I
nd
o
ne
s
ia
n J
o
urna
l o
f
E
lect
rica
l En
g
ineering
a
nd
Co
m
pu
t
er
Science
Vo
l.
38
,
No
.
3
,
J
u
n
e
2
0
2
5
,
p
p
.
1
880
~
1
8
9
5
I
SS
N:
2
5
0
2
-
4
7
5
2
,
DOI
: 1
0
.
1
1
5
9
1
/ijeecs.v
38
.i
3
.
pp
1
8
8
0
-
1
8
9
5
1880
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
ee
cs.ia
esco
r
e.
co
m
Deep learni
ng
appro
a
ches, platfor
ms
,
da
tas
ets
for b
eha
v
io
r
-
ba
sed recog
nition
:
a
surv
ey
Yunu
s
a
M
o
ha
m
m
ed
J
edda
h
1
,
Ais
ha
H
a
s
s
a
n Abda
lla
H
a
s
him
1,
2
,
O
t
hm
a
n O
m
r
a
n K
ha
lifa
1
,3
1
D
e
p
a
r
t
me
n
t
o
f
El
e
c
t
r
i
c
a
l
a
n
d
C
o
m
p
u
t
e
r
E
n
g
i
n
e
e
r
i
n
g
,
I
n
t
e
r
n
a
t
i
o
n
a
l
I
sl
a
mi
c
U
n
i
v
e
r
si
t
y
M
a
l
a
y
si
a
(
I
I
U
M
)
,
K
u
a
l
a
L
u
mp
u
r
,
M
a
l
a
y
s
i
a
2
D
e
p
ar
t
me
n
t
o
f
El
e
c
t
r
i
c
a
l
a
n
d
El
e
c
t
r
o
n
i
c
E
n
g
i
n
e
e
r
i
n
g
S
c
i
e
n
c
e
,
U
n
i
v
e
r
si
t
y
o
f
Jo
h
a
n
n
e
sb
u
r
g
,
J
o
h
a
n
n
e
sb
u
r
g
,
S
o
u
t
h
A
f
r
i
c
a
3
Li
b
y
a
L
i
b
y
a
n
C
e
n
t
e
r
f
o
r
E
n
g
i
n
e
e
r
i
n
g
R
e
s
e
a
r
c
h
a
n
d
I
n
f
o
r
m
a
t
i
o
n
Te
c
h
n
o
l
o
g
y
,
B
a
n
i
W
a
l
i
d
,
Li
b
y
a
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Au
g
28
,
2
0
2
4
R
ev
is
ed
No
v
28
,
2
0
2
4
Acc
ep
ted
Feb
27
,
2
0
2
5
Vid
e
o
s
u
rv
e
il
la
n
c
e
is
a
n
e
x
ten
siv
e
ly
u
se
d
t
o
o
l
d
u
e
to
t
h
e
h
ig
h
ra
te o
f
a
t
y
p
ica
l
b
e
h
a
v
i
o
r
a
n
d
m
a
n
y
c
a
m
e
ra
s
th
a
t
e
n
a
b
le
v
id
e
o
c
a
p
t
u
re
a
n
d
sto
ra
g
e
.
Un
fo
rt
u
n
a
tely
,
m
o
st
o
f
t
h
e
se
c
a
m
e
ra
s
a
r
e
o
p
e
ra
to
r
d
e
p
e
n
d
e
n
t
fo
r
sto
re
d
c
o
n
ten
t
a
n
a
l
y
sis.
Th
is
li
m
it
a
ti
o
n
n
e
c
e
ss
it
a
tes
th
e
p
ro
v
isio
n
o
f
a
n
a
u
to
m
a
ti
c
b
e
h
a
v
i
o
r
id
e
n
ti
fica
ti
o
n
sy
ste
m
.
T
h
is
b
e
h
a
v
io
r
id
e
n
t
ifi
c
a
ti
o
n
c
a
n
b
e
a
c
h
iev
e
d
u
sin
g
u
n
s
u
p
e
rv
ise
d
(g
e
n
e
ra
ti
v
e
)
c
o
m
p
u
ter
v
isi
o
n
m
e
th
o
d
s.
De
e
p
lea
rn
in
g
m
a
k
e
s
it
p
o
ss
ib
le
to
m
o
d
e
l
h
u
m
a
n
b
e
h
a
v
i
o
r
re
g
a
r
d
les
s
o
f
wh
e
re
t
h
e
y
c
o
u
l
d
b
e
.
We
a
tt
e
m
p
t
to
c
las
sify
c
u
rre
n
t
re
se
a
rc
h
wo
rk
t
o
re
p
o
rt
t
h
e
o
n
g
o
i
n
g
tre
n
d
s
in
h
u
m
a
n
b
e
h
a
v
io
r
re
c
o
g
n
it
io
n
u
sin
g
d
e
e
p
lea
rn
in
g
a
lg
o
rit
h
m
s.
T
h
is
p
a
p
e
r
re
v
iew
s
v
a
rio
u
s
a
sp
e
c
ts,
li
k
e
th
e
o
n
e
s
a
ss
o
c
iate
d
wit
h
m
a
c
h
in
e
lea
rn
in
g
a
n
d
d
e
e
p
lea
rn
i
n
g
m
o
d
e
ls,
h
u
m
a
n
a
c
ti
v
it
y
re
c
o
g
n
it
i
o
n
(HA
R),
d
e
e
p
lea
rn
in
g
fra
m
e
wo
rk
s/to
o
ls,
a
b
n
o
rm
a
l
b
e
h
a
v
io
r
d
a
tas
e
ts,
a
n
d
a
v
a
riet
y
o
f
o
th
e
r
c
u
rre
n
t
tren
d
s
i
n
t
h
e
field
o
f
a
u
to
m
a
ti
c
le
a
rn
in
g
.
All
th
e
se
a
re
to
g
i
v
e
th
e
r
e
se
a
rc
h
e
r
a
se
n
se
o
f
d
irec
ti
o
n
i
n
t
h
is are
a
.
K
ey
w
o
r
d
s
:
B
eh
av
io
r
r
ec
o
g
n
itio
n
Dee
p
lear
n
in
g
Hu
m
an
ac
tiv
ity
r
ec
o
g
n
itio
n
Ma
ch
in
e
lear
n
in
g
Vid
eo
s
u
r
v
eillan
ce
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
:
Yu
n
u
s
a
Mo
h
am
m
e
d
J
ed
d
a
h
Dep
ar
tm
en
t o
f
E
lectr
ical
an
d
C
o
m
p
u
ter
E
n
g
in
ee
r
in
g
,
I
n
te
r
n
atio
n
al
I
s
lam
ic
Un
iv
er
s
ity
Ma
lay
s
ia
(
I
I
UM
)
Go
m
b
ak
,
Ku
ala
L
u
m
p
u
r
,
Sela
n
g
o
r
,
Ma
lay
s
ia
E
m
ail:
y
u
n
u
s
m
j2
@
h
o
tm
ail.
co
m
1.
I
NT
RO
D
UCT
I
O
N
I
n
r
ec
en
t
y
ea
r
s
,
th
er
e
h
as
b
ee
n
an
in
cr
ea
s
e
in
r
esear
ch
f
o
cu
s
ed
o
n
AI
(
a
r
tific
ial
in
tellig
en
ce
)
,
p
ar
ticu
lar
ly
th
e
d
ee
p
lear
n
in
g
ar
ea
.
T
h
is
s
u
r
g
e
in
r
esear
ch
in
ter
est
h
as
led
to
s
ev
er
al
s
t
u
d
ies
s
u
r
v
ey
s
,
an
d
r
ev
iew
p
ap
er
s
wh
ic
h
ex
p
l
o
r
e
v
ar
io
u
s
d
ee
p
lear
n
in
g
ap
p
licatio
n
s
.
No
tab
ly
,
d
ee
p
lear
n
in
g
h
as
b
ee
n
em
p
lo
y
ed
in
th
e
f
ield
o
f
m
ed
icin
e
s
u
ch
as
ey
e
d
is
ea
s
e
d
iag
n
o
s
is
with
th
e
u
s
e
o
f
r
etin
al
f
u
n
d
u
s
i
m
ag
es
[
1
]
an
d
f
o
r
im
p
r
o
v
in
g
r
ad
io
lo
g
y
tech
n
iq
u
es
[
2
]
.
Oth
er
p
r
o
m
in
en
t
r
es
ea
r
ch
ar
ea
in
v
o
lv
es
h
u
m
an
a
ctiv
ity
r
ec
o
g
n
itio
n
(
HAR),
in
wh
ich
d
ee
p
lear
n
in
g
alg
o
r
ith
m
s
ar
e
in
cr
ea
s
in
g
ly
em
p
lo
y
ed
in
h
u
m
an
b
e
h
av
io
r
r
ec
o
g
n
itio
n
,
in
clu
d
in
g
d
etec
tio
n
o
f
a
b
n
o
r
m
al
b
eh
av
io
r
[
3
]
,
an
d
d
is
tr
ac
tio
n
r
ec
o
g
n
itio
n
[
4
]
.
Mo
s
t
r
e
ce
n
tly
,
th
e
tr
en
d
h
as
s
h
if
ted
to
war
d
s
an
o
m
alo
u
s
b
e
h
av
io
r
r
ec
o
g
n
itio
n
,
p
a
r
ticu
lar
l
y
in
s
ec
u
r
ity
a
n
d
s
u
r
v
eillan
ce
co
n
tex
ts
[
3
]
.
Desp
ite
th
e
ad
v
an
ce
m
e
n
ts
m
a
d
e,
th
er
e
ar
e
s
till
m
an
y
c
h
alle
n
g
es
an
d
o
b
s
tacle
s
th
at
r
em
ai
n
.
E
x
is
tin
g
s
u
r
v
ey
s
r
eg
u
lar
ly
f
o
cu
s
o
n
p
ar
ticu
lar
d
ee
p
lear
n
in
g
ar
ch
it
ec
tu
r
es
o
r
s
p
ec
if
ic
ap
p
licatio
n
s
,
s
u
ch
as
HA
R
,
with
o
u
t
o
f
f
e
r
in
g
a
b
r
o
a
d
o
v
er
v
iew
o
f
a
v
ailab
le
ap
p
r
o
ac
h
es,
p
latf
o
r
m
s
/to
o
ls
,
an
d
d
at
asets
f
o
r
b
eh
av
io
r
r
ec
o
g
n
itio
n
.
T
h
e
d
iv
er
s
ity
i
n
d
ata
an
d
b
eh
a
v
io
r
s
in
v
o
l
v
ed
f
u
r
th
er
lim
its
th
e
ab
ilit
y
o
f
d
ee
p
lear
n
in
g
ap
p
r
o
ac
h
es
to
g
en
er
alize
o
v
er
d
if
f
er
e
n
t
s
itu
atio
n
s
an
d
en
v
ir
o
n
m
en
ts
,
d
esp
ite
th
eir
im
p
r
ess
iv
e
p
er
f
o
r
m
an
ce
in
p
r
ed
ictin
g
a
n
d
r
ec
o
g
n
izin
g
h
u
m
an
b
eh
a
v
io
r
s
.
T
h
e
m
ain
aim
o
f
th
is
s
u
r
v
ey
i
s
to
p
r
esen
t
a
co
m
p
r
eh
e
n
s
iv
e
s
u
r
v
ey
o
f
d
ee
p
lear
n
in
g
-
b
ased
b
eh
av
i
o
u
r
r
ec
o
g
n
itio
n
ap
p
r
o
ac
h
es,
p
latf
o
r
m
s
/to
o
ls
,
an
d
d
atasets
.
T
h
is
s
u
r
v
ey
attem
p
ts
to
ass
is
t
r
es
ea
r
ch
er
s
,
esp
ec
ially
n
ew
r
esear
ch
er
s
in
t
h
e
f
ield
o
f
b
eh
av
io
r
r
ec
o
g
n
itio
n
b
y
ac
q
u
ain
tin
g
th
em
with
v
ar
i
o
u
s
d
ee
p
lear
n
in
g
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2502
-
4
7
5
2
Dee
p
lea
r
n
in
g
a
p
p
r
o
a
c
h
es,
p
l
a
tfo
r
ms,
d
a
ta
s
ets fo
r
b
eh
a
vio
r
-
b
a
s
ed
…
(
Yu
n
u
s
a
Mo
h
a
mme
d
Je
d
d
a
h
)
1881
ap
p
r
o
ac
h
es,
p
latf
o
r
m
s
,
a
n
d
d
a
taset
th
at
ca
n
p
o
ten
tially
b
e
a
d
o
p
ted
.
Ou
r
s
tu
d
y
is
in
ten
d
ed
as
a
co
m
p
r
eh
e
n
s
iv
e
s
u
r
v
ey
to
g
iv
e
an
o
v
er
v
iew
o
f
cu
r
r
en
t
ap
p
r
o
ac
h
es
an
d
tr
en
d
s
in
b
eh
av
io
r
r
ec
o
g
n
itio
n
u
s
in
g
d
ee
p
lear
n
in
g
.
I
t
d
o
es
n
o
t
f
o
llo
w
th
e
s
y
s
tem
atic
an
d
ex
h
au
s
tiv
e
m
eth
o
d
s
o
f
a
s
y
s
tem
atic
liter
atu
r
e
r
ev
iew
(
SLR)
o
r
a
s
co
p
in
g
r
ev
iew,
b
u
t
r
at
h
er
s
ee
k
s
to
b
r
o
ad
en
th
e
la
n
d
s
ca
p
e,
id
e
n
tify
co
m
m
o
n
t
h
em
es
an
d
h
ig
h
lig
h
t
k
ey
tech
n
iq
u
es.
I
n
co
n
tr
ast
to
p
r
ev
io
u
s
s
u
r
v
ey
s
wh
ich
ar
e
o
f
ten
lim
ited
in
s
co
p
e,
o
u
r
s
u
r
v
e
y
p
r
o
v
id
ed
a
b
r
o
ad
er
p
er
s
p
ec
tiv
e
o
n
th
e
u
s
es
o
f
d
ee
p
lear
n
i
n
g
s
ec
u
r
ity
an
d
s
u
r
v
eillan
ce
,
co
v
er
in
g
b
o
th
th
e
id
en
tific
atio
n
o
f
s
u
s
p
icio
u
s
o
r
ab
n
o
r
m
al
b
eh
av
io
r
s
a
n
d
th
e
r
ec
o
g
n
itio
n
o
f
n
o
r
m
al
h
u
m
an
b
eh
av
io
r
s
.
Dee
p
lear
n
in
g
o
v
er
co
m
es
th
e
lim
itatio
n
s
o
f
m
ac
h
in
e
lea
r
n
in
g
b
y
au
to
n
o
m
o
u
s
ly
ex
tr
ac
tin
g
th
e
f
ea
tu
r
es
f
r
o
m
r
aw
d
ata
an
d
ca
p
tu
r
in
g
c
o
m
p
lex
te
m
p
o
r
al
r
elatio
n
s
h
ip
s
in
d
y
n
a
m
ical
en
v
ir
o
n
m
en
ts
.
T
h
ese
ca
p
ab
ilit
ies
m
ak
e
d
ee
p
lear
n
in
g
esp
ec
ially
v
er
y
s
u
itab
le
f
o
r
h
u
m
an
b
eh
a
v
io
r
r
ec
o
g
n
itio
n
,
s
in
ce
it
m
ak
es
it
p
o
s
s
ib
le
to
ex
tr
ac
t
h
ig
h
-
le
v
el
r
ep
r
esen
tatio
n
s
f
r
o
m
m
o
tio
n
s
en
s
o
r
s
an
d
o
th
er
s
o
u
r
ce
d
ata.
T
h
is
r
esear
ch
f
ield
h
as
d
em
o
n
s
tr
ated
r
ap
id
p
r
o
g
r
ess
o
v
er
th
e
p
ast
d
ec
ad
e,
as
d
em
o
n
s
tr
ated
b
y
a
s
h
if
t
f
r
o
m
tr
ad
itio
n
al
m
ac
h
in
e
lear
n
in
g
alg
o
r
ith
m
s
to
d
ee
p
le
ar
n
in
g
a
p
p
r
o
ac
h
es f
r
o
m
2
0
0
6
t
o
2
0
1
8
[
5
]
.
T
h
is
tr
en
d
h
ig
h
lig
h
ts
h
o
w
r
esear
ch
er
s
ar
e
in
cr
ea
s
in
g
ly
ad
o
p
tin
g
a
n
d
ac
ce
p
tin
g
d
ee
p
lea
r
n
in
g
m
o
d
els.
Fig
u
r
e
1
illu
s
tr
ates
th
e
s
h
if
t
in
r
esear
c
h
er
s
’
f
o
cu
s
f
r
o
m
tr
a
d
itio
n
al
alg
o
r
ith
m
s
to
war
d
s
d
ee
p
lear
n
in
g
tech
n
iq
u
es.
At
th
e
b
eg
in
n
in
g
,
t
r
ad
itio
n
al
m
eth
o
d
s
wer
e
d
o
m
in
an
t,
p
ea
k
in
g
u
p
i
n
y
ea
r
2
0
1
3
.
Ho
wev
er
,
th
ei
r
p
r
ed
o
m
i
n
an
ce
d
r
o
p
p
ed
as
d
e
ep
lear
n
in
g
g
ain
ed
ac
ce
p
tan
ce
in
th
e
y
ea
r
2
0
1
4
,
g
o
in
g
b
e
y
o
n
d
tr
ad
itio
n
al
tech
n
iq
u
es
b
y
th
e
y
ea
r
2
0
0
5
an
d
c
o
n
tin
u
e
to
d
o
m
in
ate
o
n
war
d
s
.
Fig
u
r
e
1
.
T
r
ad
itio
n
al
m
ac
h
in
e
lear
n
in
g
v
s
d
ee
p
lear
n
i
n
g
a
lg
o
r
ith
m
s
u
s
ed
p
er
y
ea
r
[
5
]
T
h
e
co
n
tr
i
b
u
tio
n
s
o
f
o
u
r
s
u
r
v
e
y
in
clu
d
e
t
h
e
f
o
llo
win
g
:
i)
An
in
-
d
ep
t
h
r
ev
iew
o
f
d
ee
p
le
ar
n
in
g
a
p
p
r
o
ac
h
es a
n
d
p
latf
o
r
m
s
/to
o
ls
.
ii)
An
elab
o
r
ate
d
escr
ip
tio
n
o
f
th
e
m
o
s
t p
o
p
u
lar
d
atasets
r
elate
d
to
b
e
h
av
io
r
r
ec
o
g
n
itio
n
.
iii)
Dis
cu
s
s
io
n
o
n
d
ee
p
lear
n
i
n
g
m
o
d
els an
d
th
ei
r
s
p
ec
if
ic
u
s
e
ca
s
es.
iv
)
A
co
n
cise
an
d
clea
r
g
u
i
d
e
th
at
ca
n
s
er
v
e
as
a
s
tar
tin
g
p
o
in
t
f
o
r
n
ew
r
esear
ch
e
r
s
in
th
e
f
ield
o
f
d
ee
p
lear
n
in
g
f
o
r
b
e
h
av
io
r
al
an
aly
s
is
.
T
o
ac
h
iev
e
th
is
,
we
th
o
r
o
u
g
h
ly
co
n
d
u
cted
a
g
en
e
r
al
liter
atu
r
e
s
ea
r
ch
to
co
m
e
u
p
with
a
co
m
p
r
eh
e
n
s
iv
e
s
u
r
v
ey
o
f
th
e
c
u
r
r
en
t la
n
d
s
ca
p
e
in
b
eh
av
i
o
r
r
ec
o
g
n
itio
n
u
s
in
g
d
ee
p
lear
n
in
g
.
W
e
u
s
ed
tar
g
eted
k
ey
ter
m
s
s
ea
r
ch
ter
m
s
lin
k
ed
to
b
eh
a
v
io
r
r
ec
o
g
n
itio
n
,
d
ee
p
lear
n
in
g
ap
p
r
o
ac
h
es,
p
l
atf
o
r
m
s
/to
o
ls
,
an
d
d
atasets
ac
r
o
s
s
s
ev
er
al
ac
ad
e
m
ic
d
atab
ases
to
h
ar
v
est
a
w
id
e
r
an
g
e
o
f
r
elate
d
s
tu
d
ies
an
d
r
eso
u
r
ce
s
.
T
h
e
d
o
wn
lo
ad
e
d
p
ap
er
s
wer
e
b
r
o
ad
ly
ch
ec
k
ed
a
n
d
ca
teg
o
r
ize
d
b
y
co
m
m
o
n
s
u
b
jects,
lik
e
ty
p
e
o
f
alg
o
r
ith
m
,
ap
p
licatio
n
ar
ea
s
,
an
d
r
elev
a
n
t
d
ataset.
T
h
e
m
ain
f
o
c
u
s
o
f
th
is
s
u
r
v
ey
is
to
p
r
o
v
id
e
an
ad
v
an
ce
d
lev
el
ass
es
s
m
en
t
an
d
d
is
cu
s
s
io
n
o
f
th
e
v
ar
io
u
s
ap
p
r
o
ac
h
es,
em
p
h
a
s
izin
g
im
p
o
r
tan
t
u
s
e
ca
s
es
an
d
tr
en
d
s
as
o
p
p
o
s
ed
to
s
tick
in
g
t
o
r
i
g
id
s
tan
d
a
r
d
s
o
r
co
m
p
r
e
h
en
s
iv
e
in
cl
u
s
io
n
s
t
an
d
ar
d
s
.
Me
n
d
eley
was
em
p
l
o
y
ed
to
e
f
f
icien
tly
m
an
ag
e
r
ef
er
e
n
ce
s
f
r
o
m
th
e
i
d
en
tifie
d
s
o
u
r
ce
s
an
d
p
a
p
er
s
.
T
h
e
o
r
g
a
n
izatio
n
o
f
th
e
p
ap
er
is
as
f
o
llo
ws
:
in
s
ec
tio
n
2
,
r
elate
d
wo
r
k
is
p
r
esen
ted
.
in
s
ec
t
io
n
3
,
h
u
m
a
n
b
e
h
av
io
r
r
ec
o
g
n
itio
n
u
s
in
g
d
ee
p
lear
n
in
g
tech
n
iq
u
es
is
d
is
cu
s
s
ed
;
in
s
ec
tio
n
4
,
s
ev
er
al
d
ee
p
lear
n
i
n
g
p
latf
o
r
m
s
a
n
d
to
o
ls
ar
e
co
v
er
ed
;
s
ec
tio
n
5
p
r
esen
ted
d
atasets
f
o
r
b
e
h
av
io
r
r
ec
o
g
n
itio
n
;
R
esu
lts
an
d
d
is
cu
s
s
io
n
is
p
r
esen
ted
in
s
ec
tio
n
6
,
a
n
d
in
s
ec
ti
o
n
7
,
t
h
e
p
ap
e
r
is
co
n
clu
d
e
d
.
2.
RE
L
AT
E
D
WO
RK
R
ec
en
t
s
u
r
v
ey
s
h
a
v
e
e
x
p
lo
r
e
d
d
ee
p
lear
n
in
g
a
p
p
r
o
ac
h
es
f
o
r
an
o
m
al
y
d
etec
tio
n
a
n
d
h
u
m
a
n
b
e
h
av
io
r
r
ec
o
g
n
itio
n
,
esp
ec
ially
in
v
id
e
o
s
u
r
v
eillan
ce
co
n
tex
ts
.
Him
e
u
r
et
a
l.
[
6
]
p
r
esen
ted
a
th
o
r
o
u
g
h
r
ev
iew
f
o
c
u
s
in
g
o
n
em
p
lo
y
in
g
d
o
m
ain
a
d
ap
tat
io
n
an
d
d
ee
p
tr
a
n
s
f
er
lear
n
in
g
m
eth
o
d
s
to
m
a
k
e
im
p
r
o
v
em
e
n
t
in
g
en
er
aliza
tio
n
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
5
2
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
,
Vo
l.
38
,
No
.
3
,
J
u
n
e
20
25
:
1
8
8
0
-
1
8
9
5
1882
in
v
id
eo
s
u
r
v
eillan
ce
s
y
s
tem
s
.
T
h
eir
wo
r
k
is
h
o
wev
e
r
,
m
o
s
tly
lim
ited
to
s
u
r
v
eillan
ce
a
p
p
licatio
n
s
an
d
d
o
es
n
o
t
ex
p
l
o
r
e
o
t
h
er
co
n
tex
ts
.
Similar
ly
,
W
astu
p
r
an
ata
et
a
l.
[
7
]
f
o
c
u
s
es
m
ain
ly
o
n
s
u
r
v
eill
an
ce
ap
p
licatio
n
s
,
s
tr
ess
in
g
o
n
m
eth
o
d
s
f
o
r
d
ete
ctin
g
ab
n
o
r
m
al
b
eh
av
io
r
s
b
u
t
n
o
t
o
f
f
er
in
g
a
wid
e
r
o
v
er
v
ie
w
o
f
d
ee
p
lear
n
i
n
g
m
o
d
els f
o
r
d
iv
er
s
e
b
e
h
av
io
r
r
e
co
g
n
itio
n
a
p
p
licatio
n
s
.
A
n
u
m
b
er
o
f
o
th
er
s
tu
d
ies
co
n
ce
n
tr
ate
o
n
s
u
r
v
eillan
ce
v
id
eo
s
an
o
m
aly
d
etec
tio
n
u
s
in
g
d
ee
p
lear
n
in
g
.
C
h
o
u
d
h
r
y
et
a
l.
[
8
]
an
d
J
eb
u
r
et
a
l.
[
9
]
lar
g
ely
h
ig
h
lig
h
t
an
o
m
aly
d
etec
tio
n
,
m
o
s
tly
n
eg
lectin
g
ad
d
r
ess
in
g
th
e
d
iv
er
s
ity
o
f
d
atasets
,
p
latf
o
r
m
s
,
an
d
u
s
e
ca
s
es
th
at
ar
e
in
v
o
lv
ed
in
g
en
er
al
b
eh
av
io
r
r
ec
o
g
n
itio
n
.
A
s
u
r
v
ey
b
y
[
1
0
]
f
u
r
th
er
s
tr
en
g
th
en
s
th
is
tr
e
n
d
b
y
co
n
ce
n
tr
atin
g
n
ar
r
o
wly
o
n
ab
n
o
r
m
al
in
s
tan
ce
s
d
etec
tio
n
,
f
ailin
g
to
ex
p
a
n
d
t
o
o
th
er
d
o
m
ain
s
.
I
n
a
m
o
r
e
f
o
cu
s
ed
co
n
tex
t,
Sh
u
b
b
e
r
an
d
Al
-
T
a’
i
[
1
1
]
an
d
Ph
am
et
a
l.
[
1
2
]
f
o
cu
s
o
n
ce
r
tain
b
eh
a
v
io
r
r
ec
o
g
n
itio
n
u
n
d
er
tak
in
g
s
s
u
ch
as
h
u
m
an
ac
ti
o
n
r
ec
o
g
n
itio
n
an
d
v
io
len
ce
d
etec
tio
n
.
Desp
ite
p
r
o
v
id
in
g
v
alu
a
b
le
in
s
ig
h
ts
,
th
es
e
s
tu
d
ies
ar
e
r
estricte
d
to
ce
r
ta
in
ac
tiv
ities
an
d
d
o
n
o
t
c
o
n
s
id
er
t
h
e
f
u
ll
r
an
g
e
o
f
d
ee
p
lear
n
in
g
ap
p
r
o
ac
h
es
an
d
p
latf
o
r
m
s
.
I
n
ad
d
itio
n
,
[
1
3
]
f
o
cu
s
es
o
n
h
ea
lth
ca
r
e
task
s
,
p
r
im
ar
ily
an
aly
zin
g
an
o
m
aly
d
etec
tio
n
in
d
aily
ac
ti
o
n
s
b
u
t
lack
in
co
v
er
i
n
g
o
f
b
r
o
ad
er
co
n
tex
ts
lik
e
p
u
b
lic
s
ec
u
r
ity
.
C
o
n
tr
ar
y
to
th
ese
cu
r
r
en
t
s
t
u
d
ies,
o
u
r
o
wn
p
a
p
er
attem
p
ts
ad
d
r
ess
in
g
th
e
g
ap
s
i
d
e
n
tifie
d
b
y
p
r
o
v
id
i
n
g
a
b
r
o
ad
o
v
er
v
iew
o
f
d
ee
p
lear
n
in
g
a
p
p
r
o
ac
h
es,
p
latf
o
r
m
s
/to
o
ls
,
an
d
d
atasets
u
s
ed
f
o
r
b
eh
av
io
r
r
ec
o
g
n
itio
n
.
Un
lik
e
p
r
ev
io
u
s
r
ev
iews,
wh
ich
s
co
p
e
is
o
f
te
n
l
im
ited
to
s
p
ec
if
ic
ap
p
licatio
n
s
,
o
u
r
s
tu
d
y
o
f
f
er
s
a
wid
er
p
er
s
p
ec
tiv
e
o
n
th
e
ap
p
licatio
n
o
f
d
ee
p
lea
r
n
in
g
in
b
eh
a
v
io
r
r
ec
o
g
n
itio
n
,
co
m
p
r
is
in
g
s
ev
er
al
d
ee
p
lear
n
in
g
alg
o
r
ith
m
s
,
v
ar
io
u
s
p
latf
o
r
m
s
,
an
d
a
wid
e
r
an
g
e
o
f
d
atasets
.
T
h
is
a
ll
-
in
clu
s
iv
e
ap
p
r
o
ac
h
will
h
elp
in
g
u
id
in
g
r
esear
c
h
er
s
in
ch
o
o
s
in
g
th
e
r
ig
h
t m
eth
o
d
a
n
d
to
o
l f
o
r
d
if
f
er
en
t r
esear
ch
cir
c
u
m
s
tan
ce
s
an
d
r
ec
o
g
n
izes
f
u
tu
r
e
r
esear
c
h
g
u
i
d
an
ce
n
o
t b
r
o
ad
ly
c
o
v
er
e
d
in
p
r
io
r
s
tu
d
ies
.
3.
DE
E
P
L
E
A
RNING
AP
P
RO
ACH
E
S
T
O
H
UM
AN
B
E
H
AVIO
UR
R
E
CO
G
NI
T
I
O
N
Ma
n
y
r
esear
c
h
s
u
r
v
ey
s
h
a
v
e
b
ee
n
co
n
d
u
cted
o
n
h
u
m
an
ac
tiv
ity
/
b
eh
av
i
o
r
r
ec
o
g
n
itio
n
th
r
o
u
g
h
d
if
f
er
en
t
f
ac
ets,
wh
ile
o
t
h
er
r
esear
ch
er
s
f
o
c
u
s
ed
o
n
g
e
n
er
al
ass
ess
m
en
t
s
o
f
h
u
m
an
ac
tiv
ity
/
b
eh
av
io
r
.
T
h
e
v
ar
io
u
s
asp
ec
ts
s
tu
d
ied
b
y
th
e
r
esear
ch
er
s
in
clu
d
e
th
e
f
o
llo
win
g
:
th
e
alg
o
r
ith
m
ty
p
es,
th
e
m
eth
o
d
s
u
s
ed
,
th
e
ty
p
e
o
f
s
en
s
o
r
s
u
s
ed
,
th
e
d
ev
ice
ty
p
e,
an
d
th
e
ty
p
e
o
f
ac
tiv
i
ty
p
er
f
o
r
m
e
d
.
Her
e,
we
f
o
cu
s
o
n
a
s
u
r
v
ey
o
n
th
e
ap
p
r
o
ac
h
es
to
h
u
m
an
b
eh
av
i
o
r
r
ec
o
g
n
itio
n
.
On
a
g
e
n
er
al
n
o
te,
th
er
e
is
an
in
cr
ea
s
e
in
r
esear
ch
er
s
th
at
tr
y
to
ap
p
ly
d
ee
p
lear
n
in
g
ap
p
r
o
ac
h
es
in
s
tu
d
y
in
g
h
u
m
an
b
eh
av
i
o
r
.
Hu
m
a
n
b
eh
a
v
io
r
is
ex
tr
ao
r
d
in
ar
ily
co
m
p
lex
;
h
u
m
an
b
e
h
av
io
r
is
tr
ig
g
er
ed
b
y
h
ab
its
o
r
i
n
t
e
n
t
i
o
n
s
;
t
r
a
n
s
f
o
r
m
e
d
b
y
e
f
f
e
c
t
,
s
k
i
l
l
,
a
n
d
a
t
t
i
t
u
d
e
;
a
n
d
a
f
f
e
c
t
e
d
b
y
c
o
n
t
e
x
t
u
a
l
a
n
d
p
h
y
s
i
c
a
l
c
o
n
d
i
t
i
o
n
s
[
1
4
]
.
Ag
u
ileta
et
a
l.
[
1
5
]
e
x
p
lo
r
ed
a
r
a
n
g
e
o
f
tech
n
iq
u
es
an
d
ap
p
r
o
ac
h
es
p
u
t
f
o
r
th
b
y
r
esear
ch
e
r
s
to
m
er
g
e
d
ata
f
r
o
m
d
if
f
e
r
en
t
s
en
s
o
r
s
to
d
is
co
v
er
r
esear
ch
p
r
o
s
p
ec
ts
in
th
is
ar
ea
.
Desp
ite
b
ein
g
in
-
d
ep
th
a
n
d
e
x
ten
s
iv
e,
th
e
s
u
r
v
ey
n
ee
d
ed
to
p
r
o
v
id
e
d
etails ab
o
u
t th
e
im
p
lem
e
n
tatio
n
.
I
n
th
eir
s
u
r
v
ey
,
[
1
6
]
b
ased
th
e
ir
HAR
r
ev
iew
o
n
tr
ad
itio
n
al
m
ac
h
in
e
lear
n
in
g
tech
n
iq
u
es.
W
h
ile
[
1
7
]
f
o
cu
s
ed
o
n
a
d
ev
ice
-
f
r
ee
m
eth
o
d
b
ased
o
n
r
ad
io
-
f
r
e
q
u
en
c
y
id
en
tific
atio
n
(
R
FID
)
tech
n
o
lo
g
y
.
T
h
e
d
ev
ice
-
f
r
ee
m
eth
o
d
d
o
es
n
o
t
r
eq
u
ir
e
s
u
b
j
ec
ts
to
ca
r
r
y
o
r
wea
r
a
d
e
v
ic
e
f
o
r
th
e
s
en
s
o
r
s
to
r
ec
o
g
n
ize
ac
tiv
ities
.
I
n
s
tead
,
s
en
s
o
r
s
lik
e
R
FID
an
d
ca
m
er
as tag
b
o
th
th
e
o
b
jects a
n
d
th
e
en
v
ir
o
n
m
en
t.
T
h
is
ap
p
r
o
ac
h
h
as a
d
v
an
tag
es b
u
t is
co
m
p
licated
in
r
ea
l
-
life
im
p
lem
en
tatio
n
an
d
h
as
s
ig
n
if
ican
t
p
r
iv
ac
y
-
r
elate
d
is
s
u
es.
S
h
o
aib
et
a
l.
[
1
8
]
ca
teg
o
r
ized
ac
tiv
ities
in
to
tw
o
,
co
m
p
lex
ac
tiv
ities
an
d
s
i
m
p
le
ac
tiv
ities
.
Simp
le
o
n
es
ar
e
ea
s
y
to
id
e
n
tify
u
s
in
g
a
s
en
s
o
r
b
ec
au
s
e
th
ey
ar
e
n
atu
r
ally
r
ep
etitiv
e,
lik
e
s
tan
d
in
g
,
s
itti
n
g
,
r
u
n
n
in
g
,
s
m
o
k
in
g
,
ea
tin
g
,
a
n
d
m
ak
in
g
co
f
f
ee
.
W
h
ile
o
n
t
h
e
co
n
tr
ar
y
,
co
m
p
lex
ac
tiv
ities
a
r
e
d
if
f
icu
lt
to
r
ec
o
g
n
ize,
as
th
ey
ar
e
n
o
t
r
ec
u
r
r
en
t
an
d
r
eq
u
ir
e
m
o
r
e
e
f
f
o
r
t
a
n
d
d
ata.
So
m
e
d
atasets
ar
e
in
ten
d
ed
f
o
r
c
o
m
p
lex
ac
tiv
ities
,
lik
e
[
1
9
]
;
o
th
er
s
ar
e
d
esig
n
ed
o
n
ly
f
o
r
s
im
p
le
ac
tiv
ities
,
lik
e
[
2
0
]
,
wh
ile
o
t
h
er
s
ar
e
m
ad
e
f
o
r
b
o
th
ac
tiv
ity
ty
p
es,
s
u
ch
as [
2
1
]
.
B
eh
av
io
r
r
ec
o
g
n
itio
n
is
a
v
ita
l
s
u
b
-
ar
ea
o
f
HAR.
T
h
e
m
ain
id
ea
is
to
id
e
n
tify
a
p
er
s
o
n
'
s
b
eh
av
i
o
r
f
r
o
m
th
e
d
ata
g
o
tten
v
ia
d
if
f
e
r
en
t
s
en
s
o
r
s
.
B
eh
av
io
r
d
etec
tio
n
is
in
s
tr
u
m
en
tal
in
s
ev
er
al
cir
cu
m
s
tan
ce
s
,
lik
e
in
tellig
en
t su
r
r
o
u
n
d
in
g
s
(
ag
e
d
ca
r
e
ce
n
ter
s
an
d
s
m
ar
t h
o
m
es)
[
1
9
]
an
d
s
h
o
p
p
in
g
ce
n
te
r
s
.
Fo
r
ex
am
p
le,
in
a
g
ed
ca
r
e
ce
n
ter
s
,
p
atien
ts
ca
n
b
e
r
em
o
tely
m
o
n
ito
r
e
d
,
s
ig
n
if
ican
tly
r
ed
u
cin
g
th
e
co
s
t
in
v
o
l
v
ed
.
B
y
ex
ten
s
io
n
,
au
to
m
atic
b
eh
a
v
io
r
r
ec
o
g
n
itio
n
ca
n
b
e
ac
h
iev
ed
th
r
o
u
g
h
t
h
e
u
s
e
o
f
d
ee
p
lear
n
in
g
an
d
m
ac
h
in
e
lear
n
in
g
.
T
h
er
e
ar
e
s
ev
er
al
m
ac
h
in
e
lear
n
in
g
-
b
ased
m
o
d
els
lik
e
k
-
n
ea
r
est
n
eig
h
b
o
r
(
KNN)
,
s
u
itab
le
f
o
r
class
if
icatio
n
p
r
o
b
lem
s
[
2
2
]
,
s
u
p
p
o
r
t
v
ec
to
r
m
ac
h
in
e
(
SVM)
,
f
o
r
h
an
d
wr
itin
g
r
ec
o
g
n
iz
e,
id
en
tify
a
s
p
ea
k
er
,
r
ec
o
g
n
ize
f
r
a
u
d
u
len
t
cr
ed
it
c
ar
d
s
,
as
well
as
d
etec
t
f
ac
e
[
2
3
]
,
d
ec
is
io
n
tr
ee
(
DT
)
k
n
o
wn
f
o
r
e
n
h
an
ce
d
p
er
f
o
r
m
an
ce
o
u
tco
m
e’
s
v
iew
[
2
4
]
an
d
m
an
y
o
th
er
s
h
a
v
e
b
ee
n
u
s
ed
f
o
r
HAR.
Ho
wev
e
r
,
u
s
in
g
tr
a
d
itio
n
al
alg
o
r
ith
m
s
h
as
d
ec
r
ea
s
ed
g
r
e
atly
af
ter
th
e
r
em
ar
k
a
b
le
p
er
f
o
r
m
an
ce
o
f
d
ee
p
lear
n
in
g
-
b
ased
alg
o
r
ith
m
s
in
r
ec
o
g
n
izin
g
h
u
m
an
ac
tio
n
.
3
.
1
.
Dee
p
lea
rning
m
e
t
ho
ds
Sev
er
al
d
ee
p
m
o
d
els,
am
o
n
g
wh
ich
ar
e
co
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
(
C
NN
)
,
d
ee
p
n
e
u
r
al
n
etwo
r
k
(
DNN)
,
r
ec
u
r
r
en
t
n
eu
r
al
n
et
wo
r
k
(
R
NN
)
,
r
estricte
d
B
o
ltz
m
an
n
m
ac
h
in
e
(
R
B
M)
,
lo
n
g
-
s
h
o
r
t
ter
m
m
em
o
r
y
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2502
-
4
7
5
2
Dee
p
lea
r
n
in
g
a
p
p
r
o
a
c
h
es,
p
l
a
tfo
r
ms,
d
a
ta
s
ets fo
r
b
eh
a
vio
r
-
b
a
s
ed
…
(
Yu
n
u
s
a
Mo
h
a
mme
d
Je
d
d
a
h
)
1883
(
L
STM
)
,
an
d
m
an
y
h
y
b
r
id
s
d
ee
p
m
o
d
els
(
co
m
b
in
i
n
g
m
o
r
e
th
an
o
n
e
d
ee
p
s
tr
u
ctu
r
e)
,
h
a
v
e
b
ee
n
th
e
s
u
b
ject
o
f
r
esear
ch
s
tu
d
ies
[
2
5
]
-
[
2
7
]
.
C
h
o
ice
o
f
th
e
b
est
s
u
itab
le
d
ee
p
lear
n
in
g
o
r
m
ac
h
i
n
e
lear
n
in
g
m
o
d
el
r
elies
o
n
t
h
e
co
m
p
lex
ity
an
d
th
e
p
r
o
b
lem
t
y
p
e.
T
h
e
s
ec
tio
n
lo
o
k
s
at
th
e
m
o
s
t
em
p
lo
y
e
d
ty
p
es
o
f
d
ee
p
lear
n
in
g
al
g
o
r
ith
m
s
d
escr
ib
ed
in
th
e
f
o
llo
win
g
s
u
b
s
ec
tio
n
.
3
.
1
.
1
.
Dee
p
neura
l net
wo
rk
s
DNN,
as
d
ep
icted
in
F
ig
u
r
e
2
is
an
ad
v
an
ce
d
ty
p
e
o
f
a
r
tific
ial
n
eu
r
al
n
etwo
r
k
(
ANN)
[
2
8
]
.
T
h
ey
ar
e
ter
m
ed
d
ee
p
f
o
r
h
av
in
g
d
ee
p
ar
ch
itectu
r
e
b
y
p
o
s
s
ess
in
g
lo
ts
o
f
h
id
d
en
lay
er
s
s
an
d
wich
e
d
b
etwe
en
t
h
e
in
p
u
t
an
d
th
e
o
u
tp
u
t
la
y
er
s
[
2
9
]
.
ANN
h
av
in
g
s
h
allo
w
n
etwo
r
k
s
h
as
less
h
id
d
en
lay
er
s
a
n
d
is
id
en
tifie
d
as
m
u
ltil
ay
er
p
er
ce
p
tr
o
n
(
ML
P).
R
esear
ch
er
s
u
s
e
DNN
f
o
r
h
u
m
an
ac
tiv
ity
r
ec
o
g
n
itio
n
a
n
d
h
an
d
en
g
in
ee
r
in
g
f
ea
tu
r
es
ex
tr
ac
ted
f
r
o
m
s
en
s
o
r
s
.
Similar
ly
,
[
3
0
]
u
tili
ze
d
p
r
in
cip
al
co
m
p
o
n
en
t
a
n
aly
s
is
(
PC
A
)
f
o
r
f
ea
tu
r
e
ex
tr
ac
tio
n
a
n
d
DNN
to
lear
n
th
e
ac
tiv
ities
.
Pre
v
i
o
u
s
r
e
s
ea
r
ch
h
as
d
e
m
o
n
s
tr
ated
t
h
a
t
ac
tiv
ities
ca
n
b
e
id
en
tifie
d
f
r
o
m
r
aw
s
en
s
o
r
d
ata
wh
en
a
s
tr
u
ct
u
r
e
is
d
ee
p
en
o
u
g
h
an
d
h
as
en
o
u
g
h
d
ata,
n
e
g
atin
g
th
e
n
ee
d
f
o
r
cu
s
to
m
ch
ar
ac
ter
is
tics
[
3
1
]
.
A
p
o
licy
g
r
ad
ien
t
-
b
ased
r
ein
f
o
r
ce
m
en
t
lear
n
in
g
(
R
L
)
ap
p
r
o
ac
h
was
u
tili
ze
d
to
ac
co
m
p
lis
h
o
n
lin
e
tr
ain
in
g
f
o
r
a
DNN
-
b
ased
o
n
lin
e
ac
tiv
ity
id
en
tific
atio
n
s
y
s
tem
th
at
was
p
r
o
p
o
s
ed
b
y
[
3
2
]
.
Fo
r
th
e
R
L
m
ac
h
in
e
lear
n
in
g
p
ar
ad
ig
m
,
lear
n
in
g
is
ac
co
m
p
l
is
h
ed
b
y
r
ewa
r
d
s
an
d
p
u
n
is
h
m
en
ts
in
a
s
to
ch
asti
c
en
v
ir
o
n
m
en
t.
R
L
h
as a
ls
o
em
p
lo
y
ed
d
ee
p
lear
n
in
g
to
esti
m
at
e
th
e
p
o
licy
a
n
d
r
ewa
r
d
f
u
n
cti
o
n
[
3
3
]
.
Fig
u
r
e
2
.
Descr
ip
tio
n
o
f
DNN
m
o
d
el
[
3
4
]
3
.
1
.
2
.
Aut
o
-
e
nco
der
As
an
u
n
s
u
p
er
v
is
ed
lear
n
in
g
-
b
ased
,
au
to
-
en
co
d
er
(
AE
)
[
3
5
]
,
[
3
6
]
is
a
n
eu
r
al
n
etwo
r
k
[
3
7
]
,
[
3
8
]
th
at
u
s
es
th
e
b
ac
k
p
r
o
p
a
g
atio
n
al
g
o
r
ith
m
.
T
o
d
is
tin
g
u
is
h
b
etwe
en
n
o
r
m
al
an
d
ab
n
o
r
m
al
n
et
wo
r
k
ev
e
n
ts
,
[
3
9
]
p
r
o
p
o
s
ed
th
e
u
s
e
o
f
AE
an
d
s
tatis
t
ically
an
aly
s
is
-
d
r
iv
en
in
tellig
en
t
in
tr
u
s
io
n
d
etec
tio
n
s
y
s
tem
(
I
DS
)
.
T
h
r
ee
lay
er
s
m
ak
e
u
p
AE
'
s
ar
ch
itectu
r
e:
an
i
n
p
u
t,
a
h
id
d
en
(
en
c
o
d
in
g
)
lay
er
,
an
d
a
d
ec
o
d
in
g
lay
er
[
4
0
]
.
I
t
was
ap
p
lied
in
m
an
y
f
ield
s
,
lik
e
i
m
ag
e
class
if
icatio
n
,
n
atu
r
al
lan
g
u
ag
e
p
r
o
ce
s
s
in
g
,
f
ac
e
r
ec
o
g
n
itio
n
,
a
n
d
o
th
er
ar
ea
s
[
4
1
]
,
ac
h
iev
in
g
im
p
r
ess
iv
e
r
esu
lts
.
I
t
was
em
p
lo
y
ed
b
y
[
3
5
]
to
p
r
o
p
o
s
e
a
p
lu
g
-
a
n
d
-
p
lay
,
“Kitsu
n
e”
,
th
at
ca
n
id
en
tify
p
o
ten
tial
n
etwo
r
k
attac
k
s
with
o
u
t
s
u
p
er
v
is
io
n
.
Fig
u
r
e
3
s
h
o
ws
th
e
AE
’
s
s
tr
u
ctu
r
e,
s
h
o
win
g
th
e
p
r
o
ce
s
s
es o
f
en
co
d
in
g
a
n
d
d
ec
o
d
in
g
.
Fig
u
r
e
3
.
Stru
ctu
r
e
o
f
an
AE
[
4
2
]
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
5
2
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
,
Vo
l.
38
,
No
.
3
,
J
u
n
e
20
25
:
1
8
8
0
-
1
8
9
5
1884
3
.
1
.
3
C
o
nv
o
lutio
n
neura
l net
wo
rk
s
C
NN
m
o
d
els
ar
e
m
o
s
tly
e
m
p
lo
y
ed
f
o
r
th
eir
a
b
ilit
y
in
f
ea
tu
r
e
lear
n
i
n
g
f
r
o
m
n
ew
s
en
s
o
r
d
ata
[
4
3
]
.
W
h
en
th
e
in
p
u
t
d
ata
h
as
a
cl
ea
n
s
p
atial
s
tr
u
ctu
r
e,
C
NN
is
a
g
o
o
d
c
h
o
ice
(
e.
g
.
,
a
co
llec
tio
n
o
f
p
ix
els
in
a
p
h
o
to
)
.
I
n
ad
d
itio
n
,
C
NN
h
as
b
ee
n
u
s
ed
to
r
ec
o
g
n
ize
h
u
m
an
ac
tiv
ity
in
s
ev
e
r
al
s
tu
d
y
i
n
v
esti
g
atio
n
s
[
4
4
]
−
[
4
6
]
.
R
u
ed
a
et
a
l.
[
4
4
]
ass
er
ted
th
at
n
u
m
e
r
o
u
s
p
ac
k
ed
lay
er
s
o
f
co
n
v
o
l
u
tio
n
f
i
lter
s
an
d
p
o
o
lin
g
p
r
o
ce
s
s
es
ca
n
b
e
u
s
ed
to
lear
n
n
o
n
lin
ea
r
an
d
tem
p
o
r
al
s
tr
u
ctu
r
es
in
b
o
th
s
im
p
le
a
n
d
co
m
p
lex
h
u
m
an
m
o
v
em
en
ts
.
I
g
n
ato
v
[
4
5
]
s
tates
,
C
NN
was
u
ti
lized
f
o
r
in
tr
o
d
u
cin
g
a
u
s
er
-
in
d
e
p
en
d
e
n
t
an
d
o
n
lin
e
ap
p
r
o
ac
h
to
r
ec
o
g
n
izin
g
h
u
m
an
ac
tiv
ities
.
Alo
n
g
s
id
e
f
ea
tu
r
e
ex
tr
ac
tio
n
,
th
e
win
d
o
w
s
ize
wa
s
al
s
o
in
v
esti
g
ated
,
wh
er
e
it
was d
is
co
v
er
ed
th
at
a
lar
g
er
w
in
d
o
w
s
ize
o
n
ly
im
p
r
o
v
es p
er
f
o
r
m
an
ce
in
s
o
m
e
ac
tiv
ity
class
if
icatio
n
ca
s
es.
3
.
1
.
4
Rest
rict
ed
B
o
lt
zma
nn
m
a
chines
R
B
M
(
w
ith
its
b
a
s
ic
s
tr
u
ctu
r
e
s
h
o
wn
in
Fig
u
r
e
4
)
,
p
r
o
p
o
s
ed
in
1
9
8
6
b
y
Pau
l
Sm
o
len
s
k
y
[
4
7
]
,
tr
ain
e
d
to
r
ec
r
ea
te
its
in
p
u
t
f
r
o
m
th
e
f
ir
s
t
h
id
d
en
lay
er
i
n
u
n
s
u
p
er
v
is
ed
p
r
e
-
tr
ai
n
in
g
.
R
B
M
is
u
tili
z
ed
in
co
llab
o
r
ativ
e
f
ilter
in
g
[
4
8
]
,
d
im
e
n
s
io
n
ality
r
ed
u
ctio
n
[
4
9
]
,
class
if
icatio
n
[
5
0
]
,
r
eg
r
ess
io
n
[
4
9
]
,
f
ea
tu
r
e
lear
n
in
g
[
5
1
]
,
a
n
d
to
p
ic
m
o
d
ellin
g
[
4
7
]
.
I
n
th
eir
r
esear
ch
,
[
3
7
]
em
p
lo
y
ed
R
B
M
f
o
r
p
r
e
-
tr
ain
in
g
lay
er
b
y
l
ay
er
to
o
b
tain
th
e
in
itial we
ig
h
t a
n
d
o
f
f
s
et
in
th
e
ir
im
ag
e
r
ec
o
n
s
tr
u
ctio
n
.
Fig
u
r
e
4
.
T
h
e
b
asic stru
ctu
r
e
o
f
R
B
M
[
5
2
]
3
.
1
.
5
.
M
ultila
y
er
p
er
ce
ptr
o
n
ML
P
is
a
p
er
ce
p
tr
o
n
wh
er
e
g
r
o
u
p
s
with
m
an
y
p
e
r
ce
p
tio
n
s
a
r
e
d
ep
lo
y
e
d
lay
er
s
wis
e
to
f
ix
co
m
p
lex
is
s
u
es
[
5
3
]
.
ML
Ps
ar
e
th
r
ee
-
la
y
er
ed
[
5
4
]
,
wh
er
ein
th
e
f
ir
s
t
l
ay
er
is
a
n
in
p
u
t
lay
er
,
th
at
tr
a
n
s
m
its
s
ig
n
al
to
t
h
e
h
id
d
en
(
s
ec
o
n
d
)
la
y
er
,
wh
ich
also
tr
an
s
m
its
th
e
s
ig
n
al
t
o
t
h
e
o
u
tp
u
t
lay
er
.
T
h
e
p
r
im
a
r
y
u
s
e
o
f
ML
P
is
th
e
p
r
ed
ictio
n
o
f
lab
els
o
r
class
es
-
b
ased
class
if
icatio
n
s
.
A
s
im
p
le
g
en
e
r
al
id
ea
o
f
th
e
s
tr
u
ctu
r
e
o
f
ML
P
is
illu
s
tr
ated
in
Fig
u
r
e
5.
Fig
u
r
e
5
.
A
s
im
p
le
o
v
er
v
iew
o
f
ML
P
[
5
5
]
3
.
1
.
6
.
Rec
urre
nt
neura
l net
wo
rk
s
Un
lik
e
co
n
v
en
tio
n
al
alg
o
r
ith
m
s
,
R
NN
d
o
es
n
o
t
p
r
esu
m
e
th
at
th
e
d
ata
s
eq
u
en
ce
s
a
r
e
d
if
f
er
en
t
f
r
o
m
ea
ch
o
th
er
as
th
e
p
r
io
r
s
eq
u
e
n
ce
s
’
in
f
o
r
m
atio
n
is
u
tili
ze
d
to
lear
n
th
e
cu
r
r
en
t
s
eq
u
e
n
ce
[
5
6
]
,
as
it
co
u
ld
b
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2502
-
4
7
5
2
Dee
p
lea
r
n
in
g
a
p
p
r
o
a
c
h
es,
p
l
a
tfo
r
ms,
d
a
ta
s
ets fo
r
b
eh
a
vio
r
-
b
a
s
ed
…
(
Yu
n
u
s
a
Mo
h
a
mme
d
Je
d
d
a
h
)
1885
s
ee
n
in
Fig
u
r
e
6
.
T
h
er
e
f
o
r
e,
R
NN
is
f
r
eq
u
en
tly
u
tili
ze
d
to
lear
n
a
tim
e
s
er
ies’
tem
p
o
r
al
s
tr
u
ctu
r
es
an
d
d
y
n
am
ics.
G
ated
r
ec
u
r
r
en
t
u
n
it
(
GR
Us)
an
d
L
STM
ar
e
two
m
o
s
t
co
m
m
o
n
v
ar
ian
ts
o
f
R
NN
wh
er
e
L
STM
v
ar
ian
t
is
s
ee
n
as
a
b
etter
alter
n
ativ
e,
as
it
was
f
o
r
m
er
ly
r
ep
o
r
ted
to
g
iv
e
b
etter
r
esu
lts
f
o
r
p
r
o
b
lem
s
r
elate
d
to
HAR.
Fig
u
r
e
6
.
R
NN
s
tr
u
ctu
r
e
[
5
6
]
A
s
tu
d
y
b
y
[
5
7
]
in
v
esti
g
ated
th
r
ee
ty
p
es
o
f
R
NN;
L
STM
,
v
an
illa
R
NN
(
V
R
NN.
)
,
an
d
GR
U.
T
h
eir
ef
f
ec
tiv
en
ess
f
o
r
ac
tiv
ity
r
ec
o
g
n
itio
n
a
n
d
d
etec
tin
g
ab
n
o
r
m
al
d
em
en
tia
-
s
u
f
f
e
r
in
g
p
atie
n
ts
’
b
eh
av
io
r
was
ex
p
lo
r
ed
.
W
h
ile
a
d
ee
p
s
tack
ed
L
STM
was
p
r
o
p
o
s
ed
in
[
5
8
]
with
th
e
ca
p
ab
ilit
y
o
f
lear
n
in
g
an
d
g
e
n
er
alizin
g
tem
p
o
r
al
d
y
n
am
ics
f
r
o
m
r
aw
d
ata.
T
h
e
s
tr
u
ctu
r
es
o
f
th
e
G
R
U
an
d
L
STM
,
s
ch
em
es
ca
n
b
e
s
ee
n
in
Fig
u
r
e
7
.
T
h
is
d
iag
r
am
is
cr
u
cial
to
co
m
p
r
eh
en
d
th
e
ad
v
a
n
ce
s
in
R
NN
ar
ch
itectu
r
es,
wh
ic
h
h
a
v
e
b
ee
n
m
ad
e
to
im
p
r
o
v
e
th
e
ca
p
ac
ity
to
lear
n
lo
n
g
-
ter
m
r
elatio
n
s
h
ip
s
an
d
to
d
ea
l w
ith
is
s
u
es su
ch
as v
an
is
h
in
g
g
r
ad
i
en
ts
.
Fig
u
r
e
7
.
L
STM
,
GR
U
s
ch
em
es
[
5
9
]
3
.
1
.
7
.
H
y
brid
m
o
dels
T
h
ese
ar
e
ex
ce
p
tio
n
al
ca
s
es
wh
er
e
two
o
r
m
o
r
e
d
ee
p
s
tr
u
ctu
r
es
ar
e
co
m
b
in
ed
.
C
o
m
b
in
in
g
two
o
r
m
o
r
e
[
6
0
]
d
ee
p
lear
n
in
g
m
o
d
els
is
to
tak
e
ad
v
a
n
tag
e
o
f
th
e
d
ee
p
lear
n
in
g
s
tr
u
ctu
r
es
in
v
o
lv
ed
.
Fo
r
ex
am
p
le,
C
NN,
k
n
o
wn
f
o
r
ca
p
ab
ilit
ies
in
f
ea
tu
r
e
ex
tr
ac
tio
n
ca
n
b
e
co
n
ca
ten
ated
with
th
e
L
STM
m
o
d
el,
wh
ich
ca
n
lear
n
tem
p
o
r
al
d
y
n
am
ics
to
ac
co
m
p
lis
h
b
o
th
f
ea
tu
r
es.
T
h
e
a
p
p
licatio
n
s
o
f
h
y
b
r
id
m
o
d
els
ar
e
n
o
t
r
estricte
d
to
HAR.
Sti
ll,
th
ey
h
av
e
a
p
p
lied
in
a
v
a
r
iety
o
f
ap
p
licatio
n
s
d
o
m
ain
s
lik
e
v
is
u
al
r
ec
o
g
n
itio
n
[
6
1
]
,
v
o
ice
(
s
p
ee
ch
)
[
6
2
]
,
c
o
m
p
u
ter
v
is
io
n
a
n
d
n
atu
r
al
lan
g
u
a
g
e
p
r
o
ce
s
s
in
g
[
6
3
]
an
d
m
ed
ical
d
iag
n
o
s
is
[
6
4
]
.
An
ex
ten
s
iv
e
s
u
m
m
ar
y
o
f
th
e
m
ain
d
ee
p
lear
n
in
g
m
o
d
els
an
d
th
eir
ap
p
licatio
n
d
o
m
ain
s
is
g
iv
en
in
T
ab
le
1
.
Ho
wev
er
,
th
e
ch
o
ice
b
etwe
en
th
e
d
ee
p
lea
r
n
in
g
a
r
c
h
itectu
r
es
o
u
tlin
ed
in
T
a
b
le
1
d
ep
en
d
s
o
n
th
e
u
s
e
ca
s
e
at
h
an
d
an
d
th
e
av
ailab
l
e
d
ata.
I
n
n
u
ts
h
ell,
C
NN
is
r
e
g
ar
d
ed
as
th
e
b
est
ch
o
ice
wh
en
it
co
m
es
to
task
s
r
elate
d
to
im
ag
es
[
6
5
]
,
R
NN
is
id
ea
l
f
o
r
s
eq
u
en
tial
d
at
a
lik
e
s
p
ee
ch
r
ec
o
g
n
itio
n
,
a
n
d
n
atu
r
al
lan
g
u
ag
e
p
r
o
ce
s
s
in
g
[
6
6
]
.
Oth
er
s
also
h
av
e
th
eir
r
esp
ec
tiv
e
s
tr
o
n
g
u
s
e
ca
s
e
s
tr
en
g
th
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
5
2
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
,
Vo
l.
38
,
No
.
3
,
J
u
n
e
20
25
:
1
8
8
0
-
1
8
9
5
1886
T
ab
le
1
.
Dee
p
lea
r
n
in
g
m
o
d
el
s
an
d
th
eir
m
ajo
r
ap
p
licatio
n
a
r
ea
s
an
d
d
escr
ip
tio
n
A
r
c
h
i
t
e
c
t
u
r
e
M
a
j
o
r
a
p
p
l
i
c
a
t
i
o
n
a
r
e
a
s
D
e
scri
p
t
i
o
n
C
N
N
N
a
t
u
r
a
l
l
a
n
g
u
a
g
e
p
r
o
c
e
ss
i
n
g
,
d
o
c
u
m
e
n
t
a
n
a
l
y
s
i
s,
i
ma
g
e
r
e
c
o
g
n
i
t
i
o
n
,
f
a
c
e
r
e
c
o
g
n
i
t
i
o
n
,
v
i
d
e
o
a
n
a
l
y
s
i
s
U
t
i
l
i
z
e
s
c
o
n
v
o
l
u
t
i
o
n
a
l
l
a
y
e
r
s t
o
s
l
i
d
e
o
v
e
r
t
h
e
i
n
p
u
t
i
n
f
o
r
ma
t
i
o
n
f
o
r
f
e
a
t
u
r
e
l
e
a
r
n
i
n
g
.
AE
U
n
d
e
r
st
a
n
d
i
n
g
c
o
m
p
a
c
t
r
e
p
r
e
s
e
n
t
a
t
i
o
n
o
f
d
a
t
a
,
n
a
t
u
r
a
l
l
a
n
g
u
a
g
e
p
r
o
c
e
ssi
n
g
A
n
o
n
-
s
u
p
e
r
v
i
se
d
n
e
u
r
a
l
n
e
t
w
o
r
k
e
m
p
l
o
y
e
d
f
o
r
l
e
a
r
n
i
n
g
d
a
t
a
c
o
d
i
n
g
f
r
o
m
sam
p
l
e
s
.
R
N
N
S
p
e
e
c
h
a
n
d
h
a
n
d
w
r
i
t
i
n
g
r
e
c
o
g
n
i
t
i
o
n
C
o
m
p
r
i
s
e
s
o
f
l
o
o
p
s
t
h
a
t
p
r
o
v
i
d
e
i
n
f
o
r
mat
i
o
n
st
o
r
a
g
e
w
i
t
h
i
n
t
h
e
n
e
t
w
o
r
k
.
R
B
M
f
e
a
t
u
r
e
l
e
a
r
n
i
n
g
,
c
o
l
l
a
b
o
r
a
t
i
v
e
f
i
l
t
e
r
i
n
g
,
d
i
me
n
si
o
n
a
l
i
t
y
r
e
d
u
c
t
i
o
n
,
c
l
a
ss
i
f
i
c
a
t
i
o
n
,
r
e
g
r
e
ssi
o
n
,
a
n
d
t
o
p
i
c
mo
d
e
l
l
i
n
g
A
g
e
n
e
r
a
t
i
v
e
st
o
c
h
a
st
i
c
N
N
t
h
a
t
u
n
d
e
r
st
a
n
d
s
p
r
o
b
a
b
i
l
i
t
y
d
i
st
r
i
b
u
t
i
o
n
a
c
r
o
ss s
a
m
p
l
e
s.
M
LP
I
mag
e
r
e
c
o
g
n
i
t
i
o
n
,
sp
e
e
c
h
r
e
c
o
g
n
i
t
i
o
n
,
a
n
d
ma
c
h
i
n
e
t
r
a
n
s
l
a
t
i
o
n
A
f
e
e
d
f
o
r
w
a
r
d
a
r
t
i
f
i
c
i
a
l
A
N
N
3
.
2
.
Dee
p
lea
rning
o
bje
ct
de
t
ec
t
io
n a
lg
o
rit
h
m
s
Ob
ject
d
etec
tio
n
is
a
p
r
ec
i
o
u
s
an
d
p
o
p
u
la
r
co
m
p
u
ter
v
is
io
n
m
eth
o
d
t
h
at
d
ea
ls
with
o
b
ject
class
if
icatio
n
an
d
lo
ca
lizatio
n
in
an
im
ag
e
o
r
v
id
eo
.
Dee
p
lear
n
in
g
-
b
ased
o
b
ject
d
etec
ti
o
n
ap
p
r
o
ac
h
es
u
s
e
C
NN
ar
ch
itectu
r
es
s
u
ch
as
y
o
u
o
n
ly
l
o
o
k
o
n
c
e
(
YOL
O
)
,
s
in
g
le
s
h
o
t
d
etec
to
r
(
SSD
)
,
a
n
d
r
e
g
io
n
p
r
o
p
o
s
als
.
Ob
ject
d
etec
tio
n
is
ap
p
lied
in
ass
et
in
s
p
ec
tio
n
,
p
ed
estrian
d
etec
tio
n
,
s
elf
-
d
r
iv
in
g
ca
r
s
,
o
r
v
id
eo
s
u
r
v
eillan
ce
.
T
h
is
s
u
b
s
ec
tio
n
ex
p
lo
r
es
th
e
d
ee
p
lear
n
in
g
m
o
d
els
em
p
l
o
y
ed
f
o
r
o
b
ject
(
h
u
m
an
i
n
t
h
is
ca
s
e)
d
etec
tio
n
.
Fu
r
th
er
m
o
r
e
,
o
b
ject
d
etec
tio
n
ca
n
b
e
em
p
lo
y
ed
in
in
tellig
en
t
v
id
eo
an
aly
tics
(
I
VA)
wh
er
e
v
er
C
C
T
V
ca
m
er
as
ar
e
s
itu
ated
to
s
tu
d
y
h
o
w
p
e
o
p
le
in
ter
ac
t w
ith
o
n
e
a
n
o
th
er
a
n
d
th
eir
s
u
r
r
o
u
n
d
in
g
s
.
3
.
2
.
1
.
YO
L
O
YOL
Ov
1
u
tili
ze
s
a
s
in
g
le
n
eu
r
al
n
etwo
r
k
[
6
7
]
as
a
r
ea
l
-
tim
e
o
b
ject
d
etec
tio
n
s
y
s
tem
.
Acc
o
r
d
in
g
t
o
[
6
8
]
,
YOL
O
o
b
ject
d
etec
tio
n
is
o
v
er
o
n
e
th
o
u
s
an
d
tim
es
f
aster
th
an
r
e
g
io
n
-
b
ased
co
n
v
o
lu
tio
n
al
n
eu
r
a
l
n
etwo
r
k
s
(
R
-
C
NN
)
an
d
a
h
u
n
d
r
ed
tim
es
f
aster
t
h
an
f
ast
R
-
C
NN.
T
h
er
e
ex
is
t
d
if
f
er
en
t
v
er
s
io
n
s
o
f
YOL
O:
YOL
Ov
2
[
6
9
]
,
YOL
Ov
3
[
7
0
]
,
YOL
Ov
4
[
7
1
]
,
an
d
YOL
Ov
5
[
7
2
]
.
3
.
2
.
2
.
Sin
g
le
-
s
ho
t
det
ec
t
o
r
A
r
en
o
wn
ed
o
n
e
-
s
tag
e
o
b
ject
d
etec
to
r
[
7
3
]
,
SS
D
ca
n
p
r
e
d
ict
m
u
ltip
le
class
es f
aster
th
an
Y
OL
O
[
7
4
]
.
I
t
is
ea
s
y
to
tr
ain
an
d
in
co
r
p
o
r
ate
with
s
o
f
twar
e
s
y
s
tem
s
r
eq
u
ir
in
g
o
b
ject
d
etec
tio
n
m
o
d
u
l
e.
SS
D
h
as
s
u
p
er
io
r
ac
cu
r
ac
y
co
m
p
ar
e
d
to
o
t
h
er
s
in
g
le
-
s
tag
e
tech
n
iq
u
es,
ev
en
wi
th
s
m
aller
in
p
u
t im
ag
e
s
izes [
7
4
]
.
3
.
2
.
3
.
Reg
io
n
-
ba
s
ed
co
nv
o
lut
io
na
l neura
l net
wo
rk
s
E
s
tab
lis
h
ed
as
a
d
e
f
ac
to
d
e
ep
lear
n
in
g
-
b
ased
o
b
ject
d
ete
ctio
n
alg
o
r
ith
m
,
R
-
C
NN
[
7
5
]
,
[
7
6
]
ar
e
r
ev
o
lu
tio
n
a
r
y
ap
p
r
o
ac
h
es
wh
ich
em
p
lo
y
d
ee
p
m
o
d
els
to
d
et
ec
tio
n
o
f
o
b
ject.
T
h
e
m
ajo
r
w
ea
k
n
ess
o
f
R
-
C
NN
is
th
at
it
is
s
lo
w
to
im
p
lem
en
t
.
Fas
t
R
-
C
NN
was
d
ev
elo
p
ed
in
2
0
1
5
[
7
7
]
with
an
aim
to
s
ig
n
if
ican
tly
r
ed
u
ce
tr
ain
tim
e
an
d
d
r
asti
ca
lly
s
o
lv
e
th
e
lim
itatio
n
s
o
f
R
-
C
NN.
Ho
wev
er
,
YOL
O
is
s
till
a
f
as
t
er
o
p
tio
n
d
u
e
to
th
e
ea
s
e
o
f
th
e
co
d
e
[
6
8
]
.
3
.
2
.
4
.
Sq
ueez
eDe
t
Fam
o
u
s
r
esear
ch
with
in
th
e
s
co
p
e
o
f
u
n
if
ied
,
lo
w
p
o
wer
,
s
m
all
C
NN
p
r
o
p
o
s
ed
Sq
u
ee
ze
Det
f
o
r
o
b
ject
d
etec
tio
n
[
7
8
]
.
I
n
th
e
r
esear
ch
,
th
e
au
th
o
r
s
s
h
o
w
th
at
Sq
u
ee
ze
Det
is
a
lig
h
twei
g
h
t
C
NN
ex
p
licitly
d
ev
elo
p
e
d
f
o
r
r
ea
l
tim
e
o
b
ject
d
etec
tio
n
,
wh
ich
u
tili
ze
s
co
m
p
u
ter
v
is
io
n
tech
n
iq
u
es
to
d
et
ec
t
o
b
jects.
Similar
to
YOL
O,
Sq
u
ee
ze
Det
is
a
s
in
g
le
-
s
h
o
t
d
etec
t
o
r
alg
o
r
ith
m
i
n
s
p
ir
ed
b
y
YOL
O
an
d
Sq
u
ee
ze
Net
wh
ich
cr
ea
tes
an
d
class
if
ies th
e
en
tr
an
t r
eg
io
n
p
r
o
p
o
s
als in
o
n
e
n
eu
r
al
n
etwo
r
k
.
3
.
2
.
5
.
M
o
bil
eNe
t
Mo
b
ileNet
[
7
9
]
is
also
a
SSD
n
etwo
r
k
th
at
r
u
n
s
d
etec
tio
n
o
f
o
b
ject
task
s
.
Mo
b
ileNet
is
u
tili
ze
d
b
y
th
e
C
af
f
e
f
r
am
ewo
r
k
.
I
t
is
a
l
ig
h
tweig
h
t
d
ee
p
C
NN
th
at
is
m
u
ch
s
m
aller
in
s
ize
a
n
d
p
e
r
f
o
r
m
s
m
u
c
h
f
aster
th
an
m
an
y
p
o
p
u
lar
n
e
u
r
al
n
etwo
r
k
m
o
d
els.
I
ts
m
ain
ap
p
licatio
n
s
ar
e
th
e
class
if
icatio
n
an
d
d
etec
tio
n
o
f
im
ag
es.
I
t
u
tili
ze
s
d
ep
t
h
-
wis
e,
s
ep
ar
ab
le
c
o
n
v
o
lu
tio
n
s
,
w
h
ich
in
v
o
lv
e
p
er
f
o
r
m
in
g
a
s
in
g
le
co
n
v
o
lu
tio
n
o
n
ea
ch
co
lo
r
c
h
an
n
el
i
n
s
tead
o
f
co
m
b
in
in
g
all
th
r
ee
a
n
d
f
latten
in
g
it.
3
.
2
.
6
.
G
ener
a
t
iv
e
Adv
er
s
a
ria
l N
et
wo
rk
s
Gen
er
ativ
e
ad
v
er
s
ar
ial
n
etwo
r
k
s
(
GANs
)
wer
e
co
n
v
in
cin
g
ly
d
escr
ib
ed
as
a
p
o
wer
f
u
l
a
p
p
r
o
ac
h
to
u
n
s
u
p
er
v
is
ed
lear
n
i
n
g
in
m
ac
h
in
e
lear
n
in
g
,
wh
e
r
e
two
n
eu
r
al
n
etwo
r
k
s
(
th
e
g
en
er
ato
r
an
d
th
e
d
is
cr
im
in
ato
r
)
co
m
p
ete
in
a
g
am
e
lik
e
m
a
n
n
er
in
r
esear
ch
c
o
n
d
u
cted
b
y
[
8
0
]
.
T
h
e
au
th
o
r
s
ex
p
lain
ed
th
at
th
e
g
en
er
ato
r
cr
ea
tes
s
am
p
les
th
at
ar
e
tar
g
eted
at
r
ep
licatin
g
th
e
d
ata
d
is
tr
ib
u
tio
n
,
wh
er
ea
s
th
e
d
is
cr
im
in
ato
r
aim
s
to
d
if
f
er
en
tiate
b
etwe
en
tr
u
e
an
d
g
en
e
r
ated
d
ata.
T
h
is
ad
v
er
s
ar
ial
ap
p
r
o
ac
h
all
o
ws
GANs
to
ac
q
u
ir
e
co
m
p
lex
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2502
-
4
7
5
2
Dee
p
lea
r
n
in
g
a
p
p
r
o
a
c
h
es,
p
l
a
tfo
r
ms,
d
a
ta
s
ets fo
r
b
eh
a
vio
r
-
b
a
s
ed
…
(
Yu
n
u
s
a
Mo
h
a
mme
d
Je
d
d
a
h
)
1887
d
ata
d
is
tr
ib
u
tio
n
s
,
m
ak
in
g
th
e
m
s
p
ec
if
ically
ef
f
ec
tiv
e
f
o
r
task
s
lik
e
g
en
er
atin
g
r
ea
l
im
ag
es
an
d
tr
an
s
f
o
r
m
in
g
d
ata
ac
r
o
s
s
d
if
f
er
e
n
t
r
ea
lm
s
.
I
n
th
eir
i
n
v
esti
g
atio
n
,
Po
s
ilo
v
ić
et
a
l.
[
8
1
]
em
p
lo
y
e
d
GAN
to
g
en
e
r
ate
ac
tu
al
d
ata,
wh
ich
ca
n
im
p
r
o
v
e
o
b
ject
d
etec
to
r
s
.
3
.
2
.
7
.
Yo
u
o
nly
lea
rn
o
ne
re
p
re
s
ent
a
t
io
n
Yo
u
o
n
ly
lear
n
o
n
e
r
ep
r
esen
tatio
n
(
YOL
OR
)
[
8
2
]
is
a
n
o
v
el
alg
o
r
ith
m
f
o
r
d
etec
tio
n
o
f
o
b
ject,
d
if
f
er
en
t
t
o
th
e
YOL
O
d
u
e
t
o
d
if
f
e
r
en
ce
s
o
f
in
v
en
t
o
r
s
,
m
o
d
el
in
f
r
astru
ct
u
r
e,
a
n
d
a
r
ch
i
tectu
r
e.
I
t
aim
s
to
p
r
o
v
id
e
a
b
ilit
y
to
m
ac
h
in
e
le
ar
n
in
g
alg
o
r
ith
m
s
to
s
er
v
e
m
an
y
o
th
er
task
s
g
iv
en
a
s
in
g
l
e
in
p
u
t
[
8
3
]
.
W
h
ile
C
NNs
lear
n
to
an
aly
ze
in
p
u
ts
to
o
b
tain
th
e
o
u
tp
u
ts
,
YOL
OR
attem
p
ts
to
h
av
e
C
NN’
s
b
o
th
(
i
)
lear
n
g
ettin
g
o
u
tp
u
ts
an
d
,
at
t
h
e
s
am
e
tim
e
an
d
(
i
)
k
n
o
w
wh
at
all
d
if
f
e
r
en
t
o
u
tp
u
ts
co
u
ld
b
e.
I
n
s
tead
o
f
o
n
ly
a
s
in
g
le
o
u
tp
u
t,
YOL
OR
ca
n
h
av
e
a
lo
t.
4.
DE
E
P
L
E
A
RNING
T
O
O
L
S
Dee
p
lear
n
in
g
t
o
o
ls
en
a
b
le
d
at
a
r
esear
ch
er
s
t
o
b
u
ild
p
r
o
g
r
am
s
th
at
m
ay
g
iv
e
a
m
ac
h
in
e
th
e
ab
ilit
y
to
lear
n
as
a
h
u
m
an
b
r
ain
an
d
p
r
o
ce
s
s
d
ata
an
d
p
atter
n
s
b
e
f
o
r
e
m
ak
in
g
d
ec
is
io
n
s
.
T
h
e
to
o
ls
r
ely
o
n
p
r
e
d
ictiv
e
m
o
d
ellin
g
a
n
d
s
tatis
tics
,
wh
ich
h
elp
d
ata
s
cien
tis
ts
g
ath
er
,
i
n
ter
p
r
et,
a
n
d
t
h
en
a
n
aly
ze
en
o
r
m
o
u
s
a
m
o
u
n
ts
o
f
d
ata.
T
h
ese
to
o
ls
ca
n
h
elp
to
s
ea
m
less
ly
d
etec
t
o
b
jects,
r
ec
o
g
n
ize
s
p
ee
ch
,
tr
an
s
late
lan
g
u
ag
es,
an
d
m
a
k
e
d
ec
is
io
n
s
ap
p
r
o
p
r
iately
.
Als
o
r
ef
er
r
ed
to
as
d
ee
p
lear
n
in
g
p
l
atf
o
r
m
s
o
r
f
r
a
m
ewo
r
k
s
,
s
o
f
tw
ar
e
p
ac
k
ag
es
wer
e
av
ailab
le
to
th
e
r
esear
ch
er
s
to
m
itig
ate
d
ee
p
lear
n
in
g
ar
ch
it
ec
tu
r
es
co
n
s
tr
u
ctio
n
.
Ho
wev
e
r
,
s
o
m
e
y
ea
r
s
ag
o
,
non
-
d
ee
p
lear
n
in
g
ex
p
er
ts
en
co
u
n
ter
e
d
m
an
y
c
h
allen
g
es
in
m
an
ag
in
g
s
o
f
twar
e
p
ac
k
ag
es.
Su
ch
a
cir
cu
m
s
tan
ce
co
n
tin
u
e
d
u
n
til Go
o
g
le
in
2
0
1
2
p
r
o
p
o
s
ed
th
e
Dis
t
B
elief
s
y
s
tem
.
Nex
t
to
th
at,
s
o
f
twar
e
p
ac
k
ag
es
s
u
ch
as
T
en
s
o
r
Flo
w,
Dee
p
L
ea
r
n
in
g
4
j,
Mic
r
o
s
o
f
t
C
o
g
n
itiv
e
T
o
o
lk
it
(
C
NT
K)
,
T
o
r
ch
,
C
af
f
e,
H2
O
.
ai,
Neu
r
al
Desig
n
er
,
Ker
as,
an
d
Dee
p
L
e
ar
n
in
g
Kit,
h
a
v
e
wid
ely
f
u
eled
th
e
in
d
u
s
tr
y
.
4
.
1
.
Neura
l
d
esig
ner
A
u
s
er
-
f
r
ien
d
l
y
ap
p
licatio
n
d
e
v
elo
p
ed
b
y
Ar
teln
ics,
wh
ich
a
llo
ws
b
u
ild
in
g
DNNs
with
o
u
t
co
d
in
g
o
r
b
u
ild
in
g
b
lo
ck
d
iag
r
am
s
[
8
4
]
.
Am
o
n
g
s
t
its
ap
p
licatio
n
s
,
y
o
u
ca
n
also
f
in
d
alg
o
r
ith
m
s
f
o
r
d
if
f
er
e
n
t
tr
ain
in
g
s
tr
ateg
ies,
d
ata
s
tatis
tic
s
an
d
p
r
ep
ar
atio
n
,
test
in
g
an
d
d
ep
lo
y
m
en
t
o
f
th
e
m
o
d
el
o
r
e
x
p
o
r
t
th
e
r
esu
lts
to
o
t
h
er
to
o
ls
lik
e
R
o
r
Py
th
o
n
.
Fu
r
th
e
r
m
o
r
e,
n
e
u
r
al
d
esig
n
e
r
is
co
m
p
atib
le
with
th
e
m
o
s
t
f
am
iliar
d
atab
ases
an
d
d
ata
f
iles
an
d
ca
n
r
u
n
th
e
m
o
s
t
e
x
ten
s
iv
e
d
atasets
.
Mo
r
eo
v
e
r
,
y
o
u
ca
n
im
p
o
r
t
d
ata
a
n
d
u
s
e
th
e
m
ac
h
in
e
lear
n
in
g
p
latf
o
r
m
to
en
h
an
ce
p
r
o
d
u
ct
iv
ity
.
T
h
e
p
r
im
ar
y
o
b
jectiv
e
o
f
n
eu
r
al
d
esig
n
e
r
is
to
f
ac
ilit
ate
in
n
o
v
ativ
e
o
r
g
an
izatio
n
s
to
em
p
l
o
y
ar
ti
f
icial
in
tellig
en
ce
(
AI
)
,
with
a
f
o
cu
s
o
n
its
ap
p
licatio
n
s
an
d
n
o
t
b
ased
o
n
m
ath
em
atics
o
r
p
r
o
g
r
am
m
i
n
g
.
Fu
r
th
er
m
o
r
e,
n
eu
r
al
d
esi
g
n
er
s
u
p
p
o
r
ts
a
wh
o
le
m
o
d
ellin
g
cy
cle,
f
r
o
m
p
r
ep
ar
in
g
d
ata
to
m
o
d
el
p
r
o
d
u
ctio
n
[
8
5
]
.
4
.
2
.
H
2
O
.
a
i
Dev
elo
p
ed
e
m
p
lo
y
i
n
g
J
av
a
a
s
co
r
e
tech
n
o
lo
g
y
,
H2
O
p
r
o
v
i
d
es
en
o
r
m
o
u
s
f
lex
i
b
ilit
y
to
r
e
s
ea
r
ch
er
s
.
T
h
an
k
s
to
H2
O,
an
y
b
o
d
y
ca
n
ea
s
ily
ap
p
ly
p
r
ed
ictiv
e,
m
ac
h
in
e
lear
n
in
g
,
an
aly
tics
,
a
n
d
d
ee
p
lear
n
in
g
to
u
n
r
av
el
co
m
p
lex
p
r
o
b
lem
s
[
8
6
]
.
I
t
u
tili
ze
s
th
e
m
o
s
t
ac
cu
s
to
m
ed
in
ter
f
ac
e
[
8
7
]
,
an
o
p
en
-
s
o
u
r
ce
f
r
am
ew
o
r
k
with
an
ea
s
y
-
to
-
u
s
e
web
-
b
ase
d
g
r
a
p
h
ical
u
s
er
in
ter
f
ac
e
(
GU
I
)
.
T
h
is
to
o
l h
elp
s
in
r
ea
l
-
tim
e
d
ata
s
co
r
i
n
g
,
an
d
it
is
h
ig
h
ly
s
ca
lab
le.
4
.
3
.
Dee
pL
ea
rning
K
it
Ap
p
le
em
p
l
o
y
s
Dee
p
L
ea
r
n
in
g
Kit
f
r
am
ewo
r
k
o
n
m
o
s
t
o
f
th
ei
r
p
r
o
d
u
cts
s
u
c
h
as
OS
X,
iOS,
an
d
tv
OS
[
8
8
]
.
I
t
s
u
p
p
o
r
ts
th
e
p
r
e
-
tr
ai
n
ed
m
o
d
els
o
n
Ap
p
le’
s
d
ev
ice
s
with
GPUs
f
o
r
co
n
v
e
n
tio
n
a
l
OS
X
co
m
p
u
ter
s
[
8
9
]
.
I
n
ad
d
itio
n
,
it
u
s
es
d
ee
p
C
NN,
s
u
ch
as
im
ag
e
r
ec
o
g
n
itio
n
.
I
t
is
cu
r
r
e
n
tly
tr
ain
ed
with
th
e
ca
f
f
e
d
ee
p
lear
n
in
g
f
r
a
m
ewo
r
k
[
9
0
]
,
b
u
t
t
h
e
lo
n
g
-
ter
m
o
b
jectiv
e
is
to
s
u
p
p
o
r
t
th
e
u
s
e
o
f
o
th
er
m
o
d
els
s
u
ch
as
T
o
r
ch
an
d
T
en
s
o
r
Flo
w.
4
.
4
.
M
icro
s
o
f
t
C
o
g
nitiv
e
T
o
o
lk
it
T
h
is
to
o
lk
it
is
co
m
m
er
cially
u
s
ed
to
tr
ain
d
ee
p
lea
r
n
in
g
s
y
s
tem
s
to
lear
n
ac
cu
r
ately
as
th
e
h
u
m
a
n
b
r
ain
.
An
o
p
en
-
s
o
u
r
ce
wh
ich
o
f
f
er
s
ex
ce
llen
t
s
ca
lin
g
ab
ilit
i
es,
en
ter
p
r
is
e
-
lev
el
q
u
ality
,
ac
cu
r
ac
y
,
a
n
d
s
p
ee
d
.
Mic
r
o
s
o
f
t p
r
o
d
u
cts lik
e
C
o
r
ta
n
a,
Sk
y
p
e,
an
d
X
b
o
x
B
in
g
,
u
s
e
it f
o
r
AI
in
d
u
s
tr
y
-
le
v
el
g
en
er
atio
n
[
9
0
]
.
4
.
5
.
K
er
a
s
A
Py
th
o
n
-
b
ased
a
p
p
licatio
n
p
r
o
g
r
am
in
ter
f
ac
e
(
API
)
em
p
l
o
y
ed
to
r
u
n
T
e
n
s
o
r
Flo
w
[
9
1
]
,
T
h
ea
n
o
,
an
d
C
NT
K.
Ker
as
is
an
en
h
an
ce
d
lev
el
API
o
f
T
en
s
o
r
Flo
w
u
tili
ze
d
f
o
r
m
o
d
el
b
u
ild
i
n
g
an
d
p
r
o
v
es
s
im
ilar
in
s
t
a
c
k
i
n
g
l
a
y
e
r
s
.
O
n
t
h
e
o
t
h
e
r
h
a
n
d
,
i
t
h
a
s
m
i
n
i
m
a
l
f
u
n
c
t
i
o
n
a
l
i
t
y
.
I
t
w
a
s
d
e
s
i
g
n
e
d
t
o
e
n
a
b
l
e
q
u
i
c
k
e
x
p
e
r
i
m
e
n
t
s
[
9
2
]
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
5
2
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
,
Vo
l.
38
,
No
.
3
,
J
u
n
e
20
25
:
1
8
8
0
-
1
8
9
5
1888
an
d
wo
r
k
s
with
T
en
s
o
r
Flo
w
an
d
T
h
ea
n
o
.
T
h
e
m
ai
n
ad
v
a
n
tag
e
is
th
at
f
ast
r
esu
lts
ca
n
b
e
o
b
tain
ed
with
ju
s
t
o
n
e
co
n
ce
p
t.
4
.
6
.
T
ens
o
rF
lo
w
T
h
i
s
i
s
t
h
e
m
o
s
t
p
r
o
m
i
n
e
n
t
G
o
o
g
l
e
l
i
b
r
a
r
i
e
s
f
o
r
d
e
e
p
l
e
a
r
n
i
n
g
a
n
d
m
a
c
h
i
n
e
l
e
a
r
n
i
n
g
a
p
p
l
i
c
a
t
i
o
n
s
[
9
3
]
.
I
t
is
b
u
ilt
to
r
u
n
o
n
GPUs
,
C
PUs
,
an
d
in
ter
n
et
o
f
th
i
n
g
s
(
I
o
T
)
p
r
o
ce
s
s
o
r
s
s
u
ch
as
n
eu
r
al
n
etwo
r
k
s
tick
,
an
d
J
etso
n
Nan
o
.
R
elea
s
ed
in
2
0
1
5
b
y
Go
o
g
le
b
u
t
th
e
s
tab
le
v
er
s
io
n
h
as
b
ee
n
av
aila
b
le
s
in
ce
2
0
1
7
u
n
d
er
th
e
Ap
ac
h
e
Op
en
-
So
u
r
ce
L
icen
s
e
[
9
4
]
.
4
.
7
.
Co
nv
Ne
t
J
S
C
o
n
v
NetJS
en
ab
les
u
s
er
s
to
d
esig
n
an
d
r
eso
lv
e
n
e
u
r
al
n
e
two
r
k
s
b
y
u
s
in
g
J
av
aScr
ip
t
f
o
r
n
e
u
r
al
n
etwo
r
k
s
(
d
ee
p
lear
n
i
n
g
m
o
d
els)
tr
ain
in
g
en
tire
ly
in
b
r
o
wser
[
9
5
]
.
I
t
is
an
ex
p
er
im
e
n
tal
r
ein
f
o
r
ce
m
e
n
t
lear
n
in
g
m
o
d
u
le
b
u
ilt
o
n
d
ee
p
Q
l
ea
r
n
in
g
,
n
o
t
n
ee
d
i
n
g
ad
d
it
io
n
al
s
o
f
twar
e,
in
s
tallatio
n
s
,
GPUs
o
r
co
m
p
iler
s
.
Fo
r
p
r
o
ce
s
s
in
g
im
ag
es,
C
o
n
v
NetJS
ca
n
d
ef
in
e
an
d
tr
ai
n
co
n
v
o
lu
tio
n
al
n
etwo
r
k
s
.
4
.
8
.
T
o
rc
h
W
ith
a
GPU
-
b
ased
MA
T
L
A
B
-
lik
e
en
v
ir
o
n
m
e
n
t,
th
e
to
r
c
h
[
9
4
]
is
an
ex
tr
em
ely
ef
f
ec
tiv
e
o
p
en
-
s
o
u
r
ce
co
m
p
u
tin
g
p
latf
o
r
m
th
at
s
u
p
p
o
r
ts
d
ee
p
lear
n
in
g
tech
n
iq
u
es
[
9
4
]
.
I
t
is
a
h
ig
h
-
p
o
we
r
ed
N
-
d
im
e
n
s
io
n
al
ar
r
ay
th
at
f
ea
t
u
r
es
m
an
y
r
o
u
ti
n
es
f
o
r
in
d
e
x
in
g
,
tr
an
s
p
o
s
in
g
,
an
d
s
licin
g
[
9
6
]
.
I
t
h
as
o
u
ts
tan
d
in
g
s
u
p
p
o
r
t
f
o
r
GPU
an
d
is
em
b
ed
d
ab
le
t
o
co
l
lab
o
r
ate
with
An
d
r
o
id
a
n
d
iO
S.
4
.
9
.
Dee
pL
ea
rning
4
j
Dee
p
L
ea
r
n
in
g
4
j
[
9
4
]
is
an
o
p
en
-
s
o
u
r
ce
,
f
r
ee
jav
a
-
b
ased
d
ee
p
lear
n
in
g
lib
r
ar
y
wh
ich
p
r
o
v
id
es
an
ex
ten
s
iv
e
s
o
lu
tio
n
to
d
ee
p
lear
n
in
g
in
d
if
f
er
e
n
t
ap
p
licatio
n
s
s
u
ch
as
k
n
o
wled
g
e
d
is
co
v
e
r
y
a
n
d
d
ee
p
p
r
ed
ictiv
e
m
in
in
g
o
n
C
PU’
s
an
d
GPU’
s
g
r
ap
h
ics p
r
o
ce
s
s
in
g
.
I
t in
teg
r
a
tes th
e
AI
alg
o
r
ith
m
s
an
d
m
et
h
o
d
s
ap
p
r
o
p
r
iate
f
o
r
b
u
s
in
ess
in
tellig
en
ce
,
cy
b
er
f
o
r
en
s
ics,
p
r
ed
ictiv
e
a
n
aly
s
is
,
f
ac
e
r
ec
o
g
n
itio
n
,
n
etwo
r
k
in
t
r
u
s
io
n
d
etec
tio
n
a
n
d
p
r
ev
en
tio
n
,
an
o
m
al
y
d
etec
tio
n
,
r
ec
o
m
m
e
n
d
er
s
y
s
tem
s
,
an
d
lo
ts
m
o
r
e.
Mo
d
els
f
r
o
m
ad
v
a
n
ce
d
d
ee
p
-
lear
n
in
g
f
r
am
ewo
r
k
s
lik
e
C
af
f
e,
T
h
ea
n
o
,
Ker
as,
an
d
T
en
s
o
r
Flo
w
ca
n
b
e
im
p
o
r
te
d
[
9
4
]
.
4
.
1
0
.
M
AT
L
AB
dee
p lea
rni
ng
t
o
o
lbo
x
MA
T
L
AB
is
a
p
r
o
p
r
ietar
y
m
u
lti
-
p
ar
ad
ig
m
m
ath
e
m
atica
l
co
m
p
u
tin
g
en
v
ir
o
n
m
en
t,
d
e
v
elo
p
ed
b
y
Ma
th
W
o
r
k
s
th
at
allo
ws
m
an
ip
u
latio
n
s
o
f
m
atr
ix
,
p
l
o
ttin
g
f
u
n
ctio
n
s
an
d
d
ata,
im
p
lem
e
n
t
atio
n
o
f
alg
o
r
ith
m
s
,
cr
ea
tio
n
o
f
u
s
er
in
ter
f
ac
es,
a
s
well
as
in
ter
f
ac
in
g
with
p
r
o
g
r
am
s
b
u
ilt
u
s
in
g
o
t
h
er
la
n
g
u
ag
es.
MA
T
L
AB
u
tili
ze
s
to
o
lb
o
x
es,
esp
ec
ially
d
ee
p
-
lear
n
i
n
g
to
o
lb
o
x
es
to
r
u
n
p
r
o
g
r
am
s
[
8
4
]
.
I
n
2
0
1
9
,
it
was
u
s
ed
to
b
u
ild
a
GAN,
atten
tio
n
n
etwo
r
k
s
,
an
d
v
ar
iatio
n
al
au
to
e
n
co
d
e
r
s
.
C
NN,
L
STM
,
an
d
n
etwo
r
k
s
with
3
D
C
NN
lay
er
s
ca
n
all
b
e
co
m
b
in
e
d
u
s
in
g
t
h
e
MA
T
L
AB
d
ee
p
lear
n
in
g
to
o
lk
it [
9
7
]
.
A
p
latf
o
r
m
f
o
r
c
r
ea
tin
g
an
d
im
p
lem
en
tin
g
DNNs
with
alg
o
r
ith
m
s
,
p
r
e
-
tr
ain
e
d
m
o
d
els,
an
d
APPs
i
s
p
r
o
v
id
e
d
b
y
th
e
MA
T
L
AB
d
ee
p
lear
n
in
g
t
o
o
lb
o
x
.
L
STM
n
etwo
r
k
s
f
o
r
tim
e
s
er
ies
an
aly
s
is
an
d
tr
an
s
f
er
lear
n
in
g
h
av
e
b
ee
n
a
p
ar
t
o
f
th
e
MA
T
L
AB
to
o
lb
o
x
s
in
ce
2
0
1
7
[
9
7
]
.
Am
o
n
g
s
t
th
e
k
ey
p
l
u
s
o
f
MA
T
L
AB
is
th
at
th
er
e
ar
e
m
u
ltip
le
GPUs
,
clu
s
ter
co
m
p
u
tin
g
,
p
a
r
allel
an
d
clo
u
d
co
m
p
u
tin
g
,
f
o
r
s
p
ee
d
i
n
g
u
p
th
e
p
r
o
ce
s
s
es
o
f
d
ee
p
lear
n
in
g
.
T
ab
le
2
s
h
o
ws
a
s
u
m
m
ar
y
o
f
th
e
d
ee
p
le
ar
n
in
g
to
o
ls
d
is
cu
s
s
ed
.
Pro
m
i
n
en
t
d
ee
p
lear
n
in
g
to
o
ls
ar
e
p
r
esen
ted
in
T
ab
le
2
,
in
tan
d
em
with
d
etails
ab
o
u
t
th
eir
licen
s
e,
p
r
im
ar
y
p
r
o
g
r
a
m
m
in
g
lan
g
u
a
g
es,
r
elea
s
e
d
ates,
an
d
o
f
f
icial
web
s
ites
.
T
h
e
tab
le
in
ten
d
s
to
p
r
o
v
id
e
r
esear
ch
er
s
with
im
m
ed
ia
te
g
u
id
an
ce
as
th
ey
ch
o
o
s
e
th
e
r
i
g
h
t to
o
ls
to
u
s
e
in
th
eir
b
eh
a
v
io
r
-
b
ased
r
ec
o
g
n
itio
n
r
esear
ch
.
T
ab
le
2
.
Dee
p
lea
r
n
in
g
to
o
ls
s
u
m
m
ar
y
ta
b
le
F
r
a
mew
o
r
k
Li
c
e
n
s
e
C
o
r
e
l
a
n
g
u
a
g
e
R
e
l
e
a
s
e
y
e
a
r
H
o
me
p
a
g
e
N
e
u
r
a
l
D
e
s
i
g
n
e
r
P
r
o
p
r
i
e
t
a
r
y
so
f
t
w
a
r
e
C
+
+
2
0
1
4
h
t
t
p
s
:
/
/
w
w
w
.
n
e
u
r
a
l
d
e
s
i
g
n
e
r
.
c
o
m/
H
2
O
A
p
a
c
h
e
2
.
0
R
a
n
d
P
y
t
h
o
n
2
0
1
5
h
t
t
p
s
:
/
/
h
2
o
.
a
i
/
D
e
e
p
Le
a
r
n
i
n
g
K
i
t
A
p
a
c
h
e
2
.
0
C
+
+
2
0
1
5
h
t
t
p
:
/
/
d
e
e
p
l
e
a
r
n
i
n
g
k
i
t
.
o
r
g
/
M
i
c
r
o
s
o
f
t
C
o
g
n
i
t
i
v
e
To
o
l
k
i
t
M
I
T
C
+
+
2
0
1
6
h
t
t
p
s
:
/
/
l
e
a
r
n
.
mi
c
r
o
s
o
f
t
.
c
o
m/
e
n
-
u
s/
c
o
g
n
i
t
i
v
e
-
t
o
o
l
k
i
t
/
K
e
r
a
s
M
I
T
P
y
t
h
o
n
2
0
1
5
h
t
t
p
s
:
/
/
k
e
r
a
s.
i
o
/
Te
n
s
o
r
F
l
o
w
A
p
a
c
h
e
2
.
0
C
+
+
a
n
d
P
y
t
h
o
n
2
0
1
5
h
t
t
p
s
:
/
/
w
w
w
.
t
e
n
s
o
r
f
l
o
w
.
o
r
g
/
C
o
n
v
N
e
t
JS
M
I
T
Jav
a
S
c
r
i
p
t
2
0
1
4
h
t
t
p
s
:
/
/
c
s
.
st
a
n
f
o
r
d
.
e
d
u
/
p
e
o
p
l
e
/
k
a
r
p
a
t
h
y
/
c
o
n
v
n
e
t
j
s/
To
r
c
h
B
S
D
C
a
n
d
L
u
a
2
0
0
2
h
t
t
p
:
/
/
t
o
r
c
h
.
c
h
/
D
e
e
p
Le
a
r
n
i
n
g
4
j
A
p
a
c
h
e
2
.
0
Jav
a
2
0
1
4
h
t
t
p
s
:
/
/
d
e
e
p
l
e
a
r
n
i
n
g
4
j
.
o
r
g
/
M
A
TLA
B
T
o
o
l
b
o
x
P
r
o
p
r
i
e
t
a
r
y
so
f
t
w
a
r
e
C
,
C
+
+
,
J
a
v
a
,
M
A
TLA
B
1
9
9
2
h
t
t
p
s
:
/
/
ma
t
l
a
b
.
ma
t
h
w
o
r
k
s
.
c
o
m/
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2502
-
4
7
5
2
Dee
p
lea
r
n
in
g
a
p
p
r
o
a
c
h
es,
p
l
a
tfo
r
ms,
d
a
ta
s
ets fo
r
b
eh
a
vio
r
-
b
a
s
ed
…
(
Yu
n
u
s
a
Mo
h
a
mme
d
Je
d
d
a
h
)
1889
5.
B
E
H
AV
I
O
R
RE
CO
G
NIT
I
O
N
DATAS
E
T
S
L
ar
g
e
-
s
ca
le
d
atab
ase
e
x
is
ten
ce
is
a
p
r
e
r
eq
u
is
ite
f
o
r
th
e
ef
f
i
cien
t
wo
r
k
in
g
o
f
d
ee
p
lea
r
n
in
g
s
y
s
tem
.
Her
e,
we
lo
o
k
at
s
o
m
e
d
atas
ets
u
s
ed
f
o
r
b
eh
a
v
io
r
r
ec
o
g
n
i
tio
n
in
d
ee
p
lear
n
in
g
.
Un
f
o
r
t
u
n
ately
,
m
o
s
t
ea
r
ly
d
ee
p
-
lear
n
i
n
g
s
tate
-
of
-
th
e
-
a
r
t
wo
r
k
s
wer
e
tr
ain
ed
with
p
r
iv
a
te
lar
g
e
-
s
ca
le
d
atab
ases
.
A
s
it
u
atio
n
lik
e
th
is
h
as
ca
u
s
ed
r
esear
ch
er
s
to
n
ee
d
m
o
r
e
p
u
b
lic
d
atasets
to
r
ep
licate
th
e
f
in
d
in
g
s
o
f
th
ese
wo
r
k
s
o
r
m
ak
e
a
co
m
p
ar
is
o
n
o
f
th
eir
m
o
d
els.
5
.
1
.
UCF
-
C
ri
m
e
da
t
a
s
et
Sp
ec
if
ically
cr
ea
ted
to
ass
is
t
with
r
ec
o
g
n
izin
g
an
o
m
alo
u
s
ac
tiv
ities
in
d
if
f
icu
lt
en
v
ir
o
n
m
en
ts
,
th
e
UC
F
-
C
r
im
e
s
d
ataset
is
in
v
alu
ab
le
f
o
r
r
ea
l
-
wo
r
ld
an
o
m
aly
i
d
en
tific
atio
n
in
s
u
r
v
eillan
ce
v
id
eo
s
.
Acc
o
r
d
in
g
to
[
9
8
]
,
t
h
is
d
ataset
allo
ws
f
o
r
th
e
cr
ea
tio
n
an
d
test
in
g
o
f
ef
f
ec
tiv
e
d
etec
tio
n
alg
o
r
ith
m
s
b
ec
a
u
s
e
it
co
m
p
r
is
es
a
wid
e
r
an
g
e
o
f
cr
im
e
-
r
elate
d
e
v
en
ts
th
at
wer
e
r
ec
o
r
d
ed
f
r
o
m
v
ar
io
u
s
ca
m
er
a
an
g
les
an
d
s
ettin
g
s
.
T
h
e
UC
F
-
C
r
im
es
d
ataset
co
n
tr
ib
u
tes
to
th
e
d
ev
elo
p
m
en
t
o
f
a
n
o
m
aly
d
etec
tio
n
tech
n
iq
u
es
b
y
o
f
f
er
i
n
g
co
m
p
r
eh
en
s
iv
e
an
n
o
tatio
n
s
an
d
a
wid
e
r
an
g
e
o
f
s
ce
n
ar
io
s
,
to
en
h
a
n
ce
s
af
et
y
an
d
s
ec
u
r
ity
th
r
o
u
g
h
m
o
r
e
e
f
f
icien
t
s
u
r
v
eillan
ce
s
y
s
tem
s
.
UC
F
-
C
r
im
e
d
ataset
co
m
p
r
is
es
o
f
lo
n
g
u
n
cu
t
s
u
r
v
eillan
ce
v
id
e
o
s
th
at
c
o
n
tai
n
th
ir
teen
r
ea
l
-
life
ab
n
o
r
m
alities
,
co
m
p
r
is
in
g
ass
au
lt,
ab
u
s
e,
ar
r
est,
s
h
o
p
liftin
g
,
r
o
ad
ac
ci
d
en
t,
ar
s
o
n
,
f
ig
h
tin
g
,
b
u
r
g
la
r
y
,
s
h
o
o
tin
g
,
ex
p
lo
s
io
n
,
s
tealin
g
,
r
o
b
b
er
y
,
an
d
v
an
d
alis
m
.
T
h
es
e
ab
n
o
r
m
alities
wer
e
ch
o
s
en
b
ec
au
s
e
th
ey
h
av
e
a
s
u
b
s
tan
tial e
f
f
ec
t o
n
th
e
s
af
ety
o
f
th
e
p
u
b
lic.
5
.
2
.
H
B
D2
1
da
t
a
s
et
-
hu
ma
n beha
v
io
r
da
t
a
s
et
2
0
2
1
J
ay
aswal
an
d
Dix
it
[
9
9
]
is
c
o
m
p
r
is
ed
o
f
n
o
r
m
al
an
d
ab
n
o
r
m
al
ac
tio
n
v
id
eo
f
iles
p
r
ep
ar
ed
with
n
atu
r
al
an
d
ar
tific
ial
lig
h
t
co
n
d
itio
n
s
to
ea
s
e
r
ec
o
g
n
itio
n
.
T
h
e
d
ataset
is
m
ad
e
u
p
o
f
4
5
6
an
n
o
tated
v
i
d
eo
s
.
T
h
is
d
ataset
co
m
p
r
is
es
f
o
u
r
ca
teg
o
r
ies
a)
g
u
n
v
io
len
ce
,
b
)
s
ab
o
tag
e
v
io
len
ce
,
c
)
ass
au
lt
v
io
len
ce
,
an
d
d
)
n
o
r
m
al
ac
tio
n
s
.
E
v
e
r
y
ca
teg
o
r
y
co
n
tai
n
ed
o
v
er
o
n
e
h
u
n
d
r
e
d
v
id
eo
f
iles
.
E
ac
h
o
f
th
e
v
id
eo
s
is
9
s
ec
o
n
d
s
lo
n
g
,
an
d
it
is
o
b
s
er
v
ed
t
h
at
s
u
ch
a
len
g
th
is
en
o
u
g
h
to
r
ep
r
esen
t
r
ea
l
ac
tio
n
.
T
h
e
r
atio
o
f
tr
ain
i
n
g
to
test
in
g
v
id
e
o
f
iles
is
7
:3
.
I
n
o
th
er
wo
r
d
s
,
7
0
%
tr
ain
in
g
an
d
3
0
%
test
in
g
v
id
eo
s
o
f
th
e
o
r
ig
in
al
d
ataset
.
I
n
th
is
en
h
an
ce
d
v
er
s
io
n
,
t
h
e
b
r
ig
h
tn
ess
o
f
ea
ch
v
id
e
o
is
i
n
cr
ea
s
ed
at
an
a
p
p
r
o
p
r
iate
th
r
esh
o
ld
,
wh
ich
h
elp
s
ac
h
iev
e
b
etter
p
er
f
o
r
m
an
ce
in
t
h
e
tr
ain
in
g
an
d
test
in
g
s
tag
es.
5
.
3
.
M
o
L
a
I
nCa
r
AR:
da
t
a
s
et
f
o
r
a
ct
io
n r
ec
o
g
nitio
n
Mo
L
a
I
n
C
ar
AR
[
1
0
0
]
d
atase
t
is
u
s
ed
in
tr
ain
in
g
h
u
m
an
ac
tio
n
r
ec
o
g
n
itio
n
in
a
v
eh
icle
f
o
cu
s
ed
o
n
v
io
len
ce
d
etec
tio
n
.
T
h
er
m
al,
R
GB
,
d
ep
th
,
an
d
ev
en
t
-
b
ased
d
ata
wer
e
all
r
ec
o
r
d
e
d
f
o
r
t
h
is
d
ataset
with
a
r
esu
ltin
g
6
,
4
0
0
s
am
p
le
v
i
d
eo
s
an
d
o
v
er
th
r
ee
m
illi
o
n
f
r
am
es,
g
ath
er
ed
f
r
o
m
d
if
f
er
en
t
s
i
x
teen
s
u
b
jects.
T
h
e
d
ataset
in
clu
d
es f
if
ty
-
eig
h
t
ac
tio
n
class
es,
wh
ich
in
clu
d
e
n
e
u
tr
al
(
n
o
n
-
v
io
len
t)
an
d
v
io
len
t a
ctiv
ities
.
5
.
4
.
KU
-
H
AR
T
h
e
KU
-
HAR
d
ataset
p
r
o
v
i
d
es
an
ex
ten
s
iv
e
s
et
o
f
d
at
a
f
o
r
t
h
e
d
ev
el
o
p
m
e
n
t
an
d
test
in
g
o
f
r
ec
o
g
n
itio
n
alg
o
r
ith
m
s
,
wh
ich
is
a
m
ajo
r
co
n
tr
ib
u
tio
n
to
th
e
f
ield
o
f
h
u
m
an
ac
tiv
ity
r
ec
o
g
n
itio
n
.
I
n
r
esear
ch
co
n
d
u
cte
d
b
y
[
1
0
1
]
,
th
ey
p
o
s
ed
th
at
th
e
KU
-
HAR
d
ataset
is
an
o
p
en
d
ataset
co
n
tain
in
g
eig
h
teen
d
if
f
e
r
en
t
ac
tiv
ities
g
en
er
ated
f
r
o
m
n
i
n
ety
p
ar
ticip
a
n
ts
(
s
ev
en
ty
-
f
i
v
e
m
ale
an
d
f
if
teen
f
em
ale
)
with
th
e
h
elp
o
f
s
m
ar
tp
h
o
n
e
s
en
s
o
r
s
(
g
y
r
o
s
co
p
e
an
d
ac
ce
ler
o
m
eter
)
.
I
t
co
n
tai
n
s
a
wid
e
r
an
g
e
o
f
ac
tiv
ities
th
at
wer
e
ac
cu
r
ately
ca
p
tu
r
ed
a
n
d
d
o
cu
m
e
n
ted
n
u
m
er
o
u
s
asp
ec
ts
o
f
h
u
m
an
b
e
h
av
io
r
a
n
d
in
ter
ac
tio
n
s
.
T
h
e
d
ataset
co
n
s
is
ts
o
f
20
,
7
5
0
ex
tr
ac
te
d
s
u
b
s
am
p
les
an
d
1
,
9
4
5
u
n
p
r
o
ce
s
s
ed
ac
tiv
ities
g
ath
er
ed
d
ir
ec
tly
f
r
o
m
th
e
s
u
b
jects.
T
h
e
KU
-
HAR
d
ataset
co
n
tr
ib
u
tes
to
th
e
d
ev
el
o
p
m
en
t
o
f
ac
tiv
it
y
r
ec
o
g
n
itio
n
s
y
s
tem
s
b
y
o
f
f
er
in
g
lar
g
e
a
n
d
co
m
p
r
eh
e
n
s
iv
e
s
am
p
les,
wh
ich
ar
e
in
ten
d
ed
to
im
p
r
o
v
e
th
e
p
r
ec
is
io
n
an
d
p
r
ac
ticality
o
f
m
o
d
els
in
v
ar
io
u
s
co
n
tex
ts
.
5
.
5
.
M
ulti
-
g
a
it
a
nd
s
ing
le
g
a
it
da
t
a
s
et
s
Pre
v
io
u
s
r
esear
ch
in
g
ait
b
y
[
1
0
2
]
f
o
r
r
ec
o
g
n
itio
n
u
s
in
g
th
e
av
ailab
le
d
ataset
f
o
cu
s
ed
o
n
s
in
g
le
s
u
b
jects.
B
u
t in
a
r
ea
l
-
tim
e
s
i
t
u
atio
n
(
lik
e
air
p
o
r
ts
,
s
u
b
way
s
tatio
n
s
etc.
)
a
m
o
r
e
s
ig
n
if
ican
t n
u
m
b
er
o
f
p
e
r
s
o
n
s
walk
in
g
in
g
r
o
u
p
s
an
d
o
cc
lu
s
io
n
f
ac
to
r
s
af
f
ec
t
g
ait
r
ec
o
g
n
i
tio
n
p
er
f
o
r
m
an
ce
.
T
h
u
s
,
th
is
n
ew
d
ataset
wh
ich
f
o
cu
s
ed
o
n
d
y
n
am
ic
o
cc
lu
s
io
n
s
ce
n
ar
io
s
,
was
p
r
esen
ted
.
T
h
ey
b
u
ilt
two
d
is
tin
ct
ca
teg
o
r
ies
o
f
g
ait
d
ataset,
i.e
.
,
s
u
b
jects
walk
in
g
r
o
u
p
s
(
SMVDU
-
m
u
lti
-
g
ait
)
,
an
d
s
u
b
jects
m
o
v
e
in
d
iv
id
u
ally
(
S
MV
DU
-
s
in
g
le
-
g
ait
)
.
T
h
e
m
ajo
r
in
ten
t
o
f
th
is
d
ataset
is
to
ex
am
in
e
th
e
d
if
f
er
en
ce
s
in
g
ait
p
atter
n
s
b
y
th
e
tim
e
th
e
s
am
e
s
u
b
ject
walk
s
as a
g
r
o
u
p
o
r
o
n
an
i
n
d
i
v
id
u
al
b
asis
an
d
also
r
ec
o
g
n
iz
e
th
e
tar
g
et
s
u
b
ject
in
m
u
lti
-
g
a
it
.
6.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
T
h
e
s
u
r
v
ey
r
ev
e
als
th
at
C
NN
s
an
d
R
NNs
d
ee
p
lear
n
in
g
alg
o
r
ith
m
s
ar
e
th
e
m
o
s
t
o
f
ten
em
p
lo
y
ed
f
o
r
task
s
s
u
ch
as
h
u
m
an
b
e
h
av
i
o
r
r
ec
o
g
n
itio
n
d
u
e
to
t
h
eir
s
tr
en
g
th
s
in
h
an
d
lin
g
im
a
g
e
d
ata
an
d
s
eq
u
en
tial
Evaluation Warning : The document was created with Spire.PDF for Python.