I
nte
rna
t
io
na
l J
o
urna
l o
f
E
lect
rica
l a
nd
Co
m
pu
t
er
E
ng
ineering
(
I
J
E
CE
)
Vo
l.
15
,
No
.
2
,
A
p
r
il
20
25
,
p
p
.
2
2
9
1
~
2
3
0
3
I
SS
N:
2088
-
8
7
0
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijece.
v
15
i
2
.
pp
2
2
9
1
-
2
3
0
3
2291
J
o
ur
na
l
ho
m
ep
a
g
e
:
h
ttp
:
//ij
ec
e.
ia
esco
r
e.
co
m
H
a
ndwri
tt
en
t
e
x
t
recog
nition sy
ste
m using
Ras
pberr
y
P
i wi
th
O
pen
CV Tenso
r
F
lo
w
J
a
m
il Abeda
lra
him
J
a
m
il Al
s
a
y
a
y
deh
1
,
T
o
m
my
L
ee
Chui
n J
ie
1
,
Rex
B
a
ca
rr
a
2
,
B
enny
O
g
un
s
ho
la
3
,
No
o
ra
y
is
a
hb
e
M
o
hd
Ya
a
co
b
4
1
D
e
p
a
r
t
me
n
t
o
f
E
n
g
i
n
e
e
r
i
n
g
T
e
c
h
n
o
l
o
g
y
,
F
a
k
u
l
t
i
Te
k
n
o
l
o
g
i
a
n
d
K
e
j
u
r
u
t
e
r
a
a
n
El
e
k
t
r
o
n
i
k
a
n
d
K
o
m
p
u
t
e
r
,
U
n
i
v
e
r
si
t
i
Te
k
n
i
k
a
l
M
a
l
a
y
si
a
M
e
l
a
k
a
,
M
e
l
a
k
a
,
M
a
l
a
y
si
a
2
D
e
p
a
r
t
me
n
t
o
f
G
e
n
e
r
a
l
E
d
u
c
a
t
i
o
n
a
n
d
F
o
u
n
d
a
t
i
o
n
,
R
a
b
a
d
a
n
A
c
a
d
e
m
y
,
A
b
u
D
h
a
b
i
,
U
n
i
t
e
d
A
r
a
b
Em
i
r
a
t
e
s
3
D
e
p
a
r
t
me
n
t
o
f
El
e
c
t
r
i
c
a
l
a
n
d
Te
l
e
c
o
mm
u
n
i
c
a
t
i
o
n
s
E
n
g
i
n
e
e
r
i
n
g
,
F
r
o
n
t
i
e
r
i
n
st
i
t
u
t
e
o
f
t
e
c
h
n
o
l
o
g
y
,
G
l
e
n
r
o
y
C
i
t
y
,
A
u
st
r
a
l
i
a
4
C
e
n
t
e
r
f
o
r
S
o
f
t
w
a
r
e
T
e
c
h
n
o
l
o
g
y
a
n
d
M
a
n
a
g
e
m
e
n
t
(
S
O
F
TA
M
)
,
F
a
c
u
l
t
y
o
f
I
n
f
o
r
mat
i
o
n
S
c
i
e
n
c
e
a
n
d
T
e
c
h
n
o
l
o
g
y
,
U
n
i
v
e
r
si
t
y
K
e
b
a
n
g
saa
n
M
a
l
a
y
s
i
a
,
S
e
l
a
n
g
o
r
,
M
a
l
a
y
si
a
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Ma
r
1
6
,
2
0
2
4
R
ev
is
ed
Oct
2
4
,
2
0
2
4
Acc
ep
ted
No
v
2
0
,
2
0
2
4
H
a
n
d
writt
e
n
tex
t
re
c
o
g
n
it
io
n
(
HTR)
tec
h
n
o
lo
g
y
h
a
s
b
ro
u
g
h
t
a
b
o
u
t
a
re
v
o
lu
ti
o
n
i
n
t
h
e
wa
y
h
a
n
d
writ
t
e
n
d
a
ta
is
c
o
n
v
e
rted
a
n
d
a
n
a
ly
z
e
d
.
Th
is
p
ro
p
o
se
d
wo
r
k
fo
c
u
se
s
o
n
d
e
v
e
l
o
p
i
n
g
a
HTR
sy
ste
m
u
sin
g
d
e
e
p
lea
rn
in
g
th
ro
u
g
h
a
d
v
a
n
c
e
d
d
e
e
p
lea
rn
i
n
g
a
rc
h
it
e
c
tu
re
a
n
d
tec
h
n
iq
u
e
s.
Th
e
a
im
is
to
c
re
a
te
a
m
o
d
e
l
f
o
r
re
a
l
-
ti
m
e
a
n
a
l
y
sis
a
n
d
d
e
tec
ti
o
n
o
f
h
a
n
d
writt
e
n
tex
ts.
Th
e
p
ro
p
o
se
d
d
e
e
p
lea
rn
i
n
g
a
rc
h
i
tec
tu
re
th
a
t
is
c
o
n
v
o
l
u
ti
o
n
a
l
n
e
u
ra
l
n
e
two
rk
s
(CNN
s),
is
i
n
v
e
sti
g
a
ted
a
n
d
i
m
p
lem
e
n
ted
wit
h
t
o
o
ls
li
k
e
O
p
e
n
CV
a
n
d
Ten
so
rF
l
o
w.
T
h
e
m
o
d
e
l
is
train
e
d
o
n
larg
e
h
a
n
d
wri
tt
e
n
d
a
tas
e
ts
to
e
n
h
a
n
c
e
re
c
o
g
n
it
i
o
n
a
c
c
u
ra
c
y
.
T
h
e
sy
st
e
m
’s
p
e
rfo
rm
a
n
c
e
is
e
v
a
lu
a
ted
b
a
se
d
o
n
a
c
c
u
ra
c
y
,
p
re
c
isio
n
,
re
a
l
-
ti
m
e
c
a
p
a
b
il
it
ies
,
a
n
d
p
o
te
n
ti
a
l
f
o
r
d
e
p
l
o
y
m
e
n
t
o
n
p
latfo
rm
s li
k
e
Ra
sp
b
e
rry
P
i.
Th
e
a
c
tu
a
l
o
u
tco
m
e
is a ro
b
u
st HT
R
sy
ste
m
th
a
t
c
a
n
c
o
n
v
e
rt
h
a
n
d
writt
e
n
tex
t
to
d
ig
it
a
l
fo
rm
a
ts
a
c
c
u
ra
tely
.
T
h
e
d
e
v
e
lo
p
e
d
sy
ste
m
h
a
s
a
c
h
iev
e
d
a
h
ig
h
a
c
c
u
r
a
c
y
ra
te
o
f
9
1
.
5
8
%
i
n
re
c
o
g
n
izi
n
g
E
n
g
li
s
h
a
lp
h
a
b
e
ts
a
n
d
d
i
g
it
s
a
n
d
o
u
tp
e
r
fo
rm
e
d
o
t
h
e
r
m
o
d
e
ls
with
8
1
.
7
7
%
m
AP,
7
8
.
8
5
%
p
re
c
isio
n
,
7
9
.
3
2
%
re
c
a
ll
,
7
9
.
4
6
%
F
1
-
S
c
o
re
,
a
n
d
8
2
.
4
%
re
c
e
iv
e
r
o
p
e
r
a
t
i
n
g
c
h
a
ra
c
ter
is
ti
c
(R
OC)
.
T
h
i
s
re
s
e
a
rc
h
c
o
n
tr
i
b
u
tes
t
o
t
h
e
a
d
v
a
n
c
e
m
e
n
t
o
f
H
TR
tec
h
n
o
l
o
g
y
b
y
e
n
h
a
n
c
i
n
g
it
s
p
r
e
c
is
i
o
n
a
n
d
u
ti
li
t
y
.
K
ey
w
o
r
d
s
:
C
o
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
s
Dee
p
lear
n
in
g
Han
d
wr
itten
tex
t r
ec
o
g
n
itio
n
R
ea
l
-
tim
e
an
aly
s
is
R
ec
o
g
n
itio
n
ac
cu
r
ac
y
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
:
J
am
il Ab
ed
alr
ah
im
J
am
il Alsay
ay
d
eh
Dep
ar
tm
en
t o
f
E
n
g
in
ee
r
in
g
T
e
ch
n
o
lo
g
y
,
Fak
u
lti T
e
k
n
o
lo
g
i
a
n
d
Keju
r
u
te
r
aa
n
E
lek
tr
o
n
ik
a
n
d
Ko
m
p
u
te
r
,
Un
iv
er
s
iti T
ek
n
ik
al
Ma
lay
s
ia
Me
lak
a
7
6
1
0
0
Du
r
ian
T
u
n
g
g
al,
Me
lak
a
,
Ma
lay
s
ia
E
m
ail:
jam
il
@
u
tem
.
ed
u
.
m
y
1.
I
NT
RO
D
UCT
I
O
N
Han
d
wr
itten
tex
t
r
ec
o
g
n
itio
n
(
HT
R
)
tech
n
o
lo
g
y
,
p
o
we
r
ed
b
y
d
ee
p
lear
n
in
g
a
n
d
ar
tific
ial
in
tellig
en
ce
,
h
as
r
e
v
o
lu
tio
n
ized
th
e
d
ec
o
d
i
n
g
a
n
d
a
n
aly
s
is
o
f
h
a
n
d
wr
itten
d
ata
[
1
]
.
I
t
em
p
lo
y
s
o
p
tical
ch
ar
ac
ter
r
ec
o
g
n
itio
n
(
OC
R
)
,
n
eu
r
al
n
etwo
r
k
s
,
an
d
d
ee
p
l
ea
r
n
in
g
alg
o
r
ith
m
s
to
ac
cu
r
ately
id
en
tify
v
ar
io
u
s
h
an
d
wr
itin
g
s
ty
les
[
2
]
.
T
h
e
t
ec
h
n
o
lo
g
y
h
as
wid
e
-
r
an
g
i
n
g
ap
p
licatio
n
s
ac
r
o
s
s
in
d
u
s
tr
ies
s
u
ch
as
h
is
to
r
ical
d
o
cu
m
e
n
t
d
ig
itizatio
n
,
p
o
s
tal
s
er
v
ices,
f
o
r
m
s
p
r
o
ce
s
s
in
g
,
p
er
s
o
n
al
n
o
te
-
tak
in
g
,
an
d
f
in
an
ce
,
tr
an
s
f
o
r
m
in
g
tr
ad
itio
n
al
m
eth
o
d
s
o
f
p
r
o
ce
s
s
in
g
h
an
d
wr
itten
d
ata
b
y
en
h
an
cin
g
ef
f
icien
cy
,
p
r
ec
is
io
n
,
an
d
ac
ce
s
s
ib
ili
ty
.
T
h
is
wo
r
k
u
s
es
a
n
eu
r
al
n
etwo
r
k
ap
p
r
o
ac
h
to
d
ev
elo
p
an
HT
R
s
y
s
tem
th
at
p
r
o
ce
s
s
es
an
d
an
aly
ze
s
h
an
d
wr
itten
d
ata.
Neu
r
al
n
etwo
r
k
s
ar
e
tr
ai
n
ed
to
r
ec
o
g
n
ize
p
atter
n
s
an
d
m
ak
e
ac
cu
r
ate
p
r
e
d
ictio
n
s
.
T
h
e
g
o
al
is
to
cr
ea
te
an
HT
R
s
y
s
tem
th
at
ac
cu
r
ately
r
ec
o
g
n
izes
an
d
co
n
v
er
ts
h
an
d
wr
itten
tex
t
in
to
d
ig
ital
f
o
r
m
ats
[
3
]
.
T
h
e
s
y
s
tem
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
15
,
No
.
2
,
Ap
r
il
20
25
:
2
2
9
1
-
2
3
0
3
2292
u
s
es
d
ed
icate
d
h
a
r
d
war
e
f
o
r
ca
p
tu
r
in
g
a
n
d
p
r
o
ce
s
s
in
g
h
an
d
wr
itten
in
p
u
t,
w
h
ich
aid
s
in
ef
f
icien
t
tex
t
r
ec
o
g
n
itio
n
.
T
h
is
tech
n
o
lo
g
y
r
ep
r
esen
ts
a
s
ig
n
if
ican
t a
d
v
a
n
c
em
en
t in
th
e
f
iel
d
o
f
h
an
d
wr
itt
en
d
ata
p
r
o
ce
s
s
in
g
.
T
h
e
p
r
o
ce
s
s
o
f
c
o
n
v
e
r
tin
g
h
a
n
d
wr
itten
tex
t
in
to
m
ac
h
in
e
-
r
ea
d
ab
le
f
o
r
m
at
is
ch
allen
g
in
g
d
u
e
to
th
e
wid
e
r
an
g
e
o
f
i
n
d
iv
id
u
al
h
a
n
d
wr
itin
g
s
ty
les
an
d
is
s
u
es
lik
e
lo
w
r
eso
lu
tio
n
,
p
o
o
r
co
n
tr
ast,
o
r
s
ca
n
n
in
g
d
is
to
r
tio
n
s
.
E
v
en
with
ad
v
an
c
em
en
ts
in
co
m
p
u
ter
v
is
io
n
an
d
m
ac
h
in
e
lear
n
in
g
,
th
er
e
is
a
n
ee
d
f
o
r
a
s
y
s
tem
th
at
ca
n
ac
cu
r
ately
r
ec
o
g
n
iz
e
h
an
d
wr
itten
E
n
g
lis
h
alp
h
a
b
ets
an
d
n
u
m
e
r
als.
T
h
e
p
r
o
p
o
s
ed
wo
r
k
aim
s
to
ad
d
r
ess
th
e
ch
allen
g
es
o
f
HT
R
b
y
u
s
in
g
a
co
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
(
C
NN)
to
id
en
tify
an
d
ca
teg
o
r
ize
p
atter
n
s
in
h
a
n
d
wr
itten
c
h
ar
a
cter
s
.
T
h
e
g
o
al
is
to
c
r
ea
te
a
s
y
s
tem
th
at
ca
n
class
if
y
h
an
d
wr
itten
in
p
u
t
in
to
co
r
r
esp
o
n
d
in
g
ch
ar
ac
ter
s
o
r
s
u
g
g
est
th
e
clo
s
est
m
atch
wh
en
an
ex
ac
t
o
n
e
is
n
o
t
p
o
s
s
ib
le.
T
h
e
s
y
s
tem
’
s
p
er
f
o
r
m
an
ce
will
b
e
ass
ess
e
d
b
ased
o
n
ac
cu
r
ac
y
,
p
r
ec
is
io
n
,
a
n
d
r
ec
all.
T
h
is
s
y
s
tem
co
u
ld
e
n
h
an
ce
th
e
ac
cu
r
ac
y
an
d
ef
f
icie
n
cy
o
f
HT
R
,
aid
in
g
in
th
e
c
o
n
v
e
r
s
io
n
o
f
h
an
d
wr
itten
tex
t in
t
o
m
ac
h
i
n
e
-
r
ea
d
ab
le
f
o
r
m
at.
T
h
is
liter
atu
r
e
r
ev
iew
f
o
cu
s
es
o
n
th
e
latest
tech
n
iq
u
es,
ar
ch
itectu
r
es,
an
d
m
eth
o
d
o
lo
g
i
es
u
s
ed
in
h
an
d
wr
itten
r
ec
o
g
n
itio
n
s
y
s
te
m
s
,
with
a
s
p
ec
if
ic
i
n
v
esti
g
atio
n
in
to
th
e
ca
p
ab
ilit
ies
o
f
R
asp
b
er
r
y
Pi
as
a
h
ar
d
war
e
p
latf
o
r
m
an
d
th
e
u
s
e
o
f
Op
en
C
V
an
d
T
en
s
o
r
Flo
w
f
o
r
d
ee
p
lear
n
in
g
,
im
ag
e
p
r
o
c
ess
in
g
,
an
d
f
ea
tu
r
e
ex
tr
ac
tio
n
.
I
n
s
tu
d
y
[
4
]
,
i
n
th
e
HT
R
p
r
o
ce
s
s
,
o
p
tical
c
h
ar
a
cter
r
ec
o
g
n
itio
n
is
a
cr
u
cial
s
tep
th
at
tr
an
s
f
o
r
m
s
im
ag
es
o
f
tex
t
in
to
m
ac
h
in
e
-
en
co
d
e
d
tex
t,
with
s
ig
n
if
ica
n
t
ad
v
an
ce
m
en
ts
o
v
er
th
e
y
e
ar
s
co
n
tr
ib
u
tin
g
to
im
p
r
o
v
e
d
ac
cu
r
ac
y
an
d
s
p
ee
d
.
I
n
s
tu
d
y
[
5
]
,
r
eal
-
tim
e
ch
ar
a
cter
r
ec
o
g
n
itio
n
an
d
ac
cu
r
ac
y
p
er
ce
n
tag
e
a
r
e
k
ey
asp
ec
ts
in
OC
R
,
cr
u
cial
f
o
r
au
th
en
ticatin
g
u
s
er
s
as
g
e
n
u
i
n
e
in
d
iv
id
u
als,
b
u
t
c
h
allen
g
e
s
s
u
ch
as
h
an
d
lin
g
d
if
f
er
en
t
f
o
n
ts
an
d
s
ty
les,
p
o
o
r
im
ag
e
q
u
ality
,
an
d
s
k
ewe
d
o
r
r
o
tated
tex
t p
o
s
e
s
ig
n
if
ican
t h
u
r
d
les
[
6
]
,
[
7
]
.
Sev
er
al
tech
n
iq
u
es
h
av
e
b
ee
n
p
r
o
p
o
s
ed
f
o
r
HT
R
s
y
s
tem
s
,
with
a
ty
p
ical
wo
r
k
f
lo
w
in
v
o
lv
in
g
p
r
ep
r
o
ce
s
s
in
g
,
f
ea
tu
r
e
ex
tr
ac
tio
n
,
s
eg
m
en
tatio
n
,
class
if
icatio
n
,
an
d
r
ec
o
g
n
itio
n
;
a
n
ex
am
p
le
is
a
s
y
s
tem
d
ev
elo
p
e
d
u
s
in
g
c
o
n
v
o
lu
tio
n
a
l
an
d
r
ec
u
r
r
en
t
n
eu
r
al
n
etwo
r
k
s
,
tr
ain
ed
o
n
th
e
I
AM
d
ataset,
wh
ich
ac
h
iev
ed
8
4
.
5
%
ac
cu
r
ac
y
.
I
n
s
tu
d
y
[
8
]
an
d
a
n
o
th
e
r
r
esear
ch
[
9
]
ap
p
lied
C
NN
an
d
s
u
p
p
o
r
t
v
ec
t
o
r
m
ac
h
i
n
e
(
SVM)
alg
o
r
ith
m
s
f
o
r
h
an
d
wr
itin
g
r
e
co
g
n
itio
n
,
t
r
ain
ed
an
d
ev
alu
at
ed
o
n
th
e
e
x
ten
d
e
d
MN
I
ST
(
E
MN
I
ST
)
d
ataset,
ac
h
iev
in
g
a
n
o
v
er
all
s
y
s
tem
a
cc
u
r
ac
y
o
f
8
5
.
4
1
%.
Gh
o
s
h
an
d
Kr
is
ten
s
s
o
n
[
1
0
]
d
e
p
lo
y
ed
t
h
e
tex
t
d
etec
tio
n
an
d
r
ec
o
g
n
itio
n
m
o
d
el
,
co
m
b
i
n
in
g
th
e
ef
f
icien
t
an
d
ac
cu
r
ate
s
ce
n
e
tex
t
(
E
AST
)
m
o
d
el
f
o
r
tex
t
d
etec
tio
n
an
d
th
e
T
ess
er
ac
t
-
OC
R
en
g
in
e
f
o
r
tex
t
r
ec
o
g
n
itio
n
,
tr
ai
n
ed
o
n
th
e
in
ter
n
atio
n
al
co
n
f
er
en
ce
o
n
d
o
c
u
m
en
t
an
aly
s
is
an
d
r
ec
o
g
n
itio
n
d
atab
ase,
ac
h
iev
i
n
g
a
p
r
ec
is
io
n
r
ate
o
f
8
8
.
6
9
%
.
T
h
e
ac
cu
r
ac
y
o
f
HT
R
s
y
s
tem
s
is
in
f
lu
en
ce
d
b
y
in
tr
in
s
ic
an
d
ex
tr
in
s
ic
f
ac
to
r
s
[
1
1
]
.
I
n
tr
in
s
ic
f
ac
to
r
s
in
clu
d
e
th
e
v
ar
iab
ilit
y
in
h
an
d
wr
itin
g
s
ty
les,
q
u
ality
,
an
d
th
e
wr
iter
'
s
ag
e
o
r
h
ea
lth
[
1
2
]
.
E
lem
en
ts
s
u
ch
as
letter
s
h
a
p
es,
s
lan
ts
,
s
izes,
s
p
ac
in
g
,
s
tr
o
k
e
v
ar
iatio
n
s
,
an
d
wr
itin
g
s
p
ee
d
co
n
tr
ib
u
te
to
th
e
co
m
p
lex
ity
o
f
r
ec
o
g
n
itio
n
[
1
3
]
.
Po
o
r
l
y
wr
itten
o
r
illeg
ib
l
e
tex
t
ca
n
lead
to
er
r
o
r
s
.
E
x
tr
in
s
ic
f
ac
to
r
s
in
v
o
l
v
e
p
ar
tial
o
cc
lu
s
io
n
,
lig
h
tin
g
co
n
d
itio
n
s
,
p
a
p
er
o
r
wr
itin
g
in
s
tr
u
m
en
t
q
u
ality
,
wr
itin
g
o
r
s
ca
n
n
i
n
g
an
g
le,
an
d
im
ag
e
r
eso
lu
tio
n
[
1
4
]
,
[
1
5
]
.
T
h
ese
f
ac
to
r
s
ca
n
af
f
ec
t
th
e
s
y
s
tem
'
s
p
er
f
o
r
m
an
ce
b
y
m
a
k
in
g
it
ch
allen
g
in
g
to
tr
ain
m
o
d
els
th
at
g
en
er
alize
well
an
d
b
y
d
e
g
r
ad
in
g
im
a
g
e
q
u
ality
.
R
esear
ch
er
s
ad
d
r
ess
th
ese
is
s
u
es
th
r
o
u
g
h
ad
v
an
ce
d
p
r
ep
r
o
ce
s
s
in
g
,
d
ata
a
u
g
m
e
n
tatio
n
,
an
d
s
o
p
h
is
ticated
m
ac
h
i
n
e
lear
n
in
g
alg
o
r
ith
m
s
to
im
p
r
o
v
e
r
ec
o
g
n
itio
n
ac
cu
r
ac
y
.
Acc
o
r
d
i
n
g
to
r
ef
e
r
en
ce
[
1
6
]
,
a
n
ef
f
ec
tiv
e
HT
R
s
y
s
tem
s
h
o
u
ld
b
e
c
o
m
p
at
ib
le
with
s
ca
n
n
e
d
d
o
cu
m
e
n
ts
an
d
im
ag
es,
s
u
p
p
o
r
t
r
ea
l
-
tim
e
p
r
o
ce
s
s
in
g
,
b
e
r
o
b
u
s
t
to
h
an
d
wr
itin
g
s
ty
le
v
ar
i
atio
n
s
,
in
d
ep
en
d
e
n
t
o
f
tex
t
lan
g
u
ag
e
o
r
s
cr
ip
t,
an
d
ca
p
ab
le
o
f
h
a
n
d
lin
g
tex
t
f
r
o
m
d
if
f
er
en
t
o
r
ien
tatio
n
s
.
T
h
e
p
r
o
p
o
s
ed
s
y
s
tem
h
as
th
r
ee
p
r
im
ar
y
s
tep
s
:
tex
t
lo
ca
lizatio
n
,
f
ea
tu
r
e
ex
tr
ac
tio
n
,
an
d
class
if
icatio
n
.
I
n
tex
t
lo
ca
lizatio
n
,
s
ca
n
n
ed
d
o
cu
m
e
n
ts
ar
e
s
eg
m
e
n
ted
i
n
to
wo
r
d
im
ag
es.
Featu
r
e
ex
t
r
ac
tio
n
in
v
o
lv
es
ca
p
tu
r
in
g
p
i
x
el
d
en
s
ity
,
p
ix
el
in
ten
s
ity
v
ar
ian
ce
,
wo
r
d
m
ea
n
,
s
tan
d
ar
d
d
ev
iatio
n
o
f
p
ix
e
l
in
ten
s
ities
,
u
p
p
er
q
u
ar
ter
r
e
g
io
n
in
ten
s
ity
,
an
d
Ots
u
's
th
r
esh
o
ld
[
1
7
]
.
T
h
ese
f
ea
tu
r
es
ar
e
p
r
o
ce
s
s
ed
u
s
in
g
a
SVM
class
if
ier
to
d
is
tin
g
u
is
h
b
etwe
en
d
if
f
er
e
n
t
h
an
d
wr
itten
te
x
ts
.
T
h
ese
c
h
ar
ac
ter
is
tics
m
ak
e
C
NNs
a
p
o
wer
f
u
l
to
o
l
f
o
r
im
p
r
o
v
in
g
th
e
ac
c
u
r
ac
y
an
d
ef
f
ec
tiv
en
ess
o
f
HT
R
s
y
s
tem
s
[
1
8
]
.
T
h
e
im
p
lem
e
n
tatio
n
o
f
a
HT
R
s
y
s
tem
f
o
cu
s
es
o
n
th
e
E
MN
I
ST
d
ataset
wh
ich
s
er
v
es
as
th
e
f
o
u
n
d
atio
n
f
o
r
tr
ain
i
n
g
an
d
test
in
g
th
e
HT
R
m
o
d
el.
C
NN
ar
ch
itectu
r
e
is
em
p
lo
y
ed
f
o
r
its
ab
ilit
y
to
ca
p
tu
r
e
i
n
tr
icate
f
ea
tu
r
es
in
h
an
d
wr
itten
te
x
t.
Op
en
C
V
an
d
T
en
s
o
r
Flo
w
p
l
ay
p
iv
o
tal
r
o
les
in
p
r
e
-
p
r
o
ce
s
s
in
g
im
ag
es
an
d
t
r
ain
in
g
th
e
d
ee
p
lear
n
in
g
m
o
d
el,
r
esp
ec
tiv
ely
.
T
h
e
in
teg
r
atio
n
o
f
O
p
en
C
V
en
s
u
r
es
ef
f
ec
tiv
e
im
ag
e
m
an
i
p
u
latio
n
an
d
f
ea
t
u
r
e
ex
tr
ac
tio
n
,
wh
ile
T
en
s
o
r
Flo
w
f
ac
ilit
ates
th
e
tr
ain
in
g
o
f
th
e
C
NN
f
o
r
ac
cu
r
ate
tex
t
r
ec
o
g
n
itio
n
.
Fo
r
r
ea
l
-
wo
r
ld
d
ep
lo
y
m
en
t,
th
e
s
y
s
tem
is
ad
a
p
ted
t
o
r
u
n
o
n
a
R
asp
b
er
r
y
Pi
4
B
,
a
co
m
p
ac
t
a
n
d
a
f
f
o
r
d
a
b
le
em
b
ed
d
ed
p
latf
o
r
m
.
T
h
e
R
asp
b
er
r
y
Pi
C
am
er
a
Mo
d
u
le
3
is
em
p
lo
y
ed
f
o
r
ca
p
tu
r
in
g
im
a
g
es
in
r
ea
l
-
tim
e,
en
ab
lin
g
th
e
s
y
s
tem
to
r
ec
o
g
n
ize
h
an
d
wr
itten
tex
t
f
r
o
m
p
h
y
s
ical
d
o
cu
m
en
ts
o
r
s
u
r
f
ac
es.
I
n
th
is
wo
r
k
,
th
e
C
NN
m
o
d
el
h
as
d
e
m
o
n
s
tr
ated
its
ef
f
ec
tiv
en
ess
b
y
o
u
t
d
o
in
g
all
o
th
er
m
o
d
els
i
n
ev
er
y
m
ea
s
u
r
em
en
t.
Alth
o
u
g
h
th
e
SVM
an
d
co
n
n
ec
tio
n
is
t
tem
p
o
r
al
class
if
icatio
n
(
C
T
C
)
m
o
d
els
ex
h
i
b
ited
im
p
r
ess
iv
e
p
er
f
o
r
m
an
ce
in
s
o
m
e
r
esp
ec
ts
,
th
e
y
f
ell
s
h
o
r
t
o
f
th
e
C
NN.
T
h
e
r
ec
u
r
r
en
t
n
e
u
r
al
n
etwo
r
k
(
R
NN)
,
d
esp
ite
its
o
v
er
all
lo
wer
p
er
f
o
r
m
an
ce
,
ex
ce
lled
in
ter
m
s
o
f
th
e
F1
-
s
co
r
e
.
T
h
ese
o
u
tc
o
m
es
r
ein
f
o
r
ce
th
e
r
o
b
u
s
tn
ess
an
d
p
r
ee
m
in
en
ce
o
f
th
e
C
NN
m
o
d
el.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
Ha
n
d
w
r
itten
text
r
ec
o
g
n
itio
n
s
ystem
u
s
in
g
R
a
s
p
b
err
y
P
i wi
th
…
(
Ja
mil A
b
ed
a
lr
a
h
im
J
a
mil A
ls
a
ya
yd
eh
)
2293
2.
M
E
T
H
O
D
T
h
is
s
ec
tio
n
p
r
o
v
id
es
d
etailed
in
f
o
r
m
atio
n
o
n
th
e
ch
o
s
en
d
ee
p
lear
n
in
g
tech
n
iq
u
e,
C
NN,
f
o
r
HT
R
an
d
d
etec
tio
n
.
I
t
also
o
f
f
er
s
a
co
m
p
r
eh
en
s
iv
e
d
escr
ip
tio
n
o
f
th
e
m
o
d
el
tr
ain
in
g
p
r
o
ce
s
s
u
s
in
g
th
e
E
MN
I
ST
d
ataset,
wh
ich
co
n
tain
s
a
lar
g
e
n
u
m
b
e
r
o
f
h
an
d
wr
itten
ch
ar
ac
ter
s
an
d
d
ig
its
.
T
h
e
s
ec
tio
n
o
u
tlin
es
th
e
p
r
o
ce
s
s
an
d
m
eth
o
d
o
l
o
g
y
in
v
o
lv
ed
in
co
n
s
tr
u
ctin
g
th
e
HT
R
s
y
s
tem
.
T
h
is
m
eth
o
d
o
l
o
g
y
is
th
e
r
esu
lt
o
f
e
x
ten
s
iv
e
r
esear
ch
co
n
d
u
cted
th
r
o
u
g
h
l
iter
atu
r
e
r
ev
iews
an
d
r
ea
d
in
g
s
to
id
en
tify
th
e
m
o
s
t
ef
f
ec
tiv
e
d
ee
p
lear
n
in
g
m
eth
o
d
s
an
d
s
tr
ateg
ies.
T
h
e
p
r
im
ar
y
o
b
jectiv
e
o
f
th
is
m
et
h
o
d
o
lo
g
y
is
to
d
ev
elo
p
an
H
T
R
s
y
s
tem
th
at
ca
n
ac
cu
r
ately
r
ec
o
g
n
ize
an
d
in
ter
p
r
et
h
an
d
wr
itten
tex
t.
2
.
1
.
B
lo
c
k
d
ia
g
ra
m
T
h
e
b
lo
ck
d
iag
r
am
f
o
r
th
e
s
y
s
tem
in
clu
d
es
Fig
u
r
e
1
a
ca
m
er
a
m
o
d
u
le,
a
R
asp
b
er
r
y
Pi
,
a
p
o
wer
s
u
p
p
ly
,
an
d
a
n
L
C
D
m
o
n
ito
r
.
T
h
e
ca
m
er
a
m
o
d
u
le,
s
p
ec
if
ically
th
e
R
asp
b
er
r
y
Pi
C
am
er
a
Mo
d
u
le
3
,
s
er
v
es
as
th
e
in
p
u
t
d
ev
ice,
c
ap
tu
r
in
g
i
m
ag
es
o
f
th
e
h
an
d
wr
itten
tex
t
.
T
h
e
R
asp
b
er
r
y
Pi
ac
ts
as
th
e
ce
n
tr
al
p
r
o
ce
s
s
in
g
u
n
it,
r
esp
o
n
s
ib
le
f
o
r
r
u
n
n
in
g
th
e
HT
R
s
y
s
tem
.
I
t
h
an
d
les
t
ask
s
s
u
ch
as
im
ag
e
p
r
ep
r
o
ce
s
s
in
g
,
lo
ad
in
g
,
a
n
d
u
tili
zin
g
th
e
tr
ain
ed
C
NN
m
o
d
el,
an
d
p
er
f
o
r
m
in
g
OC
R
o
n
th
e
ca
p
tu
r
ed
im
ag
es.
T
h
e
p
o
wer
s
u
p
p
ly
en
s
u
r
es
th
at
th
e
R
asp
b
er
r
y
Pi
an
d
th
e
ca
m
er
a
m
o
d
u
le
r
ec
eiv
e
th
e
r
eq
u
ir
e
d
elec
tr
ical
p
o
wer
t
o
o
p
er
ate
ef
f
icien
tly
.
L
astl
y
,
th
e
L
C
D
m
o
n
ito
r
s
er
v
es
as
th
e
o
u
tp
u
t
d
ev
ice,
d
is
p
l
ay
in
g
th
e
p
r
o
ce
s
s
ed
im
ag
es
a
n
d
r
ec
o
g
n
ized
tex
t
o
u
tp
u
t to
th
e
u
s
er
.
Fig
u
r
e
1
.
B
lo
ck
d
iag
r
am
o
f
H
T
R
s
y
s
tem
2
.
2
.
H
T
R
s
y
s
t
em
f
lo
w
T
h
e
s
y
s
tem
f
lo
w
f
o
r
th
e
HT
R
d
ep
lo
y
ed
o
n
a
R
asp
b
er
r
y
P
i
4
B
with
a
ca
m
er
a
as
its
in
p
u
t
an
d
th
e
o
u
tp
u
t
s
h
o
wn
o
n
a
m
o
n
ito
r
f
o
llo
ws
a
s
er
ies
o
f
s
tep
s
,
as
i
llu
s
tr
ated
in
Fig
u
r
e
2
.
First,
th
e
R
asp
b
er
r
y
Pi
4
B
i
s
in
itialized
,
an
d
th
e
r
e
q
u
ir
e
d
d
ep
en
d
en
cies,
in
clu
d
in
g
Op
en
C
V,
T
en
s
o
r
Flo
w,
an
d
a
C
NN
clas
s
if
ier
,
ar
e
im
p
o
r
ted
.
Nex
t,
a
n
im
a
g
e
o
f
t
h
e
h
a
n
d
wr
itten
tex
t
is
ca
p
tu
r
e
d
u
s
in
g
th
e
R
asp
b
er
r
y
Pi
C
am
er
a
Mo
d
u
le
3
.
T
h
e
ca
p
tu
r
ed
im
ag
e
is
th
e
n
p
r
e
-
p
r
o
ce
s
s
ed
u
s
in
g
Op
en
C
V
to
en
h
a
n
ce
its
q
u
ality
.
On
ce
p
r
e
-
p
r
o
c
ess
ed
,
a
p
r
e
-
tr
ai
n
ed
C
NN
m
o
d
el,
tr
ain
ed
o
n
h
an
d
wr
itten
tex
t
d
ata
u
s
in
g
T
en
s
o
r
Flo
w,
is
lo
ad
ed
.
T
h
is
m
o
d
el
is
u
s
ed
to
p
er
f
o
r
m
OC
R
o
n
th
e
p
r
e
-
p
r
o
ce
s
s
ed
im
ag
e,
g
en
er
atin
g
p
r
ed
ictio
n
s
f
o
r
ea
ch
ch
ar
ac
ter
in
th
e
tex
t.
T
h
e
r
esu
lts
ar
e
th
en
p
o
s
t
-
p
r
o
ce
s
s
ed
,
wh
ic
h
m
ay
in
clu
d
e
r
em
o
v
in
g
n
o
is
e,
c
o
r
r
ec
t
in
g
s
p
ellin
g
e
r
r
o
r
s
,
o
r
f
o
r
m
att
in
g
th
e
r
ec
o
g
n
ized
tex
t.
Fin
ally
,
th
e
p
r
o
ce
s
s
ed
i
m
ag
e
an
d
th
e
r
ec
o
g
n
ized
tex
t
o
u
tp
u
t
ar
e
d
is
p
lay
ed
o
n
th
e
m
o
n
ito
r
c
o
n
n
ec
te
d
to
th
e
R
asp
b
er
r
y
Pi 4
B
,
co
m
p
letin
g
th
e
s
y
s
tem
f
lo
w.
Fig
u
r
e
2
.
Sy
s
tem
o
p
er
atin
g
f
lo
wch
ar
t
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
15
,
No
.
2
,
Ap
r
il
20
25
:
2
2
9
1
-
2
3
0
3
2294
2
.
2
.
1
.
Sy
s
t
em
pro
g
ra
mm
ing
f
lo
wcha
rt
T
h
e
p
r
o
p
o
s
ed
HT
R
s
y
s
tem
f
o
llo
ws
a
p
r
o
g
r
am
m
in
g
f
l
o
wch
a
r
t
in
Fig
u
r
e
3
to
p
r
o
ce
s
s
an
d
r
ec
o
g
n
ize
h
an
d
wr
itten
tex
t.
T
h
e
p
r
o
ce
s
s
b
eg
in
s
with
ca
p
tu
r
in
g
a
n
im
a
g
e
u
s
in
g
th
e
ca
m
er
a
m
o
d
u
le,
w
h
ich
is
th
e
n
s
en
t
to
th
e
R
asp
b
er
r
y
Pi
4
B
f
o
r
f
u
r
t
h
er
p
r
o
ce
s
s
in
g
.
T
h
e
s
y
s
tem
p
er
f
o
r
m
s
tex
t
d
etec
tio
n
o
n
th
e
ca
p
tu
r
ed
im
a
g
e.
Ne
x
t,
th
e
s
y
s
tem
ex
tr
ac
ts
f
ea
t
u
r
e
f
r
o
m
th
e
r
eg
io
n
o
f
in
ter
est
(
R
OI
)
to
f
o
cu
s
o
n
r
elev
an
t
in
f
o
r
m
atio
n
f
o
r
r
ec
o
g
n
itio
n
.
T
h
e
tex
t
r
ec
o
g
n
itio
n
p
r
o
ce
s
s
u
tili
ze
s
a
C
NN
a
s
th
e
f
o
u
n
d
at
io
n
,
wh
er
e
th
e
ex
tr
ac
ted
f
ea
tu
r
es
ar
e
f
ed
in
to
t
h
e
m
o
d
el
f
o
r
r
ec
o
g
n
itio
n
.
I
f
th
e
ca
p
tu
r
ed
tex
t
m
atch
es
an
y
o
f
th
e
tex
ts
f
r
o
m
th
e
d
ataset,
th
e
s
y
s
tem
p
r
o
ce
ed
s
to
id
en
tify
an
d
v
er
if
y
th
e
id
e
n
tity
o
f
th
e
tex
t.
Ho
wev
er
,
if
th
e
ca
p
tu
r
ed
tex
t
d
o
es
n
o
t
m
atc
h
an
y
k
n
o
wn
tex
ts
f
r
o
m
th
e
d
ataset,
it
is
co
n
s
id
er
ed
an
u
n
k
n
o
wn
tex
t,
lead
in
g
th
e
p
r
o
g
r
am
to
ter
m
in
ate.
T
h
is
p
r
o
g
r
a
m
m
in
g
f
lo
w
d
em
o
n
s
tr
ates
th
e
s
eq
u
e
n
tial
s
tep
s
in
v
o
lv
ed
in
th
e
HT
R
s
y
s
tem
,
in
clu
d
in
g
im
ag
e
ca
p
t
u
r
e,
tex
t
d
etec
tio
n
,
f
ea
tu
r
e
ex
tr
ac
tio
n
,
tex
t r
ec
o
g
n
i
tio
n
u
s
in
g
C
NN,
m
atch
in
g
an
d
v
er
if
icatio
n
,
a
n
d
h
an
d
lin
g
o
f
u
n
k
n
o
wn
tex
ts
.
Fig
u
r
e
3
.
Sy
s
tem
p
r
o
g
r
a
m
m
in
g
f
lo
wch
ar
t
2
.
3
.
Da
t
a
a
cquis
it
io
n
T
h
e
E
MN
I
ST
d
ataset
was
p
r
o
cu
r
ed
an
d
u
tili
ze
d
in
th
is
w
o
r
k
f
o
r
tr
ain
in
g
a
n
d
ev
alu
atio
n
p
u
r
p
o
s
es.
T
h
e
E
MN
I
ST
d
ataset,
an
ex
t
en
s
io
n
o
f
th
e
wid
ely
u
s
ed
MN
I
ST
d
ataset,
is
a
p
o
p
u
lar
b
e
n
ch
m
ar
k
d
ataset
in
m
ac
h
in
e
lear
n
in
g
an
d
co
m
p
u
t
er
v
is
io
n
.
I
t
co
m
p
r
is
es
a
co
m
p
r
eh
en
s
iv
e
co
llectio
n
o
f
h
an
d
wr
itten
d
ig
its
an
d
ch
ar
ac
ter
s
,
with
a
tr
ain
i
n
g
s
et
o
f
6
3
0
,
0
0
0
e
x
am
p
les
an
d
a
te
s
t
s
et
o
f
7
0
,
0
0
0
e
x
am
p
les.
T
h
e
ac
q
u
is
itio
n
o
f
th
e
E
MN
I
ST
d
ataset
ty
p
ically
i
n
v
o
lv
es
ac
ce
s
s
in
g
it
f
r
o
m
p
u
b
lic
r
ep
o
s
ito
r
ies
o
r
th
r
o
u
g
h
f
r
a
m
ewo
r
k
s
lik
e
T
en
s
o
r
Flo
w.
T
h
e
d
ataset
is
p
r
e
-
d
iv
id
ed
in
to
tr
ain
i
n
g
a
n
d
tes
t
s
ets,
en
s
u
r
in
g
a
s
tan
d
a
r
d
ized
ev
alu
atio
n
p
r
o
ce
s
s
.
T
h
e
tr
ain
in
g
s
et
is
u
s
ed
to
tr
ain
th
e
d
ee
p
lear
n
in
g
m
o
d
el,
wh
ile
th
e
test
s
et
is
em
p
lo
y
ed
to
ev
alu
ate
th
e
m
o
d
el’
s
p
er
f
o
r
m
a
n
ce
o
n
u
n
s
e
en
d
ata.
E
ac
h
s
am
p
le
in
th
e
E
MN
I
ST
d
ataset
is
a
g
r
ay
s
ca
le
im
ag
e
o
f
2
8
×
28
p
ix
els,
r
ep
r
esen
tin
g
a
s
in
g
le
d
ig
it
f
r
o
m
0
to
9
o
r
a
letter
f
r
o
m
A
to
Z
.
T
h
e
im
ag
es
h
av
e
b
ee
n
m
eticu
lo
u
s
ly
lab
eled
with
th
eir
co
r
r
esp
o
n
d
i
n
g
d
ig
it
o
r
c
h
ar
ac
ter
class
,
m
ak
in
g
th
em
s
u
itab
le
f
o
r
s
u
p
er
v
is
ed
lear
n
in
g
task
s
s
u
ch
as
ch
ar
ac
ter
r
ec
o
g
n
itio
n
.
B
y
ac
q
u
ir
in
g
an
d
u
tili
zin
g
th
e
E
MN
I
ST
d
ataset
in
th
i
s
wo
r
k
,
th
e
d
ee
p
lear
n
in
g
m
o
d
el
ca
n
b
e
t
r
ain
ed
a
n
d
ev
alu
ated
u
s
in
g
a
co
m
p
r
eh
e
n
s
iv
e
an
d
r
eliab
le
b
en
ch
m
a
r
k
d
ataset,
en
ab
lin
g
th
e
d
ev
elo
p
m
e
n
t o
f
a
n
ac
cu
r
ate
an
d
r
o
b
u
s
t
HT
R
s
y
s
tem
in
Fig
u
r
e
4.
Fig
u
r
e
4
.
E
MN
I
ST
d
ataset
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
Ha
n
d
w
r
itten
text
r
ec
o
g
n
itio
n
s
ystem
u
s
in
g
R
a
s
p
b
err
y
P
i wi
th
…
(
Ja
mil A
b
ed
a
lr
a
h
im
J
a
mil A
ls
a
ya
yd
eh
)
2295
2
.
4
.
Dee
p
lea
rning
cla
s
s
if
iers
Dee
p
lear
n
in
g
class
if
ier
s
,
s
p
e
cif
ically
C
NN,
ar
e
a
p
o
wer
f
u
l
class
if
icat
io
n
-
o
r
ien
ted
m
ac
h
i
n
e
lear
n
in
g
m
o
d
el.
C
NNs
ar
e
u
s
ed
in
th
is
s
tu
d
y
to
clas
s
if
y
p
r
ed
ictio
n
s
as
eith
er
ac
cu
r
ate
o
r
in
ac
cu
r
ate.
C
NNs
h
av
e
ac
h
iev
ed
in
n
o
v
ativ
e
p
er
f
o
r
m
an
ce
in
v
ar
io
u
s
d
o
m
ain
s
,
in
clu
d
in
g
co
m
p
u
ter
v
is
io
n
,
n
atu
r
al
lan
g
u
ag
e
p
r
o
ce
s
s
in
g
,
an
d
s
p
ee
ch
r
ec
o
g
n
itio
n
[
1
9
]
.
T
h
ese
n
etwo
r
k
s
ar
e
d
esig
n
ed
to
au
t
o
m
atica
lly
lear
n
an
d
ex
tr
ac
t
in
tr
icate
p
atter
n
s
a
n
d
ch
ar
ac
t
er
is
tics
f
r
o
m
in
p
u
t
d
ata,
m
a
k
in
g
th
em
p
a
r
ticu
lar
ly
u
s
ef
u
l
f
o
r
task
s
in
v
o
lv
in
g
co
m
p
lex
v
is
u
al
o
r
tex
tu
al
d
at
a.
T
h
is
wo
r
k
u
s
es
C
N
N
as
th
e
DL
clas
s
if
ier
to
clas
s
if
y
co
r
r
ec
t
an
d
in
co
r
r
ec
t
p
r
ed
ictio
n
s
,
en
a
b
lin
g
r
o
b
u
s
t a
n
d
r
eliab
le
class
if
icatio
n
o
u
tco
m
es
[
2
0
]
.
2
.
4
.
1
.
Cla
s
s
if
ica
t
io
n
u
s
ing
C
NN
T
h
e
p
r
o
p
o
s
ed
s
y
s
tem
u
s
es
a
C
NN
to
in
ter
p
r
et
h
an
d
wr
itten
tex
t.
T
h
e
C
NN
m
o
d
el
is
tr
ain
ed
o
n
a
d
ataset
o
f
h
an
d
wr
itten
s
am
p
le
s
,
lear
n
in
g
p
atter
n
s
in
ch
ar
ac
t
er
s
an
d
wo
r
d
s
.
C
o
n
v
o
lu
tio
n
al
an
d
p
o
o
lin
g
lay
er
s
ex
tr
ac
t
f
ea
tu
r
es
f
r
o
m
im
ag
es
f
o
r
class
if
icatio
n
.
T
h
e
m
o
d
el
ad
ju
s
ts
its
p
ar
am
eter
s
d
u
r
in
g
tr
ain
in
g
t
o
im
p
r
o
v
e
ac
cu
r
ac
y
.
T
h
is
tech
n
o
lo
g
y
ca
n
p
r
o
ce
s
s
an
d
r
ec
o
g
n
ize
h
an
d
wr
itten
tex
t
in
r
ea
l
-
tim
e,
co
n
v
er
tin
g
it
in
to
d
i
g
ital
f
o
r
m
at
[
2
1
]
.
2
.
5
.
Alg
o
rit
hm
t
r
a
ini
ng
T
h
is
wo
r
k
u
s
es
Op
en
C
V,
T
en
s
o
r
Flo
w,
an
d
C
NN
class
if
ier
s
to
tr
ain
a
HT
R
alg
o
r
ith
m
.
T
h
e
d
ataset
co
m
p
r
is
es
an
n
o
tated
h
an
d
wr
itten
tex
t
s
am
p
les.
T
h
e
C
N
N
ar
ch
itectu
r
e,
with
co
n
v
o
l
u
tio
n
al
an
d
p
o
o
lin
g
lay
er
s
,
ex
tr
ac
ts
f
ea
tu
r
e
f
r
o
m
in
p
u
t
i
m
ag
es.
T
en
s
o
r
Flo
w
is
u
s
ed
t
o
b
u
ild
an
d
tr
ain
th
e
C
NN
m
o
d
el.
T
h
e
tr
ain
in
g
p
r
o
ce
s
s
iter
ates
o
v
er
th
e
d
ataset,
ad
ju
s
tin
g
t
h
e
m
o
d
el’
s
p
a
r
a
m
eter
s
to
m
in
im
ize
p
r
e
d
ictio
n
er
r
o
r
s
.
T
h
e
Py
th
o
n
tr
ain
in
g
s
cr
ip
t
h
a
n
d
les
d
ataset
p
r
o
ce
s
s
in
g
an
d
m
o
d
el
wei
g
h
t
u
p
d
ates.
T
h
e
n
u
m
b
er
o
f
tr
ain
in
g
ep
o
ch
s
is
o
p
tim
ized
ex
p
er
im
en
tally
.
T
h
e
tr
ain
ed
C
NN
m
o
d
el
ca
n
th
en
r
ec
o
g
n
ize
n
ew
h
an
d
w
r
itten
tex
t
im
ag
es,
p
r
o
v
id
i
n
g
ca
p
a
b
ilit
ies f
o
r
d
ata
d
ig
itizatio
n
,
tr
an
s
cr
ip
tio
n
,
an
d
d
o
cu
m
e
n
t a
n
aly
s
is
.
2
.
6
.
Ra
s
pb
er
ry
P
i 4
B
T
h
e
R
asp
b
er
r
y
Pi 4
B
,
eq
u
ip
p
e
d
with
a
Qu
ad
-
co
r
e
AR
M
C
o
r
t
ex
-
A7
2
p
r
o
ce
s
s
o
r
an
d
4
GB
o
f
R
AM
,
i
s
id
ea
l
f
o
r
im
p
lem
e
n
tin
g
a
HT
R
m
o
d
el
Fig
u
r
e
5
.
I
t
f
ea
tu
r
es
th
e
B
r
o
ad
co
m
B
C
M2
7
1
1
ch
ip
an
d
h
as
a
1
.
5
GHz
clo
ck
s
p
ee
d
.
I
t
in
cl
u
d
es
3
2
K
B
L
1
ca
ch
e
f
o
r
d
ata,
4
8
KB
L
1
ca
ch
e
f
o
r
in
s
tr
u
ctio
n
s
,
an
d
1
MB
L
2
ca
ch
e
f
o
r
f
aster
d
ata
r
etr
iev
al.
T
h
e
b
o
ar
d
o
f
f
e
r
s
v
ar
io
u
s
p
o
r
ts
,
in
clu
d
i
n
g
4
USB
p
o
r
ts
,
a
n
HDM
I
p
o
r
t,
an
d
a
n
SD
ca
r
d
s
lo
t,
allo
win
g
s
ea
m
les
s
co
n
n
ec
tiv
ity
with
ex
ter
n
al
d
ev
ices.
I
t
o
p
er
ates
o
n
R
asp
b
er
r
y
Pi
OS,
wh
ich
s
u
p
p
o
r
ts
h
ig
h
-
lev
el
p
r
o
g
r
am
m
in
g
lan
g
u
ag
es
a
n
d
o
f
f
e
r
s
a
r
an
g
e
o
f
to
o
ls
an
d
lib
r
a
r
ies
f
o
r
ea
s
y
in
teg
r
atio
n
with
h
ar
d
war
e
co
m
p
o
n
en
ts
.
I
ts
co
m
p
ac
t
s
ize
an
d
ca
p
ab
ilit
ies
m
ak
e
it
a
n
e
x
ce
llen
t
c
h
o
ice
f
o
r
d
ep
lo
y
in
g
s
y
s
tem
s
in
r
eso
u
r
ce
-
co
n
s
tr
ain
ed
en
v
ir
o
n
m
en
ts
[
2
2
]
.
Fig
u
r
e
5
.
R
asp
b
er
r
y
Pi 4
B
2
.
7
.
Ra
s
pb
er
ry
P
i
Ca
m
er
a
M
o
du
le
3
T
h
e
R
asp
b
er
r
y
Pi
C
am
er
a
M
o
d
u
le
3
,
d
esig
n
ed
f
o
r
R
asp
b
er
r
y
Pi
b
o
a
r
d
s
,
f
ea
tu
r
es
a
So
n
y
I
MX
7
0
8
s
en
s
o
r
with
1
1
.
9
m
eg
a
p
ix
els
an
d
HDR
im
ag
in
g
in
Fig
u
r
e
6
.
Av
ailab
le
in
s
tan
d
ar
d
an
d
wid
e
-
an
g
le
v
ar
ia
n
ts
,
it
in
clu
d
es
ad
v
an
ce
d
f
ea
tu
r
es
lik
e
p
h
ase
d
etec
tio
n
au
to
f
o
c
u
s
an
d
2
D
d
y
n
am
ic
d
ef
ec
t
p
ix
el
c
o
r
r
ec
tio
n
.
I
t
u
s
es
a
C
SI
-
2
s
er
ial
in
ter
f
ac
e
f
o
r
d
ata
o
u
tp
u
t
an
d
is
co
m
p
atib
le
with
all
R
asp
b
er
r
y
Pi
m
o
d
els.
T
h
e
m
o
d
u
le
m
ea
s
u
r
es
25
×
24
×
1
1
.
5
m
m
an
d
co
m
es with
a
2
0
0
m
m
r
ib
b
o
n
ca
b
le
[
2
2
]
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
15
,
No
.
2
,
Ap
r
il
20
25
:
2
2
9
1
-
2
3
0
3
2296
Fig
u
r
e
6
.
R
asp
b
er
r
y
Pi
C
am
er
a
M
o
d
u
le
3
2
.
8
.
O
penCV
Op
en
C
V
is
an
o
p
en
-
s
o
u
r
ce
lib
r
ar
y
f
o
r
co
m
p
u
ter
v
is
io
n
an
d
i
m
ag
e
p
r
o
ce
s
s
in
g
in
Fig
u
r
e
7
.
I
t
p
r
o
v
id
es
to
o
ls
f
o
r
o
p
e
r
atio
n
s
lik
e
o
b
je
ct
d
etec
tio
n
,
im
ag
e
en
h
an
ce
m
en
t,
an
d
g
e
o
m
etr
ic
tr
an
s
f
o
r
m
atio
n
s
[
2
3
]
,
[
2
4
]
.
I
t
s
u
p
p
o
r
ts
f
ea
tu
r
es
s
u
ch
as
im
ag
e
p
r
o
ce
s
s
in
g
,
o
b
ject
d
etec
tio
n
,
f
ea
tu
r
e
ex
tr
ac
tio
n
,
an
d
i
n
teg
r
atio
n
with
m
ac
h
in
e
lear
n
in
g
f
r
am
ew
o
r
k
s
.
Op
en
C
V
is
u
s
ed
in
d
iv
er
s
e
f
ield
s
l
ik
e
r
o
b
o
tics
,
m
ed
ical
im
ag
in
g
,
a
n
d
s
elf
-
d
r
iv
in
g
ca
r
s
,
o
f
f
er
in
g
a
p
o
wer
f
u
l to
o
lk
it f
o
r
v
is
u
al
d
ata
an
d
c
o
m
p
u
ter
v
is
io
n
ch
allen
g
es
[
2
5
]
,
[
2
6
]
.
Fig
u
r
e
7
.
Op
e
n
C
V
2
.
9
.
T
ens
o
rF
lo
w
T
en
s
o
r
Flo
w
is
an
o
p
en
-
s
o
u
r
c
e
s
o
f
twar
e
wid
ely
u
s
ed
in
m
a
ch
in
e
lear
n
in
g
i
n
Fig
u
r
e
8
.
I
t
s
im
p
lifie
s
b
u
ild
in
g
,
tr
ai
n
in
g
,
an
d
d
ep
l
o
y
in
g
m
o
d
els
with
its
h
ig
h
-
lev
el
API
,
Ker
as
[
2
7
]
.
I
t
s
u
p
p
o
r
ts
d
is
tr
ib
u
ted
co
m
p
u
tin
g
f
o
r
f
aster
tr
ain
in
g
an
d
o
f
f
er
s
p
r
e
-
b
u
ilt
alg
o
r
ith
m
s
f
o
r
task
s
lik
e
im
ag
e
class
if
i
ca
tio
n
.
T
en
s
o
r
Flo
w
ex
ce
ls
at
h
an
d
lin
g
lar
g
e
-
s
ca
l
e
d
atasets
an
d
co
m
p
le
x
m
o
d
els,
an
d
p
r
o
v
id
es
to
o
ls
f
o
r
m
o
d
el
v
is
u
aliza
tio
n
,
d
eb
u
g
g
in
g
,
an
d
d
ep
lo
y
m
en
t.
I
ts
co
m
p
r
e
h
en
s
iv
e
f
ea
tu
r
es
a
n
d
s
tr
o
n
g
co
m
m
u
n
ity
s
u
p
p
o
r
t
m
ak
e
it
a
r
o
b
u
s
t
f
r
am
ewo
r
k
f
o
r
m
ac
h
in
e
lea
r
n
i
n
g
task
s
[
2
8
]
.
Fig
u
r
e
8
.
T
e
n
s
o
r
Flo
w
3.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
T
h
e
o
u
tc
o
m
es
an
d
an
aly
s
is
o
b
tain
ed
f
r
o
m
th
e
cr
ea
tio
n
o
f
a
HT
R
s
y
s
tem
,
u
tili
zin
g
C
NN,
ar
e
d
etailed
in
th
is
s
ec
tio
n
.
T
h
e
s
y
s
tem
u
n
d
er
wen
t
ex
te
n
s
iv
e
tr
ain
in
g
a
n
d
ev
alu
atio
n
u
s
in
g
th
e
E
MN
I
ST
d
ataset,
wh
ich
co
m
p
r
is
es
a
s
u
b
s
tan
tial
co
m
p
ilatio
n
o
f
h
an
d
wr
itten
n
u
m
er
a
ls
an
d
ch
ar
ac
ter
s
.
T
h
e
p
r
o
ce
s
s
in
v
o
lv
ed
tr
ain
in
g
th
e
C
NN
m
o
d
el
o
n
a
co
n
s
i
d
er
ab
le
co
llectio
n
o
f
6
3
0
,
0
0
0
im
ag
es
o
b
tain
ed
f
r
o
m
th
e
E
MN
I
ST
d
ataset.
Fo
llo
win
g
th
e
tr
ain
in
g
p
h
ase
,
an
ad
d
itio
n
al
s
et
o
f
7
0
,
0
0
0
im
ag
es
was
u
tili
ze
d
to
ev
alu
ate
th
e
m
o
d
el'
s
p
er
f
o
r
m
an
ce
.
T
h
e
o
u
tc
o
m
es
wer
e
v
er
if
ie
d
u
s
in
g
ess
en
tial
m
etr
ics
s
u
ch
as
r
ec
all,
ac
c
u
r
a
cy
,
an
d
p
r
ec
is
io
n
o
f
th
e
HT
R
s
y
s
tem
.
Ad
d
itio
n
ally
,
a
co
n
f
u
s
io
n
m
atr
i
x
was
em
p
lo
y
ed
t
o
p
r
o
v
id
e
a
c
o
m
p
r
e
h
en
s
iv
e
v
iew
o
f
th
e
s
y
s
tem
'
s
p
er
f
o
r
m
an
ce
,
r
ev
ea
li
n
g
its
m
er
its
an
d
o
p
p
o
r
tu
n
ities
f
o
r
en
h
an
ce
m
e
n
t.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
Ha
n
d
w
r
itten
text
r
ec
o
g
n
itio
n
s
ystem
u
s
in
g
R
a
s
p
b
err
y
P
i wi
th
…
(
Ja
mil A
b
ed
a
lr
a
h
im
J
a
mil A
ls
a
ya
yd
eh
)
2297
3
.
1
.
E
s
s
ent
ia
l
inte
rpre
t
a
t
io
n ba
s
ed
o
n k
ey
f
ind
ing
s
T
h
e
HT
R
s
y
s
tem
u
s
e
s
a
R
asp
b
er
r
y
Pi
C
am
er
a
to
ca
p
tu
r
e
im
ag
es
o
f
h
an
d
wr
itten
tex
t.
T
h
ese
im
ag
es
ar
e
p
r
o
ce
s
s
ed
b
y
a
R
asp
b
er
r
y
Pi
4
,
wh
ich
r
u
n
s
th
e
HT
R
s
y
s
tem
u
tili
zin
g
a
C
NN
m
o
d
el.
T
h
e
r
ec
o
g
n
ized
tex
t
an
d
r
elev
an
t
m
etr
ics
ar
e
d
is
p
l
ay
ed
o
n
a
m
o
n
ito
r
co
n
n
ec
ted
to
th
e
R
asp
b
er
r
y
Pi
4
,
as
s
h
o
wn
in
Fig
u
r
e
9
.
T
h
e
C
NN
m
o
d
el
d
em
o
n
s
tr
ated
im
p
r
ess
iv
e
p
er
f
o
r
m
an
ce
,
ac
h
ie
v
in
g
an
ac
cu
r
ac
y
o
f
9
1
.
5
8
%
in
r
ec
o
g
n
izin
g
E
n
g
lis
h
alp
h
ab
ets an
d
d
ig
its
.
Fig
u
r
e
9
.
Har
d
war
e
s
etu
p
3
.
2
.
Co
nfusi
o
n
m
a
t
rix
a
nd
a
na
ly
s
is
T
h
e
d
atasets
in
th
e
test
f
o
ld
e
r
s
wer
e
ev
alu
ated
to
d
eter
m
in
e
th
e
ac
c
u
r
ac
y
o
f
id
en
tify
i
n
g
g
estu
r
e
d
ata
ac
co
r
d
in
g
to
th
eir
class
es
wh
i
le
class
if
y
in
g
th
e
ten
class
es
o
f
d
ig
its
.
T
h
e
ac
cu
r
ac
y
was
ca
lcu
lated
u
s
in
g
th
e
co
n
f
u
s
io
n
m
atr
ix
d
is
p
lay
e
d
af
t
er
th
e
co
d
i
n
g
.
T
h
e
ac
cu
r
a
cy
ca
n
b
e
ca
lcu
lated
u
s
in
g
(
1
)
.
=
+
+
+
×
100%
(
1)
T
h
e
ter
m
“
tr
u
e
p
o
s
itiv
e
”
d
escr
ib
es
h
o
w
t
h
e
p
o
s
itiv
e
class
is
ac
cu
r
ately
class
if
ied
,
wh
ile
“
tr
u
e
n
e
g
ativ
e
”
r
e
f
er
s
to
th
e
ac
cu
r
ate
class
if
icatio
n
o
f
th
e
n
eg
ativ
e
class
.
A
f
alse
p
o
s
itiv
e
o
cc
u
r
s
w
h
en
a
n
e
g
ativ
e
class
is
in
co
r
r
ec
tly
f
o
r
ec
asted
as
p
o
s
itiv
e,
an
d
a
f
alse
n
eg
ativ
e
o
cc
u
r
s
wh
e
n
a
p
o
s
itiv
e
class
is
in
co
r
r
ec
tly
p
r
e
d
icted
as
n
e
g
ativ
e.
T
h
e
co
n
f
u
s
io
n
m
atr
ix
Fig
u
r
e
1
0
r
e
v
ea
ls
wh
ich
n
u
m
b
er
s
th
e
m
o
d
el
co
r
r
ec
tl
y
r
ec
o
g
n
izes
an
d
wh
ich
o
n
es
it
f
r
eq
u
e
n
tly
m
is
tak
es.
Alth
o
u
g
h
th
e
m
o
d
el
p
e
r
f
o
r
m
s
ad
m
i
r
ab
ly
,
it
o
cc
asio
n
ally
(
5
tim
es
o
u
t
o
f
1
0
,
0
0
0
)
co
n
f
u
s
es
n
u
m
b
e
r
5
with
n
u
m
b
er
3
o
r
n
u
m
b
er
2
with
n
u
m
b
e
r
7
.
W
ith
9
1
.
5
8
%
ac
cu
r
ac
y
,
th
e
C
NN
alg
o
r
ith
m
is
q
u
ite
ac
cu
r
ate.
Giv
en
th
at
th
e
ep
o
ch
s
tr
ain
ed
wer
e
1
0
0
with
class
if
icatio
n
s
b
ased
o
n
d
ig
its
0
to
9
.
T
h
is
r
esu
lt
is
co
n
s
id
er
ed
ex
ce
llen
t,
as a
cc
u
r
ac
y
m
ea
s
u
r
e
m
en
ts
b
etwe
en
7
5
% a
n
d
9
5
% a
r
e
p
r
ac
tical
an
d
id
ea
l in
p
r
ac
tice.
Fig
u
r
e
1
0
.
C
NN
co
n
f
u
s
io
n
m
a
tr
ix
af
ter
o
n
e
h
u
n
d
r
e
d
ep
o
ch
s
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
15
,
No
.
2
,
Ap
r
il
20
25
:
2
2
9
1
-
2
3
0
3
2298
3
.
3
.
Resul
t
s
o
n t
he
ha
nd
writ
t
en
re
co
g
nitio
n sy
s
t
em
T
r
ain
in
g
with
T
en
s
o
r
Flo
w
w
as
co
n
d
u
cte
d
ac
r
o
s
s
s
ev
er
al
e
p
o
ch
s
,
a
n
d
th
e
ep
o
ch
th
at
p
r
o
d
u
ce
d
th
e
h
ig
h
est
lev
el
o
f
ac
cu
r
ac
y
was
co
n
s
id
er
ed
.
T
h
e
f
ir
s
t
p
ictu
r
e
g
en
er
ated
f
r
o
m
th
e
test
s
et
i
s
s
h
o
wn
in
Fig
u
r
e
1
1
an
d
Fig
u
r
e
1
2
,
d
em
o
n
s
tr
atin
g
wh
eth
er
th
e
m
o
d
el'
s
p
r
ed
icti
o
n
was
ac
cu
r
ate.
T
h
e
m
o
d
el
p
r
ed
icted
c
o
r
r
ec
tly
,
id
en
tify
in
g
d
ig
it
s
ev
en
.
Su
b
s
eq
u
en
t
test
in
s
tan
ce
s
wer
e
p
r
ed
icted
an
d
f
o
r
ec
asted
to
o
b
s
er
v
e
th
e
m
o
d
el'
s
p
er
f
o
r
m
an
ce
an
d
er
r
o
r
s
.
Fig
u
r
e
1
1
.
Pre
d
icito
n
o
u
tp
u
t
f
r
o
m
T
en
s
o
r
Flo
w
m
o
d
el
Fig
u
r
e
1
2
.
Pre
d
ictio
n
o
u
tp
u
t t
o
ea
ch
o
f
th
eir
tr
u
e
class
3
.
4
.
Co
m
pa
riso
n o
n t
he
perf
o
rm
a
nce
o
f
f
o
ur
deep
lea
rning
m
o
dels
T
h
e
C
NN
m
o
d
el
co
n
s
is
ten
tly
o
u
tp
er
f
o
r
m
s
th
e
o
th
er
m
o
d
els
ac
r
o
s
s
all
f
iv
e
m
etr
ics:
8
1
.
7
7
%
in
m
ea
n
av
er
ag
e
p
r
ec
is
io
n
(
m
AP)
,
7
8
.
8
5
%
in
Pre
cisi
o
n
,
7
9
.
3
2
%
in
R
ec
all,
7
9
.
4
6
%
in
F1
-
Sco
r
e,
a
n
d
8
2
.
4
%
in
r
ec
eiv
er
o
p
er
atin
g
ch
a
r
ac
ter
is
ti
c
(
R
OC
)
,
as
s
h
o
wn
in
Fig
u
r
e
1
3
.
T
h
e
SVM
m
o
d
el
p
er
f
o
r
m
s
b
etter
t
h
an
C
T
C
an
d
R
N
N
in
ter
m
s
o
f
m
AP,
Pre
cisi
o
n
,
an
d
R
OC
.
Ho
wev
er
,
it
h
as
a
lo
wer
F1
-
Sco
r
e
th
an
C
T
C
an
d
R
NN.
T
h
e
C
T
C
m
o
d
el
h
as
th
e
th
i
r
d
-
b
est
p
e
r
f
o
r
m
an
ce
in
all
m
etr
ics
ex
ce
p
t
f
o
r
F1
-
Sco
r
e
,
wh
er
e
it
o
u
tp
er
f
o
r
m
s
SVM
an
d
R
NN.
T
h
e
R
NN
m
o
d
el
h
as
th
e
lo
west
v
alu
es
am
o
n
g
all
e
x
ce
p
t
f
o
r
F1
-
Sco
r
e,
wh
er
e
it
p
er
f
o
r
m
s
b
etter
th
an
SVM.
I
n
c
o
n
clu
s
io
n
,
th
e
C
NN
m
o
d
el
ap
p
ea
r
s
to
b
e
th
e
m
o
s
t
ef
f
ec
tiv
e
f
o
r
t
h
is
wo
r
k
ac
co
r
d
in
g
to
t
h
e
m
etr
ics
u
s
ed
in
th
is
co
m
p
a
r
is
o
n
.
Fig
u
r
e
1
3
.
Per
f
o
r
m
a
n
ce
o
f
C
NN,
SVM,
C
T
C
an
d
R
NN
mA
P
(
%
)
P
r
e
c
is
ion
(
%
)
R
e
c
a
ll
(
%
)
F
1
-
S
c
o
r
e
(
%
)
R
OC
(
%
)
C
NN
8
1
.
7
7
78.
85
7
9
.
3
2
79.
46
82.
4
S
V
M
7
3
.
6
3
70.
5
7
5
.
9
4
74.
18
74.
6
C
T
C
7
1
.
9
2
66.
23
6
8
.
0
4
76.
73
7
0
.
2
9
R
NN
7
0
.
9
5
62.
36
6
5
.
1
2
74.
08
6
6
.
7
9
0
10
20
30
40
50
60
70
80
90
M
e
tr
i
c
s
pe
r
c
e
n
tag
e
(
%
)
M
o
d
e
l
p
e
r
f
o
r
m
a
n
c
e
p
a
r
a
m
e
ter
s
M
e
tr
ic
s
p
e
r
fo
r
m
a
n
c
e
c
o
m
p
a
r
is
o
n
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
Ha
n
d
w
r
itten
text
r
ec
o
g
n
itio
n
s
ystem
u
s
in
g
R
a
s
p
b
err
y
P
i wi
th
…
(
Ja
mil A
b
ed
a
lr
a
h
im
J
a
mil A
ls
a
ya
yd
eh
)
2299
3
.
5
.
Co
m
pa
riso
n o
n t
he
a
cc
ura
cy
o
f
f
o
ur
deep
lea
rning
cla
s
s
if
iers
A
co
m
p
r
eh
e
n
s
iv
e
an
aly
s
is
an
d
co
m
p
ar
is
o
n
o
f
th
e
tech
n
i
q
u
es
d
is
cu
s
s
ed
will
b
e
p
r
esen
ted
at
th
e
en
d
o
f
th
is
s
ec
tio
n
.
Mo
s
t
o
f
th
e
te
ch
n
iq
u
es
u
tili
ze
d
in
th
is
s
ec
ti
o
n
ar
e
eith
er
C
NN
o
r
a
h
y
b
r
id
b
etwe
en
C
NN
an
d
o
th
er
n
e
u
r
al
n
etwo
r
k
s
.
T
h
is
s
h
o
ws
th
at
C
NN
is
p
r
o
v
en
to
b
e
ju
s
t
as
ef
f
ec
tiv
e
wh
ile
b
ei
n
g
a
s
im
p
ler
d
ee
p
lear
n
in
g
class
if
ier
.
3
.
6
.
Str
eng
t
hs
,
lim
it
a
t
io
ns
,
a
nd
un
ex
pect
ed
re
s
ults
T
ab
le
1
s
h
o
ws
th
e
ac
cu
r
ac
y
o
f
v
ar
io
u
s
SVM
-
b
ased
tex
t
r
ec
o
g
n
itio
n
tech
n
iq
u
es
o
n
MN
I
ST,
C
E
DAR
,
an
d
I
AM
On
DB
d
atasets
.
T
h
e
h
y
b
r
id
C
NN
-
SVM
m
o
d
el
p
er
f
o
r
m
s
b
est
with
8
8
.
6
%
o
n
MN
I
ST,
8
7
.
8
3
%
o
n
C
E
DAR,
an
d
8
2
.
2
4
%
o
n
I
A
M
On
DB
,
s
u
g
g
esti
n
g
th
at
c
o
m
b
in
in
g
C
NNs
an
d
SVMs
ca
n
im
p
r
o
v
e
HT
R
.
Ho
wev
er
,
ac
c
u
r
ac
y
d
r
o
p
s
o
n
t
h
e
m
o
r
e
ch
allen
g
in
g
C
E
DAR
an
d
I
AM
On
DB
d
atasets
,
in
d
i
ca
tin
g
th
e
n
ee
d
f
o
r
f
u
r
th
er
r
esear
c
h
.
C
NNs
ar
e
b
eliev
ed
to
b
e
k
e
y
in
ad
v
a
n
cin
g
HT
R
,
esp
ec
ially
in
ad
d
r
ess
in
g
SVM
-
b
ased
ap
p
r
o
ac
h
es’
lim
itatio
n
s
.
B
y
au
to
m
atica
lly
ex
tr
ac
tin
g
i
n
f
o
r
m
ativ
e
f
ea
tu
r
es,
C
NNs
elim
in
ate
th
e
n
ee
d
f
o
r
m
an
u
al
f
ea
tu
r
e
d
esig
n
,
ca
p
tu
r
in
g
s
u
b
tle
v
ar
iatio
n
s
in
wr
itin
g
s
ty
les
an
d
p
en
s
tr
o
k
es.
T
h
is
is
cr
u
cial
f
o
r
r
ec
o
g
n
izin
g
d
iv
er
s
e
r
ea
l
-
wo
r
l
d
h
a
n
d
wr
itten
te
x
t.
C
o
m
b
in
i
n
g
C
NNs
with
p
o
wer
f
u
l
cla
s
s
if
ier
s
lik
e
SVM
s
lev
er
ag
es
b
o
th
th
eir
s
tr
en
g
th
s
:
r
o
b
u
s
t
f
ea
tu
r
e
ex
tr
ac
tio
n
f
r
o
m
C
NNs
an
d
h
ig
h
class
if
icatio
n
ac
cu
r
ac
y
f
r
o
m
SVMs.
T
h
is
h
y
b
r
id
ap
p
r
o
ac
h
h
as
s
h
o
wn
p
r
o
m
is
in
g
r
esu
lts
in
d
iv
er
s
e
task
s
,
f
r
o
m
h
y
b
r
id
C
NN
-
SVM
m
o
d
els
to
en
d
-
to
-
en
d
o
n
lin
e
r
ec
o
g
n
it
io
n
with
C
NNs
an
d
R
NNs.
T
r
an
s
f
er
lear
n
in
g
,
wh
er
e
p
r
e
-
tr
ain
ed
C
NNs
ass
is
t
SVM
m
o
d
els’
f
ea
tu
r
e
ex
tr
ac
ti
o
n
,
also
s
h
o
ws
s
ig
n
if
ican
t
p
er
f
o
r
m
an
ce
im
p
r
o
v
em
en
ts
,
esp
e
cially
with
lim
ited
tr
ain
in
g
d
ata.
T
h
u
s
,
C
NNs o
f
f
er
a
p
o
wer
f
u
l to
o
l
f
o
r
a
d
v
an
ci
n
g
HT
R
.
T
ab
le
2
p
r
esen
ts
r
esu
lts
o
f
C
T
C
m
eth
o
d
s
o
n
HT
R
s
y
s
tem
s
.
T
h
e
p
r
o
p
o
s
ed
C
NN
-
b
ased
m
eth
o
d
ad
d
r
ess
es
C
T
C
’
s
lim
itatio
n
s
in
ter
m
s
o
f
ac
c
u
r
ac
y
,
r
o
b
u
s
t
n
ess
,
an
d
v
er
s
atility
.
C
NNs
ca
n
ac
h
iev
e
b
etter
ac
cu
r
ac
y
th
an
C
T
C
as
th
ey
ca
n
lear
n
m
o
r
e
c
o
m
p
lex
f
ea
tu
r
es
f
r
o
m
h
an
d
wr
itten
ch
ar
ac
t
er
s
.
T
h
ey
ar
e
m
o
r
e
r
o
b
u
s
t
to
n
o
is
e
an
d
d
is
to
r
tio
n
s
as
th
ey
ca
n
ex
t
r
ac
t
f
ea
tu
r
es
f
r
o
m
th
e
en
tire
c
h
ar
ac
ter
,
ev
en
if
it
i
s
n
o
t
p
er
f
ec
tly
alig
n
ed
o
r
h
as
n
o
is
e.
C
NNs
c
an
b
e
a
d
ap
ted
to
r
ec
o
g
n
ize
d
i
f
f
er
en
t
h
an
d
wr
itin
g
s
ty
les
an
d
lan
g
u
ag
es
with
o
u
t
r
eq
u
ir
in
g
p
r
io
r
k
n
o
wled
g
e
o
f
t
h
e
s
ty
le
o
r
lan
g
u
ag
e.
T
ab
le
1
.
R
esu
lts
o
f
tex
t r
ec
o
g
n
itio
n
tech
n
iq
u
es b
ased
o
n
SV
M
Y
e
a
r
M
e
t
h
o
d
A
c
c
u
r
a
c
y
R
e
f
e
r
e
n
c
e
2
0
2
2
C
N
N
,
S
V
M
8
5
.
4
1
%
[
9
]
2
0
1
8
S
V
M
7
8
.
6
0
%
[
1
6
]
2
0
2
3
C
N
N
,
R
N
N
,
S
V
M
8
7
.
6
8
%
[
2
9
]
2
0
1
9
S
V
M
8
7
.
8
3
%
[
3
0
]
2
0
2
1
S
V
M
8
2
.
2
4
%
[
3
1
]
2
0
2
4
C
N
N
9
1
.
5
8
%
Th
i
s
w
o
r
k
T
ab
le
2
.
R
esu
lts
o
f
tex
t r
ec
o
g
n
itio
n
tech
n
iq
u
es b
ased
o
n
C
T
C
Y
e
a
r
M
e
t
h
o
d
s
D
a
t
a
b
a
s
e
A
c
c
u
r
a
c
y
R
e
f
e
r
e
n
c
e
2
0
1
3
C
TC
,
E
M
,
D
G
W
T
O
R
L
8
9
.
0
0
%
[
3
2
]
2
0
1
7
C
TC
F
ER
ET
,
U
M
B
-
D
B
,
F
R
G
C
8
3
%
-
8
6
%
[
3
3
]
2
0
1
7
C
TC
,
S
q
2
S
q
M
e
d
i
e
v
a
l
La
t
i
n
t
e
x
t
s
7
8
.
1
0
%
[
3
4
]
2
0
1
8
C
TC
I
A
M
,
LO
B
8
6
.
6
5
%
[
3
5
]
2
0
2
4
C
N
N
EM
N
I
S
T
9
1
.
5
8
%
Th
i
s
w
o
r
k
I
n
m
o
r
e
d
etail,
C
NNs
u
s
e
co
n
v
o
lu
tio
n
an
d
p
o
o
lin
g
o
p
e
r
at
io
n
s
to
ex
t
r
ac
t
f
ea
tu
r
es,
ca
p
t
u
r
in
g
b
o
th
lo
ca
l
an
d
g
lo
b
al
f
ea
t
u
r
es.
C
T
C
,
a
lin
ea
r
m
o
d
el,
ca
n
n
o
t
e
x
tr
ac
t
s
u
ch
co
m
p
lex
f
ea
tu
r
es.
C
NNs
ar
e
m
o
r
e
r
o
b
u
s
t
th
an
C
T
C
,
wh
ich
is
s
en
s
i
tiv
e
t
o
n
o
is
e
an
d
d
is
to
r
tio
n
s
d
u
e
to
its
r
eq
u
ir
em
en
t
f
o
r
in
p
u
t
an
d
o
u
tp
u
t
s
eq
u
en
ce
s
to
b
e
alig
n
ed
.
C
NNs c
an
b
e
ad
ap
ted
to
r
ec
o
g
n
ize
d
if
f
er
en
t h
a
n
d
wr
itin
g
s
ty
les an
d
lan
g
u
ag
es,
u
n
lik
e
C
T
C
,
wh
ich
is
ty
p
ically
tr
ain
ed
o
n
a
s
p
e
cif
ic
d
ataset
an
d
m
ay
n
o
t
p
e
r
f
o
r
m
well
o
n
d
if
f
er
en
t
d
ata
s
ets.
T
h
e
p
r
o
p
o
s
ed
C
NN
-
b
ased
s
y
s
tem
ac
h
iev
es
a
n
ac
cu
r
ac
y
o
f
9
1
.
5
8
%
o
n
th
e
MN
I
ST
d
ataset,
is
r
o
b
u
s
t
to
n
o
is
e
an
d
d
is
to
r
tio
n
s
,
an
d
ca
n
r
ec
o
g
n
ize
d
if
f
e
r
en
t h
a
n
d
wr
itin
g
s
ty
les an
d
lan
g
u
ag
e
s
.
Ov
er
all,
C
NN
s
ar
e
s
u
p
er
io
r
to
C
T
C
f
o
r
HT
R
in
ter
m
s
o
f
ac
cu
r
ac
y
,
r
o
b
u
s
tn
ess
,
an
d
v
e
r
s
atility
.
T
ab
le
3
ev
alu
ates
tex
t
r
ec
o
g
n
itio
n
tech
n
iq
u
es
b
ased
o
n
R
NN
o
n
th
e
B
U
-
3
DFE
an
d
L
AF
d
atasets
.
T
h
e
p
r
o
p
o
s
ed
C
NN
-
b
ased
m
e
th
o
d
ad
d
r
ess
es
R
NN
l
im
itatio
n
s
f
o
r
tex
t
r
ec
o
g
n
itio
n
.
C
NNs
ar
e
m
o
r
e
e
f
f
icien
t
th
an
R
NNs
f
o
r
h
ig
h
-
d
im
e
n
s
io
n
al
d
ata
lik
e
im
ag
es,
less
p
r
o
n
e
to
v
a
n
is
h
in
g
an
d
e
x
p
lo
d
in
g
g
r
a
d
ien
ts
,
an
d
ca
p
tu
r
e
lo
n
g
-
r
a
n
g
e
d
e
p
en
d
e
n
cies
in
s
eq
u
en
tial
d
ata
m
o
r
e
ef
f
ec
tiv
ely
.
T
h
e
p
r
o
p
o
s
ed
C
NN
-
b
ased
m
eth
o
d
ac
h
iev
es
an
ac
cu
r
ac
y
o
f
8
7
.
3
%
o
n
th
e
B
U
-
3
DFE
d
ataset,
o
u
tp
er
f
o
r
m
in
g
th
e
b
est
R
NN
-
b
ased
m
eth
o
d
with
an
ac
cu
r
ac
y
o
f
8
4
%.
C
NNs
ar
e
m
o
r
e
ef
f
icien
t
d
u
e
to
p
ar
allel
p
r
o
ce
s
s
in
g
o
f
m
u
ltip
le
p
ix
els
an
d
u
s
e
o
f
co
n
v
o
l
u
tio
n
o
p
er
atio
n
s
.
T
h
e
y
ar
e
less
p
r
o
n
e
t
o
v
a
n
is
h
in
g
an
d
ex
p
lo
d
in
g
g
r
a
d
ien
ts
d
u
e
to
s
h
o
r
ter
p
at
h
s
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
15
,
No
.
2
,
Ap
r
il
20
25
:
2
2
9
1
-
2
3
0
3
2300
b
etwe
en
lay
er
s
a
n
d
lo
ca
l
c
o
n
n
ec
tio
n
s
.
C
NNs
ca
p
tu
r
e
lo
n
g
-
r
an
g
e
d
e
p
en
d
e
n
cies
m
o
r
e
ef
f
ec
tiv
ely
th
r
o
u
g
h
p
o
o
lin
g
lay
er
s
th
at
d
o
wn
s
am
p
le
f
ea
tu
r
e
m
ap
s
,
allo
win
g
g
l
o
b
al
f
ea
tu
r
e
ca
p
tu
r
e.
Ov
e
r
all,
C
NNs
ar
e
a
b
etter
ch
o
ice
f
o
r
f
ac
e
r
ec
o
g
n
itio
n
t
ask
s
th
an
R
NNs
d
u
e
to
th
eir
ef
f
icien
cy
,
r
esis
tan
ce
to
g
r
ad
ien
t
is
s
u
es,
an
d
ef
f
ec
tiv
e
ca
p
tu
r
e
o
f
l
o
n
g
-
r
an
g
e
d
ep
en
d
en
cies.
T
ab
le
4
ev
alu
ates
HT
R
m
eth
o
d
s
o
n
t
h
e
I
AM
d
ataset,
with
th
e
p
r
o
p
o
s
ed
C
NN
-
b
ased
m
eth
o
d
ac
h
iev
in
g
t
h
e
h
i
g
h
est
ac
cu
r
a
cy
o
f
9
1
.
5
8
%.
T
h
is
m
eth
o
d
o
u
tp
e
r
f
o
r
m
s
o
th
er
s
s
ig
n
if
ica
n
tly
b
y
ad
d
r
ess
in
g
co
m
m
o
n
C
NN
-
b
ased
HT
R
s
y
s
tem
lim
itatio
n
s
.
I
t
u
s
es
r
eg
u
lar
izatio
n
tech
n
iq
u
es
lik
e
d
r
o
p
o
u
t
an
d
b
atch
n
o
r
m
aliza
tio
n
t
o
p
r
e
v
en
t o
v
er
f
itti
n
g
an
d
im
p
r
o
v
e
g
e
n
er
aliza
tio
n
.
I
t
is
also
ef
f
icien
t
in
m
e
m
o
r
y
a
n
d
p
r
o
ce
s
s
in
g
r
eq
u
ir
em
e
n
ts
d
u
e
to
a
lig
h
tw
eig
h
t
C
NN
ar
ch
itectu
r
e.
T
h
e
m
eth
o
d
’
s
r
o
b
u
s
tn
ess
to
n
o
is
e
an
d
d
is
to
r
tio
n
s
is
en
h
an
ce
d
b
y
a
d
ata
au
g
m
en
tat
io
n
s
tr
ateg
y
th
at
g
e
n
er
ates r
ea
l
is
tic
tr
ain
in
g
s
am
p
les.
Ad
d
itio
n
ally
,
it is
v
er
s
atile
an
d
ca
n
ad
a
p
t
to
r
ec
o
g
n
ize
d
if
f
er
en
t
lan
g
u
a
g
es,
s
y
m
b
o
ls
,
an
d
h
an
d
wr
itin
g
s
ty
les,
m
a
k
in
g
it
a
p
r
o
m
is
in
g
s
o
lu
tio
n
f
o
r
v
ar
io
u
s
HT
R
ap
p
licatio
n
s
.
T
ab
le
3
.
R
esu
lts
o
f
tex
t r
ec
o
g
n
itio
n
tech
n
iq
u
es b
ased
o
n
R
N
N
Y
e
a
r
M
e
t
h
o
d
s
D
a
t
a
b
a
s
e
A
c
c
u
r
a
c
y
R
e
f
e
r
e
n
c
e
2
0
1
8
W
a
v
e
l
e
t
G
a
b
o
r
F
i
l
t
e
r
i
n
g
&
R
N
N
BU
-
3
D
F
E
8
7
.
3
0
%
[
3
6
]
2
0
2
0
R
N
N
,
R
B
F
k
e
r
n
e
l
,
P
o
l
y
n
o
m
i
a
l
k
e
r
n
e
l
LA
F
8
4
.
0
0
%
[
3
7
]
2
0
2
3
C
N
N
,
R
N
N
,
S
V
M
I
A
M
8
7
.
6
8
%
[
3
8
]
2
0
1
9
R
N
N
-
8
1
.
0
0
%
[
3
9
]
2
0
2
4
C
N
N
EM
N
I
S
T
9
1
.
5
8
%
T
h
i
s
w
o
r
k
T
ab
le
4
.
R
esu
lts
o
f
HT
R
b
ased
o
n
d
ee
p
lear
n
in
g
class
if
ier
s
Y
e
a
r
M
e
t
h
o
d
D
a
t
a
s
e
t
N
u
mb
e
r
o
f
i
m
a
g
e
s
i
n
t
r
a
i
n
i
n
g
s
e
t
N
u
mb
e
r
o
f
i
m
a
g
e
s
in
t
e
s
t
i
n
g
se
t
A
c
c
u
r
a
c
y
R
e
f
e
r
e
n
c
e
2
0
2
3
C
N
N
,
LST
M
,
R
N
N
,
C
T
C
I
A
M
6
0
0
,
0
0
0
2
0
0
,
0
0
0
8
4
.
5
0
%
[
8
]
2
0
1
8
S
V
M
I
A
M
1
0
3
,
7
8
8
1
1
,
5
3
2
8
8
.
6
0
%
[
1
6
]
2
0
2
3
EA
S
T,
T
e
ssera
c
t
-
O
C
R
I
C
D
A
R
2
0
1
9
5
,
6
0
3
4
,
5
6
3
8
8
.
6
9
%
[
2
2
]
2
0
2
3
C
N
N
,
V
G
G
-
16
I
A
M
1
1
,
0
0
0
2
,
3
5
3
8
8
.
3
6
%
[
4
0
]
2
0
1
8
C
N
N
EM
N
I
S
T
1
2
4
,
8
0
0
2
0
,
8
0
0
8
4
.
2
0
%
[
4
1
]
2
0
2
2
C
N
N
,
S
V
M
EM
N
I
S
T
4
7
,
0
0
0
1
1
,
0
0
0
8
5
.
4
1
%
[
4
2
]
2
0
2
4
C
N
N
EM
N
I
S
T
7
0
0
,
0
0
0
7
0
,
0
0
0
9
1
.
5
8
%
Th
i
s
w
o
r
k
3
.
7
.
Su
mm
a
ry
a
nd
f
uture
re
s
ea
rc
h direc
t
io
ns
T
h
is
r
esear
ch
p
r
esen
ts
a
r
o
b
u
s
t
HT
R
s
y
s
tem
u
s
in
g
a
C
NN
.
T
h
e
C
NN
m
o
d
el
ac
h
iev
ed
a
h
ig
h
ac
cu
r
ac
y
r
ate
o
f
9
1
.
5
8
%,
d
em
o
n
s
tr
atin
g
its
s
u
itab
ilit
y
f
o
r
H
T
R
ap
p
licatio
n
s
.
T
h
e
s
y
s
tem
'
s
v
er
s
atility
allo
ws
p
o
ten
tial
ap
p
licatio
n
s
in
v
ar
io
u
s
d
o
m
ai
n
s
,
s
u
ch
as
d
ig
itizin
g
h
is
to
r
ical
d
o
cu
m
e
n
ts
o
r
r
ec
o
g
n
izin
g
h
an
d
wr
itten
in
p
u
ts
o
n
m
o
b
ile
d
ev
ices.
F
u
tu
r
e
wo
r
k
will
f
o
cu
s
o
n
e
n
h
an
cin
g
th
e
s
y
s
tem
'
s
ad
ap
tab
ilit
y
to
d
if
f
er
e
n
t
lan
g
u
ag
es
an
d
h
an
d
wr
itin
g
s
ty
les,
as
well
as
im
p
r
o
v
in
g
r
ea
l
-
tim
e
p
r
o
ce
s
s
in
g
ca
p
a
b
ilit
ies
o
n
em
b
ed
d
e
d
p
latf
o
r
m
s
lik
e
R
asp
b
er
r
y
Pi.
4.
CO
NCLU
SI
O
N
T
h
is
s
tu
d
y
p
r
esen
ts
a
r
o
b
u
s
t
HT
R
s
y
s
tem
u
s
in
g
C
N
N,
im
p
lem
en
ted
with
Op
en
C
V
an
d
T
en
s
o
r
Flo
w,
an
d
d
ep
l
o
y
ed
o
n
a
R
asp
b
er
r
y
Pi
4
B
p
latf
o
r
m
.
T
h
e
s
y
s
tem
d
em
o
n
s
tr
ated
a
h
ig
h
ac
c
u
r
ac
y
r
ate
o
f
9
1
.
5
8
%
in
r
ec
o
g
n
izin
g
E
n
g
lis
h
alp
h
ab
et
s
an
d
d
ig
its
,
o
u
tp
er
f
o
r
m
in
g
o
th
er
m
o
d
els
s
u
ch
as
SVM,
C
T
C
,
an
d
R
NN.
T
h
e
f
in
d
in
g
s
h
i
g
h
lig
h
t
th
e
ef
f
ec
tiv
en
ess
o
f
C
NNs
in
HT
R
task
s
,
s
h
o
wca
s
in
g
s
ig
n
if
ican
t
im
p
r
o
v
em
en
ts
in
r
ec
o
g
n
itio
n
ac
cu
r
ac
y
c
o
m
p
ar
e
d
to
p
r
ev
i
o
u
s
m
o
d
els.
T
h
e
im
p
licatio
n
s
o
f
th
is
r
esear
ch
ar
e
s
u
b
s
tan
tial,
as
th
e
d
ev
elo
p
e
d
HT
R
s
y
s
tem
ca
n
b
e
ap
p
lied
in
v
a
r
io
u
s
d
o
m
ain
s
,
in
clu
d
in
g
d
ig
itizin
g
h
a
n
d
wr
itten
h
is
to
r
ical
d
o
cu
m
e
n
ts
,
r
ea
l
-
tim
e
tex
t
r
ec
o
g
n
itio
n
in
em
b
ed
d
ed
s
y
s
tem
s
,
an
d
en
h
a
n
cin
g
ac
ce
s
s
ib
ilit
y
to
o
ls
.
T
h
e
s
y
s
tem
'
s
ab
ilit
y
to
p
r
o
ce
s
s
an
d
r
ec
o
g
n
ize
h
an
d
wr
itten
tex
t
in
r
ea
l
-
tim
e
o
n
a
lo
w
-
co
s
t
p
latf
o
r
m
lik
e
R
asp
b
er
r
y
Pi
d
em
o
n
s
tr
ates its
p
r
ac
tical
ap
p
l
icab
ilit
y
.
Fu
tu
r
e
r
esear
ch
will
f
o
cu
s
o
n
ex
ten
d
in
g
th
e
s
y
s
tem
to
s
u
p
p
o
r
t
m
u
ltip
le
lan
g
u
ag
es
an
d
im
p
r
o
v
in
g
its
ad
ap
tab
ilit
y
to
d
if
f
er
en
t
h
an
d
wr
itin
g
s
ty
les
an
d
co
n
d
itio
n
s
.
Ad
d
itio
n
ally
,
ef
f
o
r
ts
will
b
e
m
ad
e
to
en
h
an
ce
th
e
s
y
s
tem
'
s
r
ea
l
-
tim
e
p
r
o
ce
s
s
in
g
ca
p
ab
ilit
ies
an
d
ex
p
lo
r
e
p
o
t
en
tial
ap
p
licatio
n
s
in
m
o
b
ile
d
ev
ices
an
d
o
th
e
r
r
eso
u
r
ce
-
co
n
s
tr
ain
ed
en
v
i
r
o
n
m
en
ts
.
B
y
ad
d
r
ess
in
g
th
ese
ch
allen
g
es,
th
e
HT
R
s
y
s
tem
ca
n
b
e
f
u
r
th
e
r
r
ef
in
ed
t
o
p
r
o
v
id
e
ev
en
g
r
ea
ter
ac
c
u
r
ac
y
an
d
e
f
f
icien
cy
in
HT
R
task
s
.
I
n
s
u
m
m
a
r
y
,
th
e
f
in
d
i
n
g
s
f
r
o
m
th
is
s
tu
d
y
co
n
tr
ib
u
te
to
th
e
f
ield
o
f
H
T
R
b
y
d
em
o
n
s
tr
atin
g
th
e
s
u
p
er
io
r
p
er
f
o
r
m
a
n
ce
o
f
C
NNs
an
d
th
eir
p
r
ac
tical
im
p
lem
en
tatio
n
o
n
em
b
e
d
d
ed
s
y
s
tem
s
,
o
f
f
er
in
g
a
p
r
o
m
is
in
g
s
o
lu
tio
n
f
o
r
ac
cu
r
ate
a
n
d
ef
f
icien
t
HT
R
.
Evaluation Warning : The document was created with Spire.PDF for Python.