I
A
E
S
I
n
t
e
r
n
at
io
n
al
of
A
r
t
if
ic
ia
l
I
n
t
e
ll
ig
e
n
c
e
(
I
J
-
AI
)
V
ol
.
14
, N
o.
3
,
J
une
2025
, pp.
1884
~
1899
I
S
S
N
:
2252
-
8938
,
D
O
I
:
10.11591/
ij
a
i.
v
14
.i
3
.pp
1884
-
1899
1884
Jou
r
n
al
h
om
e
page
:
ht
tp
:
//
ij
ai
.
ia
e
s
c
or
e
.c
om
C
om
p
ar
i
son
of
d
e
e
p
l
e
ar
n
i
n
g
m
od
e
l
s
:
C
N
N
an
d
V
G
G
-
16
i
n
i
d
e
n
t
i
f
yi
n
g p
or
n
ogr
ap
h
i
c
c
on
t
e
n
t
R
e
z
a C
h
an
d
r
a
1
, A
d
an
g S
u
h
e
n
d
r
a
2
, L
in
t
an
g Y
u
n
ia
r
B
an
ow
o
s
ar
i
2
, P
r
ih
an
d
ok
o
3
1
D
e
pa
r
t
m
e
nt
of
I
nf
or
m
a
t
i
c
s
M
a
na
ge
m
e
nt
, G
una
da
r
m
a
U
ni
ve
r
s
i
t
y, D
e
pok C
i
t
y,
I
ndone
s
i
a
2
D
e
pa
r
t
m
e
nt
of
I
nf
or
m
a
t
i
c
s
, G
una
da
r
m
a
U
ni
ve
r
s
i
t
y, D
e
pok C
i
t
y, I
ndone
s
i
a
3
D
e
pa
r
t
m
e
nt
of
I
nf
or
m
a
t
i
on
S
ys
t
e
m
s
, G
una
da
r
m
a
U
ni
ve
r
s
i
t
y, D
e
pok C
i
t
y, I
ndo
ne
s
i
a
A
r
t
ic
le
I
n
f
o
A
B
S
T
R
A
C
T
A
r
ti
c
le
h
is
to
r
y
:
R
e
c
e
iv
e
d
A
pr
18
,
2024
R
e
vi
s
e
d
F
e
b
12
,
2025
A
c
c
e
pt
e
d
M
a
r
15
,
2025
In
2020,
a
total
of
59,741
websites
were
blocked
by
the
Indo
nesian
government
due
to
containing
negative
content,
including
pornograph
y,
with
14,266
websites
falling
into
this
category.
However,
these
blocked
w
ebsites
could
still
be
accessed
by
the
public
using
virtual
private
networks
(
VPNs).
This
prompted
the
research
idea
to
quickly
identify
pornographic
c
ontent.
This
study
aims
to
develop
a
system
capable
of
identifying
websites
sus
pected
of
containing
pornographic
image
content,
using
a
deep
learning
a
p
proach
with
convolut
ional
neural
network
(
CNN
)
and
visual
geometry
gro
up
16
(
VGG
-
16
)
model.
The
two
models
were
then
explored
comprehensively
and
holistically
to
determine
which
model
was
most
effective
in
de
tecting
pornographic
content
quickly.
Based
on
the
findings
of
the
com
parison
between
testing
the
CNN
and
VGG
-
16
models,
research
re
sults
show
ed
that
the
best
test
results
were
obtained
in
the
eighth
experiment
using
the
CNN
model
at
an
epoch
value
level
of
50
and
a
learning
rate
of
0.001
of
0.9
487
or
94.87%.
This
can
be
interpreted
that
the
CNN
model
is
more
effective
in
detecting pornographic content quickly and a
ccurately compare
d to usi
ng the
VGG
-
16 model.
K
e
y
w
o
r
d
s
:
C
N
N
m
ode
ls
D
e
e
p l
e
a
r
ni
ng
P
or
nogr
a
phi
c
i
m
a
ge
c
ont
e
nt
VGG
-
16 mode
l
V
ir
tu
a
l
pr
iv
a
te
ne
twor
ks
This is an
open
acce
ss artic
le unde
r the
CC BY
-
SA
license.
C
or
r
e
s
pon
di
n
g A
u
th
or
:
R
e
z
a
C
h
a
ndr
a
D
e
pa
r
tm
e
nt
of
I
nf
or
m
a
ti
c
s
M
a
na
ge
m
e
nt
, G
una
da
r
m
a
U
ni
ve
r
s
it
y
M
a
r
gonda
R
a
ya
100
R
oa
d, D
e
pok
C
it
y 16424, I
ndone
s
ia
E
m
a
il
:
r
e
z
a
_c
ha
n@
s
ta
f
f
.guna
da
r
m
a
.a
c
.i
d
1.
I
N
T
R
O
D
U
C
T
I
O
N
C
ur
r
e
nt
ly
,
m
or
e
a
nd
m
or
e
I
ndone
s
ia
n
pe
opl
e
a
r
e
a
c
qui
r
in
g
c
u
lt
ur
a
l
va
lu
e
s
a
nd
nor
m
s
th
a
t
a
r
e
not
r
e
le
va
nt
to
m
or
a
l
va
lu
e
s
.
I
t
is
s
ti
ll
c
om
m
on
to
f
in
d
te
le
vi
s
io
n
s
h
ow
s
,
in
c
lu
di
ng
f
il
m
m
a
te
r
ia
l,
w
hi
c
h
of
te
n
s
how
por
nogr
a
phi
c
s
c
e
ne
s
th
a
t
a
r
e
not
in
a
c
c
or
da
nc
e
w
it
h
th
e
va
lu
e
s
a
nd
f
ir
s
t
pr
in
c
ip
le
s
of
th
e
s
ta
te
pr
in
c
ip
le
s
of
P
a
nc
a
s
il
a
.
I
n
th
e
m
id
s
t
of
c
ur
r
e
nt
a
dva
nc
e
s
in
di
gi
ta
li
z
a
ti
on,
th
e
c
ul
tu
r
e
of
m
e
di
a
pl
a
tf
or
m
s
m
us
t
a
da
pt
to
s
oc
ie
ty
'
s
ne
e
ds
.
T
e
le
vi
s
io
n
a
nd
f
il
m
m
e
di
a
,
in
c
lu
di
ng
s
m
a
r
tp
h
one
pr
ovi
de
r
s
a
nd
de
ve
lo
pe
r
s
,
ha
ve
a
s
tr
a
te
gi
c
r
ol
e
i
n di
s
s
e
m
in
a
ti
ng i
nf
or
m
a
ti
on.
P
r
obl
e
m
s
a
r
is
e
w
he
n
pr
ovi
de
r
s
of
te
le
vi
s
io
n
br
oa
dc
a
s
ts
,
f
il
m
s
,
a
nd
s
oc
ia
l
m
e
di
a
gr
ow
f
r
e
e
ly
w
it
hout
a
ny
in
te
r
ve
nt
io
n
f
r
om
th
e
gove
r
nm
e
nt
.
B
y
pr
io
r
it
iz
in
g
th
e
pr
in
c
ip
le
of
f
r
e
e
dom
of
e
xpr
e
s
s
io
n,
it
is
f
e
a
r
e
d
th
a
t
s
oc
ie
ty
c
oul
d
be
di
vi
de
d
in
to
publ
ic
opi
ni
on
a
s
pa
r
t
of
a
f
or
m
of
a
tt
it
ude
in
th
e
f
or
m
of
f
ig
ht
in
g
a
ga
in
s
t
one
a
not
he
r
a
nd
th
is
c
oul
d
e
a
s
il
y
le
a
d
to
hor
iz
ont
a
l
e
s
c
a
la
ti
on
[
1]
.
T
he
I
ndone
s
ia
n
gove
r
nm
e
nt
,
th
r
ough
th
e
le
a
de
r
s
hi
p
of
P
r
e
s
id
e
nt
J
oko
W
id
odo,
c
ont
in
ue
s
to
m
a
ke
e
f
f
or
ts
w
it
h
a
ll
it
s
pol
ic
y
in
s
tr
um
e
nt
s
to
bl
oc
k
va
r
io
us
w
e
bs
it
e
s
de
e
m
e
d
a
nd
s
u
s
pe
c
te
d
to
ha
v
e
ne
ga
ti
ve
a
nd
pr
ovoc
a
ti
v
e
c
ont
e
nt
th
a
t
c
a
n
pr
ot
e
c
t
th
e
pos
it
iv
e
th
oughts
of
th
e
publ
ic
,
to
c
r
e
a
te
a
s
a
f
e
,
po
s
it
iv
e
a
nd
hi
gh
-
qua
li
ty
in
te
r
ne
t
a
tm
os
phe
r
e
,
one
of
w
hi
c
h
is
por
nogr
a
phi
c
c
ont
e
nt
[
2]
, [
3]
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
ti
f
I
nt
e
ll
I
S
S
N
:
2252
-
8938
C
om
par
is
on of de
e
p l
e
ar
ni
ng m
ode
l
s
:
C
N
N
and V
G
G
-
16 i
n i
de
nt
if
y
in
g por
nogr
aphi
c
…
(
R
e
z
a C
hand
r
a)
1885
W
e
b us
a
g
e
i
s
gr
o
w
in
g
i
n a
w
a
y t
h
a
t
i
s
n
ot
w
e
ll
c
on
tr
ol
le
d. A
n
a
ppr
o
pr
i
a
t
e
f
r
a
m
e
w
or
k
m
o
de
l
i
s
n
e
e
de
d
to
li
m
it
t
h
e
v
a
s
t
a
m
ou
nt
of
in
f
or
m
a
ti
on
in
c
yb
e
r
s
p
a
c
e
.
W
it
h
th
e
a
d
ve
n
t
of
th
e
in
te
r
ne
t,
p
or
n
ogr
a
p
hy
be
c
a
m
e
c
h
e
a
p
,
m
a
n
y por
n s
i
te
s
c
ou
ld
b
e
e
a
s
il
y
b
ui
l
t
a
n
d
a
c
c
e
s
s
e
d.
P
or
no
gr
a
ph
y i
s
s
ti
l
l
c
o
n
s
id
e
r
e
d
il
le
ga
l
a
n
d a
g
a
i
n
s
t
th
e
l
a
w
in
I
ndo
ne
s
i
a
,
t
hu
s
e
n
c
o
ur
a
gi
ng
th
e
g
ov
e
r
n
m
e
nt
'
s
a
c
ti
ve
i
n
vo
lv
e
m
e
nt
i
n
d
e
v
e
lo
pi
n
g r
e
g
ul
a
ti
on
s
to
li
m
i
t
i
t
[
4]
–
[
6]
.
T
he
r
e
s
e
a
r
c
he
r
'
s
obs
e
r
va
ti
ons
a
r
e
th
a
t
unt
il
now
th
e
I
ndone
s
ia
n
gove
r
nm
e
nt
c
a
n
onl
y
c
a
r
r
y
out
bl
oc
ki
ng
a
c
ti
ons
ba
s
e
d
on
w
e
b
a
ddr
e
s
s
e
s
a
nd
not
ba
s
e
d
on
im
a
ge
c
ont
e
nt
.
B
a
s
e
d
on
th
e
w
e
b
pa
ge
ht
tp
s
:/
/a
dua
nkonte
n.i
d
ow
ne
d
by
T
he
M
in
is
tr
y
of
C
om
m
uni
c
a
ti
on
a
nd
I
nf
or
m
a
ti
on
(
K
om
in
f
o)
,
a
s
of
M
a
r
c
h
2020,
th
e
r
e
w
e
r
e
1,246,801
publ
ic
c
om
pl
a
in
ts
r
e
ga
r
di
ng
ne
ga
ti
ve
c
ont
e
nt
r
e
por
te
d
vi
a
th
is
pa
ge
.
W
e
b
m
oni
to
r
in
g
is
c
a
r
r
ie
d
ou
t
ba
s
e
d
on
21
c
a
te
gor
ie
s
:
c
a
te
gor
ie
s
of
hoa
xe
s
,
s
la
nde
r
,
in
te
ll
e
c
tu
a
l
pr
ope
r
ty
r
ig
ht
s
(
I
P
R
)
, vi
ol
e
n
c
e
/v
io
le
nc
e
a
ga
in
s
t
c
hi
ld
r
e
n, c
ont
e
nt
t
ha
t
vi
ol
a
te
s
s
oc
ia
l
a
nd c
ul
tu
r
a
l
va
lu
e
s
, i
nf
or
m
a
ti
on s
e
c
ur
it
y
vi
ol
a
ti
ons
,
e
xt
or
ti
on,
f
r
a
ud,
tr
a
de
in
pr
oduc
ts
w
it
h
s
pe
c
ia
l
r
e
gul
a
ti
ons
,
ga
m
bl
in
g,
por
nogr
a
phy,
e
th
ni
c
it
y,
r
e
li
gi
on,
r
a
c
e
a
nd
c
la
s
s
(
S
A
R
A
)
,
s
e
p
a
r
a
ti
s
m
,
a
nd
te
r
r
or
is
m
.
P
or
nogr
a
phy
w
e
bs
it
e
s
r
e
c
e
iv
e
d
th
e
m
os
t
c
om
pl
a
in
ts
f
r
om
t
he
publi
c
w
it
h 1,037,321 c
om
pl
a
in
ts
a
s
s
how
n i
n
F
ig
ur
e
1.
F
ig
ur
e
1. S
ta
ti
s
ti
c
s
on
publ
ic
c
om
pl
a
in
ts
r
e
ga
r
di
ng ne
ga
ti
ve
c
on
te
nt
G
iv
e
n
th
e
in
c
r
e
a
s
in
g
e
xpos
ur
e
to
por
nog
r
a
phy
in
I
ndone
s
ia
,
th
e
pot
e
nt
ia
l
ne
ga
ti
ve
ps
yc
hol
ogi
c
a
l
im
pa
c
ts
of
e
xpos
ur
e
t
o por
nogr
a
phy (
e
s
p
e
c
ia
ll
y t
hos
e
r
e
la
te
d t
o
pot
e
nt
ia
l
pr
obl
e
m
s
or
a
ddi
c
ti
on)
ha
ve
be
c
om
e
a
n
in
c
r
e
a
s
in
g
e
m
pi
r
ic
a
l
c
onc
e
r
n
in
r
e
c
e
nt
ye
a
r
s
[
7]
.
H
u
e
t
al
.
[
8]
c
a
r
r
ie
d
out
por
nogr
a
phi
c
im
a
ge
r
e
c
ogni
ti
on
w
he
r
e
t
he
c
ur
r
e
nt
m
e
th
ods
c
a
n be
c
la
s
s
if
ie
d a
s
m
ode
l,
f
e
a
tu
r
e
,
a
nd r
e
gi
on ba
s
e
d. T
he
y us
e
d s
ki
n de
te
c
ti
on f
or
por
nogr
a
phi
c
im
a
ge
r
e
c
ogni
ti
on.
T
he
w
e
a
kne
s
s
in
th
e
r
e
s
e
a
r
c
h,
pr
e
pr
oc
e
s
s
in
g
is
r
e
qui
r
e
d
in
c
la
s
s
if
yi
ng
im
a
ge
s
s
uc
h a
s
s
ki
n c
ol
or
e
xt
r
a
c
ti
on
a
nd t
he
pr
obl
e
m
s
of
ov
e
r
bl
oc
ki
ng
, m
is
s
pe
ll
in
gs
a
nd w
or
d l
is
ts
h
a
ve
not
ye
t
b
e
e
n
r
e
s
ol
ve
d.
c
onvolut
io
na
l
ne
ur
a
l
ne
twor
k
(
C
N
N
)
i
s
th
e
m
ode
l
m
o
s
t
of
te
n
us
e
d
in
im
a
ge
pr
oc
e
s
s
in
g
a
nd
c
om
put
e
r
vi
s
io
n
[
9]
.
T
he
us
in
g
C
N
N
m
ode
l
s
to
le
a
r
n
hi
gh
-
le
ve
l
f
e
a
tu
r
e
s
in
im
a
ge
da
ta
th
r
ough
c
onvolut
io
n.
C
N
N
is
ve
r
y
s
ui
ta
bl
e
f
or
obj
e
c
t
r
e
c
ogni
ti
on
w
it
h
im
a
ge
s
.
T
he
s
upe
r
io
r
it
y
of
C
N
N
in
im
a
ge
r
e
c
ogni
ti
on
is
th
e
m
a
in
a
r
gum
e
nt
a
s
w
e
ll
a
s
th
e
w
or
ld
'
s
r
e
c
ogni
ti
on
of
th
e
s
upe
r
io
r
it
y
of
de
e
p
le
a
r
ni
ng
in
s
tu
dyi
ng
la
r
ge
a
m
ount
s
of
da
ta
a
nd
it
s
a
bi
li
ty
i
n
r
e
m
ot
e
s
e
ns
in
g t
o a
c
hi
e
ve
a
dva
nc
e
d c
la
s
s
if
ic
a
ti
on a
c
c
ur
a
c
y
[
10]
–
[
12]
.
E
ve
n t
hough
t
hi
s
m
ode
l
is
ve
r
y good a
t
c
la
s
s
if
yi
ng c
e
r
ta
in
i
m
a
g
e
s
,
t
he
C
N
N
m
ode
l
i
s
i
n f
a
c
t
s
ti
ll
not
opt
im
a
ll
y
us
e
d
a
s
a
f
il
te
r
f
or
c
ont
e
nt
th
a
t
is
in
di
c
a
te
d
to
ha
ve
po
r
nogr
a
phi
c
e
le
m
e
nt
s
by
m
a
ny
r
e
s
e
a
r
c
he
r
s
,
m
os
t
of
w
hos
e
s
ol
ut
io
ns
a
r
e
s
ti
ll
c
lo
ud
-
ba
s
e
d
[
13]
–
[
15]
.
T
he
r
e
s
e
a
r
c
he
r
'
s
pe
r
s
pe
c
ti
ve
u
s
e
s
th
e
C
N
N
m
ode
l
f
or
m
or
e
a
c
c
ur
a
te
r
e
c
ogni
ti
on
of
por
nogr
a
phi
c
im
a
ge
c
ont
e
nt
.
T
he
C
N
N
m
ode
l
is
bui
lt
w
it
h
th
r
e
e
c
onvolut
io
n
l
a
ye
r
s
.
T
he
a
m
ount
of
da
ta
us
e
d
is
14,085
im
a
ge
da
ta
.
R
e
s
e
a
r
c
he
r
s
a
l
s
o
us
e
th
e
vi
s
ua
l
ge
om
e
tr
y
gr
oup
16
(
V
G
G
-
16
)
m
ode
l
a
s
a
c
om
pa
r
is
on
m
ode
l,
th
is
m
ode
l
is
a
de
ve
lo
pm
e
nt
of
th
e
C
N
N
m
ode
l
w
hi
c
h
ha
s
16
c
onvolut
io
n
l
a
ye
r
s
.
A
ga
s
ty
a
e
t
al
.
[
13]
us
e
d
th
e
C
N
N
m
ode
l
f
or
c
la
s
s
if
yi
ng
por
nogr
a
phi
c
im
a
ge
s
a
nd
de
v
e
lo
pe
d
th
e
id
e
a
in
t
hi
s
r
e
s
e
a
r
c
h.
R
e
s
e
a
r
c
he
r
s
pr
opos
e
a
s
tr
a
te
gy
f
or
im
pl
e
m
e
nt
in
g w
e
b i
m
a
ge
c
ont
e
nt
i
de
nt
if
ic
a
ti
on u
s
in
g a
d
e
e
p l
e
a
r
ni
ng a
ppr
oa
c
h (
C
N
N
a
nd
V
G
G
-
16)
a
s
a
f
il
te
r
f
or
w
e
bs
it
e
s
th
a
t
c
ont
a
in
por
nogr
a
phi
c
im
a
ge
c
ont
e
nt
.
T
he
nove
lt
y
bui
lt
in
th
is
r
e
s
e
a
r
c
h
is
a
de
e
p
le
a
r
ni
ng
a
ppr
oa
c
h.
T
he
a
dva
nt
a
ge
obt
a
in
e
d
in
th
is
r
e
s
e
a
r
c
h
is
th
a
t
it
c
a
n
id
e
nt
if
y
pr
ovoc
a
ti
ve
por
nogr
a
phi
c
im
a
ge
c
ont
e
nt
in
c
ybe
r
s
p
a
c
e
qui
c
kl
y,
he
lp
in
g
th
e
gov
e
r
nm
e
nt
pr
ovi
de
s
a
f
e
a
nd
qua
li
ty
in
te
r
ne
t
pr
ot
e
c
ti
on
f
or
th
e
publ
ic
in
a
c
c
e
s
s
in
g
w
e
bs
it
e
s
th
a
t
a
r
e
de
te
c
t
e
d
to
c
ont
a
in
por
nogr
a
phi
c
c
ont
e
nt
,
s
o
th
is
r
e
s
e
a
r
c
h
doe
s
not
r
e
qui
r
e
pr
e
pr
oc
e
s
s
in
g i
n c
la
s
s
if
yi
ng i
m
a
ge
da
ta
.
5
16
3151
10
26
667
44
0
9458
126
194170
1037321
187
3
498
0
200000
400000
600000
800000
1000000
1200000
H
oa
x
S
l
a
nde
r
HKI
C
h
i
l
d
a
b
u
s
e
/
v
i
o
l
e
n
c
e
C
ont
e
nt
t
ha
t
vi
ol
a
t
e
s
s
oc
i
a
l
a
nd c
ul
t
ur
a
l
va
l
ue
s
N
or
m
a
l
i
z
a
t
i
on
I
nf
or
m
a
t
i
on s
e
c
ur
i
t
y vi
ol
a
t
i
ons
E
xt
or
t
i
on
F
r
a
ud
T
r
a
de
pr
oduc
t
s
w
i
t
h s
pe
c
i
a
l
r
ul
e
s
G
a
m
bl
i
ng
P
or
nogr
a
phy
S
A
R
A
S
e
pa
r
a
t
i
s
m
/
da
nge
r
ous
or
ga
ni
z
a
t
i
ons
T
e
r
r
or
i
s
m
/
r
a
di
c
a
l
i
s
m
N
um
be
r
of
publ
i
c
c
om
pl
a
i
nt
s
C
a
t
e
g
or
y
of
ne
g
a
t
i
v
e
c
ont
e
nt
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
, V
ol
.
14
, N
o.
3
,
J
une
20
25
:
1884
-
1899
1886
2.
T
H
E
C
O
M
P
R
E
H
E
N
S
I
V
E
T
H
E
O
R
E
T
I
C
A
L
B
A
S
I
S
2.1. De
e
p
l
e
ar
n
in
g b
as
e
d
on
ar
t
if
ic
ia
l
n
e
u
r
al
n
e
t
w
or
k
s
T
hr
ough
it
e
r
a
ti
ve
pr
oc
e
s
s
in
g
a
nd
tr
a
in
in
g
on
la
r
ge
da
ta
s
e
ts
,
d
e
e
p
le
a
r
ni
ng
m
ode
ls
c
a
n
a
ut
om
a
ti
c
a
ll
y
e
xt
r
a
c
t
a
nd
tr
a
ns
f
or
m
f
e
a
tu
r
e
s
,
m
a
ki
ng
th
e
m
hi
ghl
y
e
f
f
e
c
ti
ve
f
or
c
om
pl
e
x
ta
s
ks
s
uc
h
a
s
im
a
ge
r
e
c
ogni
ti
on,
na
tu
r
a
l
la
ngua
ge
pr
oc
e
s
s
in
g
a
nd
s
pe
e
c
h
a
na
ly
s
i
s
.
D
e
e
p
le
a
r
ni
ng
is
th
e
m
ul
ti
-
la
ye
r
a
r
ti
f
ic
ia
l
ne
ur
a
l
ne
twor
ks
th
a
t
c
a
n
le
a
r
n
th
e
c
om
pl
e
x
non
-
li
ne
a
r
f
unc
ti
ons
.
D
e
e
p
le
a
r
ni
ng
r
e
f
e
r
s
to
da
ta
-
dr
iv
e
n
le
a
r
ni
ng
th
a
t
us
e
s
m
ul
ti
-
la
ye
r
ne
ur
a
l
ne
twor
ks
a
nd
hi
e
r
a
r
c
hi
c
a
l
pr
oc
e
s
s
in
g
to
c
om
p
ut
e
c
om
pl
e
x
pa
tt
e
r
ns
a
nd
r
e
pr
e
s
e
nt
a
ti
ons
f
r
om
la
r
ge
a
m
ount
s
of
d
a
ta
[
16]
.
T
hi
s
th
e
or
y
is
a
f
or
m
of
m
a
c
hi
n
e
le
a
r
ni
ng
pr
oc
e
s
s
th
a
t
is
ba
s
e
d
on
a
n
a
r
ti
f
ic
ia
l
ne
ur
a
l
ne
twor
k
w
it
h
m
a
ny
hi
dde
n
la
ye
r
s
a
nd
it
s
de
pt
h
in
a
na
ly
s
in
g
r
e
pr
e
s
e
nt
a
ti
ons
or
da
ta
f
e
a
tu
r
e
s
a
ut
om
a
ti
c
a
ll
y
w
hi
c
h
is
in
c
lu
de
d
in
th
e
c
a
te
gor
y
o
f
a
r
ti
f
ic
ia
l
in
t
e
ll
ig
e
nc
e
[
17]
,
[
18]
.
T
hi
s
th
e
or
y
is
a
m
a
c
hi
ne
le
a
r
ni
ng
pr
oc
e
s
s
w
it
h
th
e
c
ha
r
a
c
te
r
is
ti
c
s
of
us
in
g
s
e
ve
r
a
l
l
a
ye
r
s
of
hi
e
r
a
r
c
hi
c
a
l
non
-
li
ne
a
r
in
f
or
m
a
ti
o
n
pr
oc
e
s
s
in
g,
th
is
is
de
te
r
m
in
e
d
by
how
th
e
in
te
nde
d
a
r
c
hi
te
c
tu
r
e
a
nd
te
c
hni
que
s
a
r
e
us
e
d,
f
or
e
xa
m
pl
e
s
ynt
he
s
is
or
c
la
s
s
if
ic
a
ti
on.
D
e
e
p
le
a
r
ni
ng
is
di
vi
de
d
in
to
3
m
a
in
c
la
s
s
c
a
te
gor
ie
s
[
19]
:
i
)
uns
upe
r
vi
s
e
d
or
ge
ne
r
a
ti
ve
de
e
p
ne
twor
k
le
a
r
ni
ng.
I
n
th
e
la
te
s
t
c
a
s
e
,
th
e
us
e
of
B
a
ye
s
'
r
ul
e
s
c
a
n
c
ha
nge
th
e
c
ha
r
a
c
te
r
is
ti
c
s
of
a
ge
ne
r
a
ti
ve
ne
twor
k
in
to
a
di
s
c
r
im
in
a
ti
ve
ne
twor
k
f
or
le
a
r
ni
ng;
ii
)
de
e
p
ne
twor
k
s
upe
r
vi
s
e
d
le
a
r
ni
ng
pr
ovi
de
s
di
s
c
r
im
in
a
ti
ve
c
a
pa
bi
li
ti
e
s
f
or
pa
tt
e
r
ns
by
c
lu
s
te
r
in
g
th
e
po
s
te
r
io
r
di
s
tr
ib
ut
io
n
of
gr
oups
c
ondi
ti
one
d
on
th
e
vi
s
ib
le
pur
pos
e
s
da
ta
;
a
nd
iii
)
hybr
id
de
e
p
ne
twor
ks
(
a
s
s
is
te
d
di
s
c
r
im
in
a
ti
on)
,
of
te
n
in
a
s
ig
ni
f
ic
a
nt
w
a
y
a
nd
th
e
r
e
s
ul
t
of
ge
ne
r
a
ti
ve
de
e
p
ne
twor
ks
or
uns
u
pe
r
vi
s
e
d
l
e
a
r
ni
ng.
M
a
xi
m
iz
in
g
th
e
f
or
m
of
ne
twor
k
r
e
gul
a
r
iz
a
ti
on
in
th
e
s
upe
r
vi
s
e
d
l
e
a
r
ni
ng
c
a
te
gor
y.
G
oodf
e
l
lo
w
e
t
al
.
[
20]
a
r
gue
d
de
e
p
l
e
a
r
ni
ng
us
e
s
hi
e
r
a
r
c
hi
c
a
l
c
onc
e
pt
s
t
o
s
ol
ve
pr
obl
e
m
s
i
n c
om
put
e
r
l
e
a
r
ni
ng s
y
s
te
m
s
.
2.2.
I
m
age
p
r
o
c
e
s
s
in
g b
as
e
d
on
c
on
vol
u
t
io
n
al
n
e
u
r
al
n
e
t
w
o
r
k
m
od
e
l
T
he
C
N
N
m
ode
l
is
th
e
m
os
t
f
r
e
que
nt
ly
us
e
d
m
ode
l
f
or
im
a
ge
pr
oc
e
s
s
in
g
a
nd
c
om
put
e
r
vi
s
io
n
[
21]
.
T
hi
s
m
ode
l
is
de
s
ig
ne
d
in
s
uc
h
a
w
a
y
a
s
to
m
im
ic
th
e
s
tr
uc
tu
r
e
of
th
e
vi
s
ua
l
c
or
te
x.
S
pe
c
if
ic
a
ll
y,
a
C
N
N
ha
s
th
e
f
or
m
of
ne
ur
ons
in
a
pa
r
ti
c
ul
a
r
la
ye
r
onl
y
c
onne
c
ti
ng
to
a
s
m
a
ll
a
r
e
a
of
th
e
pr
e
vi
ous
la
y
e
r
.
T
hi
s
m
ode
l
i
s
m
os
t
of
te
n
us
e
d
f
or
im
a
ge
pr
oc
e
s
s
in
g
a
nd
c
om
put
e
r
vi
s
io
n. T
hi
s
m
ode
l
in
c
lu
de
d
in
th
e
f
or
m
of
a
de
e
p
ne
ur
a
l
ne
twor
k, t
hi
s
i
s
be
c
a
us
e
of
i
ts
c
h
a
r
a
c
te
r
is
ti
c
s
a
nd high
ne
twor
k
de
pt
h a
nd i
ts
u
s
e
i
s
of
te
n a
ppl
ie
d t
o i
m
a
g
e
da
ta
th
a
t
is
be
in
g
obs
e
r
ve
d
[
22]
.
T
he
c
la
s
s
if
ic
a
ti
on
r
e
s
ul
ts
a
r
e
pr
oduc
e
d
f
r
om
th
e
out
put
la
ye
r
w
he
r
e
e
a
c
h
pa
r
t
of
th
e
hype
r
s
pe
c
tr
a
l
im
a
ge
is
di
r
e
c
te
d
to
a
m
ode
l
th
a
t
in
di
c
a
te
s
b
a
c
k
a
nd
f
or
th
pr
opa
ga
ti
on
.
H
yp
e
r
s
pe
c
tr
a
l
im
a
g
e
c
la
s
s
if
ic
a
ti
on
is
a
c
r
uc
ia
l
s
te
p
in
hype
r
s
pe
c
tr
a
l
da
ta
pr
oc
e
s
s
in
g
[
23]
.
H
ig
hl
y
c
ons
is
te
nt
r
e
s
ul
ts
by
c
om
bi
ni
ng
f
in
ge
r
ve
in
s
a
nd f
a
c
ia
l
f
e
a
tu
r
e
t
hr
ough a
n
i
te
r
a
ti
ve
pr
e
m
iu
m
l
oc
a
l
e
xt
r
a
c
ti
on pr
oc
e
s
s
w
it
h t
hi
s
m
ode
l
[
24]
.
T
he
r
e
s
ul
ts
of
th
e
la
te
s
t
r
e
s
e
a
r
c
h
r
e
ga
r
di
ng
th
e
us
e
of
th
is
m
ode
l
in
a
bi
om
e
tr
ic
f
r
a
m
e
w
or
k
th
r
ough
c
om
bi
ni
ng
f
in
ge
r
ve
in
s
a
nd f
a
c
ia
l
f
e
a
tu
r
e
s
a
s
w
e
ll
a
s
bi
m
oda
l
la
ye
r
f
us
io
n a
nd uti
li
z
in
g t
he
A
le
xN
e
t
a
s
w
e
ll
a
s
V
G
G
-
16
m
ode
ls
pr
oduc
e
hi
gh
e
f
f
ic
a
c
y
in
d
if
f
e
r
e
nt
ia
ti
ng
f
in
ge
r
ve
in
s
im
a
ge
s
f
r
om
f
a
c
ia
l
f
e
a
tu
r
e
s
.
T
he
r
e
s
ul
ts
of
bot
h
m
ode
ls
ha
ve
a
ut
he
nt
ic
a
ti
on a
c
c
ur
a
c
y r
e
a
c
hi
ng 98.4%
[
25]
, [
26]
.
2.3.
C
la
s
s
if
yi
n
g
i
m
age
s
w
it
h
t
h
e
V
G
G
-
16
m
od
e
l
W
e
a
ls
o
u
s
e
th
e
V
G
G
-
16
m
ode
l
to
r
e
pr
e
s
e
nt
th
e
C
N
N
s
tr
uc
tu
r
e
us
in
g
13
c
onvolut
io
n
la
ye
r
s
,
5 m
a
x
pool
s
,
1 f
la
tt
e
n l
a
ye
r
,
a
nd 3 d
e
ns
e
l
a
ye
r
s
.
T
hi
s
m
ode
l
ut
il
i
z
e
s
a
c
om
pa
c
t
3
×
3
c
onvolut
io
na
l
f
il
te
r
a
s
w
e
ll
a
s
a
de
e
p
a
r
c
hi
te
c
tu
r
e
us
in
g
s
iz
e
1.
W
hi
le
th
e
pa
ddi
ng
la
ye
r
a
do
pt
s
a
2
×
2
c
onf
ig
ur
a
ti
on
a
t
a
s
te
p
s
iz
e
of
2
w
hi
le
m
a
in
ta
in
in
g
e
qua
l
pa
ddi
ng
[
27]
.
V
G
G
-
16
m
ode
l
pe
r
f
o
r
m
s
224
×
224
in
put
im
a
ge
pr
oc
e
s
s
in
g.
B
e
f
or
e
th
e
f
ul
ly
c
onne
c
te
d
la
ye
r
,
a
7
×
7
f
e
a
tu
r
e
m
a
p
c
ont
a
in
in
g
512
c
ha
nne
l
s
is
us
e
d.
T
he
n,
th
e
r
e
s
ul
ts
of
th
e
f
e
a
tu
r
e
m
a
ppi
ng
a
r
e
pr
oc
e
s
s
e
d
in
to
a
v
e
c
to
r
w
it
h
25,088
c
ha
nn
e
ls
(
7
×
7
×
512)
a
s
a
r
e
pr
e
s
e
nt
a
ti
on
of
th
e
r
e
s
ul
ti
ng
f
e
a
tu
r
e
s
[
28]
,
[
29
]
.
A
pa
r
t
f
r
o
m
th
e
C
N
N
m
ode
l,
th
is
r
e
s
e
a
r
c
h
a
ls
o
us
e
s
th
e
V
G
G
-
16
m
ode
l
a
s
a
c
om
pa
r
is
on
to
s
e
e
w
hi
c
h
m
ode
l
is
be
tt
e
r
a
t
id
e
nt
if
yi
ng por
nogr
a
phi
c
c
ont
e
nt
w
it
h da
ta
of
14,085 im
a
ge
s
.
2.4.
L
e
ve
l
of
a
c
c
e
s
s
an
d
r
e
gu
la
t
io
n
r
e
gar
d
in
g p
or
n
ogr
ap
h
y
in
I
n
d
on
e
s
ia
I
n
I
ndone
s
ia
,
th
e
r
e
is
a
c
ondi
ti
on
w
he
r
e
qui
te
a
lo
t
of
pe
opl
e
a
r
e
in
te
r
e
s
te
d
in
por
nogr
a
phy
vi
a
th
e
in
te
r
ne
t.
A
lo
ng
w
it
h
c
om
m
uni
c
a
ti
on
a
nd
in
f
or
m
a
ti
on
te
c
hnol
ogy,
th
e
s
pr
e
a
d
of
por
nogr
a
phi
c
m
e
di
a
c
a
n
be
a
c
c
e
s
s
e
d onli
ne
vi
a
t
he
i
nt
e
r
ne
t.
I
nt
e
r
ne
t
por
nog
r
a
p
hy ha
s
not
o
nl
y de
ve
lo
pe
d a
s
a
pe
r
s
ona
l
ne
e
d, but ha
s
a
ls
o
be
c
om
e
a
c
om
m
odi
ty
th
a
t
is
s
ol
d
c
om
m
e
r
c
ia
ll
y
a
nd
p
r
of
e
s
s
io
na
ll
y.
P
or
nogr
a
phi
c
ne
w
s
is
s
om
e
th
in
g
th
a
t
c
a
n
be
s
ol
d
in
th
e
m
a
s
s
m
e
di
a
,
a
pa
r
t
f
r
om
ne
w
s
a
bout
c
r
im
e
s
uc
h
a
s
c
ybe
r
-
s
e
x
a
nd
por
n
s
it
e
s
.
P
or
nogr
a
phy
is
c
la
s
s
if
ie
d
a
s
a
c
r
im
in
a
l
a
c
t
th
a
t
vi
ol
a
te
s
m
or
a
li
ty
(
Z
e
de
li
jk
he
id
)
a
nd
is
a
pr
obl
e
m
r
e
la
t
e
d
to
s
e
xu
a
li
ty
.
P
or
nogr
a
phy
in
I
ndone
s
ia
is
c
ons
id
e
r
e
d
il
le
ga
l
a
nd
vi
ol
a
te
s
th
e
la
w
[
4]
.
P
or
nogr
a
phy
is
r
e
gul
a
te
d
in
a
r
ti
c
le
1
num
be
r
(
1)
in
r
e
gul
a
ti
on
la
w
44/
2008
c
onc
e
r
ni
ng
por
nogr
a
phy
a
nd
th
e
in
f
or
m
a
ti
on
a
nd
e
le
c
tr
oni
c
tr
a
ns
a
c
ti
on
s
(
I
T
E
)
la
w
c
ha
pt
e
r
V
I
I
c
onc
e
r
ni
ng
pr
oh
ib
it
e
d
a
c
ts
,
e
s
pe
c
ia
ll
y
i
n
a
r
ti
c
le
27
pa
r
a
gr
a
ph
(
1)
a
nd
la
w
num
be
r
3
6
1999 pa
r
a
gr
a
ph 21 c
onc
e
r
ni
ng
te
le
c
om
m
uni
c
a
ti
ons
,
e
s
pe
c
i
a
ll
y point
s
r
e
ga
r
di
ng mor
a
li
ty
(
por
nog
r
a
phy)
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
ti
f
I
nt
e
ll
I
S
S
N
:
2252
-
8938
C
om
par
is
on of de
e
p l
e
ar
ni
ng m
ode
l
s
:
C
N
N
and V
G
G
-
16 i
n i
de
nt
if
y
in
g por
nogr
aphi
c
…
(
R
e
z
a C
hand
r
a)
1887
3.
M
E
T
H
O
D
3.1.
R
e
s
e
ar
c
h
ob
j
e
c
t
s
W
e
bs
it
e
s
th
a
t
w
e
r
e
a
ll
e
ge
dl
y
de
te
c
te
d
to
c
ont
a
in
por
nogr
a
phi
c
c
ont
e
nt
w
e
r
e
th
e
obj
e
c
ts
of
th
is
r
e
s
e
a
r
c
h.
T
hi
s
w
e
b
s
it
e
w
a
s
ta
ke
n
ba
s
e
d
on
a
r
ti
c
le
in
f
or
m
a
ti
on
f
r
om
th
e
of
f
ic
ia
l
K
om
in
f
o
w
e
bs
it
e
of
th
e
R
e
publ
ic
of
I
ndone
s
ia
on
t
he
w
e
b
s
it
e
ht
tp
s
:/
/t
r
us
tp
os
it
iv
e
.komi
nf
o.go.id/
w
hi
c
h c
a
n be
a
c
c
e
s
s
e
d by the
publi
c
.
T
he
r
e
a
r
e
773,517 we
bs
it
e
s
t
ha
t
a
r
e
bl
oc
ke
d, but t
he
r
e
a
r
e
s
e
ve
r
a
l
w
e
bs
it
e
s
t
ha
t
c
a
n
s
ti
ll
be
a
c
c
e
s
s
e
d u
s
in
g t
he
vi
r
tu
a
l
pr
iv
a
te
ne
twor
k
(
V
P
N
)
a
ppl
ic
a
ti
on a
nd t
he
r
e
a
r
e
a
ls
o t
hos
e
w
hos
e
dom
a
in
na
m
e
s
h
a
ve
e
xpi
r
e
d.
3.2.
R
e
s
e
ar
c
h
f
r
a
m
e
w
or
k
O
ne
of
th
e
r
e
a
s
ons
f
or
us
in
g
th
e
C
N
N
m
ode
l
f
or
im
a
ge
id
e
nt
if
ic
a
ti
on
is
be
c
a
us
e
th
i
s
m
ode
l
ha
s
th
e
m
os
t
s
ig
ni
f
ic
a
nt
r
e
s
ul
ts
in
r
e
c
ogni
z
in
g
im
a
ge
obj
e
c
ts
.
C
N
N
c
a
n
r
e
c
ogni
z
e
im
a
ge
s
w
it
h
a
lm
os
t
th
e
s
a
m
e
le
ve
l
of
a
c
c
ur
a
c
y
a
s
hum
a
ns
on
a
c
e
r
ta
in
num
be
r
of
da
ta
s
e
ts
[
30]
.
A
lg
or
it
hm
a
nd
c
ons
tr
uc
ti
on
of
a
por
nogr
a
phi
c
im
a
ge
c
ont
e
nt
id
e
nt
if
ic
a
ti
on
s
ys
te
m
:
i
)
ha
r
dw
a
r
e
de
s
ig
n,
us
in
g
a
D
G
X
A
100
m
a
c
hi
ne
e
qui
ppe
d
w
it
h
8
G
P
U
s
a
nd
ha
vi
ng
a
to
ta
l
m
e
m
or
y
of
320
G
B
a
nd
e
qui
ppe
d
w
it
h
6
N
V
I
D
I
A
N
V
S
w
it
c
he
s
be
c
a
us
e
th
e
m
a
tr
ix
ope
r
a
ti
on
c
om
put
a
ti
on
in
C
N
N
a
nd
V
G
G
-
16
is
ve
r
y
in
te
ns
iv
e
. G
P
U
s
de
s
ig
ne
d
to
pe
r
f
or
m
pa
r
a
ll
e
l
c
om
put
a
ti
ons
a
r
e
w
e
l
l
s
ui
te
d
to
th
is
ta
s
k;
ii
)
s
of
twa
r
e
s
ys
te
m
,
us
in
g
T
e
ns
or
F
lo
w
a
nd
K
e
r
a
s
a
s
a
de
e
p
le
a
r
ni
ng
f
r
a
m
e
w
or
k
to
bui
ld
a
nd
tr
a
in
C
N
N
a
nd
V
G
G
-
16
m
ode
ls
.
T
he
pr
ogr
a
m
m
in
g
la
ngua
ge
us
e
s
P
yt
hon
be
c
a
us
e
of
it
s
r
ic
h
e
c
os
ys
te
m
a
nd
th
e
la
r
ge
num
be
r
of
li
br
a
r
ie
s
a
va
il
a
bl
e
.
S
uppor
ti
ng
li
br
a
r
ie
s
us
e
O
pe
nC
V
f
or
im
a
ge
pr
oc
e
s
s
in
g,
N
um
P
y
(
num
e
r
ic
a
l
c
om
put
in
g)
,
P
a
nda
s
(
f
r
a
m
e
da
ta
)
,
M
a
tp
lo
tl
ib
(
da
ta
vi
s
ua
li
z
a
ti
on)
;
a
nd
iii
)
th
e
H
T
T
P
pr
ot
oc
ol
is
us
e
d
to
s
e
nd r
e
que
s
ts
a
nd r
e
c
e
iv
e
r
e
s
pons
e
s
f
r
om
t
he
s
e
r
ve
r
w
he
n s
c
r
a
pi
ng da
ta
f
r
om
w
e
bs
it
e
s
.
3.3.
C
on
vol
u
t
io
n
al
n
e
u
r
al
n
e
t
w
or
k
m
od
e
l
d
e
s
ig
n
I
m
a
ge
da
ta
c
ol
le
c
ti
on
b
a
s
e
d
on
r
e
f
e
r
e
nc
e
s
to
w
e
b
s
it
e
s
in
di
c
a
te
d
t
o
c
ont
a
in
por
nogr
a
phi
c
c
ont
e
nt
w
hi
c
h
a
r
e
bl
a
c
kl
is
te
d
by
K
om
in
f
o
vi
a
th
e
w
e
b
tr
us
tp
os
it
iv
e
.komi
nf
o.go.id
a
nd
a
c
ol
le
c
ti
on
of
w
e
bs
it
e
s
c
ont
a
in
in
g
non
-
por
nogr
a
phi
c
c
ont
e
nt
f
r
om
va
r
io
us
s
our
c
e
s
.
L
is
t
of
w
e
bs
it
e
a
ddr
e
s
s
e
s
c
ont
a
in
in
g por
nogr
a
phi
c
c
ont
e
nt
a
s
s
how
n
in
T
a
bl
e
1.
T
o
obt
a
in
a
n
im
a
g
e
da
ta
s
e
t,
it
is
ne
c
e
s
s
a
r
y
to
de
s
ig
n
a
w
e
b
c
r
a
w
le
r
f
or
w
e
bs
it
e
s
th
a
t
a
r
e
de
e
m
e
d t
o ha
ve
por
nogr
a
phi
c
a
nd non
-
por
nogr
a
phi
c
c
ont
e
nt
. T
he
s
ta
ge
s
a
r
e
a
s
i
n F
ig
ur
e
2
.
T
he
f
ir
s
t
s
te
p,
c
hoos
e
w
e
bs
it
e
s
th
a
t
a
r
e
in
di
c
a
te
d a
s
por
nogr
a
phi
c
a
nd
non
-
por
nogr
a
phi
c
.
T
he
n
e
nt
e
r
th
e
w
e
bs
it
e
U
R
L
a
nd
c
a
r
r
y
out
a
c
r
a
w
li
ng
pr
oc
e
s
s
w
hi
c
h
w
il
l
pul
l
a
ll
th
e
im
a
ge
s
on
th
e
ta
r
ge
te
d
w
e
bs
it
e
a
nd
s
a
ve
th
e
m
in
a
di
r
e
c
to
r
y.
W
he
n
th
e
da
ta
ha
s
f
in
i
s
he
d
c
r
a
w
li
ng,
t
he
da
ta
is
gr
oupe
d
in
to
two
d
ir
e
c
to
r
ie
s
,
na
m
e
ly
th
e
te
s
t
s
e
t
di
r
e
c
to
r
y
w
hi
c
h
is
th
e
di
r
e
c
to
r
y
f
or
th
e
te
s
t
im
a
ge
da
ta
s
e
t
w
hi
c
h
c
ons
i
s
ts
of
por
nogr
a
phi
c
im
a
ge
da
ta
s
e
ts
a
nd
non
-
por
nogr
a
phi
c
im
a
ge
da
ta
s
e
t
s
a
nd
th
e
tr
a
in
in
g
s
e
t
di
r
e
c
to
r
y
w
hi
c
h
is
th
e
di
r
e
c
to
r
y
f
or
th
e
tr
a
in
in
g
im
a
ge
da
ta
s
e
t
w
hi
c
h
c
ons
i
s
ts
of
por
nogr
a
phi
c
im
a
g
e
d
a
ta
s
e
ts
a
nd
non
-
por
nogr
a
phy,
(
F
ig
ur
e
3)
.
A
f
te
r
s
uc
c
e
s
s
f
ul
ly
obt
a
in
in
g
a
da
ta
s
e
t
of
por
nogr
a
phi
c
im
a
ge
s
a
nd
a
da
ta
s
e
t
of
non
-
por
nogr
a
phi
c
im
a
ge
s
,
w
e
th
e
n
de
s
ig
ne
d
a
C
N
N
pr
oc
e
s
s
to
tr
a
in
th
e
da
ta
s
e
t
a
nd
c
a
r
r
y
out
it
s
v
a
li
di
ty
w
it
h
te
s
t
d
a
ta
us
in
g
14,085
da
ta
w
it
h
a
tr
a
in
in
g
da
ta
c
om
pos
it
io
n
of
80%
,
th
e
r
e
m
a
in
in
g
te
s
t
da
ta
i
s
20%
.
T
he
in
it
ia
l
s
ta
ge
of
d
e
s
ig
ni
ng
th
e
C
N
N
a
r
c
hi
te
c
tu
r
a
l
m
ode
l
s
ta
r
ts
f
r
om
t
he
i
nput
da
t
a
, t
he
n goe
s
to
t
he
2
D
z
e
r
o pa
ddi
ng l
a
ye
r
w
hi
c
h i
s
u
s
e
d t
o b
e
c
om
e
th
e
out
put
s
iz
e
w
hos
e
s
ha
pe
a
nd
pr
opor
ti
ons
a
r
e
th
e
s
a
m
e
a
s
th
e
in
put
s
iz
e
.
T
hi
s
m
ode
l
c
ons
is
ts
of
3
2D
c
onvolut
io
na
l
la
ye
r
s
,
3
2D
m
a
x
pool
l
a
ye
r
s
,
1
f
la
tt
e
n
la
ye
r
a
nd
2
de
ns
e
la
ye
r
s
a
s
in
F
ig
ur
e
4
a
nd
th
e
n
id
e
nt
i
f
ie
s
im
a
ge
c
ont
e
nt
t
hr
ough s
e
ve
r
a
l
s
ta
g
e
s
a
s
i
n F
ig
ur
e
5.
T
a
bl
e
1. W
e
b
a
ddr
e
s
s
i
ndi
c
a
te
d t
o c
ont
a
in
por
nogr
a
phi
c
c
ont
e
nt
No
W
e
bs
i
t
e
U
R
L
I
P
a
ddr
e
s
s
A
c
c
e
s
s
t
i
m
e
C
r
a
w
l
i
ng t
i
m
e
1
a
s
i
a
nne
w
pi
c
s
.c
om
/
69.10.41.18
04/
10/
2019
3 hour
s
2
s
e
xpi
c
s
24.c
om
/
88.85.75.41
03/
28/
2019
3 hour
s
3
a
s
i
a
nba
be
pi
c
s
.c
om
/
109.232.227.100
04/
10/
2019
3 hour
s
4
nude
s
e
xpor
n.c
om
/
23.226.129.188
04/
10/
2019
4 hour
s
5
por
npi
c
s
.c
om
/
108.60.221.55
03/
28/
2019
4 hour
s
6
a
s
i
a
npor
npi
c
t
ur
e
.c
om
/
109.232.227.100
03/
28/
2019
3 hour
s
7
hi
gha
s
i
a
npor
n.c
om
/
162.251.111.236
03/
28/
2019
2 hour
s
8
a
l
l
ba
be
s
na
ke
d.c
om
/
185.73.221.118
03/
28/
2019
2 hour
s
9
c
hi
ne
s
e
m
i
f
l
.c
om
/
109.232.227.101
04/
10/
2019
2 hour
s
10
l
a
m
a
l
i
nks
.c
om
/
208.69.117.6
03/
28/
2019
3 hour
s
11
hdpor
npi
c
s
.c
om
/
206.54.175.4
04/
10/
2019
4 hour
s
12
f
or
c
e
s
e
xpi
c
.c
om
/
88.76.59.10
04/
10/
2019
1 hour
13
nude
xxxpi
c
t
ur
e
s
.c
om
/
78.140.153.205
03/
28/
2019
2 hour
s
14
por
npi
c
s
a
m
a
t
e
ur
.c
om
/
78.140.133.54
04/
10/
2019
2 hour
s
15
nude
bl
a
c
kgi
r
l
s
phot
os
.c
om
/
64.111.197.114
03/
28/
2019
2 hour
s
16
por
npi
c
s
.i
o/
88.85.75.40
04/
10/
2019
1 hour
17
youxx.xx.c
om
/
185.53.178.9
03/
28/
2019
2 hour
s
18
w
ow
e
r
ot
i
c
a
.c
om
/
199.80.52.155
04/
10/
2019
4 hour
s
19
ba
be
s
our
c
e
.c
om
/
99.192.136.176
03/
28/
2019
3 hour
s
20
di
r
t
ypor
nphot
os
.c
om
/
88.85.75.37
04/
10/
2019
3 hour
s
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
, V
ol
.
14
, N
o.
3
,
J
une
20
25
:
1884
-
1899
1888
F
ig
ur
e
2. I
m
a
ge
c
ont
e
nt
c
r
a
w
li
ng f
lo
w
c
ha
r
t
F
ig
ur
e
3. I
m
a
ge
da
ta
l
a
be
li
ng de
s
ig
n
F
ig
ur
e
4. C
N
N
m
ode
l
a
r
c
hi
te
c
tu
r
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
ti
f
I
nt
e
ll
I
S
S
N
:
2252
-
8938
C
om
par
is
on of de
e
p l
e
ar
ni
ng m
ode
l
s
:
C
N
N
and V
G
G
-
16 i
n i
de
nt
if
y
in
g por
nogr
aphi
c
…
(
R
e
z
a C
hand
r
a)
1889
F
ig
ur
e
5. C
N
N
m
ode
l
f
lo
w
c
ha
r
t
f
or
i
m
a
ge
i
de
nt
if
ic
a
ti
on
E
nt
e
r
in
g
th
e
li
br
a
r
ie
s
ne
e
de
d
to
s
ta
r
t
th
e
pr
oc
e
s
s
.
C
r
e
a
te
a
z
e
r
o
pa
ddi
ng
la
ye
r
w
hi
c
h
i
s
th
e
f
ir
s
t
la
ye
r
ne
e
de
d
in
th
e
pr
oc
e
s
s
,
th
e
n
th
e
f
ir
s
t
c
onvolut
io
na
l
la
ye
r
ne
e
de
d
a
f
te
r
th
a
t
th
e
pr
oc
e
s
s
c
ont
in
ue
s
w
it
h
m
a
x
pool
in
g
2
D
a
nd
th
e
n
goe
s
to
th
e
s
e
c
ond
c
onvolut
io
na
l
la
y
e
r
th
e
n
th
e
f
in
a
l
m
a
x
pool
in
g
2
D
pr
oc
e
s
s
be
f
or
e
th
e
la
ye
r
i
s
f
in
is
he
d be
in
g c
r
e
a
te
d
. D
e
te
r
m
in
e
t
he
a
lg
or
it
hm
us
e
d f
or
t
he
opt
im
iz
e
r
, na
m
e
ly
A
da
m
, de
te
r
m
in
e
s
t
he
num
be
r
of
e
poc
hs
a
nd de
te
r
m
in
e
s
t
he
l
e
a
r
ni
ng r
a
te
va
lu
e
;
s
ta
r
t
t
he
da
ta
t
r
a
in
in
g pr
oc
e
s
s
;
s
a
ve
t
h
e
r
e
s
ul
ts
of
t
he
tr
a
in
in
g da
ta
i
n t
he
f
or
m
of
a
n h5 model; and the
C
N
N
pr
oc
e
s
s
i
s
c
om
pl
e
te
.
3.4.
V
G
G
-
16
m
od
e
l
d
e
s
ig
n
T
he
de
s
ig
n of
t
hi
s
m
od
e
l
a
im
s
t
o c
om
pa
r
e
t
h
e
i
m
a
ge
i
d
e
nt
if
ic
a
ti
on r
e
s
ul
ts
f
r
om
t
he
C
N
N
m
ode
l.
T
hi
s
m
ode
l
us
e
s
13
c
onvolut
io
n
la
ye
r
s
,
5
m
a
x
pool
la
ye
r
s
,
1
f
la
tt
e
n l
a
ye
r
a
nd
3
de
ns
e
la
ye
r
s
. T
he
f
lo
w
i
s
th
e
s
a
m
e
a
s
t
he
C
N
N
m
ode
l
b
e
c
a
u
s
e
t
he
V
G
G
-
16 mode
l
is
a
de
r
iv
a
ti
ve
of
t
he
C
N
N
m
ode
l
, s
e
e
F
ig
ur
e
6.
F
ig
ur
e
6. F
lo
w
a
r
c
hi
te
c
tu
r
e
in
t
he
V
G
G
-
16
m
ode
l
4.
R
E
S
U
L
T
S
A
N
D
D
I
S
C
U
S
S
I
O
N
4.1. I
d
e
n
t
if
ic
at
io
n
of
im
age
c
on
t
e
n
t
i
d
e
n
t
if
ie
d
as
p
or
n
ogr
ap
h
ic
w
it
h
t
h
e
C
N
N
m
od
e
l
T
o
c
r
a
w
l
im
a
ge
s
,
a
m
ul
ti
th
r
e
a
di
n
g
m
e
th
od
is
u
s
e
d
th
r
oug
h
s
e
ve
r
a
l
s
t
e
ps
in
c
lu
di
n
g:
i
)
de
t
e
r
m
in
in
g
w
e
b
s
it
e
s
th
a
t
a
r
e
in
di
c
a
te
d
a
s
por
no
gr
a
phi
c
a
nd
non
-
p
or
nogr
a
phi
c
,
n
a
m
e
l
y
de
t
e
r
m
in
in
g
th
e
d
e
s
ti
na
t
io
n
w
e
b
a
ddr
e
s
s
,
a
nd
r
e
tr
i
e
vi
n
g
w
e
b
p
a
ge
in
f
or
m
a
ti
on
;
ii
)
c
a
r
r
y
out
th
e
c
r
a
w
li
ng
pr
oc
e
s
s
s
ta
r
ti
ng
u
s
in
g
th
e
m
ul
ti
t
hr
e
a
di
ng
m
e
th
od on th
e
w
e
bs
it
e
by r
e
tr
i
e
vi
ng i
m
a
ge
d
a
ta
o
n t
he
w
e
bs
i
te
vi
a
f
or
l
in
ks
i
n l
in
ks
, i
nput
ti
ng i
m
a
g
e
l
in
k
s
, a
n
d
e
xt
r
a
c
ti
ng
im
a
g
e
da
t
a
;
a
nd
iii
)
th
e
r
e
s
ul
t
s
of
th
e
c
r
a
w
li
ng
pr
oc
e
s
s
w
e
r
e
s
uc
c
e
s
s
f
u
ll
y s
a
ve
d
. T
o
tr
a
in
a
nd
t
e
s
t
th
e
im
a
ge
d
a
ta
s
e
t,
th
e
C
N
N
a
lg
or
it
hm
i
s
us
e
d.
T
hi
s
a
l
gor
it
hm
w
a
s
c
om
pi
l
e
d
u
s
in
g
t
he
P
yt
hon
pr
ogr
a
m
m
in
g
la
ngu
a
ge
m
od
e
l
v
e
r
s
i
on
3.6
a
nd
th
e
K
e
r
a
s
li
br
a
r
y.
T
he
s
t
e
ps
f
or
de
v
e
lo
pi
n
g
a
C
N
N
m
od
e
l
a
r
e
a
s
f
ol
lo
w
s
:
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
, V
ol
.
14
, N
o.
3
,
J
une
20
25
:
1884
-
1899
1890
i
)
in
s
e
r
ti
ng
t
he
N
um
P
y
,
T
e
n
s
or
F
lo
w
,
T
e
n
s
or
B
oa
r
d
a
nd
K
e
r
a
s
li
br
a
r
ie
s
;
ii
)
C
N
N
in
it
i
a
li
z
a
ti
on;
iii
)
b
ui
ld
in
g
th
e
C
N
N
m
od
e
l
i
nc
lu
d
e
s
bu
il
di
ng
th
e
f
ir
s
t
l
a
ye
r
by
de
te
r
m
in
i
ng
p
a
dd
in
g,
a
c
ti
v
a
ti
o
n
f
un
c
ti
on
a
nd
m
a
x
pool
i
ng;
bui
ld
th
e
s
e
c
ond
l
a
ye
r
by
de
te
r
m
in
i
ng
th
e
a
c
ti
va
ti
on
a
nd
m
a
x
po
ol
in
g
f
un
c
ti
on
s
;
a
nd
bu
il
d
th
e
th
ir
d
l
a
y
e
r
by
de
te
r
m
in
in
g
th
e
a
c
ti
va
ti
on
a
nd
m
a
x
p
ool
in
g
f
un
c
ti
on
s
;
iv
)
a
dd
f
la
tt
e
n;
v
)
a
dd
2
f
ul
ly
c
o
nne
c
te
d
l
a
ye
r
s
;
vi
)
a
dd
opt
im
iz
e
r
;
vi
i
)
c
a
ll
t
he
T
e
n
s
or
B
o
a
r
d
l
ib
r
a
r
y
f
or
gr
a
phi
c
s
;
vi
ii
)
f
i
tt
in
g
pr
o
c
e
s
s
;
ix
)
tr
a
in
in
g
pr
oc
e
s
s
;
a
nd
x
)
s
a
ve
th
e
tr
a
in
in
g
r
e
s
ul
ts
.
I
n
te
s
ti
ng
th
e
id
e
nt
if
i
c
a
ti
on
of
im
a
g
e
c
ont
e
nt
th
a
t
is
in
di
c
a
ti
ve
of
por
n
ogr
a
p
hy
th
r
oug
h
C
N
N
m
ode
l
tr
a
in
i
ng
,
9
tr
i
a
l
s
w
e
r
e
c
a
r
r
ie
d
out
u
s
in
g
di
f
f
e
r
e
nt
e
po
c
h
va
r
ia
bl
e
va
lu
e
s
,
di
f
f
e
r
e
nt
a
c
ti
v
a
ti
on
f
unc
ti
on
s
,
di
f
f
e
r
e
nt
l
e
a
r
ni
ng r
a
t
e
s
,
s
o t
h
a
t
th
e
b
e
s
t
r
e
s
ul
ts
w
e
r
e
obt
a
in
e
d f
r
o
m
t
he
m
ode
l
th
a
t
ha
d
be
e
n bui
lt
,
a
s
f
o
ll
ow
s
.
4.1.1. F
ir
s
t
t
e
s
t
r
e
s
u
lt
s
of
c
on
vol
u
t
io
n
al
n
e
u
r
al
n
e
t
w
or
k
m
od
e
l
B
a
s
e
d
on
th
e
pa
r
a
m
e
te
r
s
:
i)
num
be
r
of
la
ye
r
s
=
1;
ii
)
le
a
r
ni
ng
r
a
te
=
0.001;
ii
i)
e
poc
h=
25;
iv
)
s
te
ps
pe
r
e
poc
h=
100;
a
nd
v)
a
c
ti
va
ti
on
f
unc
ti
on
a
t
th
e
f
ul
l
c
onne
c
ti
on
la
ye
r
=
r
e
c
ti
f
ie
d
li
ne
a
r
uni
ts
(
R
e
L
u
)
a
nd
s
ig
m
oi
d
.
U
s
in
g
th
e
pa
r
a
m
e
te
r
num
be
r
of
c
onvolut
io
n
la
ye
r
s
is
1,
th
e
le
a
r
n
in
g
r
a
te
ha
s
a
va
lu
e
of
0.001,
e
poc
h
h
a
s
a
va
lu
e
of
25,
s
te
ps
pe
r
e
po
c
h
ha
s
a
va
lu
e
of
100
a
nd
s
e
le
c
ti
ng
th
e
R
e
L
u
a
c
ti
va
ti
on
f
unc
ti
on
in
th
e
f
ul
l
c
onne
c
ti
on
la
ye
r
pr
oduc
e
s
a
n
a
c
c
ur
a
c
y
va
lu
e
of
0.8341,
t
he
l
os
s
va
lu
e
r
e
a
c
he
s
0.
3691, va
li
d a
c
c
ur
a
c
y
ha
s
a
v
a
lu
e
of
0.8299 a
nd
a
va
li
d
lo
s
s
va
lu
e
of
0.2804.
F
ig
ur
e
7
s
how
s
f
ir
s
t
r
e
s
ul
t
s
of
id
e
nt
if
ic
a
ti
on
of
im
a
ge
c
ont
e
nt
id
e
nt
if
ie
d
a
s
por
nogr
a
phi
c
w
it
h t
he
C
N
N
m
ode
l
.
4.1.2. S
e
c
on
d
t
e
s
t
r
e
s
u
lt
s
c
on
vol
u
t
io
n
al
n
e
u
r
al
n
e
t
w
or
k
m
od
e
l
B
a
s
e
d on the
p
a
r
a
m
e
te
r
s
:
i)
numbe
r
of
l
a
ye
r
s
=
1;
i
i)
l
e
a
r
ni
ng r
a
te
=
0.0001; i
ii
)
e
poc
h=
25;
i
v)
s
te
ps
p
e
r
e
poc
h=
100;
a
nd
v)
a
c
ti
va
ti
on
f
unc
ti
on
a
t
th
e
f
ul
l
c
onne
c
ti
on
la
ye
r
=
R
e
L
u
a
nd
s
ig
m
oi
d.
U
s
e
d
p
a
r
a
m
e
te
r
s
f
or
th
e
num
be
r
of
c
onvolut
io
n
la
ye
r
s
of
1,
th
e
le
a
r
ni
ng
r
a
te
w
a
s
0.0001,
th
e
e
poc
h
w
a
s
25,
th
e
s
te
p
s
pe
r
e
po
c
h
w
a
s
100
a
nd
th
e
R
e
L
u
a
c
ti
va
ti
on
f
unc
ti
on
on
th
e
f
ul
l
c
onne
c
ti
on
la
ye
r
pr
oduc
e
d
a
n
a
c
c
ur
a
c
y
va
lu
e
of
0.8228,
a
lo
s
s
va
lu
e
of
0.386,
a
v
a
li
d
va
lu
e
of
a
c
c
ur
a
c
y
i
s
0.8038
a
nd
a
v
a
li
d
lo
s
s
v
a
lu
e
is
0.4707.
F
ig
ur
e
8
s
how
s
s
e
c
ond
r
e
s
ul
ts
of
i
de
nt
if
ic
a
ti
on of
i
m
a
ge
c
ont
e
nt
i
de
nt
if
ie
d a
s
por
nogr
a
phi
c
w
it
h t
he
C
N
N
m
ode
l.
F
ig
ur
e
7. F
ir
s
t
r
e
s
ul
ts
of
i
de
nt
if
ic
a
ti
on of
i
m
a
ge
c
ont
e
nt
i
de
nt
if
ie
d a
s
por
nogr
a
phi
c
w
it
h t
he
C
N
N
m
ode
l
F
ig
ur
e
8.
S
e
c
ond r
e
s
ul
ts
of
i
de
nt
if
ic
a
ti
on of
i
m
a
ge
c
ont
e
nt
i
de
nt
if
ie
d a
s
por
nogr
a
phi
c
w
it
h t
he
C
N
N
m
ode
l
4.1.3.
T
h
ir
d
t
e
s
t
r
e
s
u
lt
s
c
on
vol
u
t
io
n
al
n
e
u
r
al
n
e
t
w
or
k
m
od
e
l
B
a
s
e
d
on
th
e
p
a
r
a
m
e
te
r
s
:
i
)
num
be
r
of
c
onvolut
io
n
la
ye
r
s
=
1;
ii
)
le
a
r
ni
ng
r
a
te
=
0.001;
iii
)
e
poc
h
s
=
50;
iv
)
s
te
ps
pe
r
e
poc
h=
100;
a
nd
v
)
a
c
ti
va
ti
on
f
unc
ti
on
a
t
th
e
f
ul
l
c
onne
c
ti
on
la
ye
r
=
R
e
lu
a
nd
s
ig
m
oi
d
.
U
s
e
d
th
e
pa
r
a
m
e
te
r
s
of
th
e
num
be
r
of
c
onvolut
io
n
la
ye
r
s
b
e
in
g
1,
th
e
l
e
a
r
ni
ng
r
a
te
is
0.001,
th
e
e
poc
h
is
25,
th
e
s
te
p
s
pe
r
e
poc
h a
r
e
100 a
nd t
he
R
e
lu
a
c
ti
va
ti
on f
unc
ti
on on the
f
ul
l
c
onne
c
ti
on l
a
ye
r
pr
oduc
e
s
a
n a
c
c
ur
a
c
y va
lu
e
of
0.8744,
a
lo
s
s
va
lu
e
of
0.2838,
a
va
li
d
va
lu
e
.
a
c
c
ur
a
c
y
is
0.8289
a
nd
va
li
d
lo
s
s
v
a
lu
e
is
0.2281.
F
ig
ur
e
9
s
how
s
th
ir
d r
e
s
ul
ts
of
i
de
nt
if
ic
a
ti
on of
i
m
a
ge
c
ont
e
nt
i
de
nt
if
ie
d a
s
por
nogr
a
phi
c
w
it
h t
he
C
N
N
m
ode
l.
4.1.4. F
ou
r
t
h
t
e
s
t
r
e
s
u
lt
s
c
on
vol
u
t
io
n
al
n
e
u
r
al
n
e
t
w
or
k
m
od
e
l
B
a
s
e
d on the
pa
r
a
m
e
te
r
s
:
i)
numbe
r
of
c
onvolut
io
n
l
a
ye
r
s
=
2;
i
i)
l
e
a
r
ni
ng r
a
te
=
0.0001; i
ii
)
e
poc
h=
25;
iv
)
s
te
ps
p
e
r
e
poc
h
=
100;
a
nd
v)
a
c
ti
va
ti
on
f
unc
ti
on a
t
th
e
f
ul
l
c
onne
c
ti
on
la
ye
r
=
R
e
L
u
a
nd s
ig
m
oi
d.
U
s
in
g
th
e
pa
r
a
m
e
te
r
num
be
r
of
c
onvolut
io
n
la
ye
r
s
is
2,
t
he
le
a
r
ni
ng
r
a
te
is
0.0001,
th
e
e
poc
h
is
25,
th
e
s
te
ps
pe
r
e
poc
h
is
100
a
nd
th
e
R
e
L
u
a
c
ti
va
ti
on
f
unc
ti
on
on
th
e
f
ul
l
c
onn
e
c
ti
on
la
ye
r
pr
oduc
e
s
a
n
a
c
c
ur
a
c
y
va
lu
e
of
0.8503,
th
e
lo
s
s
r
a
te
va
lu
e
r
e
a
c
he
s
0.3475,
va
li
d
a
c
c
ur
a
c
y
ha
s
a
va
lu
e
of
0.7509
a
nd
a
va
li
d
lo
s
s
va
lu
e
of
0.2809.
F
ig
ur
e
10
s
how
s
f
our
th
r
e
s
ul
ts
of
i
de
nt
if
ic
a
ti
on of
i
m
a
ge
c
ont
e
nt
i
de
nt
if
ie
d a
s
por
nogr
a
phi
c
w
it
h t
he
C
N
N
m
ode
l.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
ti
f
I
nt
e
ll
I
S
S
N
:
2252
-
8938
C
om
par
is
on of de
e
p l
e
ar
ni
ng m
ode
l
s
:
C
N
N
and V
G
G
-
16 i
n i
de
nt
if
y
in
g por
nogr
aphi
c
…
(
R
e
z
a C
hand
r
a)
1891
F
ig
ur
e
9. T
hi
r
d r
e
s
ul
ts
of
i
de
nt
if
ic
a
ti
on of
i
m
a
ge
c
ont
e
nt
i
de
nt
if
ie
d a
s
por
nogr
a
phi
c
w
it
h t
he
C
N
N
m
ode
l
F
ig
ur
e
10. F
our
th
r
e
s
ul
ts
of
i
de
nt
if
ic
a
ti
on of
im
a
ge
c
ont
e
nt
i
de
nt
if
ie
d a
s
por
nogr
a
phi
c
w
it
h t
he
C
N
N
m
ode
l
4.1.5.
F
if
t
h
t
e
s
t
r
e
s
u
lt
s
c
on
vol
u
t
io
n
al
n
e
u
r
al
n
e
t
w
or
k
m
od
e
l
B
a
s
e
d
on
th
e
p
a
r
a
m
e
te
r
s
:
i
)
num
be
r
of
c
onvolut
io
n
la
ye
r
s
=
2;
ii
)
le
a
r
ni
ng
r
a
te
=
0.001;
iii
)
e
poc
h
s
=
50;
iv
)
s
te
ps
p
e
r
e
poc
h
=
100;
a
nd
v
)
a
c
ti
va
ti
on
f
unc
ti
on a
t
th
e
f
ul
l
c
onne
c
ti
on
la
ye
r
=
R
e
L
u
a
nd
s
ig
m
oi
d
.
U
s
in
g
th
e
pa
r
a
m
e
te
r
num
be
r
of
c
onvolut
io
n
la
y
e
r
s
is
2,
th
e
le
a
r
ni
ng
r
a
te
ha
s
a
va
lu
e
of
0.001,
e
po
c
h
ha
s
a
va
lu
e
of
50,
s
te
ps
pe
r
e
poc
h
ha
s
a
va
lu
e
of
100
a
nd
s
e
le
c
ti
ng
th
e
R
e
L
u
a
c
ti
va
ti
on
f
unc
ti
on
in
th
e
f
u
ll
c
onne
c
ti
on
la
ye
r
pr
oduc
e
s
a
n
a
c
c
ur
a
c
y
va
lu
e
of
0.9066,
th
e
lo
s
s
va
lu
e
r
e
a
c
he
s
0
.218,
a
va
li
d
va
lu
e
of
a
c
c
ur
a
c
y
of
0.8435
a
nd
th
e
va
li
d
lo
s
s
va
lu
e
i
s
0.2342.
F
ig
ur
e
11
s
how
s
f
if
th
r
e
s
ul
ts
o
f
id
e
nt
if
ic
a
ti
on
of
im
a
ge
c
ont
e
nt
id
e
nt
if
ie
d
a
s
por
nogr
a
phi
c
w
it
h t
he
C
N
N
m
ode
l.
4.1.6. S
ix
t
h
t
e
s
t
r
e
s
u
lt
s
c
on
vol
u
t
io
n
al
n
e
u
r
al
n
e
t
w
or
k
m
od
e
l
B
a
s
e
d
on
th
e
pa
r
a
m
e
te
r
s
:
i)
num
be
r
of
c
onvolut
io
n
la
ye
r
s
=
2;
ii
)
le
a
r
ni
ng
r
a
te
=
0.0001;
ii
i)
e
poc
hs
=
50;
iv
)
s
te
p
s
pe
r
e
po
c
h=
100;
a
nd
v)
a
c
ti
va
ti
on
f
u
nc
ti
on
a
t
th
e
f
ul
l
c
onne
c
ti
on
l
a
ye
r
=
R
e
L
u
a
nd
s
ig
m
oi
d.
U
s
in
g
th
e
pa
r
a
m
e
te
r
num
be
r
of
c
onvolut
io
n
la
y
e
r
s
i
s
2,
th
e
le
a
r
ni
ng
r
a
te
is
0.0001,
th
e
e
poc
h
is
50,
th
e
s
te
ps
pe
r
e
poc
h i
s
100 a
nd t
he
R
e
L
u a
c
ti
va
ti
on f
unc
ti
on on t
he
f
ul
l
c
onne
c
ti
on l
a
ye
r
pr
oduc
e
s
a
n a
c
c
ur
a
c
y
va
lu
e
of
0.8741,
w
hi
le
th
e
lo
s
s
va
lu
e
r
e
a
c
he
s
0.3001,
va
li
d
a
c
c
ur
a
c
y
ha
s
a
va
lu
e
of
0.7673
a
nd
a
va
li
d
lo
s
s
va
lu
e
of
0.4083.
F
ig
ur
e
12
s
how
s
s
ix
th
r
e
s
ul
ts
of
id
e
nt
if
ic
a
ti
on
of
im
a
ge
c
ont
e
nt
id
e
nt
if
ie
d
a
s
por
nogr
a
phi
c
w
it
h t
he
C
N
N
m
ode
l.
F
ig
ur
e
11. F
if
th
r
e
s
ul
ts
of
i
de
nt
if
ic
a
ti
on of
i
m
a
ge
c
ont
e
nt
i
de
nt
if
ie
d a
s
por
nogr
a
phi
c
w
it
h t
he
C
N
N
m
ode
l
F
ig
ur
e
12. S
ix
th
r
e
s
ul
ts
of
i
de
nt
if
ic
a
ti
on of
i
m
a
ge
c
ont
e
nt
i
de
nt
if
ie
d a
s
por
nogr
a
phi
c
w
it
h t
he
C
N
N
m
ode
l
4.1.7.
S
e
ve
n
t
h
t
e
s
t
r
e
s
u
lt
s
c
on
vol
u
t
io
n
al
n
e
u
r
al
n
e
t
w
or
k
m
o
d
e
l
B
a
s
e
d on the
pa
r
a
m
e
te
r
s
:
i
)
numbe
r
of
c
onvolut
io
n
l
a
ye
r
s
=
3;
ii
)
l
e
a
r
ni
ng r
a
te
=
0.0001;
iii
)
e
poc
h=
25;
iv
)
s
te
ps
p
e
r
e
poc
h
=
100;
a
nd
v
)
a
c
ti
va
ti
on
f
unc
ti
on a
t
th
e
f
ul
l
c
onne
c
ti
on
la
ye
r
=
R
e
L
u
a
nd
s
ig
m
oi
d
.
U
s
in
g
th
e
pa
r
a
m
e
te
r
num
be
r
of
c
onvolut
io
n
la
ye
r
s
is
3,
th
e
le
a
r
ni
ng
r
a
te
is
0.0001,
th
e
e
poc
h
is
25,
th
e
s
te
ps
pe
r
e
poc
h
is
100
a
nd
th
e
R
e
L
u
a
c
ti
va
ti
on
f
unc
ti
on
on
th
e
f
ul
l
c
onn
e
c
ti
on
la
ye
r
pr
oduc
e
s
a
n
a
c
c
ur
a
c
y
va
lu
e
of
0.8331,
th
e
lo
s
s
va
lu
e
i
s
0.3752,
va
li
d
a
c
c
ur
a
c
y
ha
s
th
e
va
lu
e
is
0.8254
,
a
nd
th
e
va
li
d
lo
s
s
va
lu
e
is
0.6289.
F
ig
ur
e
1
3
s
ho
w
s
s
e
ve
nt
h r
e
s
ul
ts
of
i
de
nt
if
ic
a
ti
on of
i
m
a
ge
c
ont
e
nt
i
de
nt
if
ie
d a
s
p
or
nogr
a
phi
c
w
it
h t
he
C
N
N
m
ode
l.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
, V
ol
.
14
, N
o.
3
,
J
une
20
25
:
1884
-
1899
1892
4.1.8. E
ig
h
t
h
t
e
s
t
r
e
s
u
lt
s
c
on
vol
u
t
io
n
al
n
e
u
r
al
n
e
t
w
or
k
m
od
e
l
B
a
s
e
d
on
th
e
p
a
r
a
m
e
te
r
s
:
i)
num
be
r
of
c
onvolut
io
n
la
ye
r
s
=
3;
ii
)
le
a
r
ni
ng
r
a
te
=
0.001;
ii
i)
e
poc
h
s
=
50;
iv
)
s
te
ps
p
e
r
e
poc
h
=
100;
a
nd
v)
a
c
ti
va
ti
on
f
unc
ti
on a
t
th
e
f
ul
l
c
onne
c
ti
on
la
ye
r
=
R
e
L
u
a
nd s
ig
m
oi
d.
U
s
in
g
th
e
pa
r
a
m
e
te
r
num
be
r
of
c
onvolut
io
n
la
y
e
r
s
is
3,
t
he
le
a
r
ni
ng
r
a
te
ha
s
a
va
lu
e
of
0.001,
e
po
c
h
ha
s
a
va
lu
e
of
50,
s
te
ps
pe
r
e
poc
h
ha
s
a
va
lu
e
of
100
a
nd
s
e
le
c
ti
ng
th
e
R
e
L
u
a
c
ti
va
ti
on
f
unc
ti
on
in
th
e
f
u
ll
c
onne
c
ti
on
la
ye
r
pr
oduc
e
s
a
n a
c
c
ur
a
c
y va
lu
e
of
0.9487, los
s
r
e
a
c
he
s
a
va
lu
e
of
0.1787, va
li
d a
c
c
ur
a
c
y
is
0.8831, a
nd va
li
d l
os
s
va
lu
e
is
0.6308.
F
ig
ur
e
14
s
how
s
e
ig
ht
h
r
e
s
ul
ts
of
id
e
nt
if
ic
a
ti
on
of
im
a
ge
c
ont
e
nt
id
e
nt
if
ie
d
a
s
por
nogr
a
phi
c
w
it
h t
he
C
N
N
m
ode
l.
F
ig
ur
e
13. S
e
ve
nt
h r
e
s
ul
ts
of
i
de
nt
if
ic
a
ti
on of
im
a
ge
c
ont
e
nt
i
de
nt
if
ie
d a
s
por
nogr
a
phi
c
w
it
h t
he
C
N
N
m
ode
l
F
ig
ur
e
14. E
ig
ht
h r
e
s
ul
ts
of
i
de
nt
if
ic
a
ti
on of
im
a
ge
c
ont
e
nt
i
de
nt
if
ie
d a
s
por
nogr
a
phi
c
w
it
h t
he
C
N
N
m
ode
l
4.1.9.
N
in
t
h
t
e
s
t
r
e
s
u
lt
s
c
on
vol
u
t
io
n
al
n
e
u
r
al
n
e
t
w
or
k
m
od
e
l
B
a
s
e
d
on
th
e
pa
r
a
m
e
te
r
s
:
i
)
num
be
r
of
c
onvolut
io
n
la
ye
r
s
=
3;
ii
)
le
a
r
ni
ng
r
a
te
=
0.001;
iii
)
e
poc
h=
25;
iv
)
s
te
ps
p
e
r
e
poc
h
=
100;
a
nd
v
)
a
c
ti
va
ti
on
f
unc
ti
on a
t
th
e
f
ul
l
c
onne
c
ti
on
la
ye
r
=
R
e
L
u
a
nd
s
ig
m
oi
d
.
U
s
in
g
th
e
pa
r
a
m
e
te
r
s
of
th
e
num
be
r
of
c
onvolut
io
n
la
ye
r
s
to
ta
li
ng
3,
le
a
r
ni
ng
r
a
te
w
it
h
a
va
lu
e
of
0.001,
e
poc
h
w
it
h
a
va
lu
e
of
25, s
te
ps
pe
r
e
poc
h w
it
h a
va
lu
e
of
100
a
nd t
he
R
e
L
u
a
c
ti
va
ti
on f
unc
ti
on on the
f
ul
l
c
onne
c
ti
on l
a
ye
r
pr
oduc
e
s
a
n
a
c
c
ur
a
c
y
v
a
lu
e
of
0.8122,
th
e
lo
s
s
va
lu
e
r
e
a
c
he
s
0.
417,
th
e
va
li
d
a
c
c
ur
a
c
y
va
lu
e
r
e
a
c
he
s
0.7914
,
a
nd va
li
d l
os
s
ha
s
a
va
lu
e
of
0.561.
F
ig
u
r
e
1
5
s
how
s
ni
nt
h
r
e
s
ul
ts
of
i
de
nt
if
ic
a
ti
on o
f
i
m
a
ge
c
ont
e
nt
i
de
nt
i
f
ie
d
a
s
por
nogr
a
phi
c
w
it
h t
he
C
N
N
m
ode
l.
F
ig
ur
e
1
5
. N
in
th
r
e
s
ul
ts
of
i
de
nt
if
ic
a
ti
on of
i
m
a
ge
c
ont
e
nt
i
de
nt
if
ie
d a
s
por
nogr
a
phi
c
w
it
h t
he
C
N
N
m
ode
l
4.2. Re
s
u
lt
s
of
id
e
n
t
if
ic
at
io
n
of
i
m
age
c
on
t
e
n
t
i
d
e
n
t
if
ie
d
as
p
or
n
ogr
ap
h
ic
w
it
h
t
h
e
V
G
G
-
16 m
od
e
l
V
G
G
m
ode
l
tr
a
in
in
g
is
c
a
r
r
ie
d
out
by
te
s
ti
ng
us
in
g
di
f
f
e
r
e
nt
e
poc
h
va
r
ia
bl
e
va
lu
e
s
,
di
f
f
e
r
e
nt
a
c
ti
va
ti
on
f
unc
ti
ons
,
di
f
f
e
r
e
nt
le
a
r
ni
ng
r
a
te
s
,
s
o
th
a
t
th
e
be
s
t
r
e
s
ul
ts
a
r
e
o
bt
a
in
e
d
f
r
om
th
e
m
ode
l
th
a
t
ha
s
be
e
n
bui
lt
,
a
s
f
ol
lo
w
s
:
4.
2
.1.
F
ir
s
t
t
e
s
t
r
e
s
u
lt
s
of
V
G
G
-
16 m
od
e
l
T
he
t
r
ia
l
w
a
s
c
a
r
r
ie
d out ba
s
e
d on the
pa
r
a
m
e
te
r
s
:
i
)
l
e
a
r
ni
ng r
a
t
e
=
0.001;
ii
)
e
poc
h
=
100;
iii
)
s
te
p
s
pe
r
e
poc
h=
100;
a
nd
iv
)
a
c
ti
va
ti
on
f
unc
ti
on
a
t
th
e
f
ul
l
c
onne
c
ti
on
la
ye
r
=
R
e
L
u,
R
e
L
u,
a
nd
s
of
tm
a
x
.
U
s
in
g
th
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
ti
f
I
nt
e
ll
I
S
S
N
:
2252
-
8938
C
om
par
is
on of de
e
p l
e
ar
ni
ng m
ode
l
s
:
C
N
N
and V
G
G
-
16 i
n i
de
nt
if
y
in
g por
nogr
aphi
c
…
(
R
e
z
a C
hand
r
a)
1893
le
a
r
ni
ng
r
a
te
pa
r
a
m
e
te
r
s
of
0.001,
e
poc
h
of
100,
s
te
ps
p
e
r
e
p
oc
h
of
100
a
nd
th
e
R
e
L
u,
R
e
L
u,
a
nd
s
of
tm
a
x
a
c
ti
va
ti
on f
unc
ti
ons
on t
he
f
ul
l
c
onne
c
ti
on l
a
ye
r
pr
oduc
e
s
a
n a
c
c
ur
a
c
y va
lu
e
of
0.5347, a
va
li
d a
c
c
ur
a
c
y v
a
lu
e
of
0.4375, a
l
os
s
va
lu
e
of
0.6909
,
a
nd
th
e
va
li
d l
o
s
s
v
a
lu
e
a
l
s
o r
e
a
c
he
d 0.7083.
F
ig
ur
e
1
6
s
how
s
f
ir
s
t
r
e
s
ul
t
s
of
id
e
nt
if
ic
a
ti
on of
i
m
a
ge
c
ont
e
nt
i
de
nt
if
ie
d a
s
por
nogr
a
phi
c
w
it
h t
he
V
G
G
-
16 mode
l.
F
ig
ur
e
16. F
ir
s
t
r
e
s
ul
ts
of
id
e
nt
if
ic
a
ti
on of
i
m
a
ge
c
ont
e
nt
i
de
nt
if
ie
d a
s
por
nogr
a
phi
c
w
it
h t
he
V
G
G
-
16
m
ode
l
4.2.2. S
e
c
on
d
t
e
s
t
r
e
s
u
lt
s
T
he
t
r
ia
l
w
a
s
c
a
r
r
ie
d out ba
s
e
d on th
e
pa
r
a
m
e
te
r
s
:
i)
l
e
a
r
ni
ng r
a
te
=
0.001; i
i)
e
poc
hs
=
50;
i
ii
)
s
te
p
s
pe
r
e
poc
h=
30;
a
nd
iv
)
a
c
ti
va
ti
on
f
unc
ti
on
a
t
th
e
f
ul
l
c
onne
c
ti
on
l
a
ye
r
=
R
e
L
u,
R
e
L
u,
a
nd
s
ig
m
oi
d.
U
s
in
g
th
e
le
a
r
ni
ng
r
a
te
pa
r
a
m
e
te
r
s
w
e
r
e
0.001,
e
poc
h
w
a
s
50
,
s
te
ps
p
e
r
e
poc
h
w
e
r
e
30
a
nd
th
e
R
e
L
u,
R
e
L
u,
a
nd
s
ig
m
oi
d
a
c
ti
va
ti
on
f
unc
ti
ons
i
n t
he
f
ul
l
c
onne
c
ti
on l
a
ye
r
pr
oduc
e
d a
n a
c
c
ur
a
c
y va
l
ue
of
0.5198, a
va
li
d a
c
c
ur
a
c
y v
a
lu
e
of
0.4844,
a
lo
s
s
va
lu
e
of
7.74,
a
nd
a
va
li
d
lo
s
s
va
lu
e
of
7.0517.
F
ig
ur
e
17 s
how
s
s
e
c
o
nd
r
e
s
ul
ts
of
id
e
nt
if
ic
a
ti
on
of
im
a
ge
c
ont
e
nt
i
de
nt
if
ie
d a
s
por
nogr
a
phi
c
w
it
h t
he
V
G
G
-
16 mode
l.
F
ig
ur
e
17. S
e
c
ond r
e
s
ul
ts
of
i
de
nt
if
ic
a
ti
on of
i
m
a
ge
c
ont
e
nt
i
de
nt
if
ie
d a
s
por
nogr
a
phi
c
w
it
h t
he
V
G
G
-
16
m
ode
l
4.
2
.3.
T
h
ir
d
t
e
s
t
r
e
s
u
lt
s
T
he
t
r
ia
l
w
a
s
c
a
r
r
ie
d out ba
s
e
d on th
e
pa
r
a
m
e
te
r
s
:
i
)
l
e
a
r
ni
ng r
a
te
=
0.001;
ii
)
e
poc
hs
=
50;
iii
)
s
te
p
s
pe
r
e
poc
h=
50;
a
nd
iv
)
a
c
ti
va
ti
on f
unc
ti
on a
t
th
e
f
ul
l
c
onne
c
ti
on l
a
ye
r
=
R
e
L
u,
R
e
L
u,
a
nd
R
e
L
u
. U
s
in
g t
he
l
e
a
r
ni
ng
r
a
te
pa
r
a
m
e
te
r
s
w
e
r
e
0.001, e
poc
h w
a
s
50, s
te
p
s
pe
r
e
poc
h w
e
r
e
30 a
nd t
he
R
e
L
u
,
R
e
L
u,
a
nd
R
e
L
u
a
c
ti
va
ti
on
f
unc
ti
ons
on
th
e
f
ul
l
c
onne
c
ti
on
la
ye
r
p
r
oduc
e
d
a
n
a
c
c
ur
a
c
y
va
lu
e
of
0.4512,
a
va
li
d
a
c
c
ur
a
c
y
va
lu
e
of
0.5188
,
a
lo
s
s
va
lu
e
of
8.8448
,
a
nd
a
v
a
li
d
lo
s
s
v
a
lu
e
of
7.5554.
F
ig
ur
e
1
8
s
how
s
th
ir
d
r
e
s
ul
ts
of
id
e
nt
if
ic
a
ti
on
of
im
a
ge
c
ont
e
nt
i
de
nt
if
ie
d a
s
por
nogr
a
phi
c
w
it
h t
he
V
G
G
-
16 mode
l.
4.2.4. F
ou
r
t
h
t
e
s
t
r
e
s
u
lt
s
T
he
tr
ia
l
w
a
s
c
a
r
r
ie
d
out
ba
s
e
d
on
th
e
pa
r
a
m
e
te
r
s
:
i)
le
a
r
ni
ng
r
a
te
=
0.0001;
ii
)
e
poc
h
s
=
50;
ii
i)
s
te
p
s
pe
r
e
poc
h=
50;
a
nd i
v)
a
c
ti
va
ti
on f
unc
ti
on a
t
th
e
f
ul
l
c
onne
c
ti
on l
a
ye
r
=
R
e
L
u, R
e
L
u, a
nd R
e
L
u. U
s
in
g t
he
l
e
a
r
ni
ng
r
a
te
pa
r
a
m
e
te
r
s
w
e
r
e
0.0001,
e
poc
h
w
a
s
50,
s
te
ps
p
e
r
e
poc
h
w
e
r
e
50
a
nd
th
e
R
e
L
u,
R
e
L
u,
a
nd
R
e
L
u
a
c
ti
va
ti
on
f
unc
ti
ons
i
n t
he
f
ul
l
c
onne
c
ti
on l
a
ye
r
pr
oduc
e
d a
n a
c
c
ur
a
c
y va
l
ue
of
0.4674, a
va
li
d a
c
c
ur
a
c
y v
a
lu
e
of
0.5437,
th
e
lo
s
s
va
lu
e
w
a
s
1.3211,
a
nd
th
e
va
li
d
lo
s
s
va
lu
e
i
s
3.5258.
F
ig
ur
e
19
s
how
s
f
our
th
r
e
s
ul
ts
of
id
e
nt
i
f
ic
a
ti
on
of
i
m
a
g
e
c
ont
e
nt
i
de
nt
if
ie
d a
s
por
nogr
a
phi
c
w
it
h t
he
V
G
G
-
16
m
ode
l.
Evaluation Warning : The document was created with Spire.PDF for Python.