Co
m
pu
t
er
Science
a
nd
I
nfo
r
m
a
t
io
n T
ec
hn
o
lo
g
ies
Vo
l.
7
,
No
.
2
,
J
u
ly
20
26
,
p
p
.
1
67
~
1
78
I
SS
N:
2722
-
3
2
2
1
,
DOI
:
1
0
.
1
1
5
9
1
/csi
t
.
v
7
i
2
.
p
1
67
-
1
78
167
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ia
e
s
p
r
ime.
co
m/in
d
ex
.
p
h
p
/csi
t
Perf
o
rma
nce
ev
a
lua
tion
of
th
e
dee
p
lea
rning
sy
stem
f
o
r
weed
recog
niza
tion
Abd
Abra
him
M
o
s
s
l
a
h
1
,
Rey
a
dh
H
a
zim
M
a
hd
i
2
,
H
a
s
s
a
n
K
a
s
s
im
Alba
ha
dil
y
3
1
C
o
l
l
e
g
e
of
I
sl
a
mi
c
S
c
i
e
n
c
e
,
U
n
i
v
e
r
s
i
t
y
of
A
n
b
a
r
,
A
n
b
a
r
,
I
r
a
q
2
C
o
l
l
e
g
e
of
S
c
i
e
n
c
e
,
U
n
i
v
e
r
s
i
t
y
of
M
u
st
a
n
si
r
i
y
a
h
,
B
a
g
h
d
a
d
,
I
r
a
q
3
C
o
l
l
e
g
e
of
S
c
i
e
n
c
e
,
C
o
mp
u
t
e
r
S
c
i
e
n
c
e
D
e
p
a
r
t
m
e
n
t
B
a
g
h
d
a
d
,
M
u
s
t
a
n
s
i
r
i
y
a
h
U
n
i
v
e
r
si
t
y
,
B
a
g
h
d
a
d
,
I
r
a
q
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
No
v
1
4
,
2
0
2
5
R
ev
is
ed
Ma
r
3
,
2
0
2
6
Acc
ep
ted
Ma
y
1
6
,
2
0
2
6
Nu
m
e
ro
u
s
a
p
p
r
o
a
c
h
e
s
b
a
se
d
on
m
a
c
h
in
e
lea
rn
in
g
h
a
v
e
e
m
e
rg
e
d
in
re
c
e
n
t
y
e
a
rs
to
e
n
h
a
n
c
e
c
ro
p
p
ro
tec
ti
o
n
e
fficie
n
c
y
.
On
e
e
x
a
m
p
le
is
th
e
u
ti
li
z
a
ti
o
n
of
d
e
e
p
n
e
u
ra
l
n
e
two
rk
s
(DN
Ns
)
to
d
iffere
n
ti
a
te
b
e
twe
e
n
v
a
ri
o
u
s
we
e
d
ty
p
e
s
in
a
c
tu
a
l
e
v
e
n
ts
sc
e
n
a
rio
s.
Ne
v
e
rth
e
les
s,
th
e
se
m
e
th
o
d
s
o
ften
n
e
e
d
su
b
sta
n
ti
a
l
in
p
u
t
fr
o
m
e
x
p
e
rts
wh
o
w
o
rk
it
e
ra
ti
v
e
ly
to
d
e
sig
n
th
e
r
o
b
u
st
d
e
e
p
lea
rn
i
n
g
sy
ste
m
.
To
sim
p
li
f
y
su
c
h
p
r
o
c
e
ss
a
n
d
c
o
n
se
rv
e
re
so
u
rc
e
s,
re
se
a
rc
h
e
rs
h
a
v
e
e
x
p
lo
re
d
a
fre
sh
m
e
th
o
d
k
n
o
w
n
as
a
u
to
m
a
ted
de
ep
lea
rn
in
g
o
ur
tec
h
n
o
lo
g
y
’
s
re
c
o
g
n
iza
ti
o
n
of
we
e
d
s
t
h
ro
u
g
h
t
h
e
u
se
of
m
a
c
h
in
e
lea
rn
i
n
g
wa
s
e
v
a
lu
a
ted
u
sin
g
p
la
n
t
se
e
d
li
n
g
s
a
n
d
we
e
d
c
o
ll
e
c
ti
o
n
s
fro
m
p
lan
ts
d
a
tas
e
t
to
a
d
d
re
ss
a
issu
e
of
we
e
d
re
c
o
g
n
iza
ti
o
n
.
T
h
e
stu
d
y
c
o
m
p
a
re
d
v
a
rio
u
s
c
o
n
fi
g
u
ra
ti
o
n
s
,
in
c
lu
d
in
g
p
lan
t
se
g
m
e
n
tatio
n
,
u
s
in
g
a
c
o
ll
e
c
ti
o
n
of
c
las
sifiers
in
p
lac
e
of
S
o
ftma
x
,
a
n
d
trai
n
in
g
wit
h
d
a
tas
e
ts
th
a
t
c
o
n
tain
n
o
ise
.
Th
e
fin
d
in
g
s
in
d
ica
ted
e
n
su
ri
n
g
p
e
rfo
rm
a
n
c
e
,
with
F1
-
sc
o
re
s
of
9
3
.
1
%
a
n
d
9
0
.
2
%
b
a
se
d
on
t
h
e
d
a
tas
e
t
u
ti
li
se
d
.
T
h
e
se
re
su
lt
s
a
li
g
n
to
g
e
th
e
r
with
a
u
to
m
a
ted
m
a
c
h
in
e
lea
rn
in
g
(
Au
t
o
M
L
-
li
n
k
e
d
)
st
u
d
ies
,
wh
il
e
fa
ll
s
h
o
rt
of
m
a
n
u
a
ll
y
f
in
e
-
tu
n
e
d
d
e
e
p
-
lea
rn
in
g
-
b
a
se
d
s
y
ste
m
s
c
re
a
ted
th
ro
u
g
h
h
u
m
a
n
sp
e
c
ialists.
To
c
o
n
c
lu
d
e
,
e
x
p
lo
ri
n
g
th
e
p
o
te
n
ti
a
l
of
c
o
m
b
in
i
n
g
m
a
n
u
a
l
e
x
p
e
rt
wo
rk
a
n
d
a
u
to
m
a
ted
d
e
e
p
lea
rn
i
n
g
c
o
u
ld
be
a
p
ro
m
isin
g
d
irec
ti
o
n
fo
r
e
n
h
a
n
c
in
g
e
fficie
n
c
y
in
p
lan
t
d
e
fe
n
c
e
.
K
ey
w
o
r
d
s
:
Au
to
ML
Deep
lear
n
in
g
Hy
p
er
p
ar
a
m
eter
s
Sin
g
u
lar
v
alu
e
d
ec
o
m
p
o
s
itio
n
W
ee
d
s
T
h
is
is
an
o
p
e
n
a
c
c
e
ss
a
rticle
u
n
d
e
r
th
e
CC
BY
-
SA
li
c
e
n
se
.
C
o
r
r
e
s
p
o
nd
ing
A
uth
o
r
:
Ab
d
Ab
r
ah
im
Mo
s
s
lah
C
o
lleg
e
of
I
s
lam
ic
Scien
ce
,
U
n
iv
er
s
ity
of
An
b
ar
An
b
ar
,
Fallu
jah
,
I
r
aq
E
m
ail:
aisl
.
ab
id
e@
u
o
an
b
ar
.
ed
u
.
iq
1.
I
NT
RO
D
UCT
I
O
N
In
r
ec
e
n
t
tim
es,
t
h
e
n
e
g
ativ
e
i
m
p
ac
t
of
wee
d
s
h
as
led
to
s
ig
n
if
ican
t
g
l
o
b
al
c
r
o
p
lo
s
s
es,
an
d
th
is
tr
e
n
d
is
ex
p
ec
ted
to
co
n
tin
u
e
in
th
e
f
u
tu
r
e
[
1
]
.
W
h
ile
tr
ad
itio
n
al
m
eth
o
d
s
in
v
o
lv
e
d
th
e
u
s
e
of
p
esti
cid
es
to
tack
le
th
is
is
s
u
e,
t
he
eu
r
o
p
ea
n
u
n
io
n
(
E
U)
is
in
cr
ea
s
in
g
ly
ad
o
p
tin
g
a
p
o
licy
aim
ed
at
d
ec
r
ea
s
in
g
th
e
u
s
ag
e
of
p
la
n
t
p
r
o
tectio
n
p
r
o
d
u
cts,
o
win
g
to
ap
p
r
eh
e
n
s
io
n
s
r
eg
ar
d
i
n
g
ch
e
m
ical
r
esid
u
es
on
cr
o
p
s
,
en
v
ir
o
n
m
en
tal
p
o
llu
tio
n
,
an
d
th
e
p
o
ten
tial
f
o
r
d
r
u
g
[
2
]
.
As
p
ar
t
of
th
is
p
o
licy
,
i
n
th
e
co
m
in
g
d
ec
ad
e,
th
e
EU
aim
s
to
r
ed
u
ce
p
esti
cid
e
ap
p
licatio
n
by
50%
[
3
]
.
As
a
r
esu
lt,
au
to
m
atic
weed
co
n
tr
o
l
is
b
ein
g
o
b
s
er
v
ed
as
a
p
r
o
m
is
in
g
s
o
lu
tio
n
to
r
ed
u
ce
th
e
r
elian
ce
on
c
h
e
m
ical
h
er
b
icid
es
f
o
r
weed
m
an
ag
em
en
t
[
4
]
.
T
h
e
r
ec
e
n
t
p
r
o
g
r
ess
in
im
ag
e
class
if
icatio
n
m
eth
o
d
s
o
f
f
er
s
an
o
p
p
o
r
tu
n
ity
to
en
h
an
ce
au
t
o
m
atic
weed
co
n
tr
o
l.
W
h
ile
th
e
r
e
h
as
b
ee
n
a
p
au
s
e
in
th
e
a
d
o
p
tio
n
of
t
h
ese
m
eth
o
d
s
in
th
e
s
ec
to
r
of
ag
r
icu
ltu
r
e,
th
eir
u
s
e
is
r
a
p
id
ly
g
ain
in
g
m
o
m
en
tu
m
.
I
m
ag
e
an
aly
s
is
b
ased
on
m
ac
h
in
e
lear
n
in
g
o
f
f
er
s
a
s
p
ee
d
y
,
non
-
in
v
asiv
e,
an
d
n
o
n
-
d
estru
ctiv
e
s
o
lu
tio
n
f
o
r
ad
d
r
ess
in
g
weed
g
r
o
wth
.
L
ea
r
n
in
g
p
r
o
to
co
ls
h
av
e
b
ee
n
u
s
ed
in
th
e
ag
r
icu
ltu
r
al
f
ield
to
id
en
tify
wee
d
s
an
d
id
en
tify
d
is
ea
s
es
th
at
af
f
ec
t
p
lan
ts
[
5
]
,
[
6
]
.
Am
o
n
g
v
ar
i
o
u
s
ap
p
r
o
ac
h
es,
co
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
s
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
7
2
2
-
3
2
2
1
C
o
m
p
u
t Sci
I
n
f
T
ec
h
n
o
l
,
Vo
l.
7
,
No
.
2
,
J
u
ly
20
26
:
1
67
-
1
78
168
(
C
NNs
)
ar
e
p
r
esen
tly
th
e
m
o
s
t
k
n
o
wn
due
to
th
eir
ab
ilit
y
to
o
v
er
co
m
e
ce
r
tain
ch
allen
g
es,
th
is
in
clu
d
es
f
ac
to
r
s
s
u
ch
as
s
im
ilar
ities
b
e
twee
n
d
if
f
er
e
n
t
class
es
with
in
a
p
lan
t
f
am
ily
,
as
well
as
s
ig
n
if
ican
t
v
ar
iatio
n
s
with
in
a
class
due
to
b
ac
k
g
r
o
u
n
d
,
co
lo
r
,
o
cc
lu
s
io
n
,
p
o
s
e,
an
d
li
g
h
tin
g
co
n
d
itio
n
s
.
B
esid
es
th
eir
ex
ce
llen
t
class
if
icatio
n
p
er
f
o
r
m
an
ce
,
s
o
m
e
s
tu
d
ies
h
av
e
h
ig
h
lig
h
ted
t
h
e
p
o
ten
tial
of
d
ee
p
n
eu
r
al
n
e
two
r
k
s
(
DNNs)
f
o
r
r
ea
l
-
tim
e
weed
co
n
t
r
o
l
in
a
g
r
icu
ltu
r
e,
b
ased
on
th
eir
in
f
er
en
ce
tim
es
[
7
]
.
E
v
en
wit
h
v
ar
io
u
s
p
r
o
p
o
s
ed
tech
n
iq
u
es,
alth
o
u
g
h
d
ee
p
lea
r
n
in
g
m
o
d
els
h
a
v
e
b
ee
n
ap
p
lied
in
ag
r
icu
ltu
r
e,
im
p
lem
en
ti
n
g
th
ese
s
o
lu
tio
n
s
f
u
lly
is
s
till
ch
allen
g
in
g
,
p
r
im
ar
ily
due
to
th
e
co
m
p
lex
ity
of
th
e
ag
r
ic
u
ltu
r
al
en
v
ir
o
n
m
en
t
.
T
h
is
n
ec
ess
itate
s
th
e
u
s
e
of
co
m
p
lex
m
ac
h
i
n
e
v
is
io
n
alg
o
r
ith
m
s
th
at
r
eq
u
ir
e
iter
ativ
e
f
in
e
-
tu
n
in
g
[
8
]
.
B
u
ild
in
g
a
f
itti
n
g
d
ee
p
lear
n
in
g
-
b
ased
s
y
s
tem
in
v
o
l
v
es
in
teg
r
atin
g
a
more
co
m
p
o
n
en
ts
,
as
f
ea
tu
r
e
d
etec
tio
n
,
f
ea
tu
r
e
elicitatio
n
,
a
n
d
class
if
ier
.
T
h
e
task
n
ec
ess
i
tates
ex
p
er
ien
ce
in
s
elec
tin
g
s
u
itab
le
m
o
d
el
ar
c
h
itectu
r
es,
ex
p
er
tis
e
in
m
ath
em
atics,
im
ag
e
an
aly
s
is
,
an
d
co
d
in
g
[
9
]
,
[
1
0
]
.
As
a
r
esu
lt,
ac
h
iev
in
g
th
e
b
e
s
t
p
o
s
s
ib
le
s
y
s
tem
p
er
f
o
r
m
an
ce
r
e
q
u
ir
es
co
n
s
id
er
ab
le
ex
p
er
im
e
n
tatio
n
tim
e,
an
d
a
team
of
ex
p
er
ts
is
n
ee
d
ed
by
m
an
u
ally
test
in
g
d
if
f
er
en
t
m
o
d
els
an
d
co
n
f
i
g
u
r
atio
n
.
B
ased
on
th
e
ab
o
v
e,
th
e
p
la
n
t
n
ee
d
s
to
be
r
etr
ai
n
ed
in
iter
ativ
e
p
r
o
ce
s
s
es,
as
th
e
d
if
f
er
en
ce
s
in
co
n
d
itio
n
s
ar
e
clea
r
b
etwe
en
p
ests
an
d
cr
o
p
s
ac
co
r
d
in
g
to
r
eg
io
n
s
an
d
r
e
g
io
n
s
.
C
o
n
s
eq
u
en
tly
,
th
e
ab
ilit
y
to
p
r
o
d
u
ce
au
to
m
atica
lly
a
Pr
o
s
ag
er
L
ea
r
n
i
n
g
tailo
r
ed
f
o
r
e
v
er
y
u
n
iq
u
e
s
itu
ated
,
ev
en
b
ef
o
r
e
in
d
iv
id
u
als
with
o
u
t
ex
ten
s
iv
e
ex
p
er
tis
e,
wo
u
ld
be
ex
tr
em
ely
b
en
ef
icial.
T
h
ese
s
y
s
tem
s
ar
e
d
ev
is
ed
to
a
u
to
m
atica
lly
ev
alu
at
e
m
u
ltip
le
p
ip
elin
e
co
n
f
ig
u
r
ati
o
n
s
an
d
en
h
an
ce
p
er
f
o
r
m
an
c
e
iter
ativ
ely
.
B
u
t
one
of
th
e
b
ig
g
est
is
s
u
es
with
au
to
m
ated
m
ac
h
in
e
lear
n
in
g
(
Au
to
ML
)
s
y
s
tem
s
is
th
eir
h
ig
h
d
em
a
n
d
f
o
r
co
m
p
u
tin
g
r
eso
u
r
ce
s
.
To
ad
d
r
ess
th
is
is
s
u
e,
IT
co
m
p
an
ies
s
u
ch
as
Go
o
g
le,
Mic
r
o
s
o
f
t,
an
d
Ap
p
le
h
av
e
in
tr
o
d
u
ce
d
u
s
er
-
f
r
ie
n
d
ly
Au
to
ML
clo
u
d
s
o
lu
tio
n
s
,
s
ev
er
al
co
m
m
er
cial
Au
to
ML
s
o
lu
tio
n
s
,
co
m
m
er
cial
Au
to
ML
s
o
lu
tio
n
s
o
f
f
er
ed
by
co
m
p
a
n
ies
s
u
ch
as
Go
o
g
le,
Ap
p
le
an
d
Mic
r
o
s
o
f
t,
p
r
o
v
id
e
s
im
p
le
way
s
to
u
s
e
an
d
tr
ain
m
o
d
els
with
litt
l
e
n
ee
d
f
o
r
a
r
tific
ial
in
tellig
en
ce
k
n
o
wled
g
e
.
C
o
n
v
er
s
ely
,
o
p
en
-
s
o
u
r
ce
to
o
ls
Au
to
Sk
lear
n
,
Au
to
Ker
as,
H2
O
Au
to
ML
,
Au
to
-
W
E
KA,
au
to
x
g
b
o
o
s
t,
T
POT,
an
d
OB
OE
h
av
e
em
er
g
ed
to
in
cr
ea
s
e
k
n
o
wled
g
e
of
A
u
to
ML
p
latf
o
r
m
s
’
b
en
ef
its
an
d
d
r
awb
ac
k
s
.
T
a
b
le
1
p
r
o
v
id
es
a
s
u
m
m
ar
y
of
th
ese
s
y
s
tem
s
.
T
ab
le
1
.
Ov
e
r
v
iew
o
f
v
ar
io
u
s
au
to
m
ated
d
ee
p
lear
n
i
n
g
(
a
u
to
ML
)
s
y
s
tem
s
A
u
t
o
M
L
s
y
st
e
m
Te
c
h
n
o
l
o
g
y
t
y
p
e
R
e
f
e
r
e
n
c
e
G
o
o
g
l
e
C
l
o
u
d
A
u
t
o
M
L
C
l
o
u
d
s
o
l
u
t
i
o
n
[
1
1
]
,
[
1
2
]
A
u
t
o
S
k
l
e
a
r
n
Li
b
r
a
r
y
[
1
3
]
TPO
T
Li
b
r
a
r
y
[
1
3
]
A
u
t
o
K
e
r
a
s
Li
b
r
a
r
y
[
1
4
]
O
B
O
E
Li
b
r
a
r
y
[
1
4
]
2.
RE
L
AT
E
D
WO
RK
S
In
r
ec
en
t
y
ea
r
s
,
Au
to
ML
h
as
b
ee
n
ap
p
lied
in
th
e
ag
r
icu
lt
u
r
a
l
s
ec
to
r
to
p
r
o
ce
s
s
v
ar
io
u
s
ty
p
es
of
d
ata,
in
clu
d
in
g
tim
e
s
er
ies,
s
atellite
an
d
g
r
o
u
n
d
-
b
ased
p
ictu
r
e
s
.
Fo
r
in
s
tan
ce
,
Hay
ash
i
et
a
l
.
[
1
1
]
,
e
m
p
lo
y
e
d
Au
to
ML
to
id
en
tif
y
p
est
in
s
ec
t
s
p
ec
ies,
co
n
s
tr
u
ctin
g
m
o
d
e
ls
with
im
ag
es
of
th
r
ee
a
p
h
id
s
p
ec
ies
th
at
wer
e
tr
ain
ed
in
Go
o
g
le
C
lo
u
d
Au
to
ML
Vis
io
n
.
W
ith
400
im
ag
es
p
er
class
,
th
e
m
o
d
el
ac
h
iev
ed
a
co
r
r
ec
t
r
ec
o
g
n
izatio
n
r
ate
of
o
v
er
9
6
%,
d
em
o
n
s
tr
atin
g
th
e
p
o
te
n
tial
of
r
ec
o
g
n
izatio
n
o
f
p
est
s
p
e
cies
u
s
in
g
A
u
to
ML
.
Similar
ly
,
in
[
1
2
]
,
th
e
au
th
o
r
u
tili
ze
d
A
u
to
ML
to
class
if
y
f
r
u
its
,
b
u
tter
f
lies
,
an
d
lar
v
al
h
o
s
t
p
la
n
ts
an
d
ac
h
iev
ed
an
esti
m
ated
av
e
r
a
g
e
ac
cu
r
ac
y
of
9
7
.
1
%.
In
A
u
to
ML
was
in
teg
r
ated
u
s
in
g
class
if
y
in
g
n
eu
r
al
n
etwo
r
k
tech
n
iq
u
es
r
ice
b
last
d
is
ea
s
e
b
ased
on
f
iv
e
y
e
ar
s
of
c
o
n
tin
u
o
u
s
clim
ate
d
ata,
ac
h
iev
in
g
an
8
9
%
ac
cu
r
ac
y
in
ex
ac
er
b
atio
n
ca
s
es.
Ad
d
itio
n
ally
,
L
ee
et
a
l
.
[
1
3
]
,
d
em
o
n
s
tr
ated
th
e
ef
f
ec
tiv
en
ess
of
Au
to
ML
in
cr
ea
tin
g
m
ap
s
of
Par
th
en
iu
m
g
r
ass
u
s
in
g
m
o
d
els
b
u
ilt
with
s
atellite
im
ag
es
f
r
o
m
L
an
d
s
at
8
an
d
Sen
tin
el
-
2.
T
h
e
Au
to
ML
m
o
d
el
attain
e
d
a
class
if
icatio
n
ac
cu
r
ac
y
of
7
4
%
with
L
an
d
s
at
8
an
d
8
8
.
1
5
%
with
Sen
tin
el
-
2
.
,
h
ig
h
lig
h
tin
g
th
e
u
s
ef
u
ln
ess
of
Au
to
ML
in
cr
ea
tin
g
weed
d
is
p
er
s
al
m
ap
s
u
s
in
g
s
atellite
im
ag
er
y
.
Fin
ally
,
Aco
s
ta
-
Gam
b
o
a
et
a
l.
[
1
4
]
,
co
m
p
ar
ed
Au
t
o
Ker
as
with
tr
an
s
f
er
lear
n
in
g
m
eth
o
d
s
f
o
r
h
ig
h
-
th
r
o
u
g
h
p
u
t
p
lan
t
p
h
en
o
t
y
p
in
g
in
ass
ess
in
g
wh
ea
t
lo
d
g
in
g
u
s
in
g
d
r
o
n
e
im
ag
er
y
.
Alth
o
u
g
h
p
r
e
v
io
u
s
r
esear
ch
h
as
ex
am
in
ed
Au
to
ML
,
th
er
e
r
em
ain
s
a
n
ee
d
to
ass
ess
th
e
t
ec
h
n
iq
u
es
’
ca
p
ac
ity
f
o
r
g
en
er
ality
u
s
in
g
d
iv
er
s
e
p
ictu
r
es
ca
p
tu
r
e
d
u
n
d
er
r
ea
l
-
wo
r
ld
c
o
n
d
itio
n
s
.
To
e
n
h
an
ce
ac
ce
s
s
ib
ilit
y
an
d
r
ep
r
o
d
u
cib
ilit
y
,
it
is
ess
e
n
tial
to
em
p
lo
y
u
s
e
o
p
en
-
s
o
u
r
ce
alter
n
ativ
es
in
s
tead
of
clo
u
d
-
b
ased
p
r
o
p
r
ietar
y
o
n
es.
T
h
is
s
tu
d
y
ev
alu
ates
th
e
ef
f
icac
y
of
Au
to
ML
s
y
s
tem
s
b
ased
on
o
p
en
-
s
o
u
r
ce
s
o
lu
ti
o
n
s
as
a
m
ea
n
s
of
ac
ce
ler
atin
g
an
d
s
tr
ea
m
lin
in
g
th
e
u
s
e
of
v
is
io
n
an
d
m
ac
h
in
e
lear
n
in
g
ap
p
licatio
n
s
in
ag
r
ic
u
lt
u
r
e.
T
h
e
p
r
im
a
r
y
o
b
jectiv
e
of
th
is
s
tu
d
y
is
to
d
eter
m
in
e
wh
et
h
er
Au
t
o
ML
t
ec
h
n
iq
u
es
ca
n
co
m
p
ete
with
m
an
u
ally
-
d
esig
n
ed
ar
ch
itectu
r
es.
T
h
r
ee
p
r
im
ar
y
co
n
tr
ib
u
tio
n
s
ar
e
p
r
esen
ted
in
th
is
p
ap
er
:
i
)
a
p
r
o
ce
d
u
r
e
w
ith
two
s
tag
es
th
at
u
tili
ze
s
Au
to
ML
to
d
ee
p
lear
n
in
g
c
o
m
p
o
n
en
t
ex
tr
ac
tio
n
an
d
class
if
ier
en
s
em
b
les
f
o
r
p
lan
t
id
en
tific
atio
n
;
ii
)
w
e
ex
clu
s
iv
ely
u
s
ed
o
p
e
n
-
s
o
u
r
ce
Au
to
ML
f
r
am
ewo
r
k
s
f
o
r
our
im
p
lem
e
n
tatio
n
,
alo
n
g
with
two
p
u
b
licly
Evaluation Warning : The document was created with Spire.PDF for Python.
C
o
m
p
u
t Sci
I
n
f
T
ec
h
n
o
l
I
SS
N:
2722
-
3
2
2
1
P
erfo
r
ma
n
ce
ev
a
lu
a
tio
n
o
f th
e
d
ee
p
lea
r
n
in
g
s
ystem
fo
r
w
ee
d
r
ec
o
g
n
iz
a
tio
n
(
A
b
d
A
b
r
a
h
im
Mo
s
s
la
h
a
)
169
ac
ce
s
s
ib
le
d
atasets
,
to
f
ac
ilit
ate
tr
an
s
p
ar
e
n
t
an
d
r
ep
r
o
d
u
cib
le
r
esear
ch
;
an
d
iii
)
th
is
s
tu
d
y
ai
m
s
to
e
v
alu
ate
th
e
r
eliab
ilit
y
an
d
s
u
s
ce
p
tib
ilit
y
of
Au
to
ML
s
y
s
tem
s
to
o
v
er
f
itt
in
g
u
s
in
g
n
o
is
y
d
ata
s
am
p
les.
T
h
e
m
eth
o
d
o
l
o
g
y
an
d
ex
p
er
im
en
tal
s
etu
p
a
r
e
p
r
esen
ted
in
s
ec
tio
n
2,
f
o
llo
wed
by
th
e
r
esu
lts
in
s
ec
tio
n
3.
T
h
e
im
p
licatio
n
s
of
th
e
f
in
d
in
g
s
an
d
th
e
s
u
itab
ili
ty
of
th
e
m
eth
o
d
o
lo
g
y
ar
e
th
e
p
ap
er
c
o
n
clu
d
es
by
o
u
tlin
in
g
f
u
t
u
r
e
r
esear
ch
d
ir
ec
tio
n
s
in
s
ec
tio
n
5,
as
d
is
cu
s
s
ed
in
s
ec
tio
n
4.
T
ab
le
1
p
r
o
v
id
es
an
o
v
er
v
iew
of
v
ar
i
o
u
s
au
to
m
ated
d
ee
p
lear
n
in
g
(
A
u
to
ML
)
s
y
s
tem
s
,
in
clu
d
in
g
th
o
s
e
u
s
ed
in
ag
r
icu
lt
u
r
al
ap
p
licatio
n
s
.
3.
M
E
T
H
O
D
3
.
1
.
T
he
pro
po
s
ed
a
pp
ro
a
ch
T
h
e
p
u
r
p
o
s
e
of
th
is
r
esear
ch
is
to
in
v
esti
g
ate
th
e
ef
f
ec
tiv
e
n
ess
of
Au
to
ML
in
id
e
n
tify
in
g
d
if
f
er
e
n
t
ty
p
es
of
wee
d
s
.
T
h
e
r
esear
c
h
e
r
s
aim
to
d
eter
m
i
n
e
wh
eth
e
r
Au
to
ML
can
ac
c
u
r
ately
class
if
y
an
d
d
if
f
e
r
en
tiate
am
o
n
g
weed
s
p
ec
ies
b
ased
on
th
eir
v
is
u
al
ch
ar
ac
ter
is
tics
,
s
u
ch
as
leaf
s
h
ap
e,
tex
tu
r
e
,
an
d
co
lo
r
.
T
h
e
r
esear
ch
is
im
p
o
r
tan
t
b
ec
au
s
e
id
e
n
tify
in
g
an
d
c
o
n
tr
o
llin
g
wee
d
s
is
ess
en
tial
f
o
r
c
r
o
p
m
an
a
g
em
en
t,
an
d
tr
ad
itio
n
al
m
an
u
al
m
eth
o
d
s
can
be
tim
e
-
co
n
s
u
m
in
g
an
d
co
s
tly
.
If
Au
to
ML
p
r
o
v
es
to
be
an
e
f
f
ec
tiv
e
to
o
l
f
o
r
weed
id
en
tific
atio
n
,
it
c
o
u
ld
s
ig
n
if
ican
tly
im
p
r
o
v
e
weed
m
a
n
ag
em
en
t
p
r
ac
tices
an
d
in
c
r
ea
s
e
cr
o
p
y
ield
s
.
Ad
d
itio
n
ally
,
th
e
s
tu
d
y
m
ay
co
n
tr
ib
u
te
to
t
h
e
d
ev
el
o
p
m
en
t
of
more
a
d
v
an
ce
d
an
d
au
to
m
ated
s
y
s
tem
s
f
o
r
ag
r
icu
ltu
r
al
ap
p
licatio
n
s
.
3
.
2
.
T
he
s
ing
ula
r
v
a
lue dec
o
m
po
s
it
io
n
(
SVD
)
t
heo
ry
T
h
e
SVD
is
a
m
ath
em
atica
l
tech
n
iq
u
e
u
s
ed
to
b
r
ea
k
d
o
wn
a
m
atr
ix
in
to
its
co
n
s
titu
en
t
p
ar
ts
.
T
h
ese
p
ar
ts
in
clu
d
e
a
s
et
of
v
ec
to
r
s
t
h
at
ar
e
p
er
p
en
d
icu
la
r
to
each
o
th
er
an
d
h
av
e
a
len
g
th
of
o
n
e
,
as
well
as
a
s
et
of
s
in
g
u
lar
v
alu
es
th
at
r
ep
r
esen
t
th
e
s
tr
en
g
th
of
each
v
ec
to
r
.
T
h
e
lar
g
est
s
in
g
u
lar
v
alu
e
co
r
r
e
s
p
o
n
d
s
to
th
e
m
o
s
t
im
p
o
r
tan
t
v
ec
t
o
r
,
wh
ile
th
e
s
m
allest
s
in
g
u
lar
v
alu
e
r
ep
r
esen
t
s
th
e
least
im
p
o
r
tan
t
v
ec
to
r
.
T
h
e
SVD
can
be
u
s
ed
f
o
r
v
ar
i
o
u
s
ap
p
licatio
n
s
,
s
u
ch
as
r
e
d
u
cin
g
th
e
s
ize
of
a
d
ataset
wh
ile
r
etain
in
g
im
p
o
r
tan
t
in
f
o
r
m
atio
n
,
id
en
tif
y
in
g
r
elev
a
n
t
f
ea
tu
r
es
in
a
d
a
taset,
an
d
f
ilter
in
g
o
u
t
n
o
is
e
f
r
o
m
a
s
ig
n
al.
T
h
is
tech
n
iq
u
e
is
v
alu
ab
le
f
o
r
an
al
y
zin
g
an
d
m
an
ip
u
latin
g
m
atr
i
ce
s
in
m
an
y
d
if
f
er
e
n
t
co
n
tex
t
s
.
It
is
im
p
o
r
tan
t
to
u
s
e
th
is
tech
n
iq
u
e
r
esp
o
n
s
ib
ly
an
d
not
p
r
o
m
o
te
it
as
a
to
o
l
f
o
r
s
ca
m
p
r
o
g
r
am
s
[
1
5
]
.
3
.
3
.
T
he
s
o
lutio
n
’
s
a
rc
hite
ct
ure
T
h
e
p
ap
e
r
ev
al
u
ates
a
m
eth
o
d
o
lo
g
y
th
at
c
o
m
b
in
es
two
A
u
to
ML
s
tep
s
.
T
h
e
o
b
jectiv
e
is
to
attain
co
m
p
ar
ab
le
ef
f
icac
y
to
co
n
v
e
n
tio
n
al
m
eth
o
d
s
,
lik
e
a
s
o
f
t
m
ax
im
u
m
d
ec
o
d
e
r
ato
p
a
n
eu
r
al
-
b
ased
in
f
o
r
m
atio
n
ex
tr
ac
to
r
,
as
ev
id
e
n
ce
d
by
ea
r
lier
r
esear
ch
.
A
m
eth
o
d
o
lo
g
y
co
n
s
is
ts
of
two
s
tep
s
,
wh
er
e
th
e
f
ir
s
t
s
tep
u
tili
ze
s
a
B
ay
esian
n
eu
r
al
ar
ch
itectu
r
e
s
ea
r
ch
ap
p
r
o
ac
h
to
id
en
tif
y
th
e
m
o
s
t
ef
f
ec
tiv
e
ex
tr
ac
tio
n
f
ea
tu
r
e
to
o
l
ca
p
ab
l
e
of
r
e
m
o
v
in
g
t
h
e
m
o
s
t
s
ig
n
if
ic
an
t
f
ea
tu
r
es
f
r
o
m
t
h
e
p
ict
u
r
es.
T
h
e
o
u
t
p
u
t
of
th
ese
s
tep
is
a
DNN
,
au
to
m
atica
lly
f
in
e
-
tu
n
e
d
th
r
o
u
g
h
s
ev
e
r
al
co
n
v
o
lu
tio
n
al
lay
er
s
,
T
h
e
d
ef
a
u
lt
ar
ch
itectu
r
e
s
ea
r
ch
m
et
h
o
d
is
u
s
ed
to
o
b
tain
o
p
tim
al
an
d
r
ele
v
an
t
f
ea
tu
r
es
f
r
o
m
th
e
in
p
u
t
im
ag
es.
T
h
is
ap
p
r
o
ac
h
is
d
ep
icted
in
Fig
u
r
e
1.
It
is
im
p
o
r
tan
t
to
n
o
te
th
at
th
is
tech
n
iq
u
e
s
h
o
u
l
d
not
be
p
r
o
m
o
ted
as
a
to
o
l
f
o
r
s
ca
m
p
r
o
g
r
a
m
s
,
b
ec
au
s
e
s
u
ch
an
ap
p
r
o
ac
h
is
d
esig
n
ed
to
in
itiate
th
e
a
p
p
lica
tio
n
an
d
d
ev
elo
p
m
en
t
of
im
ag
e
an
aly
s
is
an
d
m
ac
h
in
e
lear
n
in
g
tech
n
iq
u
es.
T
h
e
f
ir
s
t
s
tep
of
th
e
m
eth
o
d
o
l
o
g
y
in
v
o
lv
ed
e
x
tr
ac
tin
g
th
e
m
o
s
t
ef
f
ec
tiv
e
ch
ar
ac
ter
is
tics
of
th
e
s
o
u
r
ce
p
h
o
to
s
.
T
h
e
s
ec
o
n
d
s
tep
co
n
c
en
tr
ated
on
f
i
g
u
r
in
g
out
a
f
u
ll
p
ip
elin
e
b
ased
on
al
g
o
r
ith
m
ic
lear
n
in
g
th
at
m
ig
h
t
p
r
o
d
u
ce
th
e
g
r
ea
test
f
in
al
o
u
t
co
m
es.
Stra
teg
ies
f
o
r
p
ick
in
g
out
f
ea
tu
r
es,
r
ed
u
cin
g
s
ize,
an
d
ca
teg
o
r
izatio
n
h
av
e
b
ee
n
test
ed
in
s
id
e
th
is
p
ip
elin
e.
Par
ticu
lar
ly
,
s
ev
er
al
m
eth
o
d
s
wer
e
ass
ess
ed
:
−
By
p
r
o
jectin
g
t
h
e
d
ata
to
a
lo
wer
-
d
im
en
s
io
n
al
s
p
ac
e
u
s
in
g
th
e
SVD
of
t
h
e
d
ata
,
th
e
p
r
i
n
cip
al
co
m
p
o
n
en
ted
m
eth
o
d
[
16
]
p
r
o
v
id
es
a
p
r
o
ce
d
u
r
e
f
o
r
lin
ea
r
d
e
cr
ea
s
in
g
d
im
e
n
s
io
n
ality
th
at
l
ess
en
s
th
e
r
is
k
of
th
e
ex
ce
s
s
iv
e
f
itti
n
g
.
Un
lik
e
o
th
er
m
eth
o
d
s
of
d
im
en
s
io
n
ality
r
ed
u
ctio
n
,
p
r
io
r
to
a
d
o
p
tin
g
th
e
SVD,
in
p
u
t
f
ea
tu
r
es
wer
e
ce
n
tr
ed
t
h
o
u
g
h
not
r
esized
.
It
is
im
p
o
r
tan
t
to
u
s
e
th
is
tech
n
iq
u
e
eth
ically
an
d
n
o
t
p
r
o
m
o
te
it
as
a
to
o
l
f
o
r
s
ca
m
p
r
o
g
r
am
s
.
−
An
o
th
er
tech
n
iq
u
e
em
p
lo
y
e
d
to
less
en
o
v
er
esti
m
atio
n
an
d
r
ed
u
ce
d
im
en
s
io
n
s
is
T
VD
[
17
]
.
Ho
wev
er
,
u
n
lik
e
th
e
p
r
ev
io
u
s
m
eth
o
d
,
with
th
is
m
eth
o
d
,
th
e
d
ata
is
not
ce
n
tr
ed
b
ef
o
r
e
th
e
SVD
is
ca
lcu
lated
.
It
is
im
p
o
r
tan
t
to
n
o
te
th
at
t
h
is
tech
n
iq
u
e
s
h
o
u
ld
n
o
t
be
p
r
o
m
o
te
d
as
a
to
o
l
f
o
r
s
ca
m
p
r
o
g
r
am
s
.
−
C
o
m
p
ar
ab
le
to
th
e
in
itial
s
tr
ateg
y
,
but
em
p
l
o
y
in
g
k
er
n
els
f
o
r
not
lin
ea
r
r
e
d
u
ctio
n
of
d
im
en
s
io
n
ality
in
s
tead
of
r
eg
u
la
r
d
im
i
n
u
tio
n
of
d
im
en
s
io
n
al
is
Ker
n
el
p
r
i
n
cip
al
co
m
p
o
n
e
n
t
an
al
y
s
is
(
PC
A
)
[
18
]
.
It
i
s
g
o
al
is
to
en
h
an
ce
th
e
class
if
ier
’
s
ca
p
ac
ity
f
o
r
g
en
e
r
aliza
tio
n
by
elim
in
atin
g
r
ed
u
n
d
an
t
f
ea
tu
r
es
.
It
is
im
p
o
r
tan
t
to
u
s
e
th
is
tech
n
iq
u
e
eth
ically
an
d
not
p
r
o
m
o
te
it
as
a
to
o
l
f
o
r
s
ca
m
p
r
o
g
r
am
s
.
−
Ad
aBo
o
s
tin
g
is
an
en
s
em
b
le
lear
n
in
g
ap
p
r
o
ac
h
th
at
can
be
o
n
ly
one
class
if
ier
is
em
p
lo
y
e
d
or
as
one
of
th
e
en
s
em
b
le
co
m
p
o
n
en
ts
.
T
h
ey
m
ak
e
a
m
eth
o
d
ca
lled
b
o
o
s
tin
g
,
in
wh
ich
th
e
ch
o
ice
tr
ee
s
ar
e
r
ep
ea
ted
ly
tr
ain
ed
,
g
iv
i
n
g
m
o
r
e
weig
h
t
to
th
e
ex
am
p
les
f
o
r
wh
ich
th
e
f
o
r
ec
ast
is
in
co
r
r
ec
t.
T
h
is
te
ch
n
iq
u
e
s
h
o
u
ld
not
be
p
r
o
m
o
ted
as
a
to
o
l
f
o
r
s
ca
m
p
r
o
g
r
am
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
7
2
2
-
3
2
2
1
C
o
m
p
u
t Sci
I
n
f
T
ec
h
n
o
l
,
Vo
l.
7
,
No
.
2
,
J
u
ly
20
26
:
1
67
-
1
78
170
−
E
x
tr
a
tr
ee
s
is
a
m
ac
h
in
e
lear
n
in
g
tech
n
i
q
u
e
t
h
at
can
be
o
n
l
y
one
ca
teg
o
r
y
is
e
m
p
lo
y
e
d
or
as
co
m
p
o
n
e
n
t
in
s
id
e
a
g
r
o
u
p
.
T
h
e
r
esem
b
la
n
ce
to
er
r
atic
f
o
r
ests
,
but
u
n
lik
e
er
r
atic
f
o
r
ests
,
no
b
o
o
ts
tr
ap
s
am
p
lin
g
is
u
s
ed
.
As
a
r
esu
lt,
it
m
ay
be
more
p
r
o
n
e
to
o
v
er
f
itti
n
g
.
A
n
o
th
er
d
if
f
er
en
ce
b
etwe
en
E
x
tr
a
T
r
ee
s
an
d
r
an
d
o
m
f
o
r
ests
is
th
at
E
x
tr
a
T
r
ee
s
u
s
in
g
an
ar
b
itra
r
y
cu
t
to
cr
ea
te
n
o
d
es
with
in
th
e
b
r
an
c
h
,
th
at
can
h
elp
to
r
ed
u
ce
to
o
tig
h
t.
It
is
im
p
o
r
tan
t
to
u
s
e
th
is
tech
n
iq
u
e
eth
ically
an
d
not
p
r
o
m
o
te
it
as
a
to
o
l
f
o
r
s
ca
m
p
r
o
g
r
a
m
s
.
Fig
u
r
e
1
.
Dis
p
lay
s
s
am
p
les o
f
im
ag
es f
r
o
m
th
e
b
en
c
h
m
ar
k
d
atasets
.
T
h
e
f
ir
s
t
-
r
o
w
ex
h
ib
its
p
ictu
r
es f
r
o
m
th
e
ea
r
ly
cr
o
p
wee
d
s
d
ataset,
wh
ile
th
e
s
ec
o
n
d
-
r
o
w
f
ea
tu
r
es p
ictu
r
es f
r
o
m
t
h
e
p
lan
t seed
lin
g
s
d
ataset
T
h
e
p
u
r
p
o
s
e
of
th
is
s
tu
d
y
is
to
d
eter
m
in
e
if
u
s
in
g
p
r
e
d
icto
r
b
an
d
s
m
a
y
en
h
an
ce
e
f
f
icien
c
y
or
less
en
v
ar
ian
ce
in
th
e
class
if
icatio
n
task
r
esu
lts
.
T
h
e
g
r
o
u
p
m
eth
o
d
th
ese
wo
r
d
s
a
m
ajo
r
ity
a
p
p
r
o
v
al
m
eth
o
d
,
in
wh
ich
th
e
f
in
al
p
r
e
d
ictio
n
is
t
h
e
p
r
o
jecte
d
ca
teg
o
r
y
th
at
r
ec
eiv
es
th
e
m
o
s
t
v
o
tes
f
r
o
m
ea
c
h
class
if
ier
.
Usi
n
g
B
ay
es
o
p
tim
izatio
n
,
th
is
r
o
u
te
was
au
to
m
atica
lly
s
elec
ted
.
It
is
im
p
o
r
tan
t
to
n
o
te
th
at
o
p
en
-
s
o
u
r
ce
s
o
lu
tio
n
s
wer
e
u
tili
ze
d
in
th
e
d
ev
el
o
p
m
en
t
of
th
i
s
p
r
o
ce
s
s
,
wh
ile
th
e
f
in
is
h
ed
p
ip
e
m
ay
be
im
p
o
r
ted
an
d
put
to
u
s
e
f
o
r
a
s
elf
-
s
u
f
f
icien
t
weed
s
u
r
v
eillan
ce
s
y
s
tem
.
W
ith
r
eg
ar
d
to
co
m
p
u
tin
g
lim
itatio
n
s
an
d
d
elay
,
th
e
s
y
s
tem
can
be
im
p
lem
en
ted
as
a
s
ep
ar
ate
or
o
n
lin
e
o
p
tio
n
.
It
is
cr
u
cial
to
u
s
e
th
is
tech
n
o
l
o
g
y
eth
ically
an
d
av
o
id
p
r
o
m
o
ti
n
g
it
as
p
ar
t
of
an
y
s
ca
m
p
r
o
g
r
am
s
.
3
.
4
.
M
a
k
ing
ex
perim
ent
a
l
c
ho
ices
To
g
ain
a
b
etter
u
n
d
e
r
s
tan
d
in
g
of
t
h
e
Au
to
ML
p
r
o
ce
s
s
an
d
id
en
tify
its
ad
v
an
tag
es
an
d
li
m
itatio
n
s
,
ce
r
tain
ex
p
er
im
e
n
tal
lim
itatio
n
s
wer
e
estab
lis
h
ed
,
d
esp
ite
th
e
f
ac
t
th
at
Au
to
ML
d
o
es
n
o
t
r
eq
u
ir
e
an
y
s
p
ec
ial
s
etu
p
to
o
p
er
ate.
Du
r
i
n
g
th
e
m
eth
o
d
o
lo
g
y
e
v
alu
atio
n
,
ce
r
tain
th
e
h
y
p
e
r
p
ar
am
ete
r
s
ettin
g
of
t
h
e
Au
t
o
ML
p
r
o
ce
s
s
was
k
ep
t
co
n
s
is
ten
t.
T
ab
le
2
p
r
esen
ts
th
e
s
elec
ted
h
y
p
er
p
ar
a
m
eter
s
f
o
r
th
e
Au
t
o
ML
p
ip
elin
e
b
ased
on
th
eir
p
r
o
m
is
in
g
p
er
f
o
r
m
an
ce
a
n
d
a
v
ailab
ilit
y
of
co
m
p
u
tatio
n
al
r
eso
u
r
ce
s
.
T
h
e
B
ay
esian
o
p
tim
izatio
n
alg
o
r
ith
m
was
r
u
n
up
to
35
tim
es
to
id
en
tify
th
e
to
p
ex
tr
ac
to
r
of
f
ea
tu
r
es,
with
an
in
i
tial
b
atch
ca
p
ac
ity
of
eig
h
t
an
d
an
y
d
e
p
th
m
o
d
el
test
ed
f
o
r
an
ag
g
r
eg
ate
of
1
0
0
tim
es.
R
eg
ar
d
in
g
th
e
s
o
r
tin
g
g
r
o
u
p
,
each
m
o
d
el
was
tr
ain
e
d
f
o
r
a
m
ax
im
u
m
of
2
m
in
u
tes,
an
d
all
m
o
d
els
wer
e
tr
ain
ed
f
o
r
a
to
tal
of
20
m
in
u
tes.
Data
au
g
m
en
tatio
n
tech
n
iq
u
es
wer
e
ap
p
lied
to
th
e
im
ag
es
b
ef
o
r
e
f
ea
tu
r
e
ex
tr
ac
tio
n
,
in
clu
d
in
g
h
o
r
iz
o
n
t
al
r
o
tatio
n
,
cr
o
p
p
in
g
,
s
ca
lin
g
,
an
d
m
ir
r
o
r
in
g
.
To
im
p
r
o
v
e
th
e
Au
t
o
ML
p
r
o
g
r
a
m
’
s
ab
ilit
y
to
g
en
er
alis
e,
all
p
h
o
to
s
wer
e
r
ed
u
ce
d
to
6
5
×6
5
p
ix
els
in
o
r
d
er
to
r
em
o
v
e
a
n
y
ass
o
ciatio
n
b
etwe
en
b
o
th
p
h
o
t
o
s
ize
an
d
th
e
ac
tu
al
s
ize
of
th
e
p
la
n
ts
.
In
th
e
co
n
tr
ar
y
,
s
ev
er
al
ty
p
e
s
of
a
r
r
an
g
em
en
ts
wer
e
put
to
th
e
test
ex
p
er
im
e
n
tally
in
o
r
d
er
to
d
eter
m
in
e
w
h
ich
was
b
est.
T
h
is
p
r
o
ce
s
s
aim
ed
at
test
in
g
th
e
r
o
b
u
s
tn
ess
of
s
p
ec
if
ic
h
y
p
er
p
ar
am
eter
s
r
e
f
er
to
T
ab
le
3
in
th
e
Au
to
ML
p
ip
eli
n
e
d
esig
n
ag
ain
s
t
o
th
e
r
m
o
d
if
icatio
n
s
.
T
h
e
b
ac
k
g
r
o
u
n
d
’
s
p
r
esen
ce
in
th
e
p
h
o
to
m
ig
h
t
g
r
ea
tly
s
ig
n
if
ican
tly
af
f
ec
t
th
e
f
ea
tu
r
e
ex
tr
ac
tio
n
p
r
o
ce
s
s
.
Hen
ce
,
th
e
u
s
e
of
p
lan
t
s
eg
m
en
tatio
n
was
ev
alu
ated
.
H
u
e
-
s
atu
r
atio
n
-
v
al
u
e
(
HSV)
co
l
o
u
r
s
ch
em
e
was
u
s
ed
as
th
e
th
r
esh
o
ld
s
ap
p
r
o
ac
h
f
o
r
s
eg
m
e
n
ted
Evaluation Warning : The document was created with Spire.PDF for Python.
C
o
m
p
u
t Sci
I
n
f
T
ec
h
n
o
l
I
SS
N:
2722
-
3
2
2
1
P
erfo
r
ma
n
ce
ev
a
lu
a
tio
n
o
f th
e
d
ee
p
lea
r
n
in
g
s
ystem
fo
r
w
ee
d
r
ec
o
g
n
iz
a
tio
n
(
A
b
d
A
b
r
a
h
im
Mo
s
s
la
h
a
)
171
im
p
lem
en
tatio
n
.
M
o
r
eo
v
er
,
u
s
in
g
n
o
is
ier
d
ata
in
th
e
tr
ain
i
n
g
s
tag
e
was
ass
ess
ed
as
a
way
to
im
p
r
o
v
e
t
h
e
au
to
n
o
m
o
u
s
weed
r
ec
o
g
n
izati
o
n
s
y
s
tem
’
s
p
er
f
o
r
m
an
ce
.
T
h
e
f
ea
tu
r
e
ex
tr
ac
to
r
tr
ai
n
in
g
p
r
o
ce
s
s
also
in
clu
d
ed
ex
p
lo
r
in
g
wh
eth
er
th
e
So
f
tm
ax
alg
o
r
ith
m
a
n
d
th
e
lay
e
r
of
co
n
v
o
lu
tio
n
s
h
o
u
ld
be
c
o
u
p
led
to
g
eth
er
in
a
n
etwo
r
k
.
Fin
ally
,
o
n
ce
th
e
f
ea
tu
r
es
wer
e
ex
tr
ac
ted
,
th
e
in
p
u
t
im
ag
e
c
o
u
ld
be
class
if
ied
u
s
in
g
a
s
in
g
le
alg
o
r
ith
m
,
an
e
n
s
em
b
le
of
clas
s
if
ier
s
,
or
a
So
f
tm
ax
p
r
ed
icto
r
.
All
of
t
h
ese
o
p
tio
n
s
wer
e
ass
ess
ed
.
T
ab
le
3
lis
ts
th
e
h
y
p
er
p
ar
am
eter
s
th
at
w
er
e
ass
ess
ed
d
u
r
in
g
th
e
ev
alu
atio
n
p
r
o
ce
s
s
to
id
en
tify
th
e
m
o
s
t
o
p
tim
al
co
n
f
ig
u
r
atio
n
f
o
r
th
e
au
to
m
l
p
ip
elin
e.
T
ab
le
2
.
Fix
ed
h
y
p
e
r
p
ar
am
ete
r
s
f
o
r
th
e
e
x
p
er
im
e
n
ts
F
i
x
e
d
h
y
p
e
r
p
a
r
a
me
t
e
r
s
V
a
l
u
e
M
a
x
i
m
u
m
n
u
m
b
e
r
of
t
r
i
a
l
s
p
e
r
d
e
e
p
mo
d
e
l
35
Ep
o
c
h
s
1
0
0
I
mp
l
e
me
n
t
a
t
i
o
n
of
i
ma
g
e
mo
d
i
f
i
c
a
t
i
o
n
t
e
c
h
n
i
q
u
e
s
b
a
s
e
d
on
g
e
o
m
e
t
r
y
y
e
s
I
mag
e
si
z
e
65
×
65
S
a
mp
l
e
s
i
z
e
8
M
a
x
i
m
u
m
t
i
me
a
l
l
o
c
a
t
e
d
f
o
r
f
i
t
t
i
n
g
e
a
c
h
mo
d
e
l
2
m
i
n
To
t
a
l
t
i
me
t
a
k
e
n
to
f
i
n
d
t
h
e
b
e
st
c
l
a
ss
i
f
i
e
r
20
mi
n
T
ab
le
3
.
Hy
p
er
p
ar
a
m
eter
co
n
f
ig
u
r
atio
n
s
ass
ess
ed
f
o
r
Au
to
M
L
p
ip
elin
e
ev
alu
atio
n
V
a
r
i
a
b
l
e
s
of
t
h
e
h
y
p
e
r
p
a
r
a
m
e
t
e
r
s
e
v
a
l
u
a
t
e
d
F
e
a
t
u
r
e
e
x
t
r
a
c
t
i
o
n
Emp
l
o
y
me
n
t
of
a
f
u
l
l
y
-
c
o
n
n
e
c
t
e
d
n
e
t
w
o
r
k
{Y
e
s,
N
o
}
Ev
a
l
u
a
t
i
o
n
of
t
h
e
i
mp
a
c
t
of
se
g
m
e
n
t
i
n
g
p
l
a
n
t
r
e
g
i
o
n
s
in
i
m
a
g
e
s
on
f
e
a
t
u
r
e
e
x
t
r
a
c
t
i
o
n
{Y
e
s,
N
o
}
N
o
i
s
y
t
r
a
i
n
i
n
g
{Y
e
s,
N
o
}
Ty
p
e
of
c
l
a
s
si
f
i
e
r
{S
i
n
g
l
e
,
S
o
f
t
m
a
x
,
E
n
se
mb
l
e
}
3
.
5
.
Appl
ied
da
t
a
s
et
s
T
h
is
s
tu
d
y
u
tili
ze
d
two
m
ain
d
atasets
:
i
)
th
e
ea
r
ly
cr
o
p
weed
d
ataset,
wh
o
s
e
d
ata
wer
e
t
ak
en
f
r
o
m
th
e
r
esear
ch
[
6
]
,
co
n
s
is
ted
of
504
R
GB
p
h
o
to
s
f
ea
tu
r
in
g
f
o
u
r
d
is
tin
ct
s
p
ec
ies
d
u
r
in
g
th
eir
ea
r
ly
p
h
ases
of
g
r
o
wth
;
a
n
d
ii)
w
ith
an
ac
tu
al
r
eso
lu
tio
n
of
r
o
u
g
h
ly
10
p
ix
el
s
ev
er
y
m
illi
m
eter
,
th
e
Plan
t
S
ee
d
lin
g
s
co
llectio
n
in
clu
d
ed
R
GB
p
h
o
to
s
of
r
o
u
g
h
ly
9
6
0
d
is
tin
ct
p
lan
ts
at
v
a
r
io
u
s
s
tag
es
of
d
ev
elo
p
m
en
t
t
h
at
b
elo
n
g
e
d
to
12
d
if
f
er
en
t
s
p
ec
iesAd
d
itio
n
al
d
etails
r
eg
ar
d
in
g
t
h
is
s
et
of
d
ata
can
be
f
o
u
n
d
in
[
1
5
]
.
F
i
g
u
r
e
2
s
h
o
wca
s
es
ex
am
p
les
of
p
ictu
r
es
f
r
o
m
b
o
th
k
in
d
s
of
d
ata,
a
f
ew
of
wh
ich
h
av
e
u
n
d
er
g
o
n
e
s
eg
m
e
n
tatio
n
of
p
lan
ts
.
T
h
e
f
ir
s
t
d
ataset
f
ea
tu
r
es
v
ar
ia
b
le
illu
m
in
atio
n
co
n
d
itio
n
s
,
w
h
ich
ch
allen
g
es
t
h
e
Au
to
ML
p
r
o
g
r
am
’
s
ca
p
ac
ity
to
g
en
er
alize
an
d
d
is
r
eg
a
r
d
illu
m
in
atio
n
lev
els
wh
en
id
en
tify
in
g
cr
o
p
s
an
d
wee
d
s
.
I
m
a
g
es
f
r
o
m
in
d
o
o
r
p
lan
ts
cu
ltiv
ated
in
a
g
r
o
w
r
o
o
m
wit
h
lig
h
tin
g
t
h
at
is
ar
tific
ial
ad
d
ed
to
s
u
n
lig
h
t
m
ak
e
up
th
e
f
o
l
lo
win
g
d
ata.
Sin
ce
th
e
d
ata
wer
e
g
ath
er
e
d
in
a
l
ab
,
it
i
s
p
o
s
s
ib
le
th
at
s
o
m
e
c
h
ar
ac
ter
is
tics
an
d
m
o
r
p
h
o
lo
g
i
ca
l
tr
aits
of
p
la
n
ts
cu
ltiv
ated
o
u
ts
id
e
ar
e
a
b
s
en
t
as
in
Fig
u
r
e
1
.
Fig
u
r
e
2
.
No
is
y
s
am
p
les f
r
o
m
b
o
th
d
atasets
.
Fro
m
lef
t to
r
ig
h
t: p
ep
p
er
b
lu
r
r
y
ch
a
r
lo
ck
a
n
d
f
at
h
en
with
s
alt,
n
o
is
e
to
m
ato
with
s
alt
an
d
p
ep
p
er
n
o
is
e,
b
l
u
r
r
y
co
tto
n
3
.
6
.
Ana
ly
s
is
T
h
e
Au
to
ML
s
y
s
tem
’
s
ex
ec
u
tio
n
was
an
aly
s
ed
u
s
in
g
th
e
F1
-
s
co
r
e
(
1
)
.
R
ec
all
is
a
r
atio
of
ac
cu
r
ate
ca
teg
o
r
ies
f
o
r
th
e
in
itial
in
f
o
r
m
atio
n
s
et,
wh
ile
ac
cu
r
ac
y
is
th
e
f
in
al
r
atio
of
t
h
e
r
ig
h
t
la
b
els
in
th
e
m
o
d
el
’
s
o
u
tp
u
t.
T
h
is
s
tatis
tic
is
f
r
eq
u
e
n
tly
u
s
ed
in
class
if
icatio
n
p
r
o
b
lem
s
[
1
9
]
.
Sin
ce
b
o
t
h
d
ataset
s
in
th
is
s
tu
d
y
wer
e
m
u
lti
-
class
p
r
o
b
lem
s
with
a
class
im
b
alan
ce
,
in
co
m
p
ar
is
o
n
,
we
ca
lcu
lated
th
e
m
icr
o
-
av
er
ag
in
g
F1
g
r
a
d
e
,
wh
ich
is
a
p
r
ef
e
r
ab
le
a
g
g
r
e
g
a
tio
n
m
eth
o
d
o
v
er
th
e
m
ac
r
o
-
av
er
ag
e.
To
co
n
d
u
ct
s
tatis
tical
co
m
p
a
r
is
o
n
s
,
we
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
7
2
2
-
3
2
2
1
C
o
m
p
u
t Sci
I
n
f
T
ec
h
n
o
l
,
Vo
l.
7
,
No
.
2
,
J
u
ly
20
26
:
1
67
-
1
78
172
u
s
ed
th
e
r
o
b
u
s
t,
p
air
ed
n
o
n
-
p
ar
am
etr
ic
s
tatis
tical
test
s
[
1
]
,
[
2
0
]
.
T
h
ese
test
s
wer
e
em
p
lo
y
ed
to
p
r
ev
en
t
d
r
awin
g
to
o
o
p
tim
is
tic
ass
u
m
p
tio
n
s
.
T
h
e
in
itial
test
was
co
n
d
u
cted
to
ass
ess
co
m
p
ar
ab
le
r
esu
lts
b
etwe
en
p
ip
elin
e
s
ets
s
elec
tio
n
an
d
co
n
tr
o
lled
f
ea
t
u
r
es
ex
tr
ac
tio
n
.
Me
an
wh
ile,
th
e
A
s
ec
o
n
d
test
was
r
u
n
to
co
m
p
ar
e
id
en
tical
p
ip
elin
es
’
ca
p
ab
ilit
ie
s
on
th
e
f
r
esh
a
n
d
n
o
is
y
in
f
o
r
m
atio
n
s
ets.
1
=
2
+
(1
)
Av
o
id
o
v
er
f
itti
n
g
,
it
was
ess
en
tial
to
m
ea
s
u
r
e
th
e
v
ar
iatio
n
in
F1
-
s
co
r
e
b
etwe
en
th
e
tes
t
an
d
tr
ain
in
f
o
r
m
atio
n
s
ets,
g
iv
en
th
at
Au
to
ML
h
as
th
e
p
o
te
n
tial
to
lead
to
o
v
e
r
f
itti
n
g
.
T
h
is
allo
wed
us
to
d
eter
m
in
e
wh
ile
th
e
p
ip
elin
e
f
o
r
Au
to
M
L
p
r
o
d
u
ce
d
a
ca
teg
o
r
izer
ca
p
ab
le
m
eth
o
d
id
en
tify
i
n
g
f
r
esh
s
am
p
les
of
wee
d
s
in
s
tead
if
th
at
m
er
ely
f
it
t
h
e
tr
ain
in
g
s
et
an
d
was
th
u
s
u
n
s
u
itab
le
f
o
r
r
ea
l
-
wo
r
l
d
ap
p
licatio
n
s
[
21
]
.
Ad
d
itio
n
ally
,
t
h
e
e
v
alu
atio
n
of
th
e
r
o
b
u
s
tn
ess
was
co
n
d
u
cted
u
n
d
er
more
ch
allen
g
in
g
s
ce
n
ar
io
s
,
s
u
c
h
as
I
m
ag
es
ar
e
lo
u
d
,
h
az
y
,
an
d
s
p
r
in
k
led
with
s
alt.
T
h
e
F1
r
a
n
k
in
g
m
icr
o
-
av
er
a
g
in
g
was
ad
d
i
tio
n
ally
d
eter
m
in
e
d
u
s
in
g
co
m
p
lete
n
o
is
ier
in
f
o
r
m
atio
n
s
ets,
s
u
ch
as
d
ep
icted
in
Fig
u
r
e
2
.
T
h
e
p
r
o
b
lem
s
a
s
s
o
ciate
d
with
d
ee
p
lear
n
in
g
-
b
ased
tech
n
o
l
o
g
ies
’
r
esil
ien
ce
was
ex
ten
s
iv
ely
ex
p
lo
r
ed
in
[
22
]
.
3
.
7
.
So
f
t
wa
re
a
nd
ha
rdwa
re
T
h
is
wo
r
k
u
tili
ze
d
two
p
r
im
ar
y
s
o
f
twar
e
p
ac
k
ag
es:
“
Au
to
-
Sk
lear
n
0
.
1
0
.
0
an
d
Au
to
Ke
r
as
1
.
0
.
8
”
.
Au
to
Ker
as
is
a
s
o
f
twar
e
to
o
l
th
at
u
tili
ze
s
B
ay
es
o
p
tim
iz
atio
n
-
g
u
id
e
d
n
etwo
r
k
s
of
n
e
u
r
o
n
s
m
o
r
p
h
ic
to
o
p
tim
ize
b
o
th
ar
ch
itectu
r
e
a
n
d
h
y
p
e
r
p
ar
am
ete
r
s
f
o
r
th
e
s
elec
tio
n
of
th
e
m
o
s
t
au
s
p
icio
u
s
p
r
o
ce
d
u
r
es
at
ev
er
y
lev
el.
Ker
as
2
.
4
.
3
a
n
d
T
en
s
o
r
f
lo
w
2
.
3
.
8
b
ac
k
e
n
d
s
ar
e
u
s
ed
in
it
i
s
o
p
er
atio
n
[
23
]
.
T
h
e
s
ec
o
n
d
b
u
n
d
le
u
s
es
non
-
d
ee
p
lear
n
in
g
tec
h
n
iq
u
es
to
ac
co
m
p
lis
h
Au
t
o
ML
;
it
is
a
lib
r
ar
y
t
h
at
is
o
p
e
n
-
s
o
u
r
ce
ca
lled
Au
to
-
Sk
lear
n
.
Fo
r
tr
an
s
f
o
r
m
in
g
d
ata
an
d
au
t
o
m
ated
lear
n
in
g
,
th
is
b
u
n
d
le
u
s
es
th
e
Scik
it
-
L
ea
r
n
au
to
m
at
ed
lear
n
in
g
e
n
g
in
e
(
v
er
s
io
n
0
.
2
2
.
2
)
.
Au
to
-
Sk
lear
n
u
tili
ze
s
a
B
ay
es
Op
tim
is
a
ti
o
n
s
ea
r
ch
tech
n
i
q
u
e,
lik
e
Au
t
o
-
Ker
as,
to
q
u
ic
k
ly
id
en
tify
th
e
b
est
m
o
d
el
p
ip
elin
e
f
o
r
a
g
iv
en
c
o
llectio
n
of
c
h
ar
ac
ter
is
tics
.
Op
en
C
V
3
.
4
.
2
was
u
s
ed
as
th
e
im
ag
e
p
r
ep
r
o
ce
s
s
in
g
lib
r
ar
y
,
an
d
all
th
e
ex
p
er
im
en
ts
wer
e
co
n
d
u
ct
ed
u
s
in
g
Ub
u
n
tu
1
8
.
0
4
as
th
e
o
p
er
atin
g
s
y
s
tem
,
alo
n
g
with
a
GeFo
r
ce
R
T
X
2
0
8
0
T
i
GPU.
3
.
8
.
Alg
o
rit
hm
s
Her
e
is
a
d
escr
ip
tio
n
of
th
e
alg
o
r
ith
m
f
o
r
th
e
Su
p
er
L
ea
r
n
er
m
o
d
el
s
tack
in
g
m
et
h
o
d
:
−
Sp
lit
th
e
tr
ain
in
g
d
ata
i
n
to
K
e
q
u
ally
s
ized
f
o
l
d
s
.
−
Fo
r
each
f
o
l
d
k,
t
r
ain
N
d
iv
e
r
s
e
m
ac
h
in
e
lear
n
i
n
g
m
o
d
els
on
th
e
r
em
ain
in
g
K
-
1
f
o
ld
s
.
−
Use
each
of
th
e
N
m
o
d
els
to
p
r
ed
ict
th
e
o
u
tc
o
m
e
f
o
r
th
e
k
-
th
f
o
ld
.
−
C
o
m
b
in
e
th
e
p
r
ed
ictio
n
s
f
r
o
m
all
N
m
o
d
els
f
o
r
th
e
k
-
th
f
o
ld
to
f
o
r
m
a
n
ew
K
-
f
o
ld
d
ata
s
et.
−
T
r
ain
a
m
eta
-
lear
n
er
on
th
e
n
e
w
K
-
f
o
ld
d
ata
s
et.
−
Use
th
e
m
eta
-
lear
n
er
to
p
r
ed
ic
t
th
e
o
u
tco
m
e
f
o
r
th
e
test
d
ata.
It
is
im
p
o
r
tan
t
to
n
o
te
th
at
in
s
tep
2,
th
e
N
m
o
d
els
s
h
o
u
ld
be
d
iv
er
s
e
an
d
u
n
co
r
r
elate
d
with
each
o
th
er
,
as
th
is
wo
u
ld
lead
to
b
etter
p
er
f
o
r
m
an
ce
.
In
s
tep
5,
th
e
m
eta
-
lear
n
er
can
be
an
y
m
ac
h
in
e
lear
n
in
g
alg
o
r
ith
m
,
s
u
ch
as
lo
g
is
tic
r
eg
r
ess
io
n
or
a
n
eu
r
al
n
etwo
r
k
.
T
h
e
s
u
p
er
lear
n
er
m
eth
o
d
is
u
s
ef
u
l
b
ec
au
s
e
it
can
ad
ap
t
to
d
if
f
er
en
t
ty
p
es
of
d
ata
an
d
lear
n
to
co
m
b
in
e
th
e
s
tr
en
g
th
s
of
d
if
f
er
en
t
m
ac
h
in
e
lear
n
in
g
alg
o
r
ith
m
s
.
T
h
e
m
eth
o
d
o
lo
g
y
em
p
lo
y
ed
by
th
e
alg
o
r
ith
m
in
th
is
ar
ticle
can
be
s
u
m
m
ar
ized
as
f
o
llo
ws:
−
Data
p
r
ep
ar
atio
n
:
two
d
if
f
e
r
en
t
b
en
ch
m
a
r
k
d
atasets
co
n
tain
in
g
cr
o
p
s
,
s
ee
d
lin
g
s
,
an
d
wee
d
s
wer
e
p
r
ep
r
o
ce
s
s
ed
,
au
g
m
e
n
ted
,
an
d
d
iv
id
ed
i
n
to
tr
ain
in
g
an
d
test
in
g
s
ets.
−
I
n
teg
r
atio
n
of
Au
t
o
ML
s
y
s
tem
s
:
two
d
is
tin
ct
Au
to
ML
s
y
s
tem
s
wer
e
in
teg
r
ated
to
ev
alu
ate
th
e
m
eth
o
d
o
l
o
g
y
of
weed
id
e
n
tific
atio
n
.
T
h
ese
s
y
s
tem
s
u
s
ed
d
if
f
er
e
n
t
alg
o
r
ith
m
s
to
g
e
n
er
ate
m
ac
h
in
e
lear
n
in
g
m
o
d
els.
−
Per
f
o
r
m
an
ce
ev
alu
atio
n
:
th
e
F1
s
co
r
e
was
u
s
ed
to
id
en
ti
f
y
th
e
b
est
Au
to
ML
co
n
f
ig
u
r
atio
n
s
,
wh
ich
m
ea
s
u
r
es
th
e
ac
cu
r
ac
y
a
n
d
p
r
ec
is
io
n
of
th
e
s
y
s
tem
’
s
p
r
ed
ictio
n
s
.
T
h
e
F1
s
co
r
es
of
th
e
s
y
s
tem
s
wer
e
ev
alu
ated
on
th
e
test
in
g
s
ets.
−
Fu
tu
r
e
wo
r
k
:
th
e
s
tu
d
y
p
r
o
p
o
s
ed
p
o
ten
tial
f
u
tu
r
e
wo
r
k
to
en
h
an
ce
th
e
p
er
f
o
r
m
an
ce
a
n
d
r
o
b
u
s
tn
ess
of
th
e
Au
to
ML
s
y
s
tem
s
.
T
h
is
in
clu
d
ed
test
in
g
th
e
s
y
s
tem
s
with
n
ew
d
atasets
an
d
n
o
is
y
s
am
p
les.
−
Su
p
er
lear
n
er
alg
o
r
ith
m
:
th
e
s
tu
d
y
in
tr
o
d
u
ce
d
th
e
Su
p
er
L
ea
r
n
er
al
g
o
r
ith
m
as
a
m
eth
o
d
to
c
o
m
b
in
e
th
e
s
tr
en
g
th
s
of
v
ar
io
u
s
m
ac
h
in
e
lear
n
in
g
alg
o
r
ith
m
s
.
T
h
e
alg
o
r
ith
m
was
ex
p
lain
ed
in
d
eta
il,
in
clu
d
in
g
its
ad
v
an
tag
es
o
v
er
o
t
h
er
m
eth
o
d
s
.
In
c
o
n
clu
s
io
n
,
th
e
m
eth
o
d
o
lo
g
y
in
clu
d
ed
ev
al
u
atin
g
th
e
p
er
f
o
r
m
a
n
ce
of
Au
to
ML
s
y
s
te
m
s
f
o
r
id
en
tif
y
in
g
wee
d
s
an
d
p
r
o
p
o
s
in
g
f
u
tu
r
e
r
esear
ch
th
at
co
u
ld
en
h
an
ce
t
h
eir
ac
cu
r
ac
y
an
d
ef
f
ec
tiv
e
n
e
s
s
with
a
p
o
s
itiv
e
im
p
ac
t
on
in
c
r
ea
s
in
g
p
r
o
d
u
cti
o
n
of
f
ield
cr
o
p
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
C
o
m
p
u
t Sci
I
n
f
T
ec
h
n
o
l
I
SS
N:
2722
-
3
2
2
1
P
erfo
r
ma
n
ce
ev
a
lu
a
tio
n
o
f th
e
d
ee
p
lea
r
n
in
g
s
ystem
fo
r
w
ee
d
r
ec
o
g
n
iz
a
tio
n
(
A
b
d
A
b
r
a
h
im
Mo
s
s
la
h
a
)
173
4.
RE
SU
L
T
S
AND
D
I
SCU
SS
I
O
N
T
h
is
s
ec
tio
n
p
r
esen
ts
th
e
r
esu
lts
of
th
e
ex
p
er
im
e
n
ts
to
d
eter
m
in
e
wh
ich
Au
to
ML
p
ip
elin
e
wo
r
k
s
b
est
with
ev
er
y
s
et.
E
ac
h
p
r
o
ce
s
s
c
o
n
f
ig
was
test
ed
ten
tim
es
u
s
i
n
g
v
ar
io
u
s
r
an
d
o
m
s
ee
d
s
in
o
r
d
er
to
in
c
r
ea
s
e
th
e
r
eliab
ilit
y
of
th
e
r
esu
lts
.
T
h
e
m
ed
ian
F1
-
s
co
r
e
f
o
r
ev
er
y
co
n
f
ig
on
ev
er
y
s
et
of
d
ata
(
o
r
ig
in
al/clea
n
an
d
n
o
is
e
o
n
es)
was
th
en
p
r
o
v
i
d
ed
.
T
h
e
d
ata
was
s
p
lit
u
s
in
g
s
tr
atif
ied
s
p
litt
in
g
,
wh
er
ein
25%
of
th
e
s
am
p
les
wer
e
allo
ca
ted
f
o
r
v
alid
atio
n
,
2
5
%
f
o
r
test
in
g
,
an
d
50%
f
o
r
tr
ain
i
n
g
.
W
h
en
a
s
et
of
d
ata
is
d
esc
r
ib
ed
as
h
av
in
g
an
u
n
eq
u
al
v
e
r
s
io
n
,
it
in
d
icate
s
th
at
n
o
is
e
h
as
b
ee
n
in
clu
d
e
d
in
25%
of
th
e
tr
ain
in
g
d
ataset
s
am
p
les
(
s
o
d
iu
m
an
d
p
ep
p
er
)
an
d
50%
of
t
h
e
s
et
of
s
am
p
les
o
v
er
all.
T
h
e
o
u
tco
m
e
of
th
e
tr
ain
in
g
s
et
an
d
th
e
o
u
t
co
m
e
of
th
e
test
in
g
s
et
d
if
f
er
,
as
s
h
o
wn
by
th
e
“
O
v
er
f
itti
n
g
”
lin
e.
“
F1
-
Sco
r
e
”
co
lu
m
n
s
h
o
ws
th
e
p
e
r
f
o
r
m
an
ce
on
th
e
test
s
et.
4
.
1
.
Da
t
a
s
et
f
o
r
ea
rly
cr
o
p
wee
ds
Usi
n
g
th
e
o
r
ig
in
al
d
ataset
f
o
r
tr
ain
in
g
,
T
ab
le
4
s
h
o
ws
th
e
to
p
10
p
er
f
o
r
m
in
g
Au
to
M
L
p
ip
elin
es
b
ased
on
th
eir
F1
s
co
r
es
f
o
r
th
e
ea
r
ly
cr
o
p
wee
d
s
d
ataset.
To
en
s
u
r
e
co
n
s
is
ten
cy
,
10
test
s
wer
e
co
n
d
u
cte
d
f
o
r
ev
er
y
p
ip
elin
e
d
esig
n
u
s
in
g
v
ar
io
u
s
r
an
d
o
m
s
ee
d
lin
g
s
,
an
d
th
e
r
esu
lts
g
iv
en
F1
s
co
r
e
is
th
e
av
er
ag
e
s
co
r
e
ac
r
o
s
s
all
r
u
n
s
.
T
h
e
p
i
p
elin
e
s
wer
e
tr
ain
ed
with
a
s
tr
atif
ied
s
p
lit
of
50%
d
ata
f
o
r
25%
v
alid
atio
n
,
25%
ev
alu
atio
n
,
an
d
25%
lea
r
n
in
g
.
Fo
r
lo
u
d
d
atasets
,
50%
of
th
e
tr
ain
in
g
d
ata
was
c
o
n
tam
in
a
ted
with
eith
er
s
alt
an
d
p
ep
p
er
n
o
is
e
or
h
az
in
ess
.
Fo
llo
win
g
th
e
co
m
p
letio
n
of
t
h
e
Frie
d
m
an
ev
al
u
atio
n
at
a
0
.
1
co
n
f
id
en
ce
lev
el,
to
co
m
p
a
r
e
a
d
if
f
er
e
n
t
d
is
tr
ib
u
tio
n
,
wh
e
r
e
p
lan
t
s
eg
m
en
tati
o
n
was
u
s
ed
,
t
h
er
e
was
a
n
o
ti
ce
ab
le
d
if
f
er
e
n
ce
in
co
m
p
ar
in
g
it
with
th
e
alter
n
ati
v
e.
Ho
wev
e
r
,
th
e
So
f
tm
ax
class
if
ier
s
h
o
wed
h
i
g
h
er
v
ar
ia
n
ce
in
t
h
eir
r
esu
lts
an
d
co
u
ld
be
co
n
s
id
er
ed
less
r
o
b
u
s
t.
So
m
e
p
ip
elin
es
ac
h
iev
ed
n
ea
r
ly
100%
p
er
f
o
r
m
a
n
ce
at
th
e
ex
er
cise
g
r
o
u
n
d
,
in
d
icatin
g
p
o
ten
tial
o
v
er
f
itti
n
g
.
As
an
illu
s
tr
atio
n
,
a
t
o
p
-
r
an
k
ed
p
ip
elin
e
h
ad
an
a
v
er
ag
e
tr
ain
i
n
g
s
et
p
er
f
o
r
m
an
ce
of
9
9
.
9
8
%.
W
h
en
ev
alu
ated
on
a
n
o
is
y
test
d
ataset,
p
er
f
o
r
m
a
n
ce
d
ec
r
e
a
s
ed
in
m
o
s
t
ca
s
es,
p
h
o
to
s
th
at
ar
e
h
az
y
ar
e
more
d
if
f
icu
lt
to
ca
teg
o
r
is
e
c
o
r
r
ec
tl
y
.
So
m
e
p
ip
elin
es
m
ai
n
tain
ed
ex
ce
llen
t
o
u
t
p
u
t
f
o
r
th
e
n
o
is
e
of
s
alt
an
d
p
ep
p
er
,
w
h
ile
o
th
er
s
s
h
o
wed
a
g
lar
in
g
lack
of
r
esil
ien
ce
to
th
ese
k
in
d
s
of
n
o
is
es.
U
s
i
n
g
a
v
e
r
s
i
o
n
of
t
h
e
d
a
t
a
s
e
t
f
o
r
t
r
a
i
n
i
n
g
w
i
t
h
a
d
d
e
d
n
o
i
s
e
,
in
T
a
b
l
e
5,
we
p
r
e
s
e
n
t
t
h
e
i
m
p
a
c
t
of
t
r
a
i
n
i
n
g
A
u
t
o
M
L
p
i
p
e
l
i
n
e
s
w
i
t
h
n
o
i
s
y
s
a
m
p
l
e
s
.
S
p
e
c
i
f
i
c
a
l
l
y
,
we
t
r
a
i
n
e
d
t
h
e
m
o
d
e
l
s
on
a
d
a
t
a
s
e
t
t
h
a
t
w
a
s
c
o
m
p
o
s
e
d
of
5
0
%
c
l
e
a
n
s
a
m
p
l
e
s
,
2
5
%
s
a
l
t
a
n
d
p
e
p
p
e
r
s
a
m
p
l
e
s
,
a
n
d
2
5
%
b
l
u
r
r
y
s
a
m
p
l
e
s
.
U
s
i
n
g
a
0
.
1
p
r
o
b
a
b
i
l
i
t
y
t
h
r
e
s
h
o
l
d
,
we
r
a
n
a
F
r
i
e
d
m
a
n
e
x
a
m
a
n
d
d
i
s
c
o
v
e
r
e
d
t
h
a
t
t
h
e
i
n
i
t
i
a
l
f
o
u
r
l
i
n
e
s
h
a
d
s
i
m
i
l
a
r
p
e
r
f
o
r
m
a
n
c
e
s
b
a
s
e
d
on
t
h
e
F1
s
c
o
r
e
c
o
l
u
m
n
,
but
t
h
e
y
o
u
t
p
e
r
f
o
r
m
e
d
t
h
e
o
t
h
e
r
p
i
p
e
l
i
n
e
s
.
I
n
t
e
r
e
s
t
i
n
g
l
y
,
it
w
a
s
d
i
s
c
o
v
e
r
e
d
t
h
a
t
u
t
i
l
i
z
i
n
g
b
o
t
h
p
l
a
n
t
r
e
c
o
g
n
i
t
i
o
n
a
n
d
a
f
u
l
l
y
-
c
o
n
n
e
c
t
e
d
p
r
e
d
i
c
t
o
r
t
o
g
e
t
h
e
r
l
e
d
to
b
e
t
t
e
r
p
e
r
f
o
r
m
a
n
c
e
l
o
u
d
s
u
r
r
o
u
n
d
i
n
g
s
,
w
h
i
c
h
is
in
c
o
n
t
r
a
s
t
to
t
h
e
r
e
s
u
l
t
s
p
r
e
s
e
n
t
e
d
in
T
a
b
l
e
4.
T
h
e
F1
s
c
o
r
i
n
g
l
i
n
e
r
e
s
u
l
t
s
w
e
r
e
c
o
n
s
t
a
n
t
w
i
t
h
t
h
o
s
e
p
r
e
s
e
n
t
e
d
in
T
a
b
l
e
4,
j
u
s
t
i
f
y
it
t
r
a
i
n
i
n
g
p
o
t
e
n
t
i
a
l
l
y
r
e
s
u
l
t
in
p
i
p
e
l
i
n
e
s
p
e
r
f
o
r
m
i
n
g
s
i
m
i
l
a
r
l
y
to
t
h
o
s
e
t
r
a
i
n
e
d
on
c
o
m
p
l
e
t
e
l
y
c
l
e
a
n
d
a
t
a
s
e
t
s
w
h
e
n
u
s
e
d
w
i
t
h
n
o
i
s
y
d
a
t
a
.
P
a
r
t
i
c
u
l
a
r
p
i
p
e
l
i
n
e
s
’
v
o
l
a
t
i
l
i
t
y
a
n
d
p
e
r
f
o
r
m
a
n
c
e
,
h
o
w
e
v
e
r
,
m
i
g
h
t
d
e
c
l
i
n
e
.
B
e
c
a
u
s
e
i
n
s
t
a
n
c
e
,
t
h
e
t
o
p
-
p
e
r
f
o
r
m
i
n
g
s
y
s
t
e
m
in
T
a
b
l
e
4,
p
r
e
s
e
n
t
e
d
in
t
h
e
f
i
r
s
t
r
o
w
,
s
a
w
a
d
r
o
p
of
2
.
3
4
%
(
9
0
.
9
3
%
to
8
8
.
8
%
)
on
t
h
e
n
o
i
s
y
s
a
l
t
a
n
d
p
e
p
p
e
r
t
e
s
t
s
e
t
a
n
d
5
.
2
3
%
(
9
3
.
7
%
to
8
8
.
8
%
)
on
t
h
e
p
u
r
e
t
e
s
t
s
e
t
s
e
e
T
a
b
l
e
5;
r
o
w
3.
H
o
w
e
v
e
r
,
it
p
e
r
f
o
r
m
e
d
b
e
t
t
e
r
(
6
9
.
0
7
%
to
8
9
.
1
7
%
)
on
t
h
e
h
a
z
y
d
a
t
a
s
e
t
.
T
h
a
t
i
m
p
l
i
e
s
t
h
i
s
t
r
a
i
n
i
n
g
w
i
t
h
e
r
r
a
t
i
c
s
a
m
p
l
e
s
m
a
y
r
e
s
u
l
t
in
b
e
t
t
e
r
p
e
r
f
o
r
m
a
n
c
e
on
c
l
e
a
n
d
a
t
a
s
e
t
s
s
e
e
T
a
b
l
e
4;
r
o
w
4
a
n
d
T
a
b
l
e
5;
r
o
w
1,
but
a
l
s
o
in
d
e
c
r
e
a
s
e
d
p
e
r
f
o
r
m
a
n
c
e
s
e
e
T
a
b
l
e
4;
ro
w
1
a
n
d
T
a
b
l
e
5;
r
o
w
8.
N
o
t
a
b
l
y
,
we
o
b
s
e
r
v
e
d
a
s
i
g
n
i
f
i
c
a
n
t
i
m
p
r
o
v
e
m
e
n
t
(P
-
v
a
l
u
e
f
o
r
W
i
l
c
o
x
o
n
<
0
.
0
5
)
in
p
e
r
f
o
r
m
a
n
c
e
w
h
e
n
e
v
a
l
u
a
t
i
n
g
w
i
t
h
b
l
u
r
r
y
d
a
t
a
s
e
t
s
,
w
i
t
h
a
l
l
c
a
s
e
s
s
h
o
w
i
n
g
a
n
o
t
a
b
l
e
i
n
c
r
e
a
s
e
in
p
e
r
f
o
r
m
a
n
c
e
.
T
ab
le
4
s
h
o
ws
th
e
h
i
g
h
est
p
er
f
o
r
m
in
g
au
to
m
l
co
n
f
ig
u
r
ati
o
n
s
f
o
r
t
h
e
ea
r
ly
c
r
o
p
wee
d
s
d
ataset,
p
r
esen
ted
as
m
ea
n
an
d
s
tan
d
ar
d
d
ev
iatio
n
v
alu
es.
(
No
te
:
ps
r
ef
er
s
to
p
lan
t
s
eg
m
en
tatio
n
a
n
d
fc
r
e
f
er
s
to
f
u
lly
-
co
n
n
ec
ted
)
.
T
ab
le
5
p
r
esen
ts
th
e
h
ig
h
est
p
er
f
o
r
m
in
g
au
t
o
m
l
co
n
f
ig
u
r
atio
n
s
with
th
eir
m
ea
n
an
d
s
tan
d
ar
d
d
ev
iatio
n
f
o
r
th
e
p
lan
t seed
lin
g
s
d
ataset.
T
h
e
ab
b
r
ev
iatio
n
s
u
s
ed
in
th
e
tab
le
ar
e
p
s
f
o
r
p
la
n
t seg
m
en
tatio
n
an
d
f
c
f
o
r
f
u
lly
-
c
o
n
n
ec
ted
.
T
ab
le
4
.
T
o
p
-
p
e
r
f
o
r
m
in
g
A
u
to
ML
co
n
f
ig
u
r
atio
n
s
f
o
r
th
e
ea
r
ly
cr
o
p
wee
d
s
da
taset
PS
FC
C
l
a
s
si
f
i
e
r
F
1
S
c
o
r
e
O
v
e
r
f
i
t
t
i
n
g
S
a
l
t
F
1
B
l
u
r
F
1
Y
e
s
No
En
se
mb
l
e
9
3
.
7
±
1
.
1
3
6
.
2
8
±
1
.
0
6
9
0
.
9
3
±
5
.
5
6
6
9
.
0
7
±
1
1
.
5
4
Y
e
s
No
S
i
n
g
l
e
9
3
.
6
±
1
.
6
6
.
4
±
1
.
6
8
8
.
4
±
6
7
0
.
8
±
2
No
Y
e
s
En
se
mb
l
e
9
2
.
1
5
±
2
.
4
7
.
8
2
±
2
.
5
8
6
0
.
8
±
8
4
9
.
2
±
1
4
.
8
Y
e
s
No
S
o
f
t
ma
x
9
1
.
9
±
6
.
3
6
7
.
8
4
±
6
.
6
7
8
7
.
2
±
9
.
2
4
6
7
.
6
±
9
.
8
1
Y
e
s
Y
e
s
En
se
mb
l
e
9
1
.
3
1
±
2
.
9
4
8
.
5
8
±
3
.
0
1
8
8
.
8
±
3
.
4
2
6
2
.
6
3
±
1
0
.
8
T
ab
le
5
.
T
o
p
-
p
e
r
f
o
r
m
in
g
A
u
to
ML
co
n
f
ig
u
r
atio
n
s
f
o
r
th
e
p
lan
t seed
lin
g
s
d
ataset
PS
FC
C
l
a
s
si
f
i
e
r
F
1
S
c
o
r
e
O
v
e
r
f
i
t
t
i
n
g
S
a
l
t
F
1
B
l
u
r
F
1
Y
e
s
Y
e
s
En
se
mb
l
e
9
3
.
8
±
3
.
4
4
6
.
0
9
±
3
.
4
9
3
±
4
.
4
1
9
3
.
4
±
4
.
2
5
Y
e
s
Y
e
s
S
i
n
g
l
e
9
2
.
9
1
±
4
.
7
1
6
.
9
3
±
4
.
6
4
9
2
.
5
7
±
4
.
8
7
9
2
.
1
1
±
4
.
3
7
Y
e
s
Y
e
s
S
o
f
t
ma
x
9
2
.
0
9
±
4
.
7
3
7
.
4
6
±
4
.
8
6
9
2
.
2
7
±
4
.
8
9
0
.
8
4
±
4
.
6
5
No
No
S
i
n
g
l
e
9
1
.
6
±
4
.
8
8
8
.
0
4
±
4
.
9
6
8
8
±
4
.
3
8
9
0
.
2
±
5
.
7
Y
e
s
No
En
se
mb
l
e
8
8
.
8
±
1
.
9
6
1
1
.
0
2
±
2
.
1
8
8
.
8
±
2
.
8
5
8
9
.
1
7
±
3
.
2
2
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
7
2
2
-
3
2
2
1
C
o
m
p
u
t Sci
I
n
f
T
ec
h
n
o
l
,
Vo
l.
7
,
No
.
2
,
J
u
ly
20
26
:
1
67
-
1
78
174
4
.
2
.
Seedlin
g
s
of
pla
nts
da
t
a
s
et
Usi
n
g
th
e
o
r
ig
in
al
d
ataset
f
o
r
tr
ain
in
g
,
T
ab
le
6
s
h
o
wca
s
es
th
e
to
p
-
p
er
f
o
r
m
i
n
g
Au
to
M
L
p
ip
elin
es
b
ased
on
th
e
F1
s
co
r
e
m
etr
ic
f
o
r
th
e
in
f
o
r
m
atio
n
s
et
f
o
r
p
lan
t
s
ee
d
lin
g
s
.
T
h
e
s
etu
p
s
with
th
e
h
ig
h
est
ef
f
icien
cy
ar
e
r
ep
r
esen
te
d
by
th
e
in
itial
two
r
o
ws,
ac
co
r
d
in
g
to
th
e
Frie
d
m
an
ex
am
,
wh
i
ch
was
ca
r
r
ied
out
with
a
lev
el
of
tr
u
s
t
of
0
.
0
1
.
T
h
ese
co
n
f
ig
u
r
atio
n
s
s
h
ar
e
th
e
u
s
e
of
v
eg
etativ
e
s
eg
m
en
ts
an
d
s
teer
clea
r
of
f
u
lly
in
ter
co
n
n
ec
ted
n
etwo
r
k
s
wh
e
n
co
llectin
g
f
ea
tu
r
es.
T
h
e
g
r
ea
test
r
esu
lts
wer
e
o
b
tain
e
d
wh
en
So
f
tm
a
x
was
s
u
b
s
titu
ted
with
a
n
ew
p
r
ed
ict
o
r
(
b
o
th
g
r
o
u
p
a
n
d
s
in
g
le;
9
0
.
7
4
±
0
.
8
an
d
9
0
.
1
6
±
0
.
6
7
,
co
r
r
e
s
p
o
n
d
in
g
l
y
)
,
wh
ich
is
co
n
s
is
ten
t
with
th
e
f
in
d
in
g
s
in
T
ab
le
4.
T
h
e
r
ef
o
r
e
,
th
ese
co
n
f
ig
u
r
atio
n
s
can
be
co
n
s
id
er
ed
as
a
s
o
lid
s
tar
tin
g
p
o
in
t
f
o
r
a
d
d
itio
n
al
r
esear
ch
on
d
atasets
th
at
ar
e
f
r
ee
f
r
o
m
n
o
is
e.
T
h
e
p
lan
t
s
eg
m
en
t
is
a
u
s
ef
u
l
ch
o
ice
a
m
o
n
g
th
e
ex
tr
em
e
p
ar
a
m
eter
s
to
g
et
th
e
b
est
o
u
tco
m
es.
Fu
r
th
er
m
o
r
e,
in
s
tead
of
em
p
lo
y
in
g
th
e
f
ea
tu
r
e
ex
tr
ac
to
r
to
tr
ain
ad
d
itio
n
al
class
if
ier
s
,
S
o
f
tm
ax
,
to
b
eh
a
v
e
well
o
v
er
a
ll,
as
th
is
s
tr
ateg
y
was
u
s
ed
by
8
of
th
e
to
p
10
s
y
s
tem
s
.
Ho
wev
er
,
wh
e
n
test
ed
on
n
o
is
y
in
f
o
r
m
ati
o
n
s
ets,
ev
er
y
s
y
s
tem
s
aw
a
s
h
ar
p
d
ec
lin
e
in
ef
f
icien
cy
(
W
ilco
x
o
n
p
-
v
alu
e<
0
.
0
1
)
,
s
im
ilar
to
wh
at
h
ap
p
en
ed
with
th
e
E
ar
lier
C
r
o
p
W
ee
d
s
d
ata.
H
o
wev
er
,
it
’
s
cr
u
cial
to
r
em
em
b
er
th
at
ce
r
tain
s
tr
u
c
tu
r
es
wer
e
more
r
esis
tan
t
to
p
ar
ticu
lar
k
in
d
s
of
n
o
is
e
th
an
o
th
er
s
.
In
g
en
er
al,
t
h
e
v
alu
es
in
th
e
“
Ov
er
f
itti
n
g
”
b
o
x
ar
e
g
r
ea
ter
th
a
n
th
o
s
e
f
o
u
n
d
in
T
ab
le
4,
p
a
r
ticu
lar
ly
f
o
r
th
e
f
ir
s
t
p
air
of
p
ip
elin
es
ac
r
o
s
s
b
o
th
lis
ts
(
W
i
lco
x
o
n
p
-
v
alu
e<
0
.
0
1
)
[
16
]
.
An
u
n
f
o
cu
s
ed
v
er
s
io
n
f
o
r
t
h
e
tr
ain
d
ataset
,
T
ab
le
7
s
h
o
ws
wh
eth
er
Au
to
ML
p
r
o
ce
s
s
o
r
s
p
er
f
o
r
m
ed
b
etter
or
wo
r
s
e
wh
e
n
tau
g
h
t
with
n
o
is
y
d
ata
(
i.e
.
,
50%
clea
n
,
25%
p
ep
p
er
a
n
d
s
alt,
an
d
2
5
%
b
lu
r
r
in
g
)
.
Fo
llo
win
g
th
e
Frie
d
m
an
ev
al
u
atio
n
at
a
lev
el
of
tr
u
s
t
of
0
.
0
5
,
th
e
p
ip
es
with
th
e
b
est
p
er
f
o
r
m
an
ce
ar
e
d
is
p
lay
ed
in
th
e
t
o
p
4
r
o
ws
of
th
e
tab
le.
T
h
e
two
m
o
s
t
o
f
ten
u
s
ed
ex
tr
em
e
p
ar
am
eter
s
in
th
ese
s
tr
u
ctu
r
es
wer
e
av
o
id
in
g
a
f
u
lly
co
n
n
ec
ted
lin
k
an
d
u
s
in
g
cr
o
p
d
iv
is
io
n
.
As
an
ticip
ated
,
th
e
r
esu
lt
of
t
h
e
e
v
alu
atio
n
with
d
ata
th
at
was
n
o
is
y
s
h
o
wed
an
o
v
er
all
im
p
r
o
v
em
en
t.
No
n
eth
e
less
,
T
ab
le
5
’
s
“
Ov
er
f
itti
n
g
”
co
lu
m
n
r
ev
ea
led
g
en
er
ally
wo
r
s
e
o
u
tco
m
es
(
W
ilco
x
o
n
p
-
v
alu
e<
0
.
1
)
.
T
h
is
im
p
lies
th
at
th
e
p
ip
elin
es
m
ig
h
t
s
tr
u
g
g
le
with
ac
cu
r
ate
ca
teg
o
r
is
atio
n
.
B
ased
o
n
th
e
an
aly
s
is
o
f
th
e
r
esu
lts
p
r
esen
ted
ea
r
lier
,
it
is
ev
id
en
t
th
at
ce
r
tain
h
y
p
er
p
ar
am
eter
s
h
ad
a
g
r
ea
ter
im
p
ac
t o
n
th
e
last
p
er
f
o
r
m
an
ce
.
I
n
Fig
u
r
e
3
,
it is
co
n
s
o
lid
atin
g
th
e
o
u
tco
m
es
f
r
o
m
th
e
two
s
ets s
u
p
p
o
r
t
th
e
id
ea
th
at
ap
p
ly
in
g
v
eg
etatio
n
s
eg
m
en
tatio
n
as a
p
r
elim
in
ar
y
p
r
o
ce
s
s
in
g
m
eth
o
d
im
p
r
o
v
ed
p
er
f
o
r
m
an
ce
,
as
s
h
o
wn
in
Fig
u
r
e
3
(
a)
.
T
h
e
b
est
attr
ib
u
tes
f
r
o
m
th
e
o
r
ig
in
al
p
h
o
to
s
wer
e
ex
tr
ac
ted
b
y
th
is
n
eu
r
al
n
etwo
r
k
.
Fin
d
in
g
a
co
m
p
r
eh
en
s
iv
e
m
ac
h
in
e
lear
n
in
g
-
b
ased
p
ip
elin
e
ca
p
ab
le
o
f
p
r
o
d
u
cin
g
th
e
o
p
tim
al
o
u
tco
m
e
was
th
e
s
ec
o
n
d
s
tag
e
af
ter
th
e
ch
ar
ac
ter
is
tics
wer
e
ex
tr
ac
ted
.
I
n
co
n
tr
ast,
th
e
u
s
e
o
f
f
u
lly
-
co
n
n
ec
ted
n
etwo
r
k
s
h
ad
a
r
elativ
ely
lim
ited
im
p
ac
t
o
n
th
e
o
v
er
all
p
er
f
o
r
m
an
ce
,
as
illu
s
tr
ated
in
Fig
u
r
e
3
(
b
)
.
T
h
e
d
is
tr
ib
u
tio
n
s
o
f
th
e
p
er
f
o
r
m
an
ce
m
etr
ics
in
d
icate
th
at
en
ab
lin
g
o
r
d
is
ab
lin
g
th
e
f
u
lly
-
co
n
n
ec
ted
lay
er
d
id
n
o
t
lead
to
s
tatis
tically
s
ig
n
if
ican
t d
if
f
er
en
ce
s
.
I
n
a
s
im
ilar
v
ein
,
th
e
ty
p
e
o
f
class
if
ier
was im
p
o
r
tan
t.
Fig
u
r
e
3
(
c)
s
h
o
ws th
at
wh
ile
th
e
Sin
g
le
an
d
So
f
tm
ax
class
if
ier
s
m
ay
also
p
r
o
d
u
ce
g
o
o
d
r
esu
lts
,
th
e
E
n
s
em
b
le
tech
n
iq
u
e
h
ad
th
e
h
ig
h
est
m
ed
ian
p
er
f
o
r
m
an
ce
.
T
h
e
v
ar
ian
ce
f
o
r
th
e
Sin
g
le
class
if
ier
was
lo
wer
th
an
th
at
o
f
th
e
So
f
tm
ax
ap
p
r
o
ac
h
,
s
u
g
g
esti
n
g
th
at
it
was
a
m
o
r
e
r
eliab
le
class
if
ier
.
T
h
e
W
ilco
x
o
n
ex
am
was
u
s
ed
to
co
m
p
ar
e
th
e
v
ar
io
u
s
d
is
tr
ib
u
tio
n
s
,
an
d
ju
s
t
th
e
s
eg
m
en
tatio
n
o
f
p
lan
ts
was
f
o
u
n
d
to
b
e
s
ig
n
if
ic
an
tly
d
if
f
er
en
t
f
r
o
m
a
d
if
f
er
en
t
o
n
e
(p
-
v
alu
e<
0
.
0
1
)
.
T
h
e
o
u
tco
m
es
s
h
o
w
th
at
class
if
ier
s
ca
n
b
e
g
en
er
ated
v
ia
Au
to
ML
with
F1
s
co
r
es
ab
o
v
e
9
0
%,
wich
is
co
n
s
is
ten
t
with
o
th
er
r
elate
d
s
tu
d
ies
[
22
]
,
[
23
]
.
Ho
wev
er
,
o
u
r
p
r
ev
io
u
s
wo
r
k
ac
h
iev
ed
h
ig
h
er
p
er
f
o
r
m
an
ce
[
16
]
,
[
17
]
,
b
u
t
at
th
e
co
s
t
o
f
s
ig
n
if
ican
t
tim
e
an
d
ef
f
o
r
t
f
r
o
m
d
ee
p
lear
n
in
g
ex
p
er
ts
to
f
in
e
-
tu
n
e
th
e
n
etwo
r
k
s
.
Au
to
ML
o
f
f
er
s
a
s
o
lu
tio
n
to
s
h
o
r
ten
o
r
av
o
id
th
is
p
r
o
ce
s
s
.
Ou
r
m
ain
f
o
cu
s
in
th
is
s
tu
d
y
was
to
p
r
o
v
id
e
a
tr
u
s
two
r
th
y
ex
p
er
im
en
tal
co
n
f
ig
u
r
atio
n
f
o
r
ass
ess
in
g
Au
to
ML
q
u
ality
in
d
if
f
er
en
t
s
ce
n
ar
io
s
,
as
o
p
p
o
s
ed
to
au
to
m
atica
lly
d
eter
m
in
in
g
th
e
o
p
tim
al
ML
p
r
o
ce
s
s
es u
s
in
g
B
ay
es o
p
tim
izatio
n
.
I
t is wo
r
th
n
o
tin
g
th
at
t
h
er
e
wer
e
r
eso
u
r
ce
lim
itatio
n
s
f
o
r
th
e
lin
es th
is
s
tu
d
y
ex
am
in
ed
r
ef
er
to
T
ab
le
2
.
T
h
u
s
,
g
iv
en
m
o
r
e
tim
e
o
r
it
with
ex
p
er
im
en
ts
,
th
e
o
u
tco
m
es
m
ig
h
t
p
o
ten
tially
en
h
an
ce
in
ter
m
s
o
f
d
u
r
ab
ilit
y
v
er
s
u
s
th
e
o
v
er
f
itti
n
g
an
d
F1
s
co
r
e.
Po
ten
tial
to
p
ics
f
o
r
co
n
s
id
er
atio
n
will
in
clu
d
e
s
tr
ik
in
g
th
e
co
r
r
ec
t
b
alan
ce
b
etwe
en
Au
to
ML
,
m
an
u
al
ex
p
er
t
m
ac
h
in
e
lear
n
in
g
tu
n
in
g
,
an
d
th
e
b
est p
o
s
s
ib
le
p
er
f
o
r
m
an
ce
.
I
n
r
elatio
n
to
th
e
p
r
ev
io
u
s
in
q
u
ir
y
q
u
esti
o
n
,
it
is
wo
r
th
d
is
cu
s
s
in
g
if
th
e
p
r
ed
ictiv
e
m
o
d
elin
g
p
r
o
ce
s
s
es
co
n
s
tr
u
cted
o
n
a
b
ase
o
f
th
e
n
eu
r
al
-
b
ased
ex
tr
ac
tio
n
o
f
ch
ar
ac
ter
is
tics
s
h
o
wed
a
ce
r
tain
tr
en
d
.
B
a
s
ed
o
n
th
e
r
esu
lts
,
it
ap
p
ea
r
s
th
at
th
e
d
if
f
er
en
t
ap
p
r
o
ac
h
es
wer
e
tak
en
b
y
th
e
B
ay
es
o
p
tim
izatio
n
m
eth
o
d
,
lead
in
g
to
v
ar
io
u
s
ar
r
an
g
em
en
ts
f
o
r
p
r
ed
icto
r
tu
n
in
g
,
d
en
s
ity
r
ed
u
ctio
n
,
an
d
ch
o
ice
o
f
f
ea
tu
r
es.
T
h
e
s
u
g
g
ests
th
at
ev
en
m
in
o
r
v
ar
iatio
n
s
with
in
th
e
d
a
taset
m
ig
h
t
lead
to
s
ig
n
if
ican
tly
d
is
tin
ct
p
ip
elin
es.
T
h
e
class
if
ier
m
ig
h
t
b
e
an
ar
b
itra
r
y
f
o
r
est
o
r
a
d
ec
is
io
n
tr
ee
,
f
o
r
in
s
tan
ce
,
with
n
o
clea
r
ad
v
an
tag
e
f
o
r
eith
er
o
n
e.
T
h
is
p
h
en
o
m
en
o
n
is
clo
s
ely
co
n
n
ec
ted
to
th
e
th
eo
r
em
o
f
n
o
-
f
r
ee
m
ea
l,
wh
ich
is
esp
ec
i
ally
r
elev
an
t
in
th
e
co
n
tex
t
o
f
Au
to
ML
.
Fu
r
th
er
m
o
r
e,
o
v
er
f
itti
n
g
h
as
b
ee
n
r
ed
u
ce
d
with
th
e
u
s
e
o
f
tr
ee
g
r
o
u
p
s
,
th
e
tab
les
in
d
icate
th
at
s
o
m
e
o
v
er
f
itti
n
g
o
cc
u
r
r
ed
,
wh
ich
lim
its
th
e
s
ce
n
ar
io
s
wh
er
e
th
ese
s
y
s
tem
s
ca
n
b
e
s
af
ely
ap
p
lied
.
I
n
co
n
clu
s
io
n
,
Au
to
ML
a
d
d
s
an
ad
d
itio
n
al
lay
er
o
f
co
m
p
lex
ity
,
an
d
ac
h
iev
in
g
a
b
alan
ce
b
etwe
en
in
ter
p
r
etab
ilit
y
an
d
p
er
f
o
r
m
an
ce
is
cr
u
cial
an
d
d
ep
en
d
s
o
n
th
e
s
p
ec
if
ic
ap
p
licatio
n
an
d
it
is
ass
o
ciate
d
r
is
k
s
.
I
n
co
n
clu
s
io
n
,
Au
to
ML
h
as
th
e
p
o
ten
tial
to
ass
is
t
th
e
ag
r
o
tech
n
o
lo
g
y
co
m
m
u
n
ity
in
test
in
g
m
ac
h
in
e
-
lear
n
in
g
-
b
ased
Evaluation Warning : The document was created with Spire.PDF for Python.
C
o
m
p
u
t Sci
I
n
f
T
ec
h
n
o
l
I
SS
N:
2722
-
3
2
2
1
P
erfo
r
ma
n
ce
ev
a
lu
a
tio
n
o
f th
e
d
ee
p
lea
r
n
in
g
s
ystem
fo
r
w
ee
d
r
ec
o
g
n
iz
a
tio
n
(
A
b
d
A
b
r
a
h
im
Mo
s
s
la
h
a
)
175
s
o
lu
tio
n
s
with
r
ed
u
ce
d
im
p
lem
en
tatio
n
r
eso
u
r
ce
s
,
en
ab
lin
g
m
o
r
e
r
eso
u
r
ce
s
to
b
e
f
o
cu
s
ed
o
n
th
e
d
o
m
ain
-
s
p
ec
if
ic
p
ar
t
o
f
th
e
p
r
o
b
lem
.
Fu
r
th
er
m
o
r
e,
th
e
Au
to
ML
wo
r
k
f
lo
w
d
em
o
n
s
tr
ated
with
in
th
is
s
tu
d
y
p
o
s
s
ib
ly
r
ap
id
ly
g
en
e
r
ate
a
f
r
esh
m
o
d
el
u
s
in
g
f
r
esh
d
ata,
f
ac
ilit
atin
g
th
e
im
p
lem
en
tatio
n
o
f
ex
ce
llen
t
tech
n
o
lo
g
ies
in
r
esp
o
n
s
e
to
th
e
ev
er
-
ch
an
g
in
g
n
atu
r
e
o
f
ag
r
icu
ltu
r
e
[
18
]
,
[
2
4
]
as
s
h
o
wn
in
Fig
u
r
e
3
.
(
a)
(
b
)
(
c)
Fig
u
r
e
3
.
Dis
p
lay
s
th
e
s
tatis
tic
al
an
aly
s
is
r
esu
lts
o
f
th
r
ee
ex
p
er
im
en
tal
v
ar
iab
les:
(
a)
t
h
e
u
s
e
o
f
p
lan
t
s
eg
m
en
tatio
n
,
(
b
)
th
e
u
s
e
o
f
f
u
lly
-
co
n
n
ec
ted
n
etwo
r
k
s
,
an
d
(
c)
th
e
ty
p
e
o
f
class
if
ier
T
ab
le
6
s
h
o
ws
th
e
m
ea
n
±
s
tan
d
ar
d
d
ev
iatio
n
of
th
e
h
ig
h
est
p
er
f
o
r
m
in
g
b
est
au
to
m
l
co
n
f
i
g
u
r
atio
n
s
in
th
e
p
lan
t
s
ee
d
lin
g
s
d
ataset
.
T
ab
le
7
s
h
o
ws
th
e
h
ig
h
est
p
er
f
o
r
m
in
g
b
est
au
to
m
l
co
n
f
ig
u
r
atio
n
s
in
th
e
p
lan
t
s
ee
d
lin
g
s
d
ataset,
with
r
esu
lts
p
r
esen
ted
as
th
e
m
ea
n
v
al
u
e
pl
us
th
e
s
tan
d
ar
d
d
ev
iatio
n
.
T
ab
le
8
s
u
m
m
ar
izes
th
e
d
if
f
er
e
n
ce
s
b
etwe
en
th
e
Au
to
ML
ap
p
r
o
ac
h
es
u
s
ed
in
th
is
s
tu
d
y
an
d
r
ec
e
n
t
ad
v
a
n
ce
m
en
ts
in
au
to
m
ated
an
d
d
ee
p
lear
n
in
g
s
y
s
tem
s
.
T
ab
le
6
.
Hig
h
est
-
p
e
r
f
o
r
m
in
g
Au
to
ML
co
n
f
i
g
u
r
atio
n
s
f
o
r
th
e
p
lan
t seed
lin
g
s
d
ataset
u
n
d
e
r
n
o
is
y
im
ag
e
co
n
d
itio
n
s
PS
FC
C
l
a
s
si
f
i
e
r
F
1
S
c
o
r
e
O
v
e
r
f
i
t
t
i
n
g
S
a
l
t
F
1
B
l
u
r
F
1
Y
e
s
Y
e
s
S
o
f
t
ma
x
8
6
.
9
8
±
1
.
9
2
1
2
.
2
6
±
2
.
1
5
8
4
.
5
5
±
1
.
7
9
8
7
.
3
4
±
2
.
1
1
Y
e
s
No
En
se
mb
l
e
8
5
.
9
7
±
4
.
2
4
1
3
.
8
8
±
4
.
0
7
8
5
.
0
1
±
3
.
9
1
8
5
.
8
7
±
4
.
5
1
Y
e
s
No
S
i
n
g
l
e
8
5
.
2
9
±
5
.
2
9
1
4
.
4
±
4
.
9
8
4
.
2
8
±
4
.
5
1
8
4
.
8
1
±
5
.
0
3
No
No
En
se
mb
l
e
8
3
.
7
8
±
3
.
9
1
5
.
4
9
±
4
.
4
2
8
1
.
7
6
±
3
.
2
5
8
3
.
1
3
±
3
.
8
6
No
No
S
o
f
t
ma
x
8
0
.
4
5
±
4
.
1
3
1
7
.
6
1
±
6
.
9
4
7
8
.
7
2
±
3
.
2
2
8
0
.
0
9
±
4
N
o
t
e
:
ps
r
e
f
e
r
s
to
p
l
a
n
t
s
e
g
me
n
t
a
t
i
o
n
,
w
h
i
l
e
fc
st
a
n
d
s
f
o
r
f
u
l
l
y
-
c
o
n
n
e
c
t
e
d
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
7
2
2
-
3
2
2
1
C
o
m
p
u
t Sci
I
n
f
T
ec
h
n
o
l
,
Vo
l.
7
,
No
.
2
,
J
u
ly
20
26
:
1
67
-
1
78
176
T
ab
le
7
.
Hig
h
est
-
p
e
r
f
o
r
m
in
g
Au
to
ML
co
n
f
i
g
u
r
atio
n
s
f
o
r
th
e
ea
r
ly
cr
o
p
wee
d
s
d
ataset
u
n
d
er
n
o
is
y
im
ag
e
co
n
d
itio
n
s
PS
FC
C
l
a
s
si
f
i
e
r
F
1
S
c
o
r
e
O
v
e
r
f
i
t
t
i
n
g
S
a
l
t
F
1
B
l
u
r
F
1
Y
e
s
No
En
se
mb
l
e
9
0
.
7
4
±
0
.
8
8
.
5
1
±
1
.
2
5
7
2
.
8
9
±
7
.
5
2
7
2
.
0
6
±
1
7
.
8
3
Y
e
s
No
S
i
n
g
l
e
9
0
.
1
6
±
0
.
6
7
9
.
3
±
0
.
8
3
7
5
.
4
3
±
1
.
2
2
8
0
.
9
4
±
3
.
6
5
Y
e
s
Y
e
s
S
i
n
g
l
e
8
8
.
6
4
±
0
.
6
6
1
1
.
0
4
±
0
.
4
2
8
0
.
9
6
±
1
.
5
6
8
6
.
6
2
±
1
.
0
9
No
No
En
se
mb
l
e
8
8
.
6
3
±
1
.
2
6
8
.
1
6
±
0
.
3
6
0
.
9
4
±
1
7
.
3
3
8
3
.
5
7
±
0
.
8
3
Y
e
s
No
S
o
f
t
ma
x
8
7
.
1
7
±
2
.
6
3
6
.
3
7
±
1
.
0
2
6
7
.
7
4
±
6
.
1
8
7
1
.
5
9
±
1
7
.
2
4
N
o
t
e
:
ps
r
e
f
e
r
s
to
p
l
a
n
t
s
e
g
me
n
t
a
t
i
o
n
,
w
h
i
l
e
fc
st
a
n
d
s
f
o
r
f
u
l
l
y
-
c
o
n
n
e
c
t
e
d
.
T
ab
le
8
.
C
o
m
p
a
r
is
o
n
b
etwe
en
th
e
Au
to
ML
ap
p
r
o
ac
h
es u
s
ed
in
th
is
s
tu
d
y
an
d
m
o
d
e
r
n
s
tate
-
of
-
th
e
-
ar
t
tech
n
o
lo
g
ies
C
r
i
t
e
r
i
a
Te
c
h
n
o
l
o
g
i
e
s
u
se
d
i
n
t
h
e
st
u
d
y
M
o
d
e
r
n
s
t
a
t
e
-
of
-
t
h
e
-
a
r
t
t
e
c
h
n
o
l
o
g
i
e
s
(
2
0
2
3
–
2
0
2
4
)
M
o
d
e
l
t
y
p
e
A
u
t
o
M
L
(
A
u
t
o
K
e
r
a
s,
A
u
t
o
-
S
k
l
e
a
r
n
)
A
d
v
a
n
c
e
d
A
u
t
o
M
L+
l
a
r
g
e
l
a
n
g
u
a
g
e
m
o
d
e
l
s
(
LL
M
s)
+
h
y
b
r
i
d
d
e
e
p
l
e
a
r
n
i
n
g
S
p
e
e
d
a
n
d
p
e
r
f
o
r
ma
n
c
e
G
o
o
d
p
e
r
f
o
r
ma
n
c
e
(F1
-
sc
o
r
e
90
–
9
3
%
)
,
b
u
t
r
e
l
a
t
i
v
e
l
y
l
o
n
g
t
r
a
i
n
i
n
g
t
i
m
e
F
a
st
e
r
t
r
a
i
n
i
n
g
w
i
t
h
d
i
s
t
r
i
b
u
t
e
d
t
r
a
i
n
i
n
g
a
n
d
a
d
v
a
n
c
e
d
n
e
u
r
a
l
a
r
c
h
i
t
e
c
t
u
r
e
se
a
r
c
h
(NAS)
C
o
s
t
a
n
d
c
o
m
p
u
t
a
t
i
o
n
Lo
c
a
l
c
o
mp
u
t
a
t
i
o
n
(GPU
R
TX
2
0
8
0
T
i
)
A
d
v
a
n
c
e
d
c
l
o
u
d
c
o
m
p
u
t
i
n
g
(
e
.
g
.
,
G
o
o
g
l
e
TPU
,
A
W
S
I
n
f
e
r
e
n
t
i
a
)
w
i
t
h
o
p
t
i
m
i
z
e
d
e
n
e
r
g
y
c
o
n
su
m
p
t
i
o
n
Ea
se
of
u
se
R
e
l
a
t
i
v
e
l
y
e
a
s
y
f
o
r
d
e
v
e
l
o
p
e
r
s,
b
u
t
r
e
q
u
i
r
e
s
se
t
u
p
R
e
a
d
y
-
to
-
u
s
e
c
l
o
u
d
i
n
t
e
r
f
a
c
e
s
(
e
.
g
.
,
G
o
o
g
l
e
V
e
r
t
e
x
A
I
,
A
z
u
r
e
A
u
t
o
M
L)
w
i
t
h
No
-
C
o
d
e
/
L
o
w
-
C
o
d
e
s
u
p
p
o
r
t
S
c
a
l
a
b
i
l
i
t
y
Li
mi
t
e
d
by
l
o
c
a
l
h
a
r
d
w
a
r
e
a
n
d
d
a
t
a
s
i
z
e
H
o
r
i
z
o
n
t
a
l
l
y
a
n
d
v
e
r
t
i
c
a
l
l
y
sc
a
l
a
b
l
e
w
i
t
h
b
i
g
d
a
t
a
s
u
p
p
o
r
t
N
o
i
se
h
a
n
d
l
i
n
g
a
b
i
l
i
t
y
Te
st
e
d
w
i
t
h
n
o
i
s
y
d
a
t
a
(
sal
t
a
n
d
p
e
p
p
e
r
,
b
l
u
r
)
N
o
i
se
-
r
e
si
s
t
a
n
t
mo
d
e
l
s
u
si
n
g
c
o
n
t
r
a
s
t
i
v
e
l
e
a
r
n
i
n
g
a
n
d
a
d
v
a
n
c
e
d
d
a
t
a
a
u
g
me
n
t
a
t
i
o
n
I
n
t
e
r
p
r
e
t
a
b
i
l
i
t
y
Li
mi
t
e
d
,
e
s
p
e
c
i
a
l
l
y
w
i
t
h
e
n
s
e
mb
l
e
m
o
d
e
l
s
A
d
v
a
n
c
e
d
i
n
t
e
r
p
r
e
t
a
b
i
l
i
t
y
t
o
o
l
s:
S
H
A
P
,
LI
M
E,
e
x
p
l
a
i
n
a
b
l
e
AI
(
X
A
I
)
C
o
mm
u
n
i
t
y
su
p
p
o
r
t
a
n
d
u
p
d
a
t
e
s
O
p
e
n
-
so
u
r
c
e
c
o
mm
u
n
i
ty
(
A
u
t
o
K
e
r
a
s
,
S
c
i
k
i
t
-
l
e
a
r
n
)
La
r
g
e
c
o
mm
u
n
i
t
y
su
p
p
o
r
t
a
n
d
c
o
n
t
i
n
u
o
u
s
u
p
d
a
t
e
s
(
e
.
g
.
,
H
u
g
g
i
n
g
F
a
c
e
,
P
y
To
r
c
h
Li
g
h
t
n
i
n
g
)
A
d
v
a
n
c
e
d
a
g
r
i
c
u
l
t
u
r
a
l
a
p
p
l
i
c
a
t
i
o
n
s
W
e
e
d
a
n
d
p
l
a
n
t
r
e
c
o
g
n
i
t
i
o
n
I
n
t
e
g
r
a
t
e
d
s
y
st
e
ms:
S
mart
r
o
b
o
t
s,
d
r
o
n
e
-
b
a
se
d
a
e
r
i
a
l
sca
n
n
i
n
g
,
smar
t
i
r
r
i
g
a
t
i
o
n
s
y
st
e
ms
C
o
n
t
i
n
u
o
u
s
l
e
a
r
n
i
n
g
su
p
p
o
r
t
N
o
t
i
n
h
e
r
e
n
t
l
y
s
u
p
p
o
r
t
e
d
A
d
a
p
t
i
v
e
m
o
d
e
l
s
w
i
t
h
o
n
l
i
n
e
l
e
a
r
n
i
n
g
a
n
d
c
o
n
t
i
n
u
a
l
l
e
a
r
n
i
n
g
c
a
p
a
b
i
l
i
t
i
e
s
5.
CO
NCLU
SI
O
N
T
h
is
s
tu
d
y
ev
alu
ated
a
m
eth
o
d
o
lo
g
y
f
o
r
id
e
n
tify
in
g
wee
d
s
by
in
teg
r
atin
g
Au
to
ML
tech
n
o
lo
g
ies
an
d
b
en
ch
m
ar
k
ed
it
ag
ain
s
t
on
t
wo
s
ets
of
d
ata
with
f
o
u
r
a
n
d
th
ir
teen
class
if
icatio
n
s
of
wee
d
s
,
s
ee
d
lin
g
s
,
an
d
cr
o
p
s
,
r
esp
ec
tiv
ely
.
T
h
e
p
r
o
p
o
s
ed
m
eth
o
d
o
lo
g
y
ac
h
iev
ed
p
r
o
m
is
in
g
F1
s
co
r
es
of
90%
to
9
3
%
o
b
tain
ed
by
th
e
s
u
g
g
ested
m
eth
o
d
s
,
ac
c
o
r
d
in
g
to
th
e
i
n
f
o
r
m
atio
n
at
h
a
n
d
an
d
th
e
ex
is
ten
ce
of
n
o
is
e
o
b
s
er
v
atio
n
s
.
Fu
t
u
r
e
wo
r
k
will
ex
p
lo
r
e
m
o
r
e
co
s
tl
y
m
eth
o
d
s
,
lik
e
g
r
o
win
g
t
h
e
b
atch
s
ize
wh
ile
r
ec
eiv
in
g
in
s
tr
u
ctio
n
,
an
d
u
s
in
g
f
r
esh
d
atab
ases
,
lik
e
Dee
p
W
ee
d
s
,
to
g
ain
f
u
r
th
er
in
s
ig
h
ts
in
to
th
e
g
en
er
aliza
tio
n
ab
ilit
y
of
Au
to
ML
.
Ad
d
itio
n
ally
,
th
e
s
tu
d
y
will
ex
ten
d
ex
p
er
im
en
ts
to
test
th
e
r
o
b
u
s
tn
ess
of
th
e
s
y
s
tem
with
n
o
is
y
s
am
p
les,
in
clu
d
in
g
s
m
ea
r
in
g
n
o
is
e
r
elat
ed
to
v
eh
icle
m
o
v
em
en
t
an
d
e
v
alu
atin
g
on
test
s
ets
with
m
a
n
y
k
i
n
d
s
of
n
o
is
e.
S
in
ce
u
tili
s
in
g
d
ec
is
io
n
tr
ee
e
n
s
em
b
les
don
’
t
h
av
e
co
m
p
lete
ly
ev
a
d
ed
o
v
er
f
itti
n
g
,
th
e
s
tu
d
y
will
ex
p
lo
r
e
n
ew
m
ac
h
in
e
lear
n
in
g
p
ip
elin
es,
i
n
clu
d
in
g
r
aisi
n
g
th
e
en
s
em
b
l
e
’
s
d
ec
is
io
n
tr
ee
(
DT
)
co
u
n
t
or
u
s
in
g
a
Su
p
er
L
ea
r
n
er
m
o
d
el
s
tack
in
g
m
eth
o
d
.
T
h
e
s
tu
d
y
will
also
ev
alu
ate
class
if
ier
s
s
ep
ar
ately
an
d
co
n
s
tr
ain
th
e
B
ay
esian
ap
p
r
o
ac
h
to
a
m
o
r
e
lim
ited
s
u
b
s
et
of
im
p
r
o
v
e
th
e
in
ter
p
r
etab
ilit
y
of
th
e
o
u
tco
m
es.
T
h
ese
f
in
d
in
g
s
s
u
g
g
est
th
at
Au
to
ML
tech
n
o
lo
g
y
can
ai
d
th
e
ag
r
o
tech
n
o
lo
g
y
co
m
m
u
n
ity
by
p
r
o
v
id
in
g
h
ig
h
-
p
er
f
o
r
m
in
g
s
o
lu
tio
n
s
with
f
ewe
r
r
eso
u
r
c
es
r
eq
u
ir
e
d
f
o
r
im
p
lem
en
tati
o
n
,
f
ac
ilit
atin
g
th
e
cr
ea
tio
n
of
n
ew
m
o
d
els
f
o
r
d
y
n
am
ic
a
g
r
icu
ltu
r
al
e
n
v
ir
o
n
m
en
ts
.
ACK
NO
WL
E
DG
E
M
E
NT
T
h
e
au
th
o
r
s
w
o
u
l
d
lik
e
to
t
h
an
k
th
e
An
b
ar
of
Un
iv
er
s
it
y
an
d
Mu
s
tan
s
ir
iy
ah
Un
iv
er
s
ity
f
o
r
th
e
s
u
p
p
o
r
t
in
t
h
e
p
r
e
v
en
tio
n
wo
r
k
.
CO
NF
L
I
C
T
O
F
I
N
T
E
R
E
S
T
ST
A
T
E
M
E
NT
T
h
e
au
th
o
r
(
s
)
d
ec
lar
e
(
s
)
th
at
t
h
er
e
is
no
c
o
n
f
lict
of
in
ter
est
r
eg
ar
d
in
g
th
e
p
u
b
licatio
n
of
th
i
s
p
ap
er
.
Evaluation Warning : The document was created with Spire.PDF for Python.