I
AE
S In
t
er
na
t
io
na
l J
o
urna
l o
f
Art
if
icia
l In
t
ellig
ence
(
I
J
-
AI
)
Vo
l.
15
,
No
.
5
,
Octo
b
er
20
26
,
p
p
.
4
0
1
5
~
4
0
2
5
I
SS
N:
2
2
5
2
-
8
9
3
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijai.v
15
.i
5
.
p
p
4
0
1
5
-
4
0
2
5
4015
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
a
i
.
ia
esco
r
e.
co
m
Cla
ss
ifying
Arabic ha
ndwri
ting usi
ng
conv
o
lutiona
l neural
network
Ana
s
Q
ute
is
ha
t
1
,2
,
Ash
ra
f
Al
Sh
a
ra
h
2
,
Ahm
ed
Q
t
a
is
ha
t
3
,
T
a
re
q Ahm
a
d Ala
wneh
2
1
D
e
p
a
r
t
me
n
t
o
f
C
o
mm
u
n
i
c
a
t
i
o
n
a
n
d
C
o
m
p
u
t
e
r
E
n
g
i
n
e
e
r
i
n
g
,
F
a
c
u
l
t
y
o
f
E
n
g
i
n
e
e
r
i
n
g
,
A
l
-
A
h
l
i
y
y
a
A
mm
a
n
U
n
i
v
e
r
si
t
y
,
A
mm
a
n
,
J
o
r
d
a
n
2
D
e
p
a
r
t
me
n
t
o
f
El
e
c
t
r
i
c
a
l
En
g
i
n
e
e
r
i
n
g
,
F
a
c
u
l
t
y
o
f
E
n
g
i
n
e
e
r
i
n
g
Te
c
h
n
o
l
o
g
y
,
A
l
-
B
a
l
q
a
A
p
p
l
i
e
d
U
n
i
v
e
r
si
t
y
,
A
l
-
S
a
l
t
,
Jo
r
d
a
n
3
F
o
u
n
d
a
t
i
o
n
P
r
o
g
r
a
m,
S
o
h
a
r
U
n
i
v
e
r
si
t
y
,
S
o
h
a
r
,
O
ma
n
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Sep
2
2
,
2
0
2
5
R
ev
is
ed
J
u
l
26
,
2
0
2
6
Acc
ep
ted
Au
g
12
,
2
0
2
6
Th
is
p
a
p
e
r
p
re
se
n
ts
a
c
las
sifica
ti
o
n
o
f
Ara
b
ic
h
a
n
d
wri
tt
e
n
lett
e
rs
u
sin
g
c
o
n
v
o
lu
ti
o
n
a
l
n
e
u
ra
l
n
e
two
r
k
s
(CNN
s).
Th
e
p
a
p
e
r
a
ims
to
a
d
d
re
ss
th
e
u
n
i
q
u
e
c
h
a
ll
e
n
g
e
s
o
f
re
c
o
g
n
izi
n
g
Ara
b
ic
letters
d
u
e
to
th
e
i
r
sp
e
c
ial
c
h
a
ra
c
teristics
su
c
h
a
s
d
o
ts
a
n
d
sim
il
a
rit
ies
.
Lev
e
ra
g
i
n
g
d
e
e
p
lea
rn
in
g
,
p
a
rti
c
u
larly
CNN
s,
th
e
p
r
o
jec
t
e
m
p
lo
y
s
a
n
a
rc
h
it
e
c
tu
re
sp
e
c
ifi
c
a
ll
y
d
e
si
g
n
e
d
fo
r
ima
g
e
c
las
sifica
ti
o
n
tas
k
s,
wh
ich
is
we
ll
-
s
u
it
e
d
f
o
r
o
p
t
ica
l
c
h
a
ra
c
ter
re
c
o
g
n
it
i
o
n
(OCR)
o
f
Ara
b
ic
h
a
n
d
writt
e
n
letters
.
Th
e
d
a
tas
e
t
u
se
d
c
o
n
sists
o
f
ima
g
e
s
c
o
ll
e
c
ted
fr
o
m
t
h
e
"
Ara
b
ic
h
a
n
d
writt
e
n
a
lp
h
a
b
e
ts,
wo
rd
s,
a
n
d
p
a
ra
g
ra
p
h
s
"
(AH
AWP
)
a
n
d
"
Ara
b
ic
h
a
n
d
writt
e
n
c
h
a
ra
c
ter
d
a
tab
a
se
"
(AH
CDB),
wh
ich
h
a
v
e
b
e
e
n
c
a
re
fu
ll
y
p
r
o
c
e
ss
e
d
to
e
n
s
u
re
u
n
if
o
rm
it
y
a
n
d
d
iv
e
rsit
y
.
T
h
e
m
o
d
e
l
wa
s
b
u
il
t
u
sin
g
d
e
e
p
n
e
two
r
k
d
e
sig
n
e
r
,
a
to
o
l
in
M
ATLAB,
a
n
d
train
e
d
o
v
e
r
th
re
e
e
p
o
c
h
s to
p
re
v
e
n
t
o
v
e
rfit
ti
n
g
,
a
c
h
iev
i
n
g
a
n
a
c
c
u
ra
c
y
o
f
8
9
.
2
%
.
K
ey
w
o
r
d
s
:
Ar
ab
ic
h
an
d
wr
itin
g
Ar
tific
ial
n
eu
r
al
n
etwo
r
k
C
o
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
s
Dee
p
lear
n
in
g
Op
tical
ch
ar
ac
ter
r
ec
o
g
n
itio
n
T
h
is i
s
a
n
o
p
e
n
a
c
c
e
ss
a
rticle
u
n
d
e
r th
e
CC B
Y
-
SA
li
c
e
n
se
.
C
o
r
r
e
s
p
o
nd
ing
A
uth
o
r
:
An
as Qu
teis
h
at
Dep
ar
tm
en
t o
f
C
o
m
m
u
n
icatio
n
an
d
C
o
m
p
u
ter
E
n
g
in
ee
r
i
n
g
,
Facu
lty
o
f
E
n
g
in
ee
r
in
g
Al
-
Ah
liy
y
a
Am
m
an
Un
i
v
er
s
ity
Am
m
an
,
J
o
r
d
a
n
E
m
ail:
a.
q
u
teis
h
at@
am
m
an
u
.
e
d
u
.
jo
1.
I
NT
RO
D
UCT
I
O
N
T
h
r
o
u
g
h
o
u
t
th
e
last
1
0
0
y
ea
r
s
,
p
eo
p
le
h
a
v
e
s
tr
iv
ed
to
in
s
till
r
ea
d
in
g
ab
ilit
ies
in
to
m
ac
h
in
es
th
r
o
u
g
h
v
ar
io
u
s
m
eth
o
d
s
wh
ich
p
r
o
d
u
ce
d
d
if
f
er
e
n
t
lev
els
o
f
s
u
cc
ess
.
Ma
ch
in
e
r
ec
o
g
n
itio
n
o
f
h
an
d
wr
itin
g
r
em
ain
s
an
elu
s
iv
e
task
d
u
e
to
its
d
em
a
n
d
in
g
al
g
o
r
ith
m
s
wh
ich
m
ak
es
it
o
n
e
o
f
th
e
m
o
s
t
ad
v
a
n
ce
d
im
ag
e
r
ec
o
g
n
itio
n
p
r
o
b
lem
s
.
T
h
e
p
r
im
a
r
y
g
o
a
l
o
f
th
is
p
r
o
ce
s
s
in
v
o
lv
es
m
ac
h
in
e
r
ea
d
in
g
ab
ilit
y
,
b
u
t
th
e
in
itial
le
tter
class
if
icatio
n
s
y
s
tem
s
s
er
v
ed
as
th
e
f
o
u
n
d
atio
n
t
o
d
e
v
elo
p
a
d
v
an
ce
d
r
ea
d
in
g
ca
p
a
b
ilit
ies.
T
h
e
p
r
o
ce
s
s
k
n
o
wn
as
o
p
tical
ch
ar
ac
ter
r
ec
o
g
n
itio
n
(
OC
R
)
en
ab
les
th
e
au
to
m
ati
c
co
n
v
e
r
s
io
n
o
f
p
r
in
ted
o
r
h
an
d
wr
itten
tex
t
f
o
u
n
d
in
im
ag
es
to
d
ig
ital
f
o
r
m
s
th
a
t
p
er
m
it
u
s
er
s
to
s
ea
r
ch
,
m
o
d
i
f
y
,
s
to
r
e
co
m
p
ac
tly
,
p
r
esen
t
o
n
lin
e,
an
d
ex
ec
u
te
m
ac
h
in
e
-
b
ased
p
r
o
ce
s
s
in
g
th
r
o
u
g
h
co
g
n
itiv
e
co
m
p
u
tin
g
s
y
s
tem
s
.
OC
R
tr
an
s
f
o
r
m
s
tex
t
d
ata
m
an
ag
em
en
t
s
in
ce
it
allo
ws
q
u
ic
k
a
n
d
p
r
e
cise
co
n
v
er
s
io
n
o
f
v
ar
io
u
s
d
o
cu
m
en
t
t
y
p
es
in
to
d
i
g
ital
f
o
r
m
ats
th
r
o
u
g
h
wh
ich
u
s
er
s
ca
n
ac
h
iev
e
ap
p
licatio
n
s
lik
e
d
o
cu
m
e
n
t
wo
r
d
s
ea
r
ch
,
au
to
m
ated
p
o
s
tal
m
ail
s
o
r
ti
n
g
,
an
d
d
o
cu
m
e
n
t
ed
itin
g
f
r
o
m
p
r
in
ted
f
o
r
m
.
M
o
d
er
n
in
d
u
s
tr
y
ap
p
licatio
n
s
alo
n
g
s
id
e
ev
er
y
d
ay
s
y
s
tem
s
h
ea
v
ily
r
ely
o
n
OC
R
tech
n
o
lo
g
y
b
ec
a
u
s
e
ar
tific
ial
in
tellig
en
ce
an
d
m
ac
h
in
e
lear
n
in
g
c
o
n
tin
u
e
to
d
ev
elo
p
th
e
ir
ca
p
ab
ilit
ies
[
1
]
.
OC
R
en
ab
les
u
s
to
tr
an
s
late
tex
t
f
r
o
m
o
n
e
lan
g
u
a
g
e
to
a
n
o
th
er
u
s
in
g
tr
a
n
s
latio
n
to
o
ls
,
en
a
b
les
tr
af
f
ic
s
y
s
tem
s
to
r
ec
o
g
n
ize
an
d
r
ec
o
r
d
v
eh
icl
e
licen
s
e
p
lates,
an
d
en
ab
les
e
m
p
lo
y
ee
s
to
s
av
e
tim
e
an
d
ef
f
o
r
t
wasted
o
n
d
ata
en
tr
y
p
r
o
ce
s
s
es
[
2
]
–
[
5
]
.
W
h
il
e
OC
R
h
a
s
m
an
y
ap
p
licatio
n
s
an
d
b
en
ef
its
,
it
also
f
ac
es
s
ev
er
al
ch
allen
g
es
an
d
p
r
o
b
lem
s
.
T
e
x
t
in
d
o
cu
m
e
n
ts
with
co
m
p
lex
lay
o
u
ts
,
f
o
n
ts
,
o
r
p
o
o
r
-
q
u
ality
im
ag
es
is
d
if
f
icu
lt
to
b
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
15
,
No
.
5
,
Octo
b
er
20
26
:
4
0
1
5
-
4
0
2
5
4016
r
ec
o
g
n
ized
.
Han
d
lin
g
m
u
ltip
l
e
lan
g
u
ag
es,
v
ar
i
o
u
s
f
o
n
t
s
ty
les
an
d
h
an
d
wr
itten
tex
t
c
an
b
e
ch
allen
g
in
g
.
Als
o
,
OC
R
ca
n
b
e
af
f
ec
ted
b
y
b
ac
k
g
r
o
u
n
d
n
o
is
e
o
r
clu
tte
r
,
wh
ich
ca
n
in
te
r
f
er
e
with
th
e
tex
t
r
ec
o
g
n
itio
n
p
r
o
ce
s
s
.
Mo
r
e
o
v
er
,
i
m
p
lem
e
n
tin
g
OC
R
ca
n
b
e
ex
p
en
s
iv
e
an
d
co
m
p
lex
,
r
eq
u
ir
i
n
g
s
p
ec
ialized
h
ar
d
war
e
,
s
o
f
twar
e,
an
d
e
x
p
er
tis
e.
T
h
is
p
ap
er
f
o
c
u
s
es
o
n
b
u
ild
in
g
a
s
y
s
tem
to
r
ec
o
g
n
ize
h
an
d
wr
itten
b
asic
Ar
ab
ic
letter
s
,
w
h
ich
co
u
ld
b
e
v
er
y
u
s
ef
u
l
to
d
ev
elo
p
an
Ar
ab
ic
wr
itin
g
to
o
l
as
an
alter
n
ativ
e
to
th
e
k
ey
b
o
ar
d
,
as
s
o
m
e
p
eo
p
le
s
till
p
r
ef
er
to
wr
ite
b
y
h
an
d
.
Han
d
wr
itten
Ar
ab
ic
tex
t
h
as
u
n
iq
u
e
tech
n
ical
ch
allen
g
es,
wh
ich
h
av
e
b
ee
n
m
o
r
e
r
ec
e
n
tly
ad
d
r
ess
ed
.
Nu
m
er
o
u
s
im
ag
e
ty
p
es
h
av
e
b
ee
n
th
e
s
u
b
ject
o
f
d
if
f
er
en
t
p
r
o
p
o
s
ed
an
d
ap
p
li
ed
m
eth
o
d
s
.
T
h
er
e
ar
e
2
8
letter
s
in
th
e
Ar
ab
ic
al
p
h
ab
et,
an
d
1
5
o
f
th
em
h
av
e
v
ar
ied
s
h
ap
es.
T
h
e
s
h
ap
e
o
f
s
o
m
e
ch
ar
ac
ter
s
is
s
im
ilar
,
b
u
t
th
e
d
if
f
er
en
ce
ar
is
es
wi
th
th
e
p
o
s
itio
n
an
d
n
u
m
b
er
o
f
d
o
ts
,
wh
eth
er
ab
o
v
e
o
r
b
elo
w
th
e
ch
ar
ac
ter
s
.
Ad
d
itio
n
ally
,
a
g
r
o
u
p
o
f
two
p
o
in
ts
ca
n
b
e
wr
i
tten
as
o
n
e
o
r
two
co
n
n
ec
ted
co
m
p
o
n
en
ts
a
n
d
a
g
r
o
u
p
o
f
th
r
ee
p
o
i
n
ts
ca
n
b
e
wr
itten
as
o
n
e,
two
,
o
r
th
r
ee
co
n
n
ec
ted
c
o
m
p
o
n
en
ts
,
d
ep
e
n
d
in
g
o
n
th
e
s
ty
le
o
f
wr
itin
g
.
I
t
ca
n
b
e
co
n
s
id
er
e
d
th
at
r
ec
o
g
n
itio
n
p
r
o
ce
s
s
in
clu
d
es
th
r
ee
g
en
er
al
s
tag
es.
First
ly
,
p
r
e
-
r
ec
o
g
n
itio
n
,
wh
ich
in
v
o
lv
es
im
ag
e
p
r
o
ce
s
s
in
g
to
en
h
an
ce
th
e
q
u
ality
o
f
th
e
im
ag
e
to
m
ak
e
it
s
u
itab
le
f
o
r
OC
R
.
I
t
in
clu
d
es
n
o
is
e
r
ed
u
ctio
n
,
b
in
ar
izatio
n
,
s
k
ew
co
r
r
ec
tio
n
,
an
d
n
o
r
m
aliza
tio
n
.
Seco
n
d
ly
,
r
ec
o
g
n
itio
n
its
elf
,
th
e
co
r
e
o
f
OC
R
wh
er
e
th
e
ac
tu
al
r
ec
o
g
n
itio
n
o
f
c
h
ar
ac
ter
s
tak
es
p
la
ce
.
T
h
is
ca
n
b
e
d
o
n
e
u
s
in
g
o
n
e
o
f
two
p
r
im
a
r
y
m
eth
o
d
s
:
p
atter
n
r
ec
o
g
n
itio
n
,
wh
er
e
th
e
s
y
s
tem
c
o
m
p
ar
es
th
e
tex
t
i
n
th
e
im
ag
e
with
a
d
atab
ase
o
f
k
n
o
wn
p
atter
n
s
,
an
d
f
ea
tu
r
e
ex
t
r
ac
tio
n
,
wh
er
e
th
e
s
y
s
tem
d
ec
o
m
p
o
s
es
th
e
ch
ar
ac
ter
s
in
to
f
ea
tu
r
e
s
(
lik
e
lin
es,
lo
o
p
s
,
an
d
in
te
r
s
ec
tio
n
s
)
an
d
m
atc
h
e
s
th
em
to
k
n
o
wn
ch
ar
ac
ter
f
e
atu
r
es.
I
n
th
e
f
in
al
s
tag
e,
th
e
er
r
o
r
s
a
r
e
co
r
r
ec
ted
,
th
is
s
tep
in
v
o
lv
es
u
n
d
e
r
s
tan
d
i
n
g
th
e
c
o
n
tex
t
an
d
tex
t
is
f
o
r
m
atted
in
to
r
ea
d
ab
le
tex
t
[
6
]
.
Faizu
llah
et
a
l.
[
4
]
p
r
esen
ted
th
e
f
ir
s
t
o
m
n
i
f
o
n
t
OC
R
;
th
is
s
y
s
tem
co
u
ld
r
ec
o
g
n
ize
s
tan
d
ar
d
f
o
n
ts
.
W
h
en
d
ea
lin
g
with
Ar
a
b
ic
OC
R
,
th
e
s
y
s
tem
n
ee
d
s
to
r
ec
o
g
n
ize
cu
r
s
iv
e
s
cr
ip
ts
,
wh
ich
is
a
m
o
r
e
co
m
p
lex
p
r
o
b
lem
,
wh
er
e
d
ea
lin
g
with
letter
s
h
ap
er
is
m
o
r
e
ad
v
an
ce
d
.
Kasem
et
a
l.
[
7
]
was
ap
p
lied
th
e
OC
R
to
Ar
ab
ic
n
ewsp
ap
er
s
h
ea
d
lin
es;
th
is
is
d
o
n
e
b
e
an
aly
zi
n
g
s
u
b
-
wo
r
d
s
.
Al
-
Yo
u
s
ef
i
an
d
Ud
p
a
[
8
]
u
s
ed
B
ay
esian
class
if
icatio
n
to
is
o
late
ch
ar
ac
ter
s
th
r
o
u
g
h
m
o
m
en
ts
p
r
o
jectio
n
.
W
ith
th
e
ad
v
an
ce
s
in
d
ee
p
l
ea
r
n
in
g
A
r
ab
ic
OC
R
was
ap
p
lied
u
s
in
g
th
ese
m
eth
o
d
s
,
h
en
ce
th
e
h
an
d
lin
g
o
f
co
n
tex
tu
al
co
m
p
l
ex
ity
is
m
an
ag
ed
.
L
i
et
a
l.
[
9
]
p
r
o
p
o
s
ed
an
OC
R
b
u
ilt
o
n
en
d
to
en
d
f
ar
m
wo
r
k
.
T
h
is
m
eth
o
d
u
s
ed
v
is
io
n
tr
an
s
f
o
r
m
er
s
f
o
r
h
an
d
wr
itin
g
r
ec
o
g
n
itio
n
,
th
is
s
h
o
ws
t
h
e
b
en
ef
its
o
f
s
eq
u
en
cin
g
with
o
u
t
s
eg
m
en
tin
g
th
e
wo
r
d
s
.
Mo
r
tad
i
et
a
l.
[
1
0
]
in
tr
o
d
u
ce
d
AL
NASI
KH,
th
is
m
et
h
o
d
r
elay
e
d
o
n
a
tr
an
s
f
o
r
m
er
-
b
ased
Ar
ab
ic
OC
R
s
y
s
tem
th
e
s
y
s
tem
u
s
e
s
th
e
tr
an
s
f
o
r
m
er
-
b
ased
o
p
tical
ch
ar
ac
ter
r
ec
o
g
n
itio
n
(
T
r
OC
R
)
ar
ch
itectu
r
e
f
o
r
A
r
ab
ic
tex
t.
R
ec
en
t
m
eth
o
d
s
u
s
ed
Ar
ab
ic
-
n
o
u
g
at
ex
ten
d
ed
v
is
io
n
T
r
OC
R
to
m
ar
k
d
o
w
n
ex
t
r
ac
tio
n
,
s
h
o
w
in
g
th
e
im
p
o
r
tan
ce
o
f
d
o
cu
m
en
t
-
lev
el
OC
R
s
y
s
tem
s
f
o
r
Ar
ab
ic
tex
t
u
n
d
er
s
tan
d
i
n
g
.
R
ec
en
tly
,
e
nd
-
to
-
en
d
ar
ch
itectu
r
es
co
m
b
in
in
g
o
p
tical
an
d
co
n
tex
tu
al
r
ec
o
g
n
itio
n
s
ig
n
if
ican
tly
im
p
r
o
v
e
d
th
e
p
er
f
o
r
m
an
ce
o
f
OC
R
in
Ar
ab
ic
h
an
d
wr
itin
g
[
1
1
]
.
T
h
e
in
teg
r
ati
o
n
o
f
lar
g
e
lan
g
u
a
g
e
m
o
d
els
(
L
L
Ms)
im
p
r
o
v
ed
co
n
tex
tu
al
u
n
d
er
s
tan
d
i
n
g
in
OC
R
[
1
2
]
.
T
h
e
m
ain
c
o
n
tr
ib
u
tio
n
o
f
th
i
s
p
ap
er
is
to
b
u
ild
a
b
asic
Ar
ab
ic
letter
s
d
ataset
an
d
p
r
ep
ar
e
it
f
o
r
tr
ain
in
g
a
co
n
v
o
lu
tio
n
al
n
e
u
r
a
l
n
etwo
r
k
(
C
NN)
[
1
3
]
.
T
h
e
p
r
esen
ted
wo
r
k
will
cr
ea
te
an
A
r
ab
ic
letter
d
ataset
co
n
tain
in
g
1
4
,
0
0
0
Ar
a
b
ic
letter
s
o
b
tain
ed
f
r
o
m
two
d
if
f
er
e
n
t
d
atasets
to
ac
h
iev
e
h
ig
h
d
i
v
er
s
ity
.
T
h
is
p
r
o
ce
s
s
r
eq
u
ir
es
ad
d
itio
n
al
ef
f
o
r
t
in
th
e
p
r
o
ce
s
s
o
f
m
atch
in
g
th
e
d
if
f
er
en
t
ty
p
es
o
f
d
ata
f
r
o
m
d
if
f
er
en
t
s
o
u
r
ce
s
.
MA
T
L
AB
R
2
0
2
1
a
will
b
e
u
s
ed
to
p
er
f
o
r
m
m
o
s
t
o
f
th
e
im
ag
e
p
r
o
ce
s
s
in
g
u
n
less
o
th
er
s
o
f
twar
e
is
r
eq
u
ir
ed
.
Sectio
n
4
o
f
th
is
p
ap
e
r
will
p
r
esen
t
a
d
etailed
d
escr
ip
tio
n
o
f
th
is
p
r
o
ce
s
s
.
D
ee
p
n
etwo
r
k
d
esig
n
er
,
a
to
o
l
f
r
o
m
MA
T
L
AB
,
is
u
s
ed
to
b
u
ild
a
C
NN
f
r
o
m
s
cr
atch
,
c
o
n
f
i
g
u
r
e
its
lay
er
s
,
ad
ju
s
t
th
e
tr
ain
i
n
g
o
p
tio
n
s
,
an
d
th
en
tr
ain
it o
n
th
e
d
ataset.
T
h
e
r
em
in
d
er
o
f
th
e
p
a
p
er
is
o
r
g
an
ized
as
f
o
llo
ws
.
S
ec
tio
n
2
p
r
esen
ts
th
e
m
et
h
o
d
o
lo
g
y
u
s
ed
in
th
e
p
ap
er
.
Sectio
n
3
p
r
esen
ts
th
e
co
llectio
n
an
d
p
r
ep
ar
atio
n
o
f
t
h
e
d
ataset
u
s
ed
.
T
h
e
ex
p
e
r
im
en
ts
an
d
r
esu
lts
ar
e
g
iv
en
s
ec
tio
n
4
.
T
h
e
c
o
n
clu
s
io
n
o
f
th
e
wo
r
k
is
p
r
esen
ted
in
s
ec
tio
n
5
.
2.
M
E
T
H
O
D
T
h
e
p
r
o
p
o
s
ed
OC
R
s
y
s
tem
is
im
p
lem
en
ted
u
s
in
g
d
ee
p
lear
n
in
g
b
ec
au
s
e
o
f
its
wid
e
ly
s
p
r
ea
d
ap
p
licatio
n
s
as
m
en
tio
n
ed
a
b
o
v
e.
T
h
e
m
ain
ad
v
an
tag
e
o
f
d
ee
p
lear
n
in
g
is
th
at
it
au
to
m
atica
lly
ex
tr
ac
ts
th
e
f
ea
tu
r
es
with
o
u
t
an
y
h
u
m
an
s
u
p
er
v
is
io
n
.
Mo
r
e
s
p
ec
if
ic
ally
,
th
e
C
NNs
wer
e
u
s
ed
,
w
h
ich
ar
e
a
r
elativ
ely
n
ew
m
eth
o
d
[
1
4
]
,
[
1
5
]
.
C
NNs
d
o
n
o
t
r
eq
u
ir
e
h
u
m
an
in
ter
v
en
tio
n
to
ex
tr
ac
t
f
ea
tu
r
es
f
r
o
m
th
e
d
ata.
I
n
s
tead
,
f
e
atu
r
es
ar
e
lear
n
ed
d
ir
ec
tly
f
r
o
m
t
h
e
d
at
a
d
u
r
in
g
th
e
tr
ain
in
g
p
r
o
ce
s
s
,
r
ed
u
cin
g
th
e
n
ee
d
f
o
r
m
an
u
al
p
r
ep
r
o
ce
s
s
in
g
.
I
n
f
ac
t,
th
is
f
ea
tu
r
e
is
f
o
r
d
ee
p
lear
n
in
g
tech
n
i
q
u
es
in
g
en
er
al.
C
NN
s
lear
n
h
ier
ar
ch
ical
r
ep
r
esen
tatio
n
s
o
f
d
ata,
s
tar
tin
g
f
r
o
m
s
im
p
le
f
ea
tu
r
es
(
lik
e
d
o
ts
an
d
lin
es)
in
th
e
in
itial
lay
er
s
to
m
o
r
e
co
m
p
lex
f
ea
tu
r
es
(
lik
e
s
h
ap
es
an
d
o
b
jects)
in
d
ee
p
er
lay
e
r
s
.
T
h
is
allo
ws
C
N
Ns
to
ef
f
ec
tiv
ely
ca
p
tu
r
e
co
m
p
lex
p
atter
n
s
in
d
ata
lik
e
im
ag
es
an
d
v
id
e
o
s
.
Fig
u
r
e
1
p
r
esen
ts
t
h
e
d
if
f
er
e
n
t la
y
er
s
in
v
o
lv
ed
i
n
b
u
ild
in
g
a
C
NN.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
C
la
s
s
i
fyin
g
A
r
a
b
ic
h
a
n
d
w
r
itin
g
u
s
in
g
co
n
vo
lu
tio
n
a
l n
e
u
r
a
l n
etw
o
r
k
(
A
n
a
s
Qu
teis
h
a
t
)
4017
Fig
u
r
e
1
.
Hier
ar
c
h
ical
s
tr
u
ctu
r
e
o
f
C
NNs
C
NNs
also
r
eu
s
e
th
e
s
am
e
f
i
lter
s
ac
r
o
s
s
th
e
en
tire
im
ag
e,
wh
ich
cu
ts
d
o
wn
s
ig
n
if
ican
t
ly
o
n
th
e
n
u
m
b
er
o
f
p
ar
am
eter
s
th
at
n
ee
d
to
b
e
lear
n
ed
c
o
m
p
ar
e
d
to
tr
ad
itio
n
al
n
e
u
r
al
n
etwo
r
k
s
.
T
h
an
k
s
to
th
eir
p
o
o
lin
g
lay
er
s
,
t
h
ey
ca
n
r
ec
o
g
n
ize
p
atter
n
s
n
o
m
atter
w
h
er
e
th
e
y
ap
p
ea
r
i
n
th
e
im
a
g
e
a
f
ea
tu
r
e
t
h
at's
esp
ec
ially
u
s
ef
u
l
f
o
r
task
s
lik
e
o
b
ject
d
etec
tio
n
an
d
im
a
g
e
class
if
icatio
n
.
Po
o
lin
g
also
s
h
r
in
k
s
th
e
s
ize
o
f
th
e
d
ata
r
ep
r
esen
tatio
n
s
,
wh
ich
lo
wer
s
co
m
p
u
tatio
n
al
co
m
p
lex
it
y
an
d
b
o
o
s
ts
ef
f
icien
c
y
.
An
o
t
h
er
k
e
y
a
d
v
an
ta
g
e
is
p
ar
alleliza
tio
n
:
b
ec
au
s
e
o
f
h
o
w
C
NNs
ar
e
s
tr
u
ctu
r
ed
,
th
e
y
ca
n
p
r
o
ce
s
s
m
u
ltip
le
p
ix
els
at
o
n
ce
,
lead
in
g
to
f
aster
tr
ain
in
g
c
o
m
p
a
r
ed
to
m
eth
o
d
s
th
at
h
an
d
le
d
ata
s
eq
u
en
tially
.
T
h
is
b
ec
o
m
es
a
m
a
jo
r
ad
v
a
n
tag
e
wh
en
wo
r
k
in
g
with
lar
g
e
d
atasets
[
1
6
]
.
Desp
ite
th
eir
m
an
y
f
ea
tu
r
es,
C
NNs
also
f
ac
e
s
o
m
e
c
h
allen
g
es,
s
u
ch
as
th
e
n
ee
d
f
o
r
la
r
g
e
d
atasets
.
Ach
iev
in
g
h
ig
h
p
e
r
f
o
r
m
an
ce
in
C
NNs
r
eq
u
ir
es
m
ass
iv
e
d
ata
s
ets,
b
u
t
co
llectin
g
an
d
lab
elli
n
g
th
ese
d
atasets
is
o
f
ten
co
s
tly
an
d
tim
e
-
co
n
s
u
m
in
g
,
esp
ec
ially
in
s
ce
n
ar
io
s
w
h
er
e
th
e
d
ata
is
s
ca
r
ce
o
r
ex
p
en
s
iv
e
to
o
b
tain
.
A
m
eth
o
d
to
o
v
e
r
co
m
e
th
is
ch
al
len
g
e
is
d
ata
au
g
m
en
tatio
n
,
wh
ich
is
th
e
p
r
o
ce
s
s
o
f
ar
tific
ially
in
cr
ea
s
in
g
th
e
s
ize
an
d
d
iv
er
s
ity
o
f
th
e
tr
ain
i
n
g
d
ataset
b
y
s
u
b
jectin
g
it
to
s
o
m
e
tr
an
s
f
o
r
m
atio
n
s
s
u
ch
as
r
o
tatin
g
im
ag
es
b
y
s
m
all
an
g
les,
r
a
n
d
o
m
l
y
c
r
o
p
p
in
g
p
o
r
tio
n
s
o
f
im
ag
es,
f
lip
p
i
n
g
im
ag
es
alo
n
g
th
e
h
o
r
izo
n
tal
ax
is
,
an
d
ch
a
n
g
in
g
im
ag
e
b
r
ig
h
t
n
ess
an
d
co
n
tr
ast.
T
h
is
r
ed
u
ce
s
th
e
r
is
k
o
f
o
v
er
f
itti
n
g
b
y
p
r
e
v
en
tin
g
th
e
m
o
d
el
f
r
o
m
m
em
o
r
izin
g
s
p
ec
if
ic
d
ata
p
atter
n
s
.
T
h
e
q
u
a
lity
o
f
th
e
d
ata
is
also
im
p
o
r
ta
n
t,
n
o
t
ju
s
t
th
e
q
u
an
tity
,
as
C
NNs
ar
e
s
en
s
itiv
e
to
s
m
all
v
ar
iatio
n
s
in
in
p
u
t
d
ata,
wh
ich
ca
n
lea
d
to
s
ig
n
if
ican
t
ch
an
g
es
in
o
u
tp
u
t
an
d
a
f
f
ec
t
th
e
ac
cu
r
ac
y
if
th
e
im
ag
es
ar
e
b
lu
r
r
y
o
r
n
o
is
y
.
T
h
is
also
lead
s
to
o
v
er
f
itti
n
g
,
wh
er
e
th
e
m
o
d
el
lear
n
s
s
p
ec
if
ic
d
etails
o
r
n
o
is
e
i
n
th
e
tr
ain
in
g
d
ata,
r
esu
ltin
g
i
n
p
o
o
r
p
er
f
o
r
m
an
ce
o
n
u
n
s
ee
n
o
r
test
d
ata,
esp
ec
ially
wh
en
t
r
ain
ed
o
n
s
m
all
o
r
n
ar
r
o
w
d
atasets
.
Dr
o
p
o
u
t
is
a
r
eg
u
lar
izatio
n
tech
n
iq
u
e
t
h
at
r
ed
u
ce
s
o
v
er
f
itti
n
g
b
y
"d
r
o
p
p
in
g
"
(
d
is
ab
lin
g
)
a
r
an
d
o
m
s
u
b
s
et
o
f
n
eu
r
o
n
s
with
a
ce
r
tain
p
r
o
b
ab
ilit
y
d
u
r
in
g
ea
ch
tr
ain
in
g
iter
atio
n
.
T
h
is
f
o
r
ce
s
th
e
n
etwo
r
k
to
lear
n
m
o
r
e
r
o
b
u
s
t
an
d
g
en
e
r
alize
d
f
ea
tu
r
es
r
ath
er
th
an
r
e
ly
in
g
o
n
s
p
ec
if
ic
n
e
u
r
o
n
s
[
1
7
]
.
Af
ter
th
e
d
ata
co
llectio
n
p
h
ase,
an
o
th
er
ch
a
llen
g
e
ar
is
es,
wh
ich
is
th
e
h
ig
h
co
m
p
u
tatio
n
al
co
s
t.
T
r
ain
in
g
C
NN
m
o
d
els
r
eq
u
ir
es
s
ig
n
if
ican
t
co
m
p
u
tati
o
n
al
p
o
wer
d
u
e
to
th
e
lar
g
e
n
u
m
b
er
o
f
o
p
er
atio
n
s
in
v
o
l
v
ed
,
p
ar
ticu
lar
ly
wh
en
d
ea
lin
g
with
lar
g
e
d
atasets
a
n
d
d
ee
p
ar
ch
itectu
r
es.
T
h
is
ca
n
lead
to
lo
n
g
tr
ain
in
g
tim
e
s
an
d
th
e
n
ee
d
f
o
r
s
p
ec
ialized
h
ar
d
war
e,
s
u
ch
as
GPU
s
.
T
h
e
ad
v
en
t
o
f
a
d
v
an
ce
d
co
m
p
u
tin
g
tech
n
o
l
o
g
ies
s
u
ch
as
clo
u
d
co
m
p
u
tin
g
a
n
d
p
ar
allel
c
o
m
p
u
tin
g
p
r
o
v
id
es
ef
f
ec
tiv
e
way
s
t
o
o
v
er
co
m
e
th
is
c
h
allen
g
e.
An
o
th
er
s
o
lu
tio
n
is
to
u
s
e
tr
an
s
f
er
lear
n
in
g
tech
n
iq
u
es
to
im
p
r
o
v
e
p
er
f
o
r
m
an
ce
u
s
in
g
p
r
e
-
tr
ai
n
ed
m
o
d
els
[
1
8
]
.
T
h
er
e
ar
e
also
s
o
m
e
less
s
ig
n
if
ican
t
ch
allen
g
es
s
u
ch
as
th
at
C
NNs
ar
e
o
f
ten
c
o
n
s
id
er
ed
"
b
lack
b
o
x
es",
m
a
k
in
g
it
d
i
f
f
icu
lt
t
o
in
ter
p
r
et
h
o
w
t
h
ey
ar
r
iv
e
at
s
p
ec
if
ic
d
ec
is
io
n
s
.
T
h
is
lack
o
f
tr
an
s
p
ar
en
c
y
ca
n
b
e
a
s
ig
n
i
f
ican
t
d
r
awb
ac
k
in
ap
p
licatio
n
s
wh
er
e
u
n
d
er
s
tan
d
in
g
th
e
m
o
d
el'
s
r
ea
s
o
n
in
g
i
s
cr
u
cial
[
1
9
]
.
I
t
is
also
wo
r
th
n
o
tin
g
th
at
it
is
s
o
m
etim
es
a
g
o
o
d
id
ea
to
in
teg
r
ate
C
NNs
with
r
ec
u
r
r
en
t
n
eu
r
al
n
etwo
r
k
s
(
R
NNs
)
to
im
p
r
o
v
e
h
an
d
lin
g
o
f
s
eq
u
en
tial d
ata
s
u
ch
as c
o
n
n
ec
ted
tex
t
[
2
0
]
.
T
h
e
ar
ch
itectu
r
e
o
f
C
NNs
co
n
s
is
t
s
o
f
a
s
eq
u
en
ce
o
f
la
y
er
s
t
h
at
wo
r
k
to
g
eth
e
r
to
p
r
o
ce
s
s
in
p
u
ts
an
d
p
r
o
d
u
ce
o
u
tp
u
ts
.
T
h
e
i
n
p
u
t
lay
er
f
ir
s
t
r
ec
eiv
es
th
e
r
aw
im
a
g
e
(
th
e
m
ain
ty
p
e
o
f
in
p
u
t
d
a
ta)
p
r
o
v
id
ed
to
th
e
m
o
d
el.
T
h
is
lay
er
s
im
p
ly
p
ass
es th
e
im
ag
e
to
th
e
n
ex
t la
y
er
with
o
u
t a
n
y
co
m
p
u
tatio
n
s
,
b
u
t
it m
u
s
t b
e
ad
ju
s
ted
to
ac
co
m
m
o
d
ate
th
e
im
ag
e
p
r
o
p
er
ties
(
s
u
ch
as
h
eig
h
t,
wid
th
,
an
d
n
u
m
b
er
o
f
ch
a
n
n
els)
as
th
ese
p
r
o
p
er
ties
m
u
s
t
b
e
u
n
if
o
r
m
.
T
h
e
n
,
th
e
i
m
ag
e
is
f
ed
in
to
th
e
co
n
v
o
lu
ti
o
n
lay
er
,
wh
ich
p
er
f
o
r
m
s
a
co
n
v
o
lu
tio
n
o
p
er
atio
n
b
etwe
en
th
e
im
ag
e
an
d
o
n
e
o
r
m
o
r
e
f
ilter
s
(
also
ca
lled
k
er
n
els)
to
r
esu
lt
in
a
f
ea
tu
r
e
m
ap
co
n
tain
in
g
th
e
au
to
m
atica
lly
ex
tr
ac
ted
f
ea
tu
r
es.
T
h
e
in
p
u
t
im
ag
es
ca
n
h
av
e
o
n
e
o
r
m
o
r
e
ch
an
n
els,
an
d
th
e
f
ilter
s
h
av
e
t
h
e
s
am
e
n
u
m
b
er
o
f
c
h
an
n
els
(
ea
ch
f
ilter
h
as
a
ch
a
n
n
el
f
o
r
ea
c
h
ch
an
n
el
in
th
e
i
n
p
u
t
im
a
g
e)
.
T
h
e
im
ag
e
a
n
d
th
e
f
ilter
,
m
ea
n
in
g
ea
ch
ch
an
n
el
in
th
e
im
ag
e
with
its
co
r
r
esp
o
n
d
in
g
ch
a
n
n
el
in
th
e
f
ilter
,
ar
e
co
n
s
id
er
ed
.
B
o
th
th
e
im
ag
e
an
d
th
e
f
ilter
ar
e
tr
ea
ted
as
two
-
d
im
en
s
io
n
al
m
atr
ices,
wh
er
e
th
e
im
ag
e
m
at
r
ix
r
ep
r
esen
ts
p
ix
el
v
alu
es,
an
d
th
e
f
ilter
m
atr
i
x
c
o
n
tain
s
weig
h
t
v
alu
es.
T
h
e
f
il
ter
m
atr
ix
is
ty
p
ically
m
u
ch
s
m
aller
in
h
eig
h
t
an
d
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
15
,
No
.
5
,
Octo
b
er
20
26
:
4
0
1
5
-
4
0
2
5
4018
wid
th
.
I
n
th
e
co
n
v
o
lu
tio
n
o
p
er
atio
n
,
th
e
f
ilter
s
lid
es
o
v
er
th
e
im
ag
e
o
n
e
s
tr
id
e
at
a
tim
e
f
r
o
m
th
e
b
eg
in
n
in
g
o
f
th
e
im
ag
e
(
t
h
e
to
p
-
lef
t
co
r
n
er
)
to
th
e
e
n
d
o
f
th
e
r
o
w,
th
en
s
tar
ts
ag
ain
f
r
o
m
th
e
b
eg
in
n
in
g
o
f
th
e
r
o
w
o
n
e
s
tr
id
e
b
elo
w
th
e
p
r
e
v
io
u
s
p
o
s
itio
n
,
an
d
s
o
o
n
u
n
til it h
as g
o
n
e
th
r
o
u
g
h
th
e
e
n
tire
im
ag
e
as sh
o
wn
in
F
ig
u
r
e
2
.
Fig
u
r
e
2
.
T
h
e
p
o
s
itio
n
s
o
f
a
3
×
3
f
ilter
o
n
a
5
×
5
im
ag
e
with
a
2
×
2
s
tr
id
e
T
h
is
r
eq
u
ir
es
p
r
ec
is
io
n
i
n
ca
lc
u
latin
g
th
e
d
im
e
n
s
io
n
s
o
f
b
o
t
h
th
e
f
ilter
an
d
th
e
s
tr
id
e
s
o
t
h
at
th
ey
f
it
th
e
in
p
u
t
im
ag
e
d
im
en
s
io
n
s
.
Ass
u
m
in
g
th
at
th
e
d
im
e
n
s
io
n
s
o
f
th
e
in
p
u
t
im
ag
e
a
r
e
ℎ
×
,
th
e
d
im
en
s
io
n
s
o
f
th
e
f
ilter
ar
e
ℎ
×
,
an
d
th
e
d
im
e
n
s
io
n
s
o
f
th
e
s
tr
id
e
ar
e
ℎ
×
,
th
e
f
ea
tu
r
e
m
ap
will
b
e
a
ℎ
×
m
atr
ix
as sh
o
wn
in
(
1
)
an
d
(
2
).
ℎ
=
ℎ
−
ℎ
ℎ
+
1
(
1
)
=
−
+
1
(
2
)
T
h
ese
two
am
o
u
n
ts
m
u
s
t
b
e
n
atu
r
al
n
u
m
b
er
s
.
Oth
er
wis
e,
p
a
d
d
in
g
th
e
in
p
u
t
im
a
g
e
m
ay
b
e
n
ee
d
ed
(
c
h
an
g
e
its
d
im
en
s
io
n
s
b
y
ad
d
i
n
g
ze
r
o
s
ar
o
u
n
d
it).
Ad
d
itio
n
ally
,
ea
c
h
s
tr
id
e
d
im
en
s
io
n
m
u
s
t
n
o
t
b
e
g
r
ea
ter
th
an
its
co
r
r
esp
o
n
d
in
g
f
ilter
d
im
e
n
s
io
n
to
en
s
u
r
e
th
at
all
p
ar
ts
o
f
th
e
in
p
u
t
im
a
g
e
ar
e
p
ass
ed
th
r
o
u
g
h
.
E
ac
h
e
n
tr
y
in
th
e
m
en
tio
n
e
d
m
atr
ix
co
r
r
esp
o
n
d
s
(
in
o
r
d
e
r
)
to
o
n
e
o
f
th
e
p
o
s
itio
n
s
th
r
o
u
g
h
wh
ic
h
th
e
f
i
lter
p
ass
es.
At
ea
ch
p
o
s
itio
n
th
e
f
ilter
p
ass
es
th
r
o
u
g
h
d
u
r
in
g
its
s
tr
id
in
g
,
a
n
en
t
r
y
wis
e
p
r
o
d
u
ct
is
p
er
f
o
r
m
ed
b
et
wee
n
th
e
f
ilter
a
n
d
th
e
co
r
r
esp
o
n
d
i
n
g
p
atch
o
f
t
h
e
im
ag
e.
T
h
e
e
n
tr
y
wis
e
p
r
o
d
u
ct
o
p
e
r
atio
n
b
etwe
en
two
m
atr
ices
o
f
id
e
n
tical
d
im
en
s
io
n
s
is
p
er
f
o
r
m
ed
as
(
3
)
.
[
11
12
13
21
22
23
31
32
33
]
°
[
11
12
13
21
22
23
31
32
33
]
=
[
11
11
12
12
13
13
21
21
22
22
23
23
31
31
22
22
33
33
]
(
3
)
T
h
en
,
all
th
e
v
alu
es
r
esu
ltin
g
f
r
o
m
th
e
e
n
tr
y
wis
e
p
r
o
d
u
ct
o
p
er
atio
n
s
ar
e
s
u
m
m
e
d
with
a
d
if
f
er
en
t
v
alu
e
f
o
r
ea
ch
f
ilter
,
ca
lled
t
h
e
b
ias,
to
p
r
o
d
u
ce
th
e
co
r
r
esp
o
n
d
in
g
en
t
r
y
o
f
t
h
e
f
ea
t
u
r
e
m
ap
.
So
,
if
X
is
th
e
i
n
p
u
t
im
ag
e,
W
is
th
e
f
ilter
,
an
d
is
th
e
b
ia
s
,
th
en
ev
er
y
en
tr
y
o
f
t
h
e
f
ea
tu
r
e
m
ap
Y
is
co
m
p
u
ted
as sh
o
w
n
in
(
4
).
=
∑
(
(
,
,
)
(
+
,
+
,
)
)
+
,
,
(
4
)
T
h
e
o
u
tp
u
t
o
f
t
h
e
co
n
v
o
l
u
tio
n
lay
er
is
a
s
et
o
f
f
ea
tu
r
e
m
ap
s
,
o
n
e
f
o
r
ea
ch
f
ilter
.
T
h
ese
f
ea
tu
r
e
m
ap
s
h
ig
h
lig
h
t
th
e
d
if
f
er
e
n
t
lear
n
e
d
f
ea
tu
r
es
f
r
o
m
th
e
in
p
u
t
im
ag
e.
On
th
e
o
t
h
er
h
a
n
d
,
th
e
f
ilter
s
ar
e
lear
n
ab
le
p
ar
am
eter
s
(
lik
e
weig
h
ts
in
ar
tific
ial
n
eu
r
al
n
etwo
r
k
s
o
r
ANNs)
.
T
h
e
m
u
ltip
le
f
ilter
s
lea
r
n
d
if
f
er
en
t
p
atter
n
s
(
d
etec
t
d
if
f
er
en
t
f
ea
tu
r
es).
T
h
e
co
n
v
o
l
u
tio
n
lay
er
is
f
o
llo
wed
b
y
an
ac
tiv
atio
n
f
u
n
ctio
n
,
wh
ich
is
a
m
ath
em
atica
l
f
u
n
ctio
n
ap
p
lied
o
n
t
h
e
o
u
tp
u
t
o
f
th
e
c
o
n
v
o
lu
tio
n
lay
e
r
to
in
tr
o
d
u
c
e
n
o
n
-
lin
ea
r
ity
in
to
th
e
m
o
d
el,
e
n
ab
lin
g
it
to
lea
r
n
m
o
r
e
co
m
p
lex
p
atter
n
s
.
On
e
co
m
m
o
n
ac
tiv
ati
o
n
f
u
n
ctio
n
u
s
ed
in
C
NNs
is
th
e
r
ec
tifie
d
lin
ea
r
u
n
it
(
R
eL
U)
,
wh
ich
r
ed
u
ce
s
th
e
co
m
p
u
tatio
n
a
l lo
ad
b
y
ac
tin
g
as sh
o
wn
i
n
(
5
).
(
)
=
ma
x
(
0
,
)
(
5
)
An
o
th
er
im
p
o
r
tan
t
o
n
e
is
th
e
s
ig
m
o
id
(
)
,
wh
ich
is
u
s
ed
in
th
e
co
n
te
x
t
o
f
b
in
ar
y
class
if
icatio
n
an
d
p
r
o
b
a
b
ilis
tic
in
ter
p
r
etatio
n
s
task
s
,
wh
er
e
it tr
an
s
f
o
r
m
s
its
in
p
u
t in
to
an
o
u
tp
u
t in
r
an
g
e
o
f
0
to
1
,
b
ased
o
n
(
6
).
(
)
=
1
1
+
−
(
6
)
Of
co
u
r
s
e,
th
e
in
p
u
ts
to
th
ese
f
u
n
ctio
n
s
a
r
e
th
e
f
ea
tu
r
e
m
a
p
s
f
r
o
m
th
e
p
r
e
v
io
u
s
c
o
n
v
o
lu
tio
n
lay
er
s
,
wh
ich
a
r
e
m
atr
ices.
So
,
th
ese
f
u
n
ctio
n
s
ar
e
ap
p
lie
d
to
ea
ch
en
tr
y
o
f
t
h
e
in
p
u
t
m
atr
i
x
.
T
h
e
o
u
tp
u
t
i
s
th
en
p
ass
ed
to
th
e
p
o
o
lin
g
la
y
er
,
wh
ich
r
e
d
u
ce
s
th
e
s
ize
o
f
th
e
d
ata
to
d
ec
r
ea
s
e
th
e
co
m
p
lex
ity
o
f
th
e
m
o
d
el
an
d
in
cr
ea
s
e
its
ef
f
icien
cy
an
d
r
esis
tan
ce
to
s
m
all
v
ar
iatio
n
s
in
th
e
i
n
p
u
t
i
m
ag
es.
Ma
x
p
o
o
lin
g
,
wh
ich
s
elec
ts
th
e
m
ax
im
u
m
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
C
la
s
s
i
fyin
g
A
r
a
b
ic
h
a
n
d
w
r
itin
g
u
s
in
g
co
n
vo
lu
tio
n
a
l n
e
u
r
a
l n
etw
o
r
k
(
A
n
a
s
Qu
teis
h
a
t
)
4019
v
alu
e
f
r
o
m
a
s
p
ec
if
ic
r
eg
io
n
(
ca
lled
a
win
d
o
w)
,
is
co
m
m
o
n
l
y
u
s
ed
.
T
h
e
co
n
ce
p
t
o
f
s
tr
id
e
is
also
p
r
esen
t
h
er
e
with
th
e
o
n
ly
d
if
f
er
e
n
ce
b
ein
g
th
at
th
e
win
d
o
w
r
ep
lace
s
th
e
f
ilter
as sh
o
wn
in
F
ig
u
r
e
3
.
Fig
u
r
e
3
.
Ma
x
p
o
o
lin
g
with
a
2
×
2
win
d
o
w
an
d
a
2
×
2
s
tr
id
e
A
s
eq
u
en
ce
o
f
c
o
n
v
o
lu
tio
n
la
y
er
,
ac
tiv
atio
n
f
u
n
ctio
n
,
an
d
p
o
o
lin
g
lay
er
is
u
s
u
ally
r
ep
e
ated
m
o
r
e
th
an
o
n
ce
in
C
NNs,
im
m
ed
iately
af
ter
th
e
in
p
u
t
lay
er
.
B
u
t
in
s
tead
o
f
th
e
r
aw
im
ag
es,
th
e
in
p
u
t
o
f
ea
ch
s
u
b
s
eq
u
en
t
c
o
n
v
o
lu
tio
n
lay
er
is
th
e
o
u
tp
u
t
o
f
its
p
r
ev
io
u
s
lay
er
,
wh
ich
will
b
e
o
n
e
o
r
m
o
r
e
f
ea
tu
r
e
m
ap
s
.
Ad
d
itio
n
ally
,
m
u
ltip
le
f
ea
t
u
r
e
m
ap
s
ar
e
tr
ea
ted
as
ch
a
n
n
els
o
f
a
s
in
g
le
i
n
p
u
t
im
ag
e
wh
en
f
ed
in
to
a
co
n
v
o
l
u
tio
n
lay
er
.
Af
ter
th
em
co
m
es
th
e
r
o
le
o
f
th
e
f
latten
in
g
lay
er
to
c
o
n
v
e
r
t
th
e
m
u
lti
-
d
im
en
s
io
n
al
in
p
u
t
m
atr
ices
in
to
v
ec
to
r
s
.
T
h
is
la
y
er
m
ay
b
e
co
m
b
in
ed
with
th
e
n
ex
t
lay
er
,
wh
ich
is
th
e
f
u
ll
y
co
n
n
ec
ted
lay
er
.
T
h
e
f
u
lly
co
n
n
ec
ted
lay
er
p
r
o
ce
s
s
es
a
f
latten
ed
v
ec
to
r
f
o
r
class
if
icatio
n
o
r
d
ec
is
io
n
-
m
ak
in
g
b
ased
o
n
t
h
e
ex
tr
ac
ted
f
ea
tu
r
es.
So
,
it
is
u
ti
lized
as
a
class
if
ier
in
class
if
i
ca
tio
n
m
o
d
els.
I
n
s
id
e
th
is
lay
er
,
ev
er
y
n
e
u
r
o
n
is
co
n
n
ec
ted
t
o
ev
er
y
n
eu
r
o
n
in
t
h
e
p
r
ev
i
o
u
s
lay
er
as sh
o
wn
in
Fig
u
r
e
4
.
Fig
u
r
e
4
.
T
h
e
f
u
lly
co
n
n
ec
ted
lay
er
If
is
th
e
n
u
m
b
er
o
f
n
eu
r
o
n
s
o
f
its
p
r
ev
io
u
s
lay
er
,
is
th
e
n
u
m
b
er
o
f
its
n
eu
r
o
n
s
,
X1
×
is
it
s
in
p
u
t
m
atr
ix
,
W
×
is
its
weig
h
t
m
atr
ix
(
wh
er
e
ea
ch
r
o
w
co
n
tain
s
th
e
weig
h
ts
th
at
c
o
n
n
ec
t
o
n
e
n
e
u
r
o
n
f
r
o
m
th
e
p
r
ev
io
u
s
lay
er
to
all
n
eu
r
o
n
s
o
f
th
e
c
u
r
r
en
t
lay
e
r
i
n
o
r
d
e
r
)
,
B
1
×
is
its
b
ias
m
atr
ix
,
an
d
Y1
×
is
its
o
u
tp
u
t m
atr
ix
as sh
o
wn
i
n
(
7
)
.
=
+
(
7
)
T
h
e
n
u
m
b
er
o
f
n
eu
r
o
n
s
in
a
f
u
lly
co
n
n
ec
ted
lay
er
d
ep
en
d
s
o
n
th
e
task
(
class
if
icatio
n
o
r
r
eg
r
ess
io
n
)
.
I
n
class
if
icatio
n
m
o
d
els,
th
is
n
u
m
b
er
m
u
s
t
b
e
eq
u
al
to
th
e
n
u
m
b
er
o
f
class
es
in
th
e
m
o
d
el,
s
o
th
at
its
o
u
tp
u
t
is
a
s
et
o
f
v
alu
es,
o
n
e
f
o
r
ea
ch
class
.
T
h
e
s
o
f
tm
ax
f
u
n
ctio
n
th
en
co
n
v
er
ts
th
ese
v
alu
e
s
in
to
p
r
o
b
a
b
ilit
ies
(
v
alu
es
b
etwe
en
0
an
d
1
,
s
u
c
h
th
at
th
eir
s
u
m
eq
u
als
1
)
r
ep
r
esen
tin
g
th
e
lik
elih
o
o
d
o
f
e
ac
h
class
g
iv
en
th
e
in
p
u
t.
I
f
is
th
e
th
o
r
ig
in
al
v
alu
e
(
o
f
th
e
f
u
lly
co
n
n
ec
ted
l
ay
er
)
as sh
o
wn
in
(
8
)
.
(
)
=
∑
(
8
)
T
h
e
o
n
ly
task
lef
t
f
o
r
th
e
last
lay
er
,
th
e
o
u
tp
u
t
lay
er
,
is
to
d
eter
m
in
e
th
e
m
o
s
t
lik
ely
class
ac
co
r
d
in
g
to
th
e
p
r
o
b
a
b
ilit
ies
co
m
p
u
ted
b
y
th
e
p
r
ev
i
o
u
s
f
u
n
ctio
n
(
th
e
p
r
ed
ictio
n
f
o
r
th
e
g
iv
en
i
n
p
u
t
im
ag
e)
.
E
v
er
y
th
in
g
m
en
tio
n
ed
s
o
f
a
r
is
ab
o
u
t
wh
at
is
ca
lled
f
o
r
war
d
p
r
o
p
ag
ati
o
n
,
wh
e
r
e
d
ata
p
r
o
p
ag
ates
f
r
o
m
th
e
i
n
p
u
t
lay
er
t
o
th
e
o
u
tp
u
t
lay
er
f
o
r
p
r
ed
ict
io
n
.
B
u
t
f
o
r
tr
ain
in
g
p
u
r
p
o
s
es,
an
o
th
er
ty
p
e
o
f
d
ata
p
r
o
p
a
g
atio
n
ca
lled
b
ac
k
p
r
o
p
ag
atio
n
is
n
ee
d
ed
.
B
ac
k
p
r
o
p
ag
atio
n
is
u
s
ed
to
u
p
d
ate
th
e
weig
h
ts
an
d
b
iases
b
as
ed
o
n
th
e
d
if
f
e
r
en
ce
b
etwe
en
p
r
ed
icted
v
alu
es
a
n
d
ac
tu
al
v
alu
es.
Du
r
in
g
tr
ain
i
n
g
,
th
e
in
p
u
t
im
ag
e
is
f
ir
s
t
p
ass
ed
th
r
o
u
g
h
f
o
r
war
d
p
r
o
p
a
g
atio
n
,
w
h
ich
o
u
tp
u
ts
a
p
r
ed
icted
v
alu
e
f
o
r
t
h
e
in
p
u
t.
T
h
en
th
e
lo
s
s
is
ca
lcu
lated
u
s
in
g
a
l
o
s
s
f
u
n
ctio
n
t
o
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
15
,
No
.
5
,
Octo
b
er
20
26
:
4
0
1
5
-
4
0
2
5
4020
m
ea
s
u
r
e
th
e
d
if
f
er
e
n
ce
b
etwe
en
th
e
p
r
ed
icted
v
alu
e
an
d
th
e
ac
tu
al
v
alu
e
.
T
h
e
lo
s
s
f
u
n
ctio
n
d
e
p
en
d
s
o
n
t
h
e
task
,
b
u
t
a
co
m
m
o
n
v
e
r
s
io
n
f
o
r
class
if
icatio
n
task
s
is
th
e
cr
o
s
s
-
en
tr
o
p
y
l
o
s
s
.
I
f
is
t
h
e
p
r
ed
icted
v
alu
e
(
p
r
o
b
a
b
ilit
y
)
o
f
th
e
tr
u
e
class
o
f
an
in
p
u
t im
ag
e
,
th
en
t
h
e
cr
o
s
s
-
en
tr
o
p
y
lo
s
s
L
is
ca
lcu
lated
as sh
o
wn
in
(
9
)
.
=
−
∑
=
1
l
og
(
)
(
9
)
wh
er
e
C
is
th
e
n
u
m
b
er
o
f
class
es
(
2
8
clas
s
es),
is
1
f
o
r
co
r
r
ec
t
class
es
an
d
0
o
th
er
wis
e,
an
d
is
th
e
m
o
d
el’
s
p
r
ed
icted
p
r
o
b
ab
ilit
y
f
o
r
class
i.
Af
ter
th
e
lo
s
s
is
ca
l
cu
lated
,
th
e
g
r
ad
ien
ts
ar
e
ca
lcu
lated
in
ea
ch
lay
er
u
s
in
g
th
e
d
if
f
er
e
n
tiatio
n
to
m
e
asu
r
e
th
e
im
p
ac
t
o
f
ch
an
g
in
g
t
h
e
weig
h
ts
o
n
th
e
lo
s
s
,
an
d
th
en
d
is
tr
ib
u
te
th
ese
ch
an
g
es
to
th
e
p
r
ev
io
u
s
lay
e
r
s
,
wh
ich
is
ca
lled
b
ac
k
p
r
o
p
ag
atio
n
.
Fin
ally
,
an
o
p
tim
iza
tio
n
alg
o
r
ith
m
lik
e
s
to
ch
asti
c
g
r
ad
ien
t
d
escen
t
(
S
GD)
o
r
A
d
am
is
u
s
ed
to
u
p
d
a
te
th
e
weig
h
ts
b
ased
o
n
th
e
g
r
ad
ien
ts
ca
lcu
lated
d
u
r
in
g
b
ac
k
p
r
o
p
a
g
atio
n
.
I
f
ω
o
is
th
e
o
ld
weig
h
t,
ω
is
th
e
n
ew
weig
h
t,
is
a
f
ac
to
r
ca
lled
th
e
lear
n
in
g
r
ate
,
an
d
L
is
th
e
lo
s
s
,
th
en
t
h
e
u
p
d
ate
f
o
r
m
u
la
u
s
in
g
SGD
is
as s
h
o
wn
in
(
1
0
)
.
=
0
−
(
ω
o
)
(
10
)
T
h
is
p
r
o
ce
s
s
is
r
ep
ea
ted
with
ea
ch
b
atch
(
s
et
o
f
s
am
p
les)
o
f
th
e
tr
ain
in
g
d
ata,
g
r
ad
u
ally
i
m
p
r
o
v
i
n
g
th
e
p
er
f
o
r
m
a
n
ce
o
f
th
e
m
o
d
el
th
r
o
u
g
h
m
u
ltip
le
iter
atio
n
s
b
y
m
in
im
izin
g
th
e
lo
s
s
.
I
n
s
tead
o
f
p
ass
in
g
th
e
en
tire
tr
ain
in
g
d
ataset
th
r
o
u
g
h
th
e
n
etwo
r
k
at
o
n
ce
(
wh
ic
h
is
c
alled
an
e
p
o
ch
)
,
th
e
d
ata
is
d
iv
id
ed
in
to
s
m
aller
g
r
o
u
p
s
(
b
at
ch
es
)
.
T
h
e
w
ei
g
h
ts
ar
e
u
p
d
ate
d
a
f
te
r
p
r
o
c
ess
i
n
g
e
a
ch
b
at
ch
b
ase
d
o
n
t
h
e
a
v
e
r
a
g
e
lo
s
s
o
f
th
e
s
am
p
l
es
in
t
h
at
b
a
tc
h
.
T
h
is
ap
p
r
o
ac
h
s
p
e
ed
s
u
p
t
r
ai
n
i
n
g
a
n
d
r
e
d
u
c
es
m
em
o
r
y
u
s
a
g
e
.
A
d
d
iti
o
n
al
ly
,
p
er
f
o
r
m
an
ce
is
ev
alu
ated
o
n
a
d
ataset
(
ca
lled
a
v
alid
atio
n
d
ataset)
s
ep
ar
ate
f
r
o
m
t
h
e
tr
ain
in
g
d
ata.
T
h
is
ca
n
b
e
d
o
n
e
s
ev
er
a
l
tim
es d
u
r
in
g
o
r
af
ter
t
r
ain
in
g
a
n
d
h
elp
s
m
o
n
ito
r
t
h
e
g
en
e
r
aliza
tio
n
ab
ilit
y
an
d
p
r
ev
en
t o
v
er
f
itti
n
g
.
3.
DATAS
E
T
DE
SCRI
P
T
I
O
N
T
h
e
d
ata
u
s
ed
is
ca
r
ef
u
lly
c
o
llected
to
ac
h
iev
e
r
an
d
o
m
n
ess
an
d
b
alan
ce
f
r
o
m
two
d
atasets
,
"Ar
ab
ic
h
an
d
wr
itten
alp
h
ab
ets,
wo
r
d
s
,
an
d
p
ar
ag
r
a
p
h
s
"
(
AHAWP
)
[
2
1
]
an
d
"Ar
ab
ic
h
a
n
d
wr
itten
ch
ar
ac
ter
d
atab
ase"
(
AHCDB
)
[
2
2
]
.
I
n
th
e
f
ir
s
t
d
ataset,
th
e
s
am
p
les
ar
e
d
iv
id
ed
b
y
th
e
u
s
er
s
(
wr
iter
s
)
,
b
u
t
it
d
o
es
n
o
t
co
v
er
th
e
2
8
b
asic
Ar
a
b
ic
letter
s
(
wh
ich
th
e
m
o
d
el
aim
s
to
r
ec
o
g
n
ize)
.
I
t
co
v
er
s
o
n
l
y
1
8
o
f
th
em
with
ab
o
u
t
8
2
0
im
a
g
es
f
r
o
m
8
2
u
s
er
s
,
an
d
a
to
tal
o
f
1
4
,
7
4
0
im
ag
es.
On
th
e
o
th
er
h
an
d
,
th
e
s
ec
o
n
d
d
ataset
co
v
e
r
s
all
2
8
letter
s
with
7
5
0
im
ag
es
f
o
r
ea
ch
letter
,
f
o
r
a
to
tal
o
f
2
1
,
0
0
0
im
ag
es,
b
u
t
it
is
le
s
s
d
iv
er
s
e
as
it
s
ee
m
s
to
b
e
m
ad
e
b
y
f
ewe
r
u
s
er
s
.
So
,
to
ad
d
s
o
m
e
d
iv
er
s
ity
to
o
u
r
c
o
llectio
n
,
a
n
d
to
m
ak
e
it
in
clu
d
e
all
2
8
letter
s
,
f
o
r
ea
ch
letter
,
2
5
0
im
a
g
es
wer
e
s
elec
ted
b
elo
n
g
in
g
to
5
0
u
s
er
s
f
r
o
m
AHAWP
an
d
an
o
th
er
2
5
0
f
r
o
m
AHCD
B
,
m
ak
in
g
a
co
llectio
n
o
f
5
0
0
im
ag
es
p
e
r
letter
f
o
r
th
e
f
ir
s
t
1
8
letter
s
a
n
d
5
0
0
im
ag
es
f
r
o
m
th
e
AH
C
DB
f
o
r
th
e
o
th
er
1
0
letter
s
.
T
h
e
to
tal
n
u
m
b
e
r
o
f
im
ag
es wa
s
1
4
,
0
0
0
im
ag
es.
Sin
ce
th
e
im
ag
es
ca
m
e
f
r
o
m
d
if
f
er
en
t
s
o
u
r
ce
s
im
ag
e
p
r
e
p
r
o
ce
s
s
in
g
s
tag
e
was
r
eq
u
ir
ed
.
First
th
e
im
ag
es
wer
e
s
ep
ar
ated
in
to
f
o
l
d
er
s
,
ea
ch
n
am
e
d
af
ter
t
h
e
Ar
a
b
ic
letter
th
at
th
e
f
o
l
d
er
co
n
tain
s
its
im
ag
es.
T
h
e
im
ag
es
d
if
f
er
in
f
o
r
m
at
b
etwe
en
th
e
two
d
atasets
,
b
ein
g
PN
G
f
o
r
m
atted
in
AHAWP
,
an
d
B
itm
ap
f
o
r
m
atted
in
HACD
B
.
So
,
th
e
s
ec
o
n
d
s
tep
af
ter
co
llectin
g
th
e
im
ag
e
s
was
to
u
n
if
y
th
eir
f
o
r
m
at,
a
p
r
o
g
r
am
ca
lle
d
I
r
f
an
View
was
u
s
ed
to
co
n
v
er
t
all
th
e
im
a
g
es
in
th
e
c
o
llectio
n
to
PNG
f
o
r
m
at
n
u
m
b
e
r
ed
f
r
o
m
0
0
0
to
4
9
9
f
o
r
ea
ch
letter
.
I
n
ad
d
itio
n
,
th
e
i
m
ag
es
o
f
AHAWP
ar
e
g
r
ay
s
ca
le
an
d
ar
e
ch
an
g
ed
in
to
b
i
n
ar
y
im
ag
es
u
s
in
g
MA
T
L
AB
.
Nex
t,
th
e
em
p
ty
wh
ite
s
p
ac
es
ar
o
u
n
d
th
e
im
a
g
es
ar
e
c
r
o
p
p
ed
to
m
ak
e
th
e
m
s
m
aller
in
s
ize.
Fig
u
r
e
5
s
h
o
ws th
e
s
tep
s
in
v
o
lv
ed
in
b
u
ild
in
g
th
e
d
ataset.
Fig
u
r
e
5
.
Step
s
in
v
o
lv
ed
in
b
u
ild
in
g
th
e
d
ataset
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
C
la
s
s
i
fyin
g
A
r
a
b
ic
h
a
n
d
w
r
itin
g
u
s
in
g
co
n
vo
lu
tio
n
a
l n
e
u
r
a
l n
etw
o
r
k
(
A
n
a
s
Qu
teis
h
a
t
)
4021
I
t
is
p
r
ef
er
ab
le
(
b
u
t
n
o
t
r
e
q
u
i
r
ed
)
th
at
th
e
in
p
u
t
im
ag
es
ar
e
s
q
u
ar
e.
T
h
er
ef
o
r
e,
p
ad
d
in
g
e
ac
h
im
ag
e
with
em
p
ty
an
d
b
alan
ce
d
wh
it
e
s
p
ac
es
eith
er
f
r
o
m
th
e
r
ig
h
t a
n
d
lef
t
(
if
th
e
h
eig
h
t
is
g
r
ea
te
r
th
an
th
e
wid
th
)
o
r
f
r
o
m
th
e
to
p
a
n
d
b
o
tto
m
(
i
f
th
e
wid
th
is
g
r
ea
ter
th
a
n
th
e
h
ei
g
h
t)
wh
ile
leav
in
g
th
e
s
q
u
a
r
e
im
ag
es
as
th
ey
ar
e
with
o
u
t
ch
a
n
g
e.
Fin
ally
,
th
e
i
m
ag
es
wer
e
d
o
wn
s
ized
to
ju
s
t
2
8
×
2
8
p
ix
els
to
r
ed
u
ce
th
e
c
o
m
p
u
tatio
n
al
p
o
wer
n
ee
d
ed
to
tr
ain
th
e
m
o
d
el
in
a
r
ea
s
o
n
ab
le
tim
e.
Ad
d
itio
n
ally
,
th
is
ty
p
e
o
f
im
ag
e
d
o
e
s
n
o
t
r
eq
u
ir
e
h
ig
h
r
eso
lu
tio
n
,
as th
e
lar
g
est im
ag
e
s
ize
b
ef
o
r
e
d
o
wn
s
izin
g
was
3
4
8
×
3
4
8
(
an
e
x
ag
g
er
ated
s
ize
f
o
r
ju
s
t le
tter
s
)
.
4.
I
M
P
L
E
M
E
NT
A
T
I
O
N
A
ND
RE
SU
L
T
S
T
h
e
C
NN
was
b
u
ilt
an
d
tr
ain
ed
f
r
o
m
s
cr
atch
,
f
o
r
t
h
is
p
u
r
p
o
s
e,
th
e
d
ee
p
n
etwo
r
k
d
esig
n
er
was
u
s
ed
,
an
ap
p
licatio
n
in
MA
T
L
AB
R
2
0
2
1
a.
T
h
is
to
o
l
o
f
f
e
r
s
an
in
ter
ac
tiv
e
en
v
ir
o
n
m
en
t
f
o
r
d
e
s
ig
n
in
g
,
an
aly
zin
g
,
an
d
tr
ain
in
g
d
ee
p
lear
n
in
g
m
o
d
els
o
n
an
ea
s
y
-
to
-
u
s
e
g
r
ap
h
ical
in
ter
f
ac
e.
T
h
e
d
atase
t
was
d
iv
id
ed
in
to
two
s
ets,
8
0
%
o
f
th
e
d
ata
was
u
s
ed
f
o
r
tr
ain
in
g
an
d
2
0
%
w
as
u
s
ed
f
o
r
test
in
g
.
T
h
e
d
iv
is
io
n
o
f
th
e
d
ata
was
d
o
n
e
e
q
u
ally
am
o
n
g
all
class
es.
T
h
e
n
etwo
r
k
b
u
ilt
co
n
s
is
ts
o
f
an
in
p
u
t
lay
er
t
h
at
r
ec
eiv
es
s
in
g
le
-
ch
an
n
el
im
a
g
es
o
f
2
8
×
2
8
s
ize
(
ac
co
r
d
in
g
to
th
e
c
o
llectio
n
i
m
ag
es),
f
o
llo
wed
b
y
a
co
n
v
o
l
u
tio
n
lay
e
r
with
two
5
×
5
f
ilter
s
an
d
a
2
×
2
s
tr
id
e,
f
o
llo
wed
b
y
a
R
eL
U
ac
tiv
atio
n
f
u
n
ctio
n
,
a
n
d
th
e
n
a
m
a
x
p
o
o
lin
g
lay
e
r
with
a
2
×
2
win
d
o
w.
T
h
e
n
ag
ain
,
a
co
n
v
o
l
u
tio
n
lay
er
b
u
t
with
f
o
u
r
5
×
5
f
ilter
s
an
d
th
e
s
am
e
2
×
2
s
tr
id
e,
f
o
llo
wed
b
y
a
R
eL
U
ac
tiv
atio
n
f
u
n
ctio
n
a
n
d
th
e
n
a
m
ax
p
o
o
l
in
g
lay
er
with
th
e
s
am
e
2
×
2
win
d
o
w.
T
h
e
n
co
m
es
th
e
f
u
l
ly
co
n
n
ec
ted
lay
er
with
2
8
n
eu
r
o
n
s
(
Ar
ab
ic
letter
s
)
,
f
o
llo
wed
b
y
a
s
o
f
tm
ax
lay
e
r
an
d
f
in
ally
th
e
o
u
tp
u
t
lay
er
.
Fig
u
r
e
6
s
h
o
ws
th
e
C
NN
n
etwo
r
k
u
s
ed
to
class
if
y
th
e
d
ataset
an
d
T
a
b
le
1
s
h
o
ws
th
e
d
etails o
f
ea
ch
la
y
er
u
s
ed
.
Fig
u
r
e
6
.
T
h
e
C
NN
u
s
ed
to
cl
ass
if
y
th
e
d
ataset
T
ab
le
1
.
Deta
ils
th
e
lay
er
c
o
ef
f
icien
ts
La
y
e
r
t
y
p
e
O
u
t
p
u
t
s
i
z
e
P
a
r
a
me
t
e
r
s
A
c
t
i
v
a
t
i
o
n
I
n
p
u
t
28
×
28
×
1
-
-
C
o
n
v
2
D
-
1
26
×
26
×
32
5
×
5
k
e
r
n
e
l
,
3
2
f
i
l
t
e
r
s
R
e
LU
M
a
x
P
o
o
l
i
n
g
-
1
13
×
13
×
32
2
×
2
p
o
o
l
si
z
e
-
C
o
n
v
2
D
-
2
11
×
11
×
64
5
×
5
k
e
r
n
e
l
,
6
4
f
i
l
t
e
r
s
R
e
LU
M
a
x
P
o
o
l
i
n
g
-
2
5
×
5
×
64
2
×
2
p
o
o
l
si
z
e
-
F
l
a
t
t
e
n
1
,
6
0
0
-
-
F
u
l
l
y
c
o
n
n
e
c
t
e
d
1
2
8
1
2
8
u
n
i
t
s
R
e
LU
O
u
t
p
u
t
28
2
8
u
n
i
t
s
S
o
f
t
ma
x
As
p
r
ev
io
u
s
ly
m
en
tio
n
ed
,
th
e
d
ataset
was
d
iv
id
ed
in
to
two
s
u
b
s
ets,
with
2
0
%
(
2
,
8
0
0
im
ag
es)
allo
ca
ted
f
o
r
test
in
g
an
d
th
e
r
e
m
ain
in
g
1
1
,
2
0
0
im
ag
es u
s
ed
f
o
r
tr
ain
i
n
g
.
T
h
e
lear
n
in
g
r
ate
was
s
et
to
0
.
0
1
,
an
d
th
e
b
atch
s
ize
was
d
ef
in
ed
as
1
4
,
m
ea
n
in
g
th
e
tr
ain
in
g
p
r
o
c
ess
co
n
s
is
ted
o
f
8
0
0
iter
atio
n
s
p
er
ep
o
ch
.
Du
r
in
g
ex
p
er
im
en
tatio
n
,
it
was
o
b
s
er
v
ed
th
at
th
e
C
NN’
s
p
er
f
o
r
m
a
n
ce
d
id
n
o
t
im
p
r
o
v
e
b
e
y
o
n
d
th
e
th
ir
d
ep
o
ch
;
in
f
ac
t,
tr
ain
in
g
th
e
n
etwo
r
k
f
o
r
ad
d
itio
n
al
e
p
o
ch
s
led
to
o
v
e
r
f
itti
n
g
.
Sp
ec
if
ically
,
w
h
en
e
x
ten
d
ed
t
o
a
f
o
u
r
t
h
ep
o
ch
,
th
e
test
in
g
ac
cu
r
ac
y
b
eg
an
to
d
ec
lin
e,
in
d
icatin
g
th
e
o
n
s
et
o
f
o
v
er
f
itti
n
g
.
Fig
u
r
e
6
illu
s
tr
ates
th
e
s
y
s
tem
’
s
ac
cu
r
ac
y
a
n
d
lo
s
s
cu
r
v
es.
I
t
ca
n
b
e
o
b
s
er
v
e
d
t
h
at
ac
cu
r
ac
y
in
c
r
ea
s
ed
s
h
ar
p
l
y
a
f
ter
th
e
f
ir
s
t
ep
o
c
h
(
ar
o
u
n
d
iter
atio
n
8
0
0
)
,
c
o
n
ti
n
u
ed
to
im
p
r
o
v
e
g
r
ad
u
ally
d
u
r
in
g
th
e
s
ec
o
n
d
ep
o
ch
,
an
d
s
h
o
wed
m
i
n
im
al
en
h
an
ce
m
e
n
t
in
th
e
th
ir
d
ep
o
c
h
,
co
n
f
ir
m
in
g
th
at
th
e
m
o
d
el
h
ad
ef
f
ec
tiv
ely
co
n
v
er
g
e
d
.
I
n
t
h
e
lo
wer
s
ec
tio
n
o
f
F
ig
u
r
e
6
,
th
e
l
o
s
s
f
u
n
ctio
n
is
p
r
esen
ted
.
T
h
e
lo
s
s
f
u
n
ctio
n
i
n
d
icate
s
h
o
w
f
ar
o
f
f
th
e
m
o
d
el’
s
p
r
ed
ictio
n
s
a
r
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
15
,
No
.
5
,
Octo
b
er
20
26
:
4
0
1
5
-
4
0
2
5
4022
f
r
o
m
th
e
tr
u
e
lab
els.
T
h
e
lo
s
s
f
u
n
ctio
n
is
s
h
o
wn
in
(
8
)
,
I
n
th
e
f
ir
s
t
ep
o
ch
,
th
e
m
o
d
els
p
r
ed
icts
all
cla
s
s
es
n
ea
r
ly
eq
u
ally
,
s
o
=
1
28
≈
0
.
036
an
d
wh
en
t
h
is
v
alu
e
is
s
u
b
s
titu
ted
in
eq
u
atio
n
8
=
−
l
og
(
1
28
)
≈
3
.
33
.
T
h
is
is
s
h
o
wn
in
F
ig
u
r
e
7
.
As th
e
tr
ain
in
g
p
r
o
ce
e
d
s
th
e
lo
s
s
s
tab
ilizes ar
o
u
n
d
0
.
2
-
0
.
3
.
T
h
e
co
n
f
u
s
io
n
m
atr
ix
f
o
r
th
e
r
esu
lts
is
s
h
o
wn
in
F
ig
u
r
e
8
.
I
n
th
e
co
n
f
u
s
io
n
m
at
r
ix
it
is
v
is
ib
le
th
at
letter
s
with
d
o
t
ch
an
g
es
s
u
ch
as
ب
-
ت
-
ث
-
د
,
ي
-
ذ
-
ر
-
ط
,
ز
-
ظ
-
ع
-
غ
,
an
d
ج
-
ح
-
خ
h
av
e
h
ig
h
er
co
n
f
u
s
io
n
s
in
ce
th
ey
ar
e
v
er
y
s
im
ilar
.
T
o
f
u
r
th
er
a
n
a
ly
ze
th
e
p
er
f
o
r
m
a
n
ce
o
f
th
e
C
NN,
th
e
r
ec
all,
p
r
ec
is
io
n
,
an
d
F1
-
s
co
r
e
ar
e
ca
lcu
lated
f
o
r
ea
ch
c
h
ar
ac
ter
.
T
h
e
o
b
tain
ed
r
esu
lts
ar
e
s
h
o
wn
in
T
ab
le
2
.
Fro
m
T
ab
le
2
,
o
n
e
ca
n
s
ee
th
at
th
e
av
er
ag
e
F1
-
s
co
r
e
is
9
0
.
9
%
w
h
ich
is
h
ig
h
er
th
a
n
th
e
o
v
er
al
l
ac
cu
r
ac
y
o
f
8
9
.
2
%.
T
h
is
is
d
u
e
to
th
e
f
ac
t
th
at
th
er
e
ar
e
f
u
n
d
am
en
tal
d
if
f
er
e
n
ce
s
b
etwe
en
th
e
two
q
u
an
titi
es.
T
h
e
F1
-
s
co
r
e
b
ein
g
s
lig
h
tly
h
ig
h
er
th
an
th
e
ac
cu
r
ac
y
s
u
g
g
ests
th
at
th
e
m
o
d
el
h
as
lear
n
ed
ch
ar
ac
ter
is
tics
o
f
m
o
s
t
Ar
ab
ic
letter
s
r
ea
s
o
n
ab
ly
well,
with
o
n
ly
f
ew
ch
allen
g
in
g
ca
s
es su
ch
as
ي
an
d
ج
-
ح
-
خ
g
r
o
u
p
.
Fig
u
r
e
7
.
T
h
e
g
r
a
p
h
s
o
f
b
o
th
a
cc
u
r
ac
y
an
d
lo
s
s
Fig
u
r
e
8
.
T
h
e
co
n
f
u
s
io
n
m
atr
ix
f
o
r
t
h
e
Ar
ab
ic
letter
s
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
C
la
s
s
i
fyin
g
A
r
a
b
ic
h
a
n
d
w
r
itin
g
u
s
in
g
co
n
vo
lu
tio
n
a
l n
e
u
r
a
l n
etw
o
r
k
(
A
n
a
s
Qu
teis
h
a
t
)
4023
T
ab
le
2
.
C
lass
if
icatio
n
p
er
f
o
r
m
an
ce
f
o
r
ea
ch
Ar
a
b
ic
letter
C
l
a
s
s (Le
t
t
e
r
)
P
r
e
c
i
s
i
o
n
(
%)
R
e
c
a
l
l
(
%)
F1
-
s
c
o
r
e
(
%)
C
l
a
s
s (Le
t
t
e
r
)
P
r
e
c
i
s
i
o
n
(
%)
R
e
c
a
l
l
(
%)
F1
-
s
c
o
r
e
(
%)
ا
(
A
l
i
f
)
9
3
.
1
95
94
ض
(
D
a
d
)
91
91
91
ب
(
B
a
)
9
0
.
9
90
9
0
.
5
ط
(
Ta
h
)
93
93
93
ت
(
Ta
)
9
0
.
1
91
9
0
.
5
ظ
(
Za
h
)
9
3
.
9
92
9
2
.
9
ث
(
T
h
a
)
9
0
.
8
89
8
9
.
9
ع
(
A
i
n
)
9
1
.
9
91
9
1
.
5
ج
(
Jee
m)
88
88
88
غ
(
G
a
i
n
)
92
92
92
ح
(
H
a
)
8
8
.
8
87
8
7
.
9
ف
(
F
a
)
91
91
91
خ
(
K
h
a
)
9
0
.
5
86
8
8
.
2
ق
(
Q
a
f
)
9
1
.
1
92
9
1
.
5
د
(
D
a
l
)
9
3
.
9
93
9
3
.
5
ك
(
K
a
f
)
9
2
.
9
91
9
1
.
9
ذ
(
T
h
a
l
)
9
4
.
7
90
9
2
.
3
ل
(
La
m)
9
0
.
2
92
9
1
.
1
ر
(
Ra)
9
3
.
9
92
9
2
.
9
م
(
M
e
e
m)
92
92
92
ز
(
Za
y
)
9
3
.
9
92
9
2
.
9
ن
(
N
o
o
n
)
9
2
.
9
91
9
1
.
9
س
(
S
e
e
n
)
9
1
.
9
91
9
1
.
5
ه
(
H
a
)
9
1
.
1
92
9
1
.
5
ش
(
S
h
e
e
n
)
9
1
.
8
90
9
0
.
9
و
(
W
a
w
)
9
2
.
9
91
9
1
.
9
ص
(
S
a
d
)
9
2
.
8
90
9
1
.
4
ي
(
Y
a
a
)
84
84
84
T
ab
le
3
s
h
o
ws
a
p
er
f
o
r
m
a
n
ce
co
m
p
ar
is
o
n
o
f
d
if
f
er
en
t
lea
r
n
in
g
m
eth
o
d
s
co
m
p
ar
ed
with
th
e
C
NN.
T
h
e
tab
les
s
h
o
w
th
at
th
e
p
er
f
o
r
m
an
ce
o
f
t
h
e
C
NN
n
etwo
r
k
g
av
e
b
etter
r
esu
lts
in
co
m
p
ar
is
o
n
with
o
th
e
r
s
y
s
tem
s
.
T
h
e
p
r
o
p
o
s
ed
m
o
d
el
ac
h
iev
ed
an
ac
cu
r
ac
y
o
f
8
9
.
2
%,
wh
ich
p
r
o
v
es
th
at
th
is
ty
p
e
o
f
n
etwo
r
k
(
C
NN
n
etwo
r
k
s
)
is
m
o
r
e
s
u
it
ab
le
f
o
r
th
e
p
u
r
p
o
s
es
o
f
r
ec
o
g
n
izin
g
all
t
y
p
es
o
f
im
ag
es,
i
n
clu
d
in
g
im
ag
es
o
f
letter
s
an
d
ch
ar
ac
ter
s
as sh
o
w
n
in
Fig
u
r
e
6
.
T
ab
le
3.
Per
f
o
r
m
an
ce
co
m
p
a
r
is
o
n
o
f
d
i
f
f
er
en
t c
lass
if
ier
s
D
a
t
a
s
e
t
M
o
d
e
l
/
a
p
p
r
o
a
c
h
R
e
p
o
r
t
e
d
a
c
c
u
r
a
c
y
(
%)
H
A
C
D
B
S
u
p
p
o
r
t
v
e
c
t
o
r
ma
c
h
i
n
e
[
2
3
]
8
6
.
1
H
A
C
D
B
K
-
n
e
a
r
e
s
t
n
e
i
g
h
b
o
r
s
[
2
4
]
8
5
.
4
7
H
A
A
W
P
S
u
p
p
o
r
t
v
e
c
t
o
r
ma
c
h
i
n
e
w
i
t
h
h
a
n
d
c
r
a
f
t
e
d
f
e
a
t
u
r
e
s
[
2
5
]
8
7
.
2
H
A
C
D
B
a
n
d
A
H
A
W
P
H
i
st
o
g
r
a
m
o
f
o
r
i
e
n
t
e
d
g
r
a
d
i
e
n
t
s
(
H
O
G
)
+
m
u
l
t
i
l
a
y
e
r
p
e
r
c
e
p
t
r
o
n
(
M
LP
)
[
2
6
]
8
8
.
2
H
A
C
D
B
a
n
d
A
H
A
W
P
S
u
p
p
o
r
t
v
e
c
t
o
r
ma
c
h
i
n
e
[
2
5
]
8
7
.
5
H
A
C
D
B
a
n
d
A
H
A
W
P
P
r
o
p
o
se
d
C
N
N
[
2
5
]
8
9
.
2
5.
CO
NCLU
SI
O
N
T
h
is
p
ap
er
d
e
v
elo
p
ed
an
d
te
s
ted
a
C
NN
f
o
r
h
an
d
wr
itten
Ar
ab
ic
letter
r
ec
o
g
n
itio
n
,
ac
h
iev
in
g
an
ac
cu
r
ac
y
o
f
8
9
.
2
%
o
n
a
ca
r
e
f
u
lly
p
r
e
p
ar
ed
d
ataset.
T
h
e
d
a
taset
u
s
ed
in
th
is
p
ap
er
is
co
m
p
o
s
ed
o
f
s
elec
ted
im
ag
es
f
r
o
m
th
e
HACDB
an
d
AHAWP
d
atase
ts
.
A
to
tal
o
f
1
4
,
0
0
0
im
ag
es
wer
e
c
o
llected
an
d
wer
e
u
s
ed
f
o
r
test
in
g
an
d
t
r
ain
in
g
C
NN.
T
h
e
r
esu
lts
co
n
f
ir
m
th
e
s
u
itab
ilit
y
o
f
C
NNs
f
o
r
t
h
is
task
,
wh
il
e
also
h
ig
h
lig
h
tin
g
ch
allen
g
es
s
u
ch
as
o
v
e
r
f
itti
n
g
an
d
lim
ited
d
ataset
d
iv
er
s
ity
.
Fu
tu
r
e
w
o
r
k
co
u
ld
b
u
ild
o
n
th
is
f
o
u
n
d
atio
n
b
y
ap
p
ly
in
g
d
ee
p
er
ar
ch
itectu
r
e,
d
ata
au
g
m
e
n
tatio
n
,
an
d
tr
a
n
s
f
er
lear
n
in
g
to
im
p
r
o
v
e
ac
cu
r
ac
y
,
as
well
as
in
teg
r
atin
g
lin
g
u
is
tic
m
o
d
els f
o
r
h
an
d
lin
g
c
o
n
n
ec
ted
s
cr
ip
t a
n
d
am
b
i
g
u
o
u
s
ch
ar
ac
ter
s
.
F
UNDING
I
NF
O
R
M
A
T
I
O
N
Au
th
o
r
s
s
tate
n
o
f
u
n
d
in
g
in
v
o
lv
ed
.
AUTHO
R
CO
NT
RI
B
UT
I
O
NS ST
A
T
E
M
E
N
T
T
h
is
jo
u
r
n
al
u
s
es
th
e
C
o
n
tr
ib
u
to
r
R
o
les
T
ax
o
n
o
m
y
(
C
R
ed
iT)
to
r
ec
o
g
n
ize
in
d
iv
id
u
al
au
th
o
r
co
n
tr
ib
u
tio
n
s
,
r
ed
u
ce
au
th
o
r
s
h
ip
d
is
p
u
tes,
an
d
f
ac
ilit
ate
co
llab
o
r
atio
n
.
Na
m
e
o
f
Aut
ho
r
C
M
So
Va
Fo
I
R
D
O
E
Vi
Su
P
Fu
An
as Qu
teis
h
at
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
Ash
r
af
Al
Sh
ar
ah
✓
✓
✓
✓
✓
Ah
m
ed
Qtaish
at
✓
✓
✓
✓
✓
✓
✓
T
ar
eq
Ah
m
a
d
Alawn
eh
✓
✓
✓
✓
✓
✓
✓
C
:
C
o
n
c
e
p
t
u
a
l
i
z
a
t
i
o
n
M
:
M
e
t
h
o
d
o
l
o
g
y
So
:
So
f
t
w
a
r
e
Va
:
Va
l
i
d
a
t
i
o
n
Fo
:
Fo
r
mal
a
n
a
l
y
s
i
s
I
:
I
n
v
e
s
t
i
g
a
t
i
o
n
R
:
R
e
so
u
r
c
e
s
D
:
D
a
t
a
C
u
r
a
t
i
o
n
O
:
W
r
i
t
i
n
g
-
O
r
i
g
i
n
a
l
D
r
a
f
t
E
:
W
r
i
t
i
n
g
-
R
e
v
i
e
w
&
E
d
i
t
i
n
g
Vi
:
Vi
su
a
l
i
z
a
t
i
o
n
Su
:
Su
p
e
r
v
i
s
i
o
n
P
:
P
r
o
j
e
c
t
a
d
mi
n
i
st
r
a
t
i
o
n
Fu
:
Fu
n
d
i
n
g
a
c
q
u
i
si
t
i
o
n
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
15
,
No
.
5
,
Octo
b
er
20
26
:
4
0
1
5
-
4
0
2
5
4024
CO
NF
L
I
C
T
O
F
I
N
T
E
R
E
S
T
ST
A
T
E
M
E
NT
Au
th
o
r
s
d
ec
lar
e
th
at
t
h
er
e
is
n
o
co
n
f
lict o
f
in
ter
est.
DATA AV
AI
L
AB
I
L
I
T
Y
T
h
e
d
ata
th
at
s
u
p
p
o
r
t
th
e
f
in
d
i
n
g
s
o
f
th
is
s
tu
d
y
a
r
e
o
p
e
n
ly
a
v
ailab
le
in
Kag
g
le
an
d
Me
n
d
e
ley
D
ata
at
h
t
t
p
s
:
/
/
w
w
w
.
k
a
g
g
l
e
.
c
o
m
/
d
a
t
a
s
e
t
s
/
m
l
o
e
y
1
/
a
h
c
d
1
a
n
d
h
t
t
p
s
:
/
/
d
a
t
a
.
m
e
n
d
e
l
e
y
.
c
o
m
/
d
a
t
a
s
e
t
s
/
2
h
7
6
6
7
2
z
n
t
/
2
,
r
e
s
p
e
c
t
i
v
e
l
y
.
RE
F
E
R
E
NC
E
S
[
1
]
A
.
A
.
I
.
A
l
a
sad
i
,
M
.
S
,
A
.
S
mera
t
,
K
.
G
o
v
i
n
d
u
,
G
.
S
i
v
a
k
u
m
a
r
,
a
n
d
V
.
S
a
h
a
sr
a
n
a
mam,
“
M
a
t
h
e
ma
t
i
c
a
l
l
y
mo
d
i
f
i
e
d
d
e
e
p
l
e
a
r
n
i
n
g
mo
d
e
l
a
ssi
s
t
e
d
h
a
n
d
w
r
i
t
t
e
n
d
i
g
i
t
r
e
c
o
g
n
i
t
i
o
n
f
o
r
i
n
t
e
l
l
i
g
e
n
t
d
o
c
u
me
n
t
p
r
o
c
e
ssi
n
g
s
y
s
t
e
ms
,
”
J
o
u
r
n
a
l
o
f
M
a
c
h
i
n
e
a
n
d
C
o
m
p
u
t
i
n
g
,
v
o
l
.
5
,
n
o
.
4
,
p
p
.
2
6
6
1
–
2
6
7
1
,
O
c
t
.
2
0
2
5
,
d
o
i
:
1
0
.
5
3
7
5
9
/
7
6
6
9
/
j
m
c
2
0
2
5
0
5
2
0
4
.
[
2
]
R
.
N
a
j
a
m
a
n
d
S
.
F
a
i
z
u
l
l
a
h
,
“
A
n
a
l
y
si
s
o
f
r
e
c
e
n
t
d
e
e
p
l
e
a
r
n
i
n
g
t
e
c
h
n
i
q
u
e
s
f
o
r
A
r
a
b
i
c
h
a
n
d
w
r
i
t
t
e
n
-
t
e
x
t
O
C
R
a
n
d
p
o
s
t
-
O
C
R
c
o
r
r
e
c
t
i
o
n
,
”
A
p
p
l
i
e
d
S
c
i
e
n
c
e
s
,
v
o
l
.
1
3
,
n
o
.
1
3
,
J
u
n
.
2
0
2
3
,
d
o
i
:
1
0
.
3
3
9
0
/
a
p
p
1
3
1
3
7
5
6
8
.
[
3
]
T.
T.
H
.
N
g
u
y
e
n
,
A
.
J
a
t
o
w
t
,
M
.
C
o
u
s
t
a
t
y
,
a
n
d
A
.
D
o
u
c
e
t
,
“
S
u
r
v
e
y
o
f
p
o
s
t
-
O
C
R
p
r
o
c
e
ssi
n
g
a
p
p
r
o
a
c
h
e
s,
”
AC
M
C
o
m
p
u
t
i
n
g
S
u
r
v
e
y
s
,
v
o
l
.
5
4
,
n
o
.
6
,
p
p
.
1
–
3
7
,
J
u
l
.
2
0
2
2
,
d
o
i
:
1
0
.
1
1
4
5
/
3
4
5
3
4
7
6
.
[
4
]
S
.
F
a
i
z
u
l
l
a
h
,
M
.
S
.
A
y
u
b
,
S
.
H
u
ssa
i
n
,
a
n
d
M
.
A
.
K
h
a
n
,
“
A
s
u
r
v
e
y
o
f
O
C
R
i
n
A
r
a
b
i
c
l
a
n
g
u
a
g
e
:
a
p
p
l
i
c
a
t
i
o
n
s
,
t
e
c
h
n
i
q
u
e
s
,
a
n
d
c
h
a
l
l
e
n
g
e
s,”
A
p
p
l
i
e
d
S
c
i
e
n
c
e
s
,
v
o
l
.
1
3
,
n
o
.
7
,
A
p
r
.
2
0
2
3
,
d
o
i
:
1
0
.
3
3
9
0
/
a
p
p
1
3
0
7
4
5
8
4
.
[
5
]
C
.
A
.
S
c
h
m
i
d
t
,
K
.
H
.
F
.
-
W
e
l
l
ma
n
,
a
n
d
P
.
D
.
N
e
u
f
e
r
,
“
F
r
o
m
O
C
R
a
n
d
EC
A
R
t
o
e
n
e
r
g
y
:
p
e
r
s
p
e
c
t
i
v
e
s
o
n
t
h
e
d
e
si
g
n
a
n
d
i
n
t
e
r
p
r
e
t
a
t
i
o
n
o
f
b
i
o
e
n
e
r
g
e
t
i
c
s
st
u
d
i
e
s,
”
J
o
u
r
n
a
l
o
f
B
i
o
l
o
g
i
c
a
l
C
h
e
m
i
st
r
y
,
v
o
l
.
2
9
7
,
n
o
.
4
,
O
c
t
.
2
0
2
1
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
j
b
c
.
2
0
2
1
.
1
0
1
1
4
0
.
[
6
]
A
.
A
.
-
K
h
a
d
r
a
h
e
t
a
l
.
,
“
E
n
h
a
n
c
i
n
g
i
n
t
r
u
si
o
n
d
e
t
e
c
t
i
o
n
s
y
st
e
ms
u
si
n
g
ma
c
h
i
n
e
l
e
a
r
n
i
n
g
a
n
d
a
d
v
a
n
c
e
d
f
e
a
t
u
r
e
sel
e
c
t
i
o
n
met
h
o
d
s
,
”
El
e
c
t
r
o
n
i
c
s
,
v
o
l
.
1
5
,
n
o
.
9
,
A
p
r
.
2
0
2
6
,
d
o
i
:
1
0
.
3
3
9
0
/
e
l
e
c
t
r
o
n
i
c
s
1
5
0
9
1
8
6
0
.
[
7
]
M
.
S
.
K
a
sem
,
M
.
M
a
h
m
o
u
d
,
a
n
d
H
.
-
S
.
K
a
n
g
,
“
A
d
v
a
n
c
e
me
n
t
s
a
n
d
c
h
a
l
l
e
n
g
e
s
i
n
A
r
a
b
i
c
o
p
t
i
c
a
l
c
h
a
r
a
c
t
e
r
r
e
c
o
g
n
i
t
i
o
n
:
a
c
o
m
p
r
e
h
e
n
si
v
e
su
r
v
e
y
,
”
A
C
M
C
o
m
p
u
t
i
n
g
S
u
rv
e
y
s
,
v
o
l
.
5
8
,
n
o
.
4
,
p
p
.
1
–
3
7
,
M
a
r
.
2
0
2
6
,
d
o
i
:
1
0
.
1
1
4
5
/
3
7
6
8
1
5
0
.
[
8
]
H
.
A
l
-
Y
o
u
sefi
a
n
d
S
.
S
.
U
d
p
a
,
“
R
e
c
o
g
n
i
t
i
o
n
o
f
A
r
a
b
i
c
c
h
a
r
a
c
t
e
r
s,”
I
EEE
T
ra
n
s
a
c
t
i
o
n
s
o
n
P
a
t
t
e
r
n
A
n
a
l
y
si
s
a
n
d
Ma
c
h
i
n
e
I
n
t
e
l
l
i
g
e
n
c
e
,
v
o
l
.
1
4
,
n
o
.
8
,
p
p
.
8
5
3
–
8
5
7
,
1
9
9
2
,
d
o
i
:
1
0
.
1
1
0
9
/
3
4
.
1
4
9
5
8
5
.
[
9
]
M
.
Li
e
t
a
l
.
,
“
Tr
O
C
R
:
t
r
a
n
sf
o
r
mer
-
b
a
s
e
d
o
p
t
i
c
a
l
c
h
a
r
a
c
t
e
r
r
e
c
o
g
n
i
t
i
o
n
w
i
t
h
p
r
e
-
t
r
a
i
n
e
d
m
o
d
e
l
s,”
T
h
e
T
h
i
rt
y
-
S
e
v
e
n
t
h
AAAI
C
o
n
f
e
re
n
c
e
o
n
Art
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
(
AAAI
-
2
3
)
,
v
o
l
.
3
7
,
n
o
.
1
1
,
p
p
.
1
3
0
9
4
–
1
3
1
0
2
,
J
u
n
.
2
0
2
3
,
d
o
i
:
1
0
.
1
6
0
9
/
a
a
a
i
.
v
3
7
i
1
1
.
2
6
5
3
8
.
[
1
0
]
A
.
M
o
r
t
a
d
i
e
t
a
l
.
,
“
A
LN
A
S
I
K
H
:
a
n
A
r
a
b
i
c
O
C
R
sy
s
t
e
m
b
a
s
e
d
o
n
t
r
a
n
sf
o
r
mers
,
”
i
n
2
0
2
3
I
n
t
e
r
n
a
t
i
o
n
a
l
Mo
b
i
l
e
,
I
n
t
e
l
l
i
g
e
n
t
,
a
n
d
U
b
i
q
u
i
t
o
u
s
C
o
m
p
u
t
i
n
g
C
o
n
f
e
re
n
c
e
(
MIU
C
C
)
,
S
e
p
.
2
0
2
3
,
p
p
.
7
4
–
81
,
d
o
i
:
1
0
.
1
1
0
9
/
M
I
U
C
C
5
8
8
3
2
.
2
0
2
3
.
1
0
2
7
8
3
7
0
.
[
1
1
]
A
.
M
o
s
t
a
f
a
e
t
a
l
.
,
“
O
C
F
o
r
m
e
r
:
a
t
r
a
n
s
f
o
r
m
e
r
-
b
a
s
e
d
m
o
d
e
l
f
o
r
A
r
a
b
i
c
h
a
n
d
w
r
i
t
t
e
n
t
e
x
t
r
e
c
o
g
n
i
t
i
o
n
,
”
i
n
2
0
2
1
I
n
t
e
r
n
a
t
i
o
n
a
l
M
o
b
i
l
e
,
I
n
t
e
l
l
i
g
e
n
t
,
a
n
d
U
b
i
q
u
i
t
o
u
s
C
o
m
p
u
t
i
n
g
C
o
n
f
e
r
e
n
c
e
(
M
I
U
C
C
)
,
M
a
y
2
0
2
1
,
p
p
.
1
8
2
–
186
,
d
o
i
:
1
0
.
1
1
0
9
/
M
I
U
C
C
5
2
5
3
8
.
2
0
2
1
.
9
4
4
7
6
0
8
.
[
1
2
]
F
.
V
a
l
i
z
a
d
e
h
,
F
.
G
h
a
sem
i
a
n
,
a
n
d
E.
S
h
a
b
a
n
i
n
i
a
,
“
C
o
m
p
a
r
a
t
i
v
e
a
n
a
l
y
s
i
s
o
f
l
a
r
g
e
l
a
n
g
u
a
g
e
m
o
d
e
l
s
f
o
r
O
C
R
p
o
st
-
p
r
o
c
e
ss
i
n
g
i
n
P
e
r
si
a
n
:
f
r
o
m
P
a
r
sB
ER
T
t
o
G
P
T,
”
i
n
2
0
2
5
2
9
t
h
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
m
p
u
t
e
r
C
o
n
f
e
r
e
n
c
e
,
C
o
m
p
u
t
e
r
S
o
c
i
e
t
y
o
f
I
ra
n
(
C
S
I
C
C
)
,
F
e
b
.
2
0
2
5
,
p
p
.
1
–
6
,
d
o
i
:
1
0
.
1
1
0
9
/
C
S
I
C
C
6
5
7
6
5
.
2
0
2
5
.
1
0
9
6
7
4
7
2
.
[
1
3
]
Y
.
B
e
n
g
i
o
,
Y
.
Le
C
u
n
,
C
.
N
o
h
l
,
a
n
d
C
.
B
u
r
g
e
s,
“
Le
R
e
c
:
a
N
N
/
H
M
M
h
y
b
r
i
d
f
o
r
o
n
-
l
i
n
e
h
a
n
d
w
r
i
t
i
n
g
r
e
c
o
g
n
i
t
i
o
n
,
”
N
e
u
r
a
l
C
o
m
p
u
t
a
t
i
o
n
,
v
o
l
.
7
,
n
o
.
6
,
p
p
.
1
2
8
9
–
1
3
0
3
,
N
o
v
.
1
9
9
5
,
d
o
i
:
1
0
.
1
1
6
2
/
n
e
c
o
.
1
9
9
5
.
7
.
6
.
1
2
8
9
.
[
1
4
]
M
.
A
.
-
Ja
b
e
r
,
N
.
M
a
k
h
o
u
l
,
M
.
A
.
-
J
a
b
e
r
,
a
n
d
R
.
B
e
a
l
e
,
“
C
o
s
t
p
r
e
d
i
c
t
i
o
n
f
o
r
p
r
o
d
u
c
t
d
e
v
e
l
o
p
m
e
n
t
u
si
n
g
h
y
b
r
i
d
d
e
e
p
l
e
a
r
n
i
n
g
mo
d
e
l
:
a
met
a
-
h
e
u
r
i
st
i
c
m
o
d
e
l
,
”
Mu
l
t
i
m
e
d
i
a
T
o
o
l
s
a
n
d
A
p
p
l
i
c
a
t
i
o
n
s
,
v
o
l
.
8
4
,
n
o
.
2
7
,
p
p
.
3
3
3
0
7
–
3
3
3
3
0
,
D
e
c
.
2
0
2
4
,
d
o
i
:
1
0
.
1
0
0
7
/
s
1
1
0
4
2
-
024
-
2
0
4
3
7
-
y.
[
1
5
]
L.
A
l
z
u
b
a
i
d
i
e
t
a
l
.
,
“
R
e
v
i
e
w
o
f
d
e
e
p
l
e
a
r
n
i
n
g
:
c
o
n
c
e
p
t
s,
C
N
N
a
r
c
h
i
t
e
c
t
u
r
e
s
,
c
h
a
l
l
e
n
g
e
s,
a
p
p
l
i
c
a
t
i
o
n
s
,
f
u
t
u
r
e
d
i
r
e
c
t
i
o
n
s,
”
J
o
u
rn
a
l
o
f
Bi
g
D
a
t
a
,
v
o
l
.
8
,
n
o
.
1
,
M
a
r
.
2
0
2
1
,
d
o
i
:
1
0
.
1
1
8
6
/
s
4
0
5
3
7
-
0
2
1
-
0
0
4
4
4
-
8.
[
1
6
]
T.
K
a
t
t
e
n
b
o
r
n
,
J.
Le
i
t
l
o
f
f
,
F
.
S
c
h
i
e
f
e
r
,
a
n
d
S
.
H
i
n
z
,
“
R
e
v
i
e
w
o
n
c
o
n
v
o
l
u
t
i
o
n
a
l
n
e
u
r
a
l
n
e
t
w
o
r
k
s
(
C
N
N
)
i
n
v
e
g
e
t
a
t
i
o
n
r
e
mo
t
e
sen
si
n
g
,
”
I
S
PRS
J
o
u
rn
a
l
o
f
Ph
o
t
o
g
ra
m
m
e
t
ry
a
n
d
Re
m
o
t
e
S
e
n
si
n
g
,
v
o
l
.
1
7
3
,
p
p
.
2
4
–
4
9
,
M
a
r
.
2
0
2
1
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
i
sp
r
sj
p
r
s.2
0
2
0
.
1
2
.
0
1
0
.
[
1
7
]
J.
G
u
p
t
a
,
S
.
P
a
t
h
a
k
,
a
n
d
G
.
K
u
m
a
r
,
“
D
e
e
p
l
e
a
r
n
i
n
g
(
C
N
N
)
a
n
d
t
r
a
n
sf
e
r
l
e
a
r
n
i
n
g
:
a
r
e
v
i
e
w
,
”
J
o
u
r
n
a
l
o
f
P
h
y
si
c
s:
C
o
n
f
e
re
n
c
e
S
e
r
i
e
s
,
v
o
l
.
2
2
7
3
,
n
o
.
1
,
M
a
y
2
0
2
2
,
d
o
i
:
1
0
.
1
0
8
8
/
1
7
4
2
-
6
5
9
6
/
2
2
7
3
/
1
/
0
1
2
0
2
9
.
[
1
8
]
K
.
S
h
a
h
e
e
d
e
t
a
l
.
,
“
D
S
-
C
N
N
:
a
p
r
e
-
t
r
a
i
n
e
d
X
c
e
p
t
i
o
n
mo
d
e
l
b
a
se
d
o
n
d
e
p
t
h
-
w
i
se
se
p
a
r
a
b
l
e
c
o
n
v
o
l
u
t
i
o
n
a
l
n
e
u
r
a
l
n
e
t
w
o
r
k
f
o
r
f
i
n
g
e
r
v
e
i
n
r
e
c
o
g
n
i
t
i
o
n
,
”
E
x
p
e
rt
S
y
st
e
m
s w
i
t
h
A
p
p
l
i
c
a
t
i
o
n
s
,
v
o
l
.
1
9
1
,
A
p
r
.
2
0
2
2
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
e
sw
a
.
2
0
2
1
.
1
1
6
2
8
8
.
[
1
9
]
Y
.
Zh
a
n
g
e
t
a
l
.
,
“
C
a
r
r
y
i
n
g
o
u
t
C
N
N
c
h
a
n
n
e
l
p
r
u
n
i
n
g
i
n
a
w
h
i
t
e
b
o
x
,
”
I
EEE
T
r
a
n
s
a
c
t
i
o
n
s
o
n
N
e
u
r
a
l
N
e
t
w
o
r
k
s
a
n
d
L
e
a
rn
i
n
g
S
y
s
t
e
m
s
,
v
o
l
.
3
4
,
n
o
.
1
0
,
p
p
.
7
9
4
6
–
7
9
5
5
,
O
c
t
.
2
0
2
3
,
d
o
i
:
1
0
.
1
1
0
9
/
TN
N
LS
.
2
0
2
2
.
3
1
4
7
2
6
9
.
[
2
0
]
H
.
S
.
G
i
l
l
a
n
d
B
.
S
.
K
h
e
h
r
a
,
“
A
n
i
n
t
e
g
r
a
t
e
d
a
p
p
r
o
a
c
h
u
si
n
g
C
N
N
-
R
N
N
-
LSTM
f
o
r
c
l
a
ssi
f
i
c
a
t
i
o
n
o
f
f
r
u
i
t
i
mag
e
s,”
M
a
t
e
ri
a
l
s
T
o
d
a
y
:
Pro
c
e
e
d
i
n
g
s
,
v
o
l
.
5
1
,
p
p
.
5
9
1
–
5
9
5
,
2
0
2
2
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
ma
t
p
r
.
2
0
2
1
.
0
6
.
0
1
6
.
[
2
1
]
M
.
K
h
a
n
,
“
A
r
a
b
i
c
h
a
n
d
w
r
i
t
t
e
n
a
l
p
h
a
b
e
t
s,
w
o
r
d
s
a
n
d
p
a
r
a
g
r
a
p
h
s
p
e
r
u
ser
(
A
H
A
W
P
)
d
a
t
a
set
,
”
Me
n
d
e
l
e
y
D
a
t
a
,
V
1
,
2
0
2
1
,
d
o
i
:
1
0
.
1
7
6
3
2
/
2
h
7
6
6
7
2
z
n
t
.
1
.
[
2
2
]
A
.
La
w
g
a
l
i
,
M
.
A
n
g
e
l
o
v
a
,
a
n
d
A
.
B
o
u
r
i
d
a
n
e
,
“
H
A
C
D
B
:
h
a
n
d
w
r
i
t
t
e
n
A
r
a
b
i
c
c
h
a
r
a
c
t
e
r
s
d
a
t
a
b
a
se
f
o
r
a
u
t
o
ma
t
i
c
c
h
a
r
a
c
t
e
r
r
e
c
o
g
n
i
t
i
o
n
,
”
i
n
2
0
1
3
4
t
h
E
u
r
o
p
e
a
n
W
o
rks
h
o
p
o
n
V
i
s
u
a
l
I
n
f
o
rm
a
t
i
o
n
Pr
o
c
e
s
si
n
g
,
EU
VI
P
2
0
1
3
,
2
0
1
3
,
p
p
.
2
5
5
–
259.
[
2
3
]
M
.
N
.
A
l
J
a
r
r
a
h
,
M
.
M
.
Zy
o
u
t
,
a
n
d
R
.
D
u
w
a
i
r
i
,
“
A
r
a
b
i
c
h
a
n
d
w
r
i
t
t
e
n
c
h
a
r
a
c
t
e
r
s
r
e
c
o
g
n
i
t
i
o
n
u
s
i
n
g
c
o
n
v
o
l
u
t
i
o
n
a
l
n
e
u
r
a
l
n
e
t
w
o
r
k
,
”
i
n
2
0
2
1
1
2
t
h
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
r
e
n
c
e
o
n
I
n
f
o
rm
a
t
i
o
n
a
n
d
C
o
m
m
u
n
i
c
a
t
i
o
n
S
y
st
e
m
s
(
I
C
I
C
S
)
,
M
a
y
2
0
2
1
,
p
p
.
1
8
2
–
1
8
8
,
d
o
i
:
1
0
.
1
1
0
9
/
I
C
I
C
S
5
2
4
5
7
.
2
0
2
1
.
9
4
6
4
5
9
6
.
[
2
4
]
M
.
L
o
e
y
,
A
.
E
l
-
S
a
w
y
,
a
n
d
H
.
E
L
-
B
a
k
r
y
,
“
D
e
e
p
l
e
a
r
n
i
n
g
a
u
t
o
e
n
c
o
d
e
r
a
p
p
r
o
a
c
h
f
o
r
h
a
n
d
w
r
i
t
t
e
n
a
r
a
b
i
c
d
i
g
i
t
s
r
e
c
o
g
n
i
t
i
o
n
,
”
2
0
1
7
,
a
rXi
v
:
1
7
0
6
.
0
6
7
2
0
.
[
2
5
]
A
.
La
w
g
a
l
i
,
“
A
s
u
r
v
e
y
o
n
A
r
a
b
i
c
c
h
a
r
a
c
t
e
r
r
e
c
o
g
n
i
t
i
o
n
,
”
I
n
t
e
rn
a
t
i
o
n
a
l
J
o
u
rn
a
l
o
f
S
i
g
n
a
l
Pro
c
e
ss
i
n
g
,
I
m
a
g
e
Pr
o
c
e
ssi
n
g
a
n
d
Pa
t
t
e
r
n
Re
c
o
g
n
i
t
i
o
n
,
v
o
l
.
8
,
n
o
.
2
,
p
p
.
4
0
1
–
4
2
6
,
F
e
b
.
2
0
1
5
,
d
o
i
:
1
0
.
1
4
2
5
7
/
i
j
si
p
.
2
0
1
5
.
8
.
2
.
3
7
.
[
2
6
]
A
.
M
e
z
g
h
a
n
i
,
R
.
M
a
a
l
e
j
,
M
.
El
l
e
u
c
h
,
a
n
d
M
.
K
h
e
r
a
l
l
a
h
,
“
R
e
c
e
n
t
a
d
v
a
n
c
e
s
o
f
M
L
a
n
d
D
L
a
p
p
r
o
a
c
h
e
s
f
o
r
A
r
a
b
i
c
h
a
n
d
w
r
i
t
i
n
g
r
e
c
o
g
n
i
t
i
o
n
:
a
r
e
v
i
e
w
,
”
I
n
t
e
r
n
a
t
i
o
n
a
l
J
o
u
r
n
a
l
o
f
H
y
b
r
i
d
I
n
t
e
l
l
i
g
e
n
t
S
y
s
t
e
m
s
,
v
o
l
.
1
9
,
n
o
.
1
-
2
,
p
p
.
6
1
–
7
8
,
Ju
l
.
2
0
2
3
,
d
o
i
:
1
0
.
3
2
3
3
/
H
I
S
-
2
3
0
0
0
5
.
Evaluation Warning : The document was created with Spire.PDF for Python.