
p
r
esen
ts
r
ep
r
esen
tativ
e
L
I
ME
-
b
ased
s
alien
cy
m
ap
s
f
o
r
th
e
th
r
ee
s
tr
awb
er
r
y
d
is
ea
s
e
class
es.
Fig
u
r
e
5
(
a)
h
ig
h
lig
h
ts
th
e
s
alien
t
r
eg
io
n
s
ass
o
ciate
d
with
g
r
ey
m
o
u
l
d
,
wh
er
e
th
e
m
o
d
el
p
r
im
ar
ily
f
o
cu
s
e
s
o
n
f
u
n
g
al
t
ex
tu
r
e.
Fig
u
r
e
5
(
b
)
illu
s
tr
ates
th
e
r
eg
io
n
s
c
o
n
tr
ib
u
tin
g
to
th
e
p
r
e
d
ictio
n
o
f
p
o
wd
er
y
m
ild
ew
b
y
em
p
h
asi
s
in
g
th
e
ch
ar
ac
ter
is
tic
wh
ite
p
o
wd
er
y
s
u
r
f
ac
e.
Fig
u
r
e
5
(
c)
id
en
tifie
s
th
e
s
alien
t
r
e
g
io
n
s
co
r
r
esp
o
n
d
in
g
to
an
th
r
ac
n
o
s
e
f
r
u
it
r
o
t
b
y
lo
ca
li
s
i
n
g
th
e
b
lack
lesi
o
n
z
o
n
es r
esp
o
n
s
ib
le
f
o
r
th
e
d
is
ea
s
e
p
r
ed
ictio
n
.
(
a)
(
b
)
(
c)
Fig
u
r
e
5
.
L
I
ME
-
b
ased
s
alien
c
y
m
ap
v
is
u
ali
s
atio
n
s
f
o
r
d
is
ea
s
e
p
r
ed
ictio
n
(
a)
g
r
ey
m
o
u
ld
,
(
b
)
p
o
wd
er
y
m
ild
ew,
an
d
(
c
)
an
th
r
ac
n
o
s
e
f
r
u
it r
o
t
5.
DIS
CU
SS
I
O
N
R
es
u
l
ts
s
h
o
w
t
h
a
t
t
h
e
h
y
b
r
i
d
G
N
N
–
M
L
P
m
o
d
e
l
o
u
t
p
e
r
f
o
r
m
s
t
h
e
e
v
a
l
u
at
e
d
t
r
a
d
i
t
i
o
n
al
m
a
c
h
i
n
e
-
l
e
a
r
n
i
n
g
,
C
N
N
-
b
as
e
d
,
a
n
d
Y
O
L
O
-
b
as
e
d
b
as
e
li
n
e
s
.
I
ts
a
c
c
u
r
a
c
y
a
n
d
m
a
c
r
o
-
F
1
s
c
o
r
e
i
n
d
i
c
a
t
e
t
h
a
t
i
t
c
a
n
r
e
p
r
e
s
e
n
t
i
r
r
e
g
u
l
a
r
s
p
a
t
i
al
c
h
a
r
a
c
t
e
r
is
t
i
cs
o
f
s
t
r
a
w
b
e
r
r
y
d
i
s
e
ase
s
y
m
p
t
o
m
s
t
h
at
a
r
e
d
i
f
f
i
c
u
l
t
f
o
r
g
r
i
d
-
b
a
s
e
d
C
N
Ns
t
o
e
n
c
o
d
e
e
x
p
l
i
ci
t
l
y
.
P
r
e
v
i
o
u
s
s
t
r
aw
b
e
r
r
y
s
t
u
d
ie
s
h
a
v
e
s
h
o
w
n
t
h
a
t
C
NN
-
b
a
s
e
d
s
y
s
t
e
m
s
c
a
n
a
c
h
i
e
v
e
s
t
r
o
n
g
d
e
t
e
c
t
i
o
n
p
e
r
f
o
r
m
a
n
c
e
[
6
]
,
[
1
2
]
,
[
1
3
]
,
w
h
i
l
e
Y
O
L
O
-
b
a
s
e
d
d
e
t
e
c
t
o
r
s
p
r
o
v
i
d
e
e
f
f
e
c
ti
v
e
l
o
c
a
l
i
s
a
t
i
o
n
[
1
1
]
,
[
2
8
]
.
T
h
e
p
r
o
p
o
s
e
d
m
o
d
e
l
e
x
t
e
n
d
s
t
h
e
s
e
a
p
p
r
o
a
c
h
e
s
b
y
e
x
p
l
i
c
itl
y
m
o
d
e
l
l
i
n
g
l
e
s
i
o
n
-
to
-
l
e
s
i
o
n
s
p
a
t
i
al
r
e
la
t
i
o
n
s
h
i
p
s
t
h
r
o
u
g
h
g
r
a
p
h
r
e
a
s
o
n
i
n
g
.
T
h
i
s
d
e
s
i
g
n
i
s
c
o
n
s
is
t
e
n
t
wi
t
h
g
r
a
p
h
-
b
a
s
e
d
l
e
a
r
n
i
n
g
s
t
u
d
ie
s
t
h
at
d
em
o
n
s
t
r
a
t
e
t
h
e
v
a
l
u
e
o
f
n
o
n
-
E
u
c
l
i
d
e
a
n
r
e
p
r
e
s
e
n
t
a
t
io
n
s
f
o
r
p
l
a
n
t
s
t
r
u
c
t
u
r
e
s
[
9
]
,
[
1
5
]
,
[
2
7
]
.
T
h
e
s
t
a
b
l
e
5
-
f
o
l
d
c
r
o
s
s
-
v
a
l
i
d
at
i
o
n
p
e
r
f
o
r
m
a
n
c
e
a
l
s
o
i
n
d
i
c
at
e
s
g
o
o
d
g
e
n
e
r
a
l
i
s
a
t
i
o
n
a
c
r
o
s
s
d
a
ta
s
p
l
i
t
s
.
M
o
r
e
o
v
e
r
,
t
h
e
L
I
M
E
a
n
d
G
r
a
d
-
C
A
M
o
u
t
p
u
ts
h
i
g
h
l
i
g
h
t
b
i
o
l
o
g
i
c
a
ll
y
r
e
l
e
v
a
n
t
l
e
s
i
o
n
a
r
e
as
t
h
at
j
u
s
t
i
f
y
t
r
u
s
t
w
o
r
t
h
y
b
y
-
d
e
s
i
g
n
a
g
r
i
c
u
l
t
u
r
a
l
A
I
[
1
7
]
,
[
2
2
]
.
T
h
er
e
ar
e
v
ar
io
u
s
ad
v
a
n
tag
es
o
f
th
e
p
r
o
p
o
s
ed
f
r
am
ewo
r
k
.
First,
th
e
GNN
in
g
r
ed
ien
t
m
o
d
el
lear
n
s
p
atter
n
s
o
f
ad
jace
n
c
y
am
o
n
g
lesi
o
n
s
(
i.e
.
,
lesi
o
n
f
r
ag
m
e
n
ta
tio
n
an
d
d
y
n
am
ic
e
v
o
lu
tio
n
)
th
at
s
in
g
le
im
ag
es
ca
n
n
o
t
s
u
f
f
icien
tly
e
n
co
d
e,
wh
ich
C
NNs
f
ail
to
r
e
p
r
esen
t
r
o
b
u
s
tly
.
Seco
n
d
,
t
h
e
s
y
n
th
etic
h
y
p
e
r
s
p
ec
tr
al
p
r
ep
r
o
ce
s
s
in
g
en
h
a
n
ce
s
s
en
s
itiv
ity
to
ea
r
ly
b
io
ch
e
m
ical
c
h
an
g
es
,
c
o
n
f
ir
m
in
g
r
esu
lts
s
h
o
wn
i
n
p
r
ev
io
u
s
h
y
p
er
s
p
ec
tr
al
a
n
d
m
u
ltis
p
ec
tr
al
s
tu
d
ies
[
2
]
,
[
1
4
]
,
[
1
5
]
,
[
1
6
]
.
T
h
ir
d
,
th
e
ML
P
f
u
s
io
n
h
ea
d
:
t
h
e
s
em
an
tic
im
ag
es
u
s
ed
f
o
r
g
lo
b
al
s
em
an
tic
cu
es
h
elp
to
s
tab
ilis
e
d
ec
is
io
n
b
o
u
n
d
ar
ies
.
L
astl
y
,
in
te
g
r
at
ed
XAI
r
en
d
er
s
th
e
p
r
ed
ictio
n
s
tr
an
s
p
ar
e
n
t
an
d
tr
u
s
two
r
th
y
,
co
m
b
atin
g
t
h
e
in
ter
p
r
etab
ilit
y
g
a
p
o
b
s
er
v
ed
in
co
n
tem
p
o
r
ar
y
liter
atu
r
e
[
5
]
,
[
1
0
]
,
[
1
7
]
,
[
2
2
]
.
T
h
e
a
n
aly
s
is
h
as
s
o
m
e
lim
i
tatio
n
s
.
T
h
e
d
ataset
d
o
es
n
o
t
in
clu
d
e
m
u
lti
-
y
ea
r
tem
p
o
r
al
v
ar
iatio
n
th
at
m
ay
b
e
im
p
o
r
ta
n
t
f
o
r
p
er
f
o
r
m
an
ce
in
ex
tr
em
e
en
v
ir
o
n
m
en
tal
co
n
d
itio
n
s
.
Alth
o
u
g
h
s
y
n
th
etic
h
y
p
er
s
p
ec
tr
al
s
im
u
latio
n
is
s
u
f
f
icien
t,
it’s
o
n
ly
a
n
ap
p
r
o
x
im
atio
n
o
f
th
e
r
ea
l
o
n
e
.
T
h
e
m
o
d
el
h
as
n
o
t
b
ee
n
ev
a
lu
ated
in
th
e
p
r
esen
ce
o
f
s
ig
n
if
ican
t
o
cc
lu
s
io
n
s
o
r
co
m
p
licated
co
-
in
f
ec
tio
n
s
itu
atio
n
s
,
wh
ich
ca
n
af
f
ec
t
its
o
v
er
all
ac
cu
r
ac
y
i
n
r
ea
l
life
.
I
n
th
e
f
u
tu
r
e
,
th
e
s
y
s
tem
ca
n
b
e
f
u
r
th
er
d
e
v
elo
p
e
d
b
y
ac
q
u
ir
in
g
m
u
lti
-
s
ea
s
o
n
r
ea
l
-
f
ield
d
ata
,
ad
d
in
g
ac
tu
a
l
m
u
ltis
p
ec
tr
al
o
r
h
y
p
er
s
p
e
ctr
al
s
en
s
o
r
s
,
an
d
co
n
s
id
er
in
g
d
is
ea
s
e
s
ev
er
ity
esti
m
atio
n
an
d
tem
p
o
r
al
d
ev
el
o
p
m
en
t
m
o
d
ellin
g
.
R
u
n
n
in
g
th
e
m
o
d
el
o
n
lo
w
-
en
d
ed
g
e
d
e
v
ices
an
d
co
u
p
lin
g
it
with
I
o
T
-
b
ased
s
en
s
in
g
s
y
s
tem
s
,
as
in
v
esti
g
ated
in
r
elate
d
wo
r
k
[
2
3
]
,
[
2
4
]
,
m
a
y
b
e
an
o
th
e
r
p
o
te
n
tial v
en
u
e
f
o
r
r
ea
l f
ield
u
s
e.
6.
CO
NCLU
SI
O
N
I
n
th
is
s
tu
d
y
,
a
h
y
b
r
id
GNN
–
ML
P
m
o
d
el
co
m
b
in
ed
with
s
y
n
th
etic
h
y
p
er
s
p
ec
tr
al
s
im
u
latio
n
an
d
ex
p
lain
ab
le
AI
was
p
r
o
p
o
s
ed
t
o
s
o
lv
e
th
e
is
s
u
e
o
f
r
ea
l
-
f
ield
p
o
s
itio
n
s
tr
awb
er
r
y
d
is
ea
s
e
d
e
tectio
n
.
T
h
e
m
o
d
el
ac
h
iev
ed
h
ig
h
ac
cu
r
ac
y
,
p
r
ec
is
io
n
,
r
ec
all
,
an
d
m
ac
r
o
-
F1
p
er
f
o
r
m
a
n
ce
,
in
d
icatin
g
s
tr
o
n
g
g
en
er
ali
s
atio
n
to
h
eter
o
g
en
e
o
u
s
d
atasets
.
T
h
e
en
h
an
ce
m
e
n
t
-
b
ased
,
g
r
ap
h
-
b
ased
r
elatio
n
al
m
o
d
ellin
g
a
n
d
s
em
an
tic
f
u
s
io
n
p
r
o
p
o
s
al
n
etwo
r
k
was
ab
le
to
ac
cu
r
ately
ca
p
tu
r
e
ir
r
e
g
u
lar
le
s
io
n
s
tr
u
ctu
r
es
t
h
at
tr
ad
itio
n
al
C
NN/YOL
O
b
ased
Evaluation Warning : The document was created with Spire.PDF for Python.