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.
41
,
No
.
3
,
Ma
r
ch
20
26
,
p
p
.
1
1
0
5
~
1
1
1
6
I
SS
N:
2
502
-
4
7
52
,
DOI
: 1
0
.
1
1
5
9
1
/ijee
cs
.v
41.
i
3
.
p
p
1
1
0
5
-
1
1
1
6
1105
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
ee
cs
.
ia
esco
r
e.
co
m
An inno
v
a
tive dee
p learning
bas
ed
a
ppro
a
ch f
o
r ano
ma
ly
detec
tion in in
telli
g
ent
v
ideo
surv
eil
la
nce
M
eg
ha
G
.
P
a
llewa
r
1
,
Vij
a
y
a
R.
P
a
wa
r
2
1
D
e
p
a
r
t
me
n
t
o
f
El
e
c
t
r
o
n
i
c
s a
n
d
T
e
l
e
c
o
mm
u
n
i
c
a
t
i
o
n
E
n
g
i
n
e
e
r
i
n
g
,
C
O
EP T
e
c
h
n
o
l
o
g
i
c
a
l
U
n
i
v
e
r
s
i
t
y
,
P
u
n
e
,
I
n
d
i
a
2
D
e
p
a
r
t
me
n
t
o
f
El
e
c
t
r
o
n
i
c
s a
n
d
T
e
l
e
c
o
mm
u
n
i
c
a
t
i
o
n
E
n
g
i
n
e
e
r
i
n
g
,
B
h
a
r
a
t
i
V
i
d
y
a
p
e
e
t
h
’
s
C
o
l
l
e
g
e
o
f
E
n
g
i
n
e
e
r
i
n
g
f
o
r
W
o
me
n
,
S
a
v
i
t
r
i
b
a
i
P
h
u
l
e
P
u
n
e
U
n
i
v
e
r
si
t
y
,
P
u
n
e
,
I
n
d
i
a
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Ap
r
2
6
,
2
0
2
4
R
ev
is
ed
Oct
1
5
,
2
0
2
5
Acc
ep
ted
Dec
1
3
,
2
0
2
5
No
wa
d
a
y
s
,
a
n
o
m
a
ly
d
e
tec
ti
o
n
h
a
s
g
a
in
e
d
v
it
a
l
im
p
o
rtan
c
e
a
s
se
c
u
rit
y
is
a
m
a
jo
r
c
o
n
c
e
rn
e
v
e
ry
wh
e
re
.
T
h
is
wo
rk
f
o
c
u
se
s
o
n
d
e
v
e
lo
p
in
g
a
n
i
n
telli
g
e
n
t
v
id
e
o
su
rv
e
il
lan
c
e
sy
ste
m
c
a
p
a
b
le
o
f
d
e
tec
ti
n
g
a
n
o
m
a
lo
u
s
a
c
ti
v
it
i
e
s
i
n
v
id
e
o
s,
u
ti
l
izin
g
t
h
e
UCF
Crime
d
a
tas
e
t
a
s
th
e
p
rima
ry
so
u
rc
e
.
Th
e
p
ro
p
o
se
d
m
o
d
e
l
e
m
p
lo
y
e
d
a
m
u
lt
istag
e
m
e
th
o
d
u
n
i
ti
n
g
t
h
e
c
o
n
v
o
l
u
ti
o
n
a
l
n
e
u
ra
l
n
e
two
rk
s
(CNN
)
a
n
d
l
o
n
g
sh
o
rt
-
term
m
e
m
o
ry
(LS
TM
)
n
e
two
r
k
s.
In
t
h
e
p
ro
p
o
se
d
a
p
p
ro
a
c
h
,
v
id
e
o
fra
m
e
s
se
rv
e
a
s
in
p
u
t
to
t
h
e
CNN
,
wh
ich
p
ro
c
e
ss
e
s
th
e
m
to
e
x
trac
t
k
e
y
fe
a
tu
re
s.
Th
e
se
fe
a
tu
re
s
a
re
th
e
n
p
a
ss
e
d
to
a
n
LS
TM
n
e
two
rk
to
c
a
p
tu
re
tem
p
o
ra
l
d
e
p
e
n
d
e
n
c
ies
a
n
d
id
e
n
ti
f
y
a
n
o
m
a
lo
u
s
e
v
e
n
ts
o
v
e
r
ti
m
e
.
Th
is
CNN
-
LS
TM
a
rc
h
it
e
c
tu
re
su
c
c
e
ss
fu
ll
y
d
e
t
e
c
ts
twe
lv
e
d
isti
n
c
t
t
y
p
e
s
o
f
a
n
o
m
a
lo
u
s
a
c
ti
v
it
ies
:
a
b
u
se
,
a
rre
st,
a
rso
n
,
a
ss
a
u
lt
,
b
u
r
g
lary
,
e
x
p
lo
si
o
n
,
fig
h
t,
r
o
a
d
a
c
c
id
e
n
t,
r
o
b
b
e
ry
,
ste
a
li
n
g
,
s
h
o
p
li
fti
n
g
,
a
n
d
v
a
n
d
a
li
sm
.
Th
e
d
a
tas
e
t
is
d
i
v
id
e
d
in
to
p
o
rt
io
n
s
f
o
r
train
in
g
,
tes
ti
n
g
,
a
n
d
v
a
li
d
a
ti
o
n
,
a
lo
n
g
with
c
r
o
ss
-
v
a
li
d
a
ti
o
n
to
e
n
su
re
m
o
d
e
l
g
e
n
e
ra
li
z
a
ti
o
n
.
T
h
e
sy
ste
m
a
c
h
iev
e
s
a
n
a
c
c
u
ra
c
y
o
f
9
8
.
6
%
,
re
flec
ti
n
g
a
sig
n
ifi
c
a
n
t
imp
r
o
v
e
m
e
n
t
o
f
4
-
5
%
o
v
e
r
e
x
isti
n
g
sy
ste
m
s.
Th
is
d
e
m
o
n
stra
tes
th
e
ro
b
u
st
n
e
ss
o
f
th
e
p
ro
p
o
se
d
m
e
th
o
d
i
n
d
e
tec
ti
n
g
a
n
o
m
a
lo
u
s b
e
h
a
v
io
r
in
v
id
e
o
d
a
ta.
K
ey
w
o
r
d
s
:
An
o
m
aly
d
etec
tio
n
C
o
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
s
Dee
p
lear
n
in
g
Hu
m
an
ac
tiv
ity
d
etec
tio
n
L
o
n
g
s
h
o
r
t
-
ter
m
m
e
m
o
r
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
:
Me
g
h
a
G.
Pallewar
Dep
ar
tm
en
t o
f
E
lectr
o
n
ics
an
d
T
elec
o
m
m
u
n
icatio
n
E
n
g
in
ee
r
in
g
,
C
OE
P T
ec
h
n
o
lo
g
ical
Un
iv
er
s
ity
Pu
n
e
4
1
1
0
0
5
,
Ma
h
ar
ash
tr
a,
I
n
d
ia
E
m
ail:
m
s
y
an
n
awa
r
@
g
m
ail.
co
m
1.
I
NT
RO
D
UCT
I
O
N
An
o
m
aly
d
ete
cti
o
n
is
t
h
e
t
ec
h
n
i
q
u
e
co
m
m
o
n
l
y
u
s
ed
t
o
id
en
t
if
y
d
ata
p
at
te
r
n
s
o
r
i
n
c
id
e
n
c
es
t
h
at
d
r
asti
ca
l
ly
d
iv
er
g
e
f
r
o
m
th
e
p
r
e
d
ic
te
d
n
o
r
m
s
.
T
h
ese
an
o
m
alies
m
a
y
r
es
u
l
t
f
r
o
m
ch
a
n
g
e
s
i
n
t
h
e
u
n
d
e
r
l
y
i
n
g
p
h
e
n
o
m
e
n
o
n
o
r
f
r
o
m
th
e
a
p
p
e
ar
an
ce
o
f
p
r
e
v
i
o
u
s
l
y
u
n
s
e
en
s
c
en
ar
io
s
i
n
t
h
e
m
o
n
it
o
r
ed
en
v
i
r
o
n
m
en
t
[
1
]
-
[
3
]
.
T
h
e
p
r
o
ce
s
s
o
f
i
d
e
n
t
if
y
i
n
g
a
n
d
lo
ca
lizi
n
g
a
n
o
m
a
lo
u
s
i
n
s
t
an
ce
s
i
n
s
ec
u
r
it
y
f
o
o
ta
g
e
ca
n
b
e
d
o
n
e
m
a
n
u
all
y
,
s
o
m
e
wh
at
m
ec
h
an
ica
ll
y
,
o
r
e
n
t
ir
el
y
a
u
t
o
m
at
ica
ll
y
.
T
h
e
m
aj
o
r
it
y
o
f
t
r
a
d
iti
o
n
al
s
u
r
v
eill
a
n
ce
s
y
s
te
m
s
a
r
e
o
p
e
r
ate
d
m
a
n
u
al
ly
an
d
o
n
l
y
b
y
h
u
m
a
n
s
.
T
h
is
m
e
th
o
d
d
e
m
a
n
d
s
co
n
s
t
an
t
att
en
ti
o
n
a
n
d
c
o
n
ce
n
t
r
a
ti
o
n
i
n
o
r
d
e
r
t
o
j
u
d
g
e
e
v
e
n
ts
o
r
b
e
h
a
v
i
o
r
to
d
e
te
r
m
in
e
wh
et
h
e
r
t
h
e
y
a
r
e
u
n
u
s
u
al
o
r
s
u
s
p
ic
io
u
s
.
B
u
t
t
h
is
p
r
o
c
e
d
u
r
e
is
q
u
ite
l
a
b
o
r
-
i
n
t
en
s
iv
e
s
in
ce
o
n
e
m
u
s
t
m
o
n
it
o
r
s
e
v
e
r
al
v
i
d
e
o
f
e
ed
s
a
t
t
h
e
s
a
m
e
ti
m
e
.
T
h
e
o
p
e
r
at
o
r
s
'
m
en
tal
l
o
a
d
o
f
te
n
r
e
s
u
lts
i
n
l
o
w
er
f
o
c
u
s
an
d
o
p
e
r
ati
o
n
al
i
n
a
d
e
q
u
a
cies
[
4
]
.
I
n
o
r
d
er
to
e
n
s
u
r
e
p
u
b
lic
s
af
et
y
i
n
u
r
b
a
n
a
r
e
as,
v
i
d
e
o
m
o
n
i
t
o
r
i
n
g
is
c
r
u
cia
l.
V
id
e
o
s
u
r
v
ei
lla
n
c
e
p
la
y
s
a
cr
u
c
ial
r
o
le
i
n
en
s
u
r
i
n
g
t
h
e
s
af
et
y
o
f
i
n
d
i
v
i
d
u
als
.
Vi
d
e
o
d
ata
p
r
o
d
u
ce
d
b
y
s
e
cu
r
i
ty
s
y
s
t
em
s
c
o
n
ta
in
s
a
v
ast
am
o
u
n
t
o
f
v
al
u
a
b
l
e
i
n
f
o
r
m
at
i
o
n
.
T
h
e
o
u
t
p
u
ts
f
r
o
m
t
h
ese
s
y
s
te
m
s
ar
e
r
ic
h
i
n
c
o
n
te
n
t
,
o
f
f
er
i
n
g
i
m
p
o
r
ta
n
t
in
s
i
g
h
ts
e
m
b
e
d
d
e
d
wi
th
in
t
h
e
f
o
o
t
a
g
e
.
T
h
e
n
e
e
d
f
o
r
p
u
b
li
c
s
af
et
y
em
p
h
as
iz
es
th
e
a
n
a
ly
s
i
s
a
n
d
p
r
o
c
ess
i
n
g
o
f
v
i
d
e
o
d
ata
.
T
h
e
p
r
i
m
a
r
y
ai
m
is
t
o
e
n
h
an
ce
s
e
cu
r
i
ty
i
n
p
u
b
l
ic
ar
ea
s
,
wit
h
th
e
o
b
je
cti
v
e
o
f
s
t
r
ea
m
li
n
i
n
g
an
d
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
52
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
,
Vo
l.
41
,
No
.
3
,
Ma
r
ch
20
26
:
1
1
0
5
-
1
1
1
6
1106
s
im
p
li
f
y
i
n
g
t
h
e
tas
k
s
o
f
s
e
c
u
r
it
y
p
er
s
o
n
n
el
.
A
n
al
y
z
in
g
a
n
d
in
te
r
p
r
et
i
n
g
th
is
d
ata
b
e
c
o
m
es
i
n
c
r
e
asi
n
g
ly
im
p
o
r
t
an
t
as
t
h
e
n
ee
d
f
o
r
p
u
b
l
ic
s
a
f
et
y
i
n
c
r
e
ases
.
T
h
e
d
e
v
el
o
p
e
d
s
y
s
te
m
’
s
g
o
al
is
t
o
r
ed
u
ce
t
h
e
w
o
r
k
lo
a
d
f
o
r
s
ec
u
r
it
y
s
e
r
v
i
ce
s
a
n
d
i
m
p
r
o
v
e
p
u
b
l
ic
a
r
e
a
s
ec
u
r
it
y
.
I
n
t
y
p
ic
al
s
u
r
v
e
illa
n
ce
s
ce
n
ar
io
s
,
s
ec
u
r
it
y
p
e
r
s
o
n
n
el
o
b
s
e
r
v
e
a
d
is
p
la
y
co
n
n
e
cte
d
t
o
v
i
d
e
o
ca
m
e
r
as
,
l
o
o
k
i
n
g
f
o
r
an
o
m
al
o
u
s
e
v
en
ts
s
u
c
h
as
t
r
a
f
f
ic
ac
ci
d
e
n
ts
,
f
i
r
es
,
ex
p
l
o
s
i
o
n
s
,
r
o
b
b
er
ies
,
a
n
d
s
ta
m
p
e
d
es
.
H
o
we
v
e
r
,
r
el
y
i
n
g
o
n
h
u
m
an
e
v
al
u
a
ti
o
n
al
o
n
e
is
f
ar
f
r
o
m
i
d
e
al
,
as
i
t
d
e
m
a
n
d
s
s
u
s
t
ai
n
e
d
f
o
c
u
s
o
v
e
r
e
x
te
n
d
e
d
p
er
io
d
s
,
a
tas
k
th
at
is
b
o
t
h
m
e
n
t
all
y
e
x
h
a
u
s
ti
n
g
an
d
p
r
o
n
e
to
e
r
r
o
r
.
T
h
u
s
,
a
n
a
u
t
o
m
ate
d
s
y
s
te
m
is
c
r
u
cia
l
f
o
r
d
ete
cti
n
g
an
d
r
ec
o
g
n
izi
n
g
ab
n
o
r
m
a
l
h
u
m
an
a
ct
iv
i
ties
i
n
r
e
al
-
ti
m
e
,
wit
h
th
e
a
d
d
e
d
ca
p
a
b
i
lit
y
o
f
n
o
tif
y
i
n
g
r
el
ev
an
t
a
u
t
h
o
r
i
tie
s
i
m
m
ed
iat
el
y
.
A
u
t
o
m
at
ic
all
y
m
o
n
it
o
r
i
n
g
C
C
T
V
ca
m
e
r
as
t
o
i
d
e
n
ti
f
y
s
u
s
p
i
ci
o
u
s
a
cti
v
i
ties
a
n
d
g
e
n
e
r
a
ti
n
g
t
im
ely
r
es
p
o
n
s
es
c
an
p
l
a
y
a
v
ita
l
r
o
le
i
n
p
r
ev
en
ti
n
g
ac
c
id
en
ts
o
r
m
iti
g
ati
n
g
d
a
m
a
g
e.
B
y
eli
m
i
n
ati
n
g
t
h
e
n
e
ed
f
o
r
c
o
n
s
ta
n
t
h
u
m
a
n
m
o
n
it
o
r
in
g
,
au
t
o
m
at
e
d
s
y
s
t
em
s
o
f
f
er
a
h
i
g
h
l
y
ef
f
i
ci
en
t
s
o
l
u
ti
o
n
f
o
r
v
i
d
eo
a
n
o
m
al
y
d
e
tec
ti
o
n
,
s
a
v
i
n
g
ti
m
e
a
n
d
l
a
b
o
r
wh
ile
f
r
e
ei
n
g
o
p
er
at
o
r
s
f
r
o
m
te
d
i
o
u
s
t
ask
s
.
As
a
r
esu
lt,
m
a
n
y
r
ese
ar
c
h
e
r
s
h
av
e
s
h
if
te
d
f
o
cu
s
to
s
e
m
i
-
a
u
t
o
m
ati
c
s
u
r
v
ei
lla
n
ce
s
y
s
tem
s
t
h
at
m
i
n
im
iz
e
h
u
m
an
i
n
te
r
v
e
n
t
io
n
.
T
h
ese
s
y
s
te
m
s
c
a
n
m
a
k
e
d
ec
is
io
n
s
i
n
d
e
p
en
d
e
n
t
ly
,
wit
h
o
u
t
r
e
q
u
ir
in
g
c
o
n
ti
n
u
o
u
s
h
u
m
an
o
v
e
r
s
i
g
h
t.
I
n
f
u
ll
y
a
u
t
o
m
at
ed
s
u
r
v
ei
lla
n
c
e,
t
h
e
k
e
y
c
h
alle
n
g
e
g
o
es
b
e
y
o
n
d
r
ec
o
g
n
izi
n
g
a
b
n
o
r
m
al
ev
e
n
ts
in
v
i
d
e
o
s
t
r
ea
m
s
—
i
t
als
o
i
n
v
o
l
v
es
a
cc
u
r
at
el
y
d
et
ec
ti
n
g
a
n
d
l
o
ca
liz
in
g
t
h
es
e
k
in
d
s
o
f
s
i
tu
a
ti
o
n
s
an
d
h
u
m
an
ac
ti
o
n
s
.
T
h
e
ex
tr
ac
t
io
n
o
f
r
o
b
u
s
t
a
n
d
d
is
c
r
i
m
i
n
at
iv
e
f
ea
t
u
r
es
is
ess
e
n
ti
al
t
o
s
o
l
v
i
n
g
t
h
i
s
c
o
m
p
le
x
p
r
o
b
l
em
ef
f
e
cti
v
e
ly
.
D
et
ec
t
in
g
u
n
u
s
u
al
an
d
a
b
n
o
r
m
al
e
v
e
n
ts
p
r
es
en
t
s
s
ev
e
r
al
c
h
al
le
n
g
es
,
l
ar
g
e
ly
d
u
e
t
o
t
h
e
v
a
r
y
i
n
g
p
at
te
r
n
s
i
n
d
i
f
f
e
r
e
n
t
s
ce
n
es.
As
s
ce
n
ar
io
s
a
n
d
a
p
p
li
ca
ti
o
n
s
e
v
o
lv
e,
s
o
d
o
t
h
e
t
y
p
es
o
f
a
n
o
m
a
lies
.
V
id
e
o
an
o
m
al
y
d
et
ec
t
io
n
s
y
s
te
m
s
a
r
e
wi
d
el
y
em
p
l
o
y
e
d
a
c
r
o
s
s
d
i
v
e
r
s
e
s
ec
t
o
r
s
,
in
cl
u
d
in
g
h
e
alt
h
c
ar
e
s
e
r
v
i
ce
s
,
s
e
cu
r
it
y
at
ai
r
p
o
r
ts
,
s
h
o
p
p
i
n
g
m
al
ls
,
o
r
r
ai
lwa
y
s
t
ati
o
n
s
,
a
n
d
s
u
r
v
ei
ll
an
ce
f
o
r
tr
af
f
i
c
m
o
n
it
o
r
i
n
g
o
r
th
e
f
t
d
ete
cti
o
n
[
5
]
-
[
10]
.
A
b
n
o
r
m
al
a
cti
v
it
y
r
ec
o
g
n
iti
o
n
h
as
v
ast
p
o
te
n
ti
al
ac
r
o
s
s
m
a
n
y
f
i
el
d
s
,
es
p
e
cia
ll
y
a
s
d
ata
c
o
n
ti
n
u
es
t
o
ex
p
a
n
d
i
n
c
o
m
p
l
e
x
it
y
an
d
e
n
co
m
p
ass
v
a
r
i
o
u
s
f
o
r
m
s
o
f
i
n
f
o
r
m
a
ti
o
n
.
T
i
m
el
y
d
ete
cti
o
n
o
f
a
b
n
o
r
m
a
liti
es
ca
n
p
r
e
v
e
n
t
m
a
ch
in
e
r
y
f
ail
u
r
es
,
i
m
p
r
o
v
e
p
er
f
o
r
m
a
n
ce
,
c
o
n
t
r
o
l
d
is
ea
s
e
o
u
t
b
r
e
ak
s
,
o
r
e
v
e
n
s
av
e
l
i
v
es
.
A
n
i
n
t
ell
ig
e
n
t
v
i
d
e
o
s
u
r
v
ei
lla
n
c
e
s
y
s
t
em
is
a
l
way
s
u
s
e
f
u
l
f
o
r
m
u
lti
p
l
e
o
p
e
r
a
tio
n
s
ass
o
ci
at
ed
[
1
1
]
.
Du
b
ey
et
a
l
.
[
1
2
]
s
u
g
g
ested
3
D
d
ee
p
m
u
ltip
le
in
s
tan
ce
lear
n
in
g
with
R
esNet
(
MI
L
R
)
tech
n
iq
u
e
u
s
ed
in
th
e
ex
tr
ac
tio
n
o
f
tem
p
o
r
a
l
an
d
s
p
atial
ch
ar
ac
ter
is
tics
f
r
o
m
th
e
v
id
eo
s
.
T
h
e
a
n
o
m
a
ly
s
co
r
e
was
th
en
o
b
tain
ed
em
p
lo
y
in
g
th
ese
f
e
atu
r
es.
Hu
an
g
et
a
l
.
[
1
3
]
c
r
ea
ted
th
e
tem
p
o
r
al
s
p
atial
c
o
n
v
o
lu
ti
o
n
al
n
eu
r
al
n
etwo
r
k
(
T
SC
NN)
,
a
co
n
v
o
lu
t
io
n
al
n
eu
r
al
n
etwo
r
k
(
C
NN)
th
at
ac
ce
p
ts
co
n
tin
u
o
u
s
f
r
am
e
s
eq
u
en
ce
s
as
in
p
u
t.
T
h
e
n
etwo
r
k
was
cr
ea
ted
to
d
etec
t
h
u
m
an
ac
tiv
ity
in
r
ea
l
tim
e
u
s
in
g
3
D
co
n
v
o
lu
tio
n
.
T
h
e
lo
west
d
er
iv
ed
ac
cu
r
ac
y
is
8
1
.
8
%,
wh
ile
th
e
av
er
ag
e
class
if
icatio
n
ac
cu
r
a
cy
is
9
4
.
6
%.
T
h
e
n
,
p
r
o
m
in
en
t
wo
r
k
is
d
o
n
e
b
y
T
an
g
et
a
l
.
[
1
4
]
wh
o
estab
lis
h
ed
a
C
NN
f
o
r
h
u
m
an
ac
tiv
ity
r
ec
o
g
n
itio
n
(
HAR)
th
at
m
ak
es u
s
e
o
f
L
e
g
o
f
ilter
s
.
T
h
e
is
s
u
es
with
s
tan
d
ar
d
C
NN,
s
u
ch
as
th
e
r
eq
u
i
r
em
en
t
f
o
r
tem
p
o
r
al
d
im
en
s
io
n
s
in
p
r
o
c
ess
in
g
u
n
its
(
f
ilter
s
)
an
d
u
n
it
s
h
ar
in
g
ac
r
o
s
s
s
ev
er
a
l
s
en
s
o
r
s
,
ar
e
r
eso
lv
ed
b
y
em
p
lo
y
in
g
L
eg
o
C
NN.
T
h
u
s
,
o
n
e
ca
n
c
r
ea
te
a
m
o
r
e
ef
f
ec
tiv
e
HAR
m
o
d
el
b
y
s
u
b
s
titu
tin
g
L
eg
o
f
ilter
s
f
o
r
th
e
s
tan
d
ar
d
f
ilter
s
.
T
h
e
estab
lis
h
ed
m
o
d
el
is
th
en
ap
p
lied
to
f
iv
e
p
u
b
licly
av
aila
b
le
d
atasets
,
an
d
th
e
o
u
tco
m
es
ar
e
co
n
tr
asted
.
Ku
m
ar
an
d
Sailaja
[
1
5
]
p
r
o
p
o
s
ed
a
d
ee
p
lear
n
in
g
b
ased
ap
p
r
o
ac
h
b
y
u
s
in
g
a
b
i
d
ir
ec
tio
n
al
lo
n
g
s
h
o
r
t
-
ter
m
m
em
o
r
y
(
Bi
-
L
STM
)
n
et
wo
r
k
co
m
b
in
ed
with
s
k
eleto
n
ac
ti
v
ity
f
o
r
ec
asti
n
g
(
SAF)
to
r
e
co
g
n
ize
a
n
o
m
alo
u
s
h
u
m
an
a
ctiv
ities
.
T
h
e
p
o
s
e
esti
m
atio
n
p
er
f
o
r
m
ed
u
s
in
g
th
e
h
u
m
an
s
u
b
ject'
s
s
k
eleto
n
jo
in
t
p
o
in
ts
.
Fu
r
th
e
r
th
e
d
e
v
elo
p
m
en
ts
ar
e
m
ajo
r
ly
f
o
cu
s
ed
o
n
th
e
v
ar
iety
o
f
d
ee
p
lear
n
in
g
tec
h
n
iq
u
es
co
m
b
i
n
atio
n
s
alwa
y
s
p
r
o
g
r
ess
with
t
h
e
s
ca
lab
ilit
y
an
d
ac
cu
r
ac
y
o
f
th
e
s
y
s
tem
.
Few
c
o
m
m
o
n
l
y
u
s
ed
m
eth
o
d
o
lo
g
ies
ar
e
ASR
Net,
C
o
n
v
2
D,
co
n
v
o
lu
tio
n
al
lo
n
g
s
h
o
r
t
-
ter
m
m
em
o
r
y
(
C
o
n
v
L
STM
)
,
B
i
-
L
STM
with
C
NN
[
3
]
,
[
1
6
]
-
[
2
1
]
.
T
h
e
o
u
tco
m
es
co
m
in
g
with
th
ese
m
eth
o
d
o
l
o
g
ies
ar
e
h
av
in
g
lim
itatio
n
s
r
elate
d
with
lim
ited
d
ataset,
co
n
s
tr
ain
ed
s
ce
n
ar
io
f
o
r
d
etec
tio
n
,
lim
ited
f
o
cu
s
ed
ac
tiv
ities
an
d
g
o
o
d
ac
cu
r
ac
y
o
n
ly
f
o
r
s
im
p
le
ac
tiv
it
y
b
ase.
T
h
i
s
r
ev
iew
o
f
th
e
liter
atu
r
e
in
d
icate
s
th
at
au
to
m
atic
v
id
eo
s
u
r
v
eillan
ce
s
y
s
tem
s
wi
th
d
ef
in
ed
an
d
r
es
tr
icted
cr
iter
ia
h
av
e
b
ee
n
r
e
p
o
r
ted
to
h
av
e
g
o
o
d
ac
cu
r
ac
y
.
T
h
er
e
f
o
r
e,
m
ac
h
i
n
e
lear
n
in
g
-
b
ased
class
if
ier
s
with
an
u
n
c
o
n
s
tr
ain
ed
p
ar
a
m
eter
ap
p
r
o
ac
h
ar
e
u
s
ed
in
th
e
cu
r
r
en
t sy
s
tem
.
T
h
e
p
r
im
ar
y
ch
allen
g
es c
an
b
e
id
en
tifie
d
as f
o
llo
ws:
-
T
h
e
ch
allen
g
es
f
ac
ed
with
e
n
v
ir
o
n
m
e
n
tal
co
n
d
itio
n
s
s
u
ch
a
s
lig
h
tin
g
co
n
d
itio
n
s
wh
ich
ch
an
g
es
with
d
ay
-
n
ig
h
t,
in
d
o
o
r
-
o
u
t
d
o
o
r
f
o
o
ta
g
e
in
v
id
eo
.
-
B
ac
k
g
r
o
u
n
d
co
m
p
lex
ity
is
also
a
m
ajo
r
is
s
u
e
.
-
I
t
is
ch
allen
g
in
g
to
id
en
tif
y
a
ctiv
ities
in
v
o
lv
in
g
m
an
y
p
eo
p
le
th
an
ac
tiv
ities
in
v
o
lv
in
g
a
s
in
g
le
p
er
s
o
n
o
r
two
p
eo
p
le
.
-
S
ca
lab
ilit
y
is
s
u
es
ar
is
e
wh
en
m
an
ag
in
g
lar
g
e
-
s
ca
le
s
u
r
v
ei
llan
ce
s
y
s
tem
s
with
n
u
m
er
o
u
s
ca
m
er
as
an
d
en
o
r
m
o
u
s
v
o
lu
m
es o
f
d
ata
.
-
A
ctiv
ities
th
at
lo
o
k
s
im
ilar
(
e.
g
.
,
f
ig
h
tin
g
v
s
.
ab
u
s
in
g
o
r
r
o
b
b
er
y
v
s
.
b
u
r
g
l
ar
y
)
ca
n
b
e
d
if
f
icu
lt
to
d
is
tin
g
u
is
h
.
A
k
e
y
ch
all
en
g
e
i
n
a
b
n
o
r
m
a
l
i
n
ci
d
en
t
class
i
f
i
ca
ti
o
n
is
t
h
e
l
ac
k
o
f
s
u
f
f
ici
en
t
d
a
tase
ts
co
n
t
ai
n
i
n
g
r
el
e
v
a
n
t
v
id
eo
s
.
F
ew
e
x
is
ti
n
g
d
at
ase
ts
i
n
cl
u
d
e
cr
u
cia
l e
v
e
n
t t
y
p
es
s
u
ch
as
r
o
b
b
e
r
y
,
a
b
u
s
e
,
o
r
h
a
r
ass
m
en
t.
Hi
g
h
-
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:
2
5
0
2
-
4
7
52
A
n
in
n
o
v
a
tive
d
ee
p
le
a
r
n
in
g
b
a
s
ed
a
p
p
r
o
a
ch
fo
r
a
n
o
ma
ly
d
e
tectio
n
in
in
tellig
en
t
…
(
Meg
h
a
G.
P
a
llew
a
r
)
1107
q
u
ali
ty
d
a
tase
ts
ar
e
th
e
f
o
u
n
d
a
tio
n
o
f
s
u
cc
ess
f
u
l
n
e
u
r
al
n
et
w
o
r
k
m
o
d
e
ls
,
as
t
h
e
y
e
n
a
b
l
e
th
e
s
e
s
y
s
te
m
s
t
o
lea
r
n
an
d
a
cc
u
r
at
el
y
c
lass
i
f
y
i
n
c
id
e
n
ts
.
T
o
b
u
il
d
an
a
r
c
h
it
ec
t
u
r
e
th
a
t d
eli
v
e
r
s
t
r
u
t
h
f
u
l
r
esu
lts
,
a
d
a
t
ab
ase
m
u
s
t
b
e
u
s
e
d
wh
i
ch
c
o
v
er
s
al
m
o
s
t
al
l
a
b
n
o
r
m
al
e
v
en
ts
.
S
o
t
h
e
c
u
r
r
e
n
t
m
et
h
o
d
o
lo
g
y
e
m
p
l
o
y
s
th
e
U
C
F
C
r
im
e
Da
ta
b
as
e,
wh
i
ch
i
n
cl
u
d
es
1
3
a
b
n
o
r
m
al
ac
ti
v
i
ties
,
s
u
c
h
as
:
a
b
u
s
e,
a
r
r
e
s
t,
ar
s
o
n
,
ass
a
u
l
t,
b
u
r
g
la
r
y
,
e
x
p
l
o
s
i
o
n
,
f
i
g
h
t
,
r
o
a
d
ac
c
id
en
t,
r
o
b
b
e
r
y
,
s
h
o
o
ti
n
g
,
s
teal
in
g
,
s
h
o
p
li
f
ti
n
g
,
a
n
d
v
a
n
d
a
lis
m
.
T
h
is
c
o
m
p
r
e
h
e
n
s
i
v
e
d
atas
et
s
u
p
p
o
r
ts
t
h
e
d
e
v
e
lo
p
m
e
n
t
o
f
a
p
r
o
f
i
cie
n
t
s
u
r
v
eil
la
n
c
e
s
y
s
te
m
c
ap
ab
le
o
f
a
u
t
o
n
o
m
o
u
s
ly
id
e
n
ti
f
y
in
g
s
u
s
p
i
cio
u
s
b
e
h
av
io
r
.
B
y
f
o
c
u
s
i
n
g
o
n
ch
all
en
g
es
r
e
lat
e
d
to
b
a
ck
g
r
o
u
n
d
c
o
m
p
l
ex
it
y
,
ca
m
e
r
a
a
n
g
les
,
a
n
d
e
n
v
i
r
o
n
m
en
t
al
c
o
n
d
iti
o
n
s
,
t
h
e
s
y
s
te
m
le
v
e
r
ag
es
a
v
e
r
s
atil
e
d
a
taset
t
o
im
p
r
o
v
e
p
r
o
ce
s
s
in
g
ac
cu
r
a
cy
.
An
o
m
al
y
d
et
ec
ti
o
n
h
as
b
r
o
a
d
ap
p
l
ic
ati
o
n
s
,
f
r
o
m
i
d
en
ti
f
y
in
g
t
r
a
f
f
ic
ac
ci
d
e
n
ts
t
o
f
la
g
g
i
n
g
s
u
s
p
ic
io
u
s
ac
ti
v
i
ties
.
C
o
n
v
o
lu
tio
n
al
n
e
u
r
a
l
n
e
tw
o
r
k
s
(
C
NNs)
e
x
ce
l
at
d
e
tect
in
g
a
n
o
m
al
o
u
s
ac
ti
v
it
ies
i
n
v
i
d
e
o
s
u
r
v
e
ill
an
ce
b
y
ef
f
i
cie
n
tl
y
l
ea
r
n
in
g
h
ie
r
a
r
c
h
i
ca
l
f
e
at
u
r
es
f
r
o
m
r
aw
d
at
a,
al
lo
wi
n
g
f
o
r
p
r
e
cis
e
i
d
en
ti
f
ic
ati
o
n
o
f
c
o
m
p
l
ex
b
e
h
a
v
i
o
r
s
.
T
h
e
p
r
o
p
o
s
ed
s
y
s
te
m
u
s
es
a
co
n
v
o
l
u
ti
o
n
al
n
e
u
r
al
n
etw
o
r
k
–
l
o
n
g
s
h
o
r
t
-
t
er
m
m
e
m
o
r
y
(
C
NN
-
L
S
T
M
)
m
o
d
el,
w
h
e
r
e
C
NNs
h
a
n
d
le
f
e
atu
r
e
e
x
t
r
ac
t
io
n
a
n
d
t
h
e
l
o
n
g
s
h
o
r
t
-
t
er
m
m
em
o
r
y
(
L
S
T
M
)
m
an
ag
e
class
if
ica
ti
o
n
,
aim
in
g
m
o
s
t
r
ec
en
t
c
h
all
en
g
e
s
in
t
h
es
e
y
e
a
r
s
.
T
h
e
r
est
o
f
th
e
p
ap
er
is
s
tr
u
ctu
r
e
d
as
f
o
l
lo
ws
to
p
r
o
v
id
e
a
co
m
p
r
eh
e
n
s
iv
e
o
v
er
v
iew
o
f
th
e
wo
r
k
.
Sectio
n
2
d
escr
ib
es
th
e
p
r
esen
t
m
eth
o
d
o
lo
g
y
.
Sectio
n
3
elab
o
r
ates
o
n
th
e
ex
p
er
i
m
en
tatio
n
.
Sectio
n
4
d
is
cu
s
s
es
p
er
f
o
r
m
a
n
ce
a
n
aly
s
is
an
d
r
esu
lts
.
Sectio
n
5
p
r
e
s
en
ts
th
e
co
n
clu
s
io
n
an
d
f
u
t
u
r
e
s
co
p
e.
2.
M
E
T
H
O
D
T
h
e
p
r
o
ce
s
s
f
o
r
v
id
e
o
-
b
ased
an
o
m
aly
d
etec
tio
n
with
C
NN
an
d
L
STM
is
s
h
o
wn
in
Fig
u
r
e
1
.
T
h
is
m
eth
o
d
o
l
o
g
y
c
o
m
p
r
is
es si
x
s
tag
es:
-
Data
s
et
s
elec
tio
n
:
th
e
f
ir
s
t
s
tag
e
in
v
o
lv
es
s
elec
tin
g
a
d
ataset
th
at
en
c
o
m
p
ass
es
a
wid
e
v
a
r
i
ety
o
f
ac
tiv
ities
.
Fo
r
th
is
s
tu
d
y
,
th
e
UC
F
cr
im
e
d
atab
ase
is
u
s
ed
.
Ma
in
tain
ed
b
y
th
e
U
n
iv
er
s
ity
o
f
C
en
tr
a
l
Flo
r
id
a
(
UC
F)
an
d
th
e
UC
F
Po
lice
Dep
ar
tm
en
t,
th
is
d
ataset
in
clu
d
es
v
id
e
o
s
o
f
v
a
r
io
u
s
cr
im
in
al
ac
tiv
ities
s
u
ch
as
th
ef
t,
ass
au
lt,
an
d
v
an
d
alis
m
,
ca
p
tu
r
ed
o
n
ca
m
p
u
s
.
T
h
e
d
ataset
is
ex
ten
s
iv
e,
with
ap
p
r
o
x
im
ately
7
5
v
id
e
o
s
p
er
ac
tiv
ity
,
r
esu
ltin
g
in
a
d
iv
er
s
e
co
llectio
n
.
-
Vid
eo
to
f
r
am
e
c
o
n
v
e
r
s
io
n
:
v
id
eo
s
f
r
o
m
th
e
d
ataset
ar
e
co
n
v
er
ted
in
to
in
d
i
v
id
u
al
f
r
a
m
es
f
o
r
f
u
r
th
er
p
r
o
ce
s
s
in
g
.
E
ac
h
ac
tiv
ity
v
id
e
o
g
en
er
ates a
p
p
r
o
x
im
ately
6
,
0
0
0
to
7
,
0
0
0
f
r
am
es.
-
Pre
-
p
r
o
ce
s
s
in
g
:
th
e
r
etr
iev
ed
f
r
am
es
u
n
d
e
r
g
o
s
ev
e
r
al
p
r
e
p
r
o
ce
s
s
in
g
tech
n
iq
u
es
to
e
n
h
an
c
e
q
u
ality
.
T
h
ese
tech
n
iq
u
es
in
cl
u
d
e
e
d
g
e
d
etec
tio
n
,
n
o
is
e
elim
in
atio
n
,
an
d
b
a
ck
g
r
o
u
n
d
r
em
o
v
al.
T
h
e
h
is
to
g
r
am
o
f
o
r
ien
ted
g
r
a
d
ien
ts
(
Ho
G)
is
ap
p
lied
to
t
h
e
p
r
ep
r
o
ce
s
s
ed
f
r
am
es to
o
b
t
ain
f
ilter
ed
im
ag
es.
-
Featu
r
e
ex
tr
ac
tio
n
:
th
e
p
r
ep
r
o
ce
s
s
ed
an
d
f
ilter
ed
im
ag
es
ar
e
th
en
f
ed
in
to
a
m
u
lti
-
s
tag
e
C
NN
-
L
STM
m
o
d
el.
T
h
is
m
o
d
el
u
s
es a
lay
e
r
ed
s
tr
u
ctu
r
e
t
o
ex
tr
ac
t a
n
d
an
aly
ze
d
etailed
f
ea
tu
r
e
attr
ib
u
te
s
.
-
Actio
n
d
etec
tio
n
:
t
h
e
m
u
lti
-
la
y
er
ed
C
NN
-
L
STM
ar
ch
itectu
r
e
p
r
o
ce
s
s
es
th
e
f
ea
tu
r
e
attr
ib
u
t
es
to
d
etec
t
an
d
class
if
y
ac
t
io
n
s
ef
f
ec
tiv
ely
.
-
An
o
m
aly
d
etec
tio
n
:
f
in
ally
,
t
h
e
s
y
s
tem
ev
alu
ates
th
e
p
r
o
ce
s
s
ed
d
ata
to
id
en
tify
an
d
f
lag
an
o
m
al
o
u
s
ac
tiv
ities
b
ased
o
n
th
e
lear
n
e
d
p
atter
n
s
.
T
h
e
s
tu
d
y
f
o
cu
s
es
o
n
an
o
m
al
y
d
etec
tio
n
in
v
id
e
o
s
tr
ea
m
s
,
with
an
e
m
p
h
asis
o
n
id
e
n
tify
i
n
g
c
o
m
p
lex
ac
tiv
ities
.
W
h
ile
p
r
ev
io
u
s
m
eth
o
d
s
h
av
e
ac
h
iev
ed
g
o
o
d
r
e
s
u
lts
,
th
ey
p
r
im
ar
ily
f
o
cu
s
ed
o
n
s
im
p
ler
,
s
in
g
le
-
v
iewe
d
ac
tiv
ities
.
T
h
is
s
tu
d
y
aim
s
to
im
p
r
o
v
e
ac
cu
r
ac
y
in
m
o
r
e
ch
allen
g
in
g
en
v
i
r
o
n
m
en
ts
th
r
o
u
g
h
co
m
p
r
eh
e
n
s
iv
e
ex
p
e
r
im
en
tati
o
n
.
T
h
e
em
p
ir
ical
an
al
y
s
is
i
s
c
ar
r
ied
o
u
t b
y
u
s
in
g
th
e
f
o
llo
wi
n
g
s
tag
es
:
Fig
u
r
e
1
.
T
h
e
p
r
o
p
o
s
ed
m
o
d
el
f
o
r
an
o
m
alo
u
s
ac
tiv
ity
d
etec
tio
n
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
52
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
,
Vo
l.
41
,
No
.
3
,
Ma
r
ch
20
26
:
1
1
0
5
-
1
1
1
6
1108
2
.
1
.
UCF
cri
m
e
da
t
a
ba
s
e
I
n
ter
m
s
o
f
s
ca
le
an
d
c
o
m
p
lex
ity
,
th
e
UC
F
C
r
im
e
Data
b
ase,
cr
ea
ted
b
y
Su
ltan
i
et
a
l.
[
2
2
]
,
r
ep
r
esen
ts
a
s
ig
n
if
ican
t
im
p
r
o
v
em
e
n
t
o
v
er
ea
r
lier
d
ata
b
ases
.
I
t
co
n
tai
n
s
r
aw
C
C
T
V
f
o
o
tag
e
s
h
o
wi
n
g
th
ir
teen
ty
p
ical
r
ea
l
-
life
cr
im
e
s
ce
n
es,
in
clu
d
i
n
g
r
o
b
b
e
r
y
,
ex
p
lo
s
io
n
,
f
ig
h
t,
th
ef
t,
s
h
o
p
liftin
g
,
a
n
d
v
an
d
ali
s
m
.
T
h
e
c
o
llectio
n
in
clu
d
es
ar
o
u
n
d
2
,
000
v
id
e
o
s
co
m
p
r
is
in
g
1
2
8
h
o
u
r
s
o
f
f
o
o
ta
g
e.
T
h
ese
v
id
eo
s
h
av
e
an
av
er
ag
e
r
u
n
tim
e
o
f
f
o
u
r
m
in
u
tes,
wh
ich
m
ak
es
th
e
d
at
aset
m
o
r
e
v
alu
ab
le
f
o
r
class
if
y
in
g
an
o
m
alies
b
as
ed
o
n
v
id
e
o
s
.
T
h
e
av
er
ag
es
o
f
7
,
2
4
7
f
r
am
es
ar
e
g
en
er
ated
f
o
r
ea
ch
v
id
eo
.
T
h
e
d
ataset
is
d
iv
id
ed
in
to
two
s
tr
ea
m
s
as
n
o
r
m
al
r
ec
o
r
d
in
g
s
an
d
r
aw
r
ea
l
-
wo
r
ld
r
ec
o
r
d
in
g
s
.
I
n
itially
,
t
h
e
r
ese
a
r
c
h
co
n
ce
n
t
r
at
e
d
o
n
cl
ass
if
y
i
n
g
a
n
d
p
r
e
d
ic
ti
n
g
i
n
ci
d
e
n
ts
o
f
ab
u
s
e
an
d
f
ig
h
t
in
g
.
As
th
e
s
t
u
d
y
p
r
o
g
r
ess
ed
,
it
e
x
p
a
n
d
e
d
t
o
c
o
v
e
r
s
ix
d
is
t
in
ct
a
cti
o
n
s
a
n
d
u
l
ti
m
at
el
y
e
x
te
n
d
e
d
t
o
in
c
lu
d
e
a
ll
1
2
a
cti
o
n
s
r
e
p
r
ese
n
te
d
in
t
h
e
d
a
tas
et.
T
h
e
r
esu
lt
s
we
r
e
ac
h
ie
v
ed
u
s
in
g
th
e
s
a
m
e
s
et
o
f
al
g
o
r
it
h
m
s
th
r
o
u
g
h
o
u
t
t
h
es
e
d
i
f
f
e
r
e
n
t
r
ese
ar
ch
p
h
ases
.
2
.
2
.
M
ultista
g
e
a
pp
ro
a
ch
wi
t
h la
y
er
ed
CNN
-
L
ST
M
I
n
o
r
d
e
r
t
o
g
et
f
ea
t
u
r
es
f
r
o
m
s
p
a
tia
l
a
n
d
te
m
p
o
r
al
d
at
a
C
NN
a
n
d
L
S
T
M
ar
e
c
o
m
b
in
e
d
as
C
NN
h
a
n
d
les
s
p
atial
f
ea
t
u
r
es
wel
l
an
d
L
ST
M
w
o
r
k
s
p
e
r
f
ec
t
ly
wit
h
t
em
p
o
r
al
f
e
at
u
r
es
[
2
3
]
.
T
h
e
p
r
o
p
o
s
e
d
w
o
r
k
ex
ec
u
te
d
b
y
C
NN
-
L
STM
a
r
c
h
itec
tu
r
e
w
h
ic
h
i
n
cl
u
d
es t
h
r
ee
2
D
c
o
n
v
o
l
u
ti
o
n
l
ay
er
s
a
n
d
t
h
e
n
a
L
S
T
M
l
a
y
e
r
.
T
h
e
ty
p
ic
al
C
NN
a
r
ch
ite
ct
u
r
e
is
s
h
o
w
n
in
F
i
g
u
r
e
2
.
T
h
e
l
ay
e
r
s
o
f
c
o
n
v
o
lu
ti
o
n
al
n
etw
o
r
k
s
h
a
r
e
e
q
u
al
n
u
m
b
er
o
f
f
ilt
e
r
s
an
d
k
e
r
n
e
ls
.
T
h
e
i
n
p
u
t
t
o
th
e
v
er
y
i
n
it
ial
c
o
n
v
o
lu
ti
o
n
lay
e
r
co
n
s
is
ts
o
f
im
ag
es
wit
h
s
ize
o
f
2
2
4
×2
2
4
×
3
(
h
ei
g
h
t,
w
id
th
,
a
n
d
n
u
m
b
er
o
f
ch
an
n
els)
.
T
h
e
i
n
p
u
t
is
f
u
r
t
h
e
r
h
a
n
d
le
d
b
y
m
u
l
ti
p
le
c
o
n
v
o
lu
ti
o
n
al
l
ay
er
s
s
t
ac
k
e
d
to
g
et
h
er
,
wi
th
r
e
cti
f
i
ed
li
n
ea
r
u
n
it
(
R
eL
U
)
s
er
v
i
n
g
as
t
h
e
ac
tiv
ati
o
n
f
u
n
ct
io
n
f
o
r
ea
c
h
l
a
y
er
,
i
n
v
o
l
v
i
n
g
4
,
0
9
6
n
o
d
es.
E
a
ch
la
y
e
r
co
n
t
ai
n
s
1
6
f
il
t
e
r
s
a
n
d
a
k
e
r
n
e
l
o
f
3
×
3
s
i
z
e
is
u
s
e
d
.
E
v
e
r
y
c
o
n
v
o
l
u
t
io
n
l
ay
er
is
f
o
l
lo
we
d
b
y
m
a
x
p
o
o
l
in
g
la
y
e
r
.
T
h
e
C
N
N
in
c
lu
d
es
m
a
x
-
p
o
o
li
n
g
o
p
e
r
a
ti
o
n
s
,
w
h
ic
h
h
el
p
d
e
cr
ea
s
e
t
h
e
d
i
m
en
s
i
o
n
s
o
f
th
e
o
u
t
p
u
t
f
e
at
u
r
e
v
ec
t
o
r
s
,
m
a
k
i
n
g
th
e
r
ep
r
ese
n
ta
ti
o
n
m
o
r
e
m
an
ag
ea
b
l
e
a
n
d
c
o
m
p
u
tati
o
n
a
ll
y
e
f
f
i
ci
en
t.
S
p
at
ial
f
e
at
u
r
es
e
x
t
r
a
ct
ed
b
y
C
N
N
a
r
e
s
u
b
s
eq
u
en
tl
y
s
e
n
t
t
o
L
S
T
M
lay
e
r
s
.
Af
te
r
t
h
e
C
NN
l
a
y
e
r
,
a
L
S
T
M
la
y
e
r
u
s
e
d
wh
i
ch
h
as
3
×
3
k
e
r
n
e
l
s
izes
an
d
h
as
6
4
f
ilt
e
r
s
.
T
h
is
la
y
e
r
c
a
p
t
u
r
es
s
p
ati
o
-
te
m
p
o
r
al
i
n
f
o
r
m
ati
o
n
b
y
p
r
o
c
ess
i
n
g
s
eq
u
e
n
t
ia
l
d
at
a,
w
h
i
c
h
is
ess
e
n
ti
al
f
o
r
v
i
d
eo
-
b
as
ed
an
o
m
al
y
d
et
ec
t
io
n
.
T
h
e
n
etw
o
r
k
ar
ch
it
ec
t
u
r
e
t
h
e
n
c
o
n
s
is
ts
of
a
d
r
o
p
o
u
t
la
y
e
r
,
a
f
l
atte
n
l
a
y
e
r
,
a
n
d
a
d
e
n
s
e
la
y
e
r
.
An
a
d
d
ed
p
ai
r
o
f
d
r
o
p
o
u
t
a
n
d
d
en
s
e
l
ay
er
s
is
u
s
e
d
at
t
h
e
co
n
c
lu
s
io
n
o
f
t
h
e
n
e
tw
o
r
k
.
At
th
e
en
d
o
f
t
h
e
n
e
tw
o
r
k
,
a
d
d
iti
o
n
al
p
ai
r
o
f
d
r
o
p
o
u
t
a
n
d
d
e
n
s
e
la
y
er
s
is
e
m
p
l
o
y
e
d
an
d
t
h
e
s
o
f
tm
ax
ac
ti
v
at
io
n
f
u
n
cti
o
n
is
em
p
l
o
y
e
d
w
it
h
e
v
er
y
d
e
n
s
e
la
y
e
r
.
T
h
is
la
y
er
is
r
es
p
o
n
s
ib
le
f
o
r
m
a
k
i
n
g
th
e
f
i
n
al
c
lass
i
f
i
ca
t
io
n
d
ec
is
i
o
n
s
a
n
d
c
o
n
t
ai
n
s
ap
p
r
o
x
im
ate
ly
1
,
5
0
0
n
o
d
es.
Fig
u
r
e
2
.
T
h
e
C
NN
a
r
c
h
i
tec
tu
r
e
T
a
b
le
s
1
a
n
d
2
s
h
o
w
th
e
o
u
t
p
u
t
s
h
a
p
e
o
f
ea
c
h
l
ay
er
,
d
et
aili
n
g
th
e
o
u
t
p
u
t
s
h
a
p
es
wi
th
t
o
t
a
l
n
u
m
b
e
r
o
f
p
a
r
a
m
e
te
r
s
.
T
h
e
L
ST
M
a
r
c
h
i
t
ec
t
u
r
e
,
s
h
o
w
n
in
Fi
g
u
r
e
3
,
is
a
ty
p
e
o
f
r
e
cu
r
r
e
n
t
n
eu
r
a
l
n
et
w
o
r
k
s
(
R
NNs
)
.
T
h
es
e
ar
e
g
en
e
r
all
y
u
s
e
d
t
o
m
a
n
a
g
e
th
e
v
an
is
h
in
g
g
r
a
d
ie
n
t
p
r
o
b
le
m
i
n
t
r
a
d
it
io
n
a
l
R
NNs
b
y
e
n
a
b
li
n
g
t
h
e
ca
p
t
u
r
e
o
f
ab
i
d
i
n
g
d
e
p
e
n
d
e
n
c
ies
i
n
s
e
q
u
en
t
ial
d
ata
.
E
ac
h
L
STM
ce
l
l
c
o
n
tai
n
s
t
h
r
e
e
g
at
es:
t
h
e
in
p
u
t
g
at
e,
wh
ic
h
d
ec
id
es
wh
a
t
in
f
o
r
m
a
ti
o
n
t
o
a
d
d
;
t
h
e
f
o
r
g
et
g
ate
,
w
h
ic
h
d
et
er
m
i
n
es
wh
a
t
to
d
is
ca
r
d
;
an
d
t
h
e
o
u
t
p
u
t
g
at
e,
w
h
i
c
h
s
ele
cts
wh
a
t
i
n
f
o
r
m
ati
o
n
t
o
p
ass
f
o
r
w
ar
d
.
T
h
es
e
g
ates
,
c
o
n
t
r
o
ll
ed
b
y
s
i
g
m
o
i
d
an
d
t
an
h
ac
t
iv
ati
o
n
s
,
al
lo
w
L
S
T
Ms
t
o
r
et
ai
n
o
r
f
o
r
g
et
i
n
f
o
r
m
at
io
n
o
v
e
r
tim
e.
As
a
r
es
u
l
t,
L
ST
Ms
a
r
e
h
i
g
h
l
y
e
f
f
ec
t
iv
e
f
o
r
tem
p
o
r
al
s
e
q
u
e
n
c
e
m
o
d
eli
n
g
to
d
ete
ct
a
n
o
m
ali
es
wh
i
ch
ar
e
u
s
e
d
f
o
r
v
id
eo
an
al
y
s
is
.
T
e
m
p
o
r
al
d
e
p
e
n
d
e
n
ci
es
ar
e
ca
p
t
u
r
e
d
b
y
t
h
e
L
STM
l
ay
e
r
s
t
h
r
o
u
g
h
th
ei
r
s
e
q
u
e
n
ti
al
p
r
o
ce
s
s
in
g
o
f
t
h
e
s
p
atia
l
in
f
o
r
m
a
ti
o
n
[
1
7
]
,
[
2
4
]
,
[
2
5
]
.
Un
d
e
r
s
t
a
n
d
in
g
lo
n
g
-
r
a
n
g
e
d
ep
e
n
d
e
n
ci
es
i
n
t
h
e
d
ata
is
m
a
d
e
p
o
s
s
i
b
l
e
b
y
th
e
i
n
te
r
n
a
l
m
em
o
r
y
s
t
ate
t
h
a
t
t
h
e
L
ST
M
l
ay
e
r
s
m
ai
n
t
ai
n
.
R
e
g
r
ess
io
n
,
cl
ass
i
f
i
ca
ti
o
n
,
a
n
d
o
t
h
e
r
t
ask
s
ca
n
b
e
p
er
f
o
r
m
e
d
wi
th
th
e
f
in
al
o
u
t
p
u
t
o
f
t
h
e
L
ST
M
lay
e
r
s
.
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:
2
5
0
2
-
4
7
52
A
n
in
n
o
v
a
tive
d
ee
p
le
a
r
n
in
g
b
a
s
ed
a
p
p
r
o
a
ch
fo
r
a
n
o
ma
ly
d
e
tectio
n
in
in
tellig
en
t
…
(
Meg
h
a
G.
P
a
llew
a
r
)
1109
T
ab
le
1
.
L
a
y
er
-
wis
e
s
u
m
m
ar
y
o
f
C
NN
m
o
d
el
La
y
e
r
(
ty
p
e
)
P
a
r
a
me
t
e
r
s
R
e
s
c
a
l
i
n
g
1
0
C
o
n
v
2
D
448
M
a
x
p
o
o
l
i
n
g
2
D
0
C
o
n
v
2
D
1
4640
M
a
x
p
o
o
l
i
n
g
2
D
1
0
C
o
n
v
2
D
2
18496
M
a
x
p
o
o
l
i
n
g
2
D
2
0
F
l
a
t
t
e
n
0
D
e
n
se
3965056
D
e
n
se
1
1548
T
o
t
a
l
p
a
r
a
m
e
t
e
r
s
3990188
T
ab
le
2
.
L
a
y
er
-
wis
e
s
u
m
m
ar
y
o
f
L
STM
m
o
d
el
L
a
y
e
r
(
t
y
p
e
)
N
u
m
b
e
r
o
f
p
a
r
a
m
e
t
e
r
s
I
n
p
u
t
l
a
y
e
r
0
LSTM
1
,
2
8
7
,
1
6
8
LSTM
1
9
7
,
1
2
0
LSTM
4
9
,
4
0
8
F
l
a
t
t
e
n
0
D
e
n
se
4
9
1
,
7
7
6
D
r
o
p
o
u
t
0
D
e
n
se
3
2
,
8
9
6
D
r
o
p
o
u
t
0
D
e
n
se
2
5
8
Fig
u
r
e
3
.
T
h
e
L
STM
m
o
d
el
L
STM
ar
ch
itectu
r
es
ty
p
ically
in
co
r
p
o
r
ate
th
r
ee
f
u
n
d
am
en
tal
ty
p
es
o
f
lo
g
ical
g
ates
th
at
s
er
v
e
as
th
e
b
u
ild
in
g
b
lo
c
k
s
o
f
th
e
s
y
s
tem
.
T
h
e
ess
en
tial
elem
en
ts
wi
th
in
an
L
STM
ar
e
h
t
an
d
c
t
,
wh
ich
p
lay
a
cr
u
cial
r
o
le
in
its
f
u
n
ctio
n
in
g
.
T
h
ese
two
u
n
its
,
h
t
an
d
c
t
,
a
r
e
an
aly
ze
d
an
d
co
m
p
u
ted
u
s
in
g
(
1
)
-
(
5
)
:
Fo
r
g
et
g
ate
:
f
t
=σ
(
W
f
˖
[
h
t−
1
,x
t
]
+
b
f
)
(
1
)
I
n
p
u
t
g
ate
:
i
t
=σ
(
W
i
˖
[
h
t−
1
,x
t
]
+
b
i
)
(
2
)
C˜
t
=ta
n
h
(
W
C
˖
[
h
t−
1
,x
t
]
+b
C
)
C
ell
s
tate
u
p
d
ate
:
C
t
=f
t
*C
t
−
1
+i
t
C˜
t
(
3
)
Ou
tp
u
t
ga
te
:
o
t
=σ
(
W
o
˖
[
h
t−
1
,x
t
]
+b
o
)
(
4
)
Hid
d
en
State
:
h
t
=o
t
*
tan
h
(
C
t
)
(
5
)
Her
e
,
th
e
s
ig
m
o
id
al
f
u
n
ctio
n
i
s
d
en
o
ted
b
y
σ
,
th
e
h
y
p
er
b
o
lic
tan
g
en
t
is
n
o
t
h
in
g
b
u
t
tan
h
,
a
n
d
˖
g
i
v
es
th
e
m
u
ltip
licatio
n
o
f
m
atr
ix
w
eig
h
ts
W
i
an
d
r
esp
ec
tiv
e
g
ate
v
alu
es.
T
h
e
L
STM
m
o
d
el
s
u
m
m
ar
y
is
p
r
o
v
id
ed
in
T
ab
le
2
.
T
o
en
s
u
r
e
c
o
n
tin
u
o
u
s
p
r
o
ce
s
s
in
g
,
th
e
C
NN
-
L
ST
M
m
u
ltis
tag
e
ar
ch
itectu
r
e,
as
s
h
o
wn
in
Fig
u
r
e
4
,
s
tar
ts
b
y
ex
tr
ac
tin
g
f
r
am
es
f
r
o
m
v
id
eo
cli
p
s
,
with
a
th
r
esh
o
l
d
s
et
at
4
0
f
r
am
es.
E
ac
h
im
a
g
e
g
e
n
e
r
ates
f
r
am
es
wit
h
s
p
ec
if
ic
h
eig
h
t
a
n
d
wi
d
t
h
,
wh
ic
h
ar
e
f
e
d
to
t
h
e
n
e
u
r
al
n
et
wo
r
k
.
Du
r
i
n
g
p
r
e
-
p
r
o
c
ess
i
n
g
,
t
h
e
i
n
p
u
t
d
at
a
w
er
e
cr
o
p
p
e
d
a
n
d
a
d
j
u
s
t
e
d
.
T
h
e
to
ta
l d
atas
et
is
d
i
v
i
d
e
d
w
it
h
a
s
p
lit
o
f
7
0
%
f
o
r
t
r
ai
n
i
n
g
,
2
0
%
f
o
r
v
ali
d
at
io
n
,
a
n
d
1
0
%
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
52
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
,
Vo
l.
41
,
No
.
3
,
Ma
r
ch
20
26
:
1
1
0
5
-
1
1
1
6
1110
f
o
r
t
esti
n
g
.
T
h
e
C
NN
m
o
d
el
ex
tr
ac
ts
f
ea
tu
r
es
f
r
o
m
th
e
f
r
am
es,
p
r
o
d
u
cin
g
2
5
,
0
8
8
f
ea
tu
r
es p
er
f
r
am
e,
r
esu
ltin
g
in
4
0
s
ets
o
f
f
ea
tu
r
e
v
ec
to
r
s
.
T
h
ese
v
ec
to
r
s
ar
e
th
en
s
eq
u
en
tially
p
r
o
ce
s
s
ed
b
y
th
e
L
STM
n
etwo
r
k
.
T
h
e
f
i
n
al
o
u
tp
u
t
is
o
b
tain
ed
f
r
o
m
th
e
d
e
n
s
e
lay
er
at
th
e
last
s
tag
e
o
f
th
e
C
NN
-
L
STM
m
o
d
el.
On
c
e
th
e
t
r
ai
n
i
n
g
p
r
o
ce
s
s
was
c
o
m
p
let
e
d
u
s
i
n
g
t
h
e
tr
ai
n
i
n
g
d
at
a,
t
h
e
m
o
d
el
was
e
v
a
lu
a
te
d
o
n
t
h
e
tes
ti
n
g
d
at
a,
a
n
d
t
h
e
a
cc
u
r
ac
y
a
n
d
co
n
f
u
s
io
n
m
at
r
i
x
w
e
r
e
c
o
m
p
u
t
ed
.
Fig
u
r
e
4
.
Sy
s
tem
ar
c
h
it
ec
tu
r
e
o
f
C
NN
-
L
STM
m
o
d
el
3.
E
XP
E
R
I
M
E
N
T
A
T
I
O
N
E
x
ten
s
iv
e
ex
p
er
im
en
tatio
n
w
as
co
n
d
u
cte
d
u
s
in
g
a
Py
th
o
n
3
.
7
.
0
en
v
ir
o
n
m
e
n
t
with
a
2
.
7
GHz
I
n
tel
C
o
r
e
i7
-
4
6
0
0
U
C
PU,
8
GB
R
AM
,
an
d
a
6
4
-
b
it
o
p
er
atin
g
s
y
s
tem
.
T
h
e
test
in
g
in
v
o
lv
ed
d
if
f
er
en
t
f
r
am
e
co
u
n
ts
o
f
2
0
,
4
0
,
an
d
6
0
f
r
am
es.
E
ac
h
ac
tio
n
class
was
tr
ain
ed
wit
h
5
0
f
r
am
es
p
r
i
o
r
t
o
test
in
g
.
T
h
e
ex
p
er
im
en
tatio
n
v
ar
ied
th
e
n
u
m
b
er
o
f
ep
o
ch
s
(
1
0
,
2
0
,
an
d
3
0
)
,
alo
n
g
with
ad
ju
s
tm
en
ts
to
b
atch
s
izes,
f
il
ter
co
u
n
ts
,
an
d
th
e
n
u
m
b
er
o
f
d
en
s
e
lay
er
s
.
Ultim
ately
,
r
esu
lts
wer
e
g
ath
er
ed
u
s
in
g
3
0
ep
o
c
h
s
.
I
n
itially
,
th
e
f
o
cu
s
wa
s
o
n
class
if
y
in
g
two
ac
tiv
ities
,
wh
ich
was
later
ex
p
a
n
d
ed
to
s
ix
ac
tiv
ity
class
if
icatio
n
s
.
T
h
e
m
o
d
el
p
er
f
o
r
m
e
d
ex
ce
p
tio
n
ally
well
with
s
ix
cla
s
s
es.
B
u
ild
in
g
o
n
th
is
s
u
cc
es
s
,
th
e
m
o
d
el
was e
x
ten
d
ed
to
class
if
y
1
2
ac
tiv
ities
,
in
clu
d
in
g
:
ab
u
s
e,
ar
r
est,
ar
s
o
n
,
ass
au
lt,
b
u
r
g
lar
y
,
ex
p
lo
s
io
n
,
f
ig
h
t,
r
o
ad
ac
cid
en
t,
r
o
b
b
er
y
,
s
h
o
o
tin
g
,
s
tealin
g
,
s
h
o
p
liftin
g
,
a
n
d
v
an
d
alis
m
.
Fi
g
u
r
e
5
illu
s
tr
ate
s
th
e
r
esu
lts
f
o
r
ea
ch
class
an
d
th
e
lab
elin
g
o
f
t
h
e
1
2
ac
tiv
ity
class
es,
r
esp
ec
tiv
ely
.
Ad
d
itio
n
ally
,
th
e
m
o
d
el
was te
s
ted
wit
h
in
p
u
ts
o
f
6
0
,
8
0
,
an
d
1
0
0
f
r
a
m
es p
er
class
.
Fig
u
r
e
5
.
T
h
e
1
2
class
ac
tiv
ity
lab
ellin
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:
2
5
0
2
-
4
7
52
A
n
in
n
o
v
a
tive
d
ee
p
le
a
r
n
in
g
b
a
s
ed
a
p
p
r
o
a
ch
fo
r
a
n
o
ma
ly
d
e
tectio
n
in
in
tellig
en
t
…
(
Meg
h
a
G.
P
a
llew
a
r
)
1111
T
h
e
m
eth
o
d
o
lo
g
y
is
wid
ely
r
e
f
er
en
ce
d
f
o
r
s
ce
n
e
an
aly
s
is
an
d
s
u
s
p
icio
u
s
ac
tiv
ity
d
etec
tio
n
[
2
6
]
-
[
3
0
]
.
T
h
e
m
o
d
el’
s
p
e
r
f
o
r
m
an
ce
wa
s
ev
alu
ated
u
s
in
g
th
e
ar
ea
u
n
d
er
th
e
cu
r
v
e
(
AUC)
to
m
ea
s
u
r
e
its
o
v
er
all
ab
ilit
y
to
d
is
tin
g
u
is
h
b
etwe
en
class
es,
an
d
ac
cu
r
ac
y
to
d
eter
m
in
e
th
e
p
r
o
p
o
r
tio
n
o
f
co
r
r
ec
tly
c
lass
if
ied
in
s
tan
ce
s
.
T
ab
le
3
c
o
m
p
a
r
es
th
e
v
i
d
eo
class
if
icatio
n
r
esu
lts
with
v
ar
io
u
s
m
eth
o
d
o
lo
g
ies,
in
clu
d
in
g
SVM,
C
NN
-
B
i
L
STM
,
Den
s
eNe
t,
an
d
R
esNet.
T
h
e
p
r
esen
t
s
y
s
tem
ac
h
iev
e
d
a
p
er
f
o
r
m
a
n
ce
ac
cu
r
ac
y
o
f
9
8
.
6
%
an
d
an
AUC
o
f
9
0
%.
T
h
e
c
o
m
p
a
r
is
o
n
h
ig
h
lig
h
ts
th
at
th
e
cu
r
r
e
n
t
m
u
ltis
tag
e
ap
p
r
o
ac
h
d
eliv
er
s
s
u
p
er
i
o
r
ac
cu
r
ac
y
f
o
r
t
h
e
UC
F
C
r
im
e
d
ataset
co
m
p
ar
ed
to
o
th
er
m
et
h
o
d
o
lo
g
ies
.
T
ab
le
3
.
R
esu
lt o
f
v
id
eo
class
if
icatio
n
M
o
d
e
l
u
se
d
Y
e
a
r
D
a
t
a
s
e
t
A
c
c
u
r
a
c
y
AUC
C
N
N
-
B
i
LS
T
M
[
2
8
]
2023
U
C
F
1
1
U
C
F
5
0
U
C
F
c
r
i
m
e
9
8
.
9
9
6
.
0
4
6
1
.
0
4
--
C
u
b
i
c
S
V
M
K
N
N
[
2
9
]
2021
C
I
F
A
R
1
0
0
9
9
.
2
4
--
V
A
E
[
3
0
]
2021
U
C
S
D
p
e
d
1
A
v
e
n
u
e
--
9
2
.
3
8
2
.
1
D
e
n
se
N
e
t
1
2
1
[
3
1
]
2023
U
C
F
c
r
i
m
e
--
8
6
.
6
3
R
e
s
N
e
t
(
M
I
L
R
)
[
1
2
]
2019
U
C
F
c
r
i
m
e
--
ST
-
G
C
N
[
3
2
]
2023
U
C
S
D
P
e
d
1
U
C
S
D
P
e
d
2
S
h
a
n
g
h
a
i
T
e
c
h
8
9
.
1
-
-
8
5
.
5
9
7
.
9
8
3
.
8
C
N
N
-
L
S
T
M
(
o
u
r
m
o
d
e
l
)
2023
U
C
F
c
r
i
m
e
9
8
.
6
90
4.
RE
SU
L
T
S AN
D
P
E
RF
O
RM
ANCE AN
AL
Y
SI
S
T
h
e
s
y
s
t
em
ca
n
b
e
a
n
a
ly
ze
d
u
s
in
g
d
if
f
e
r
e
n
t
p
a
r
am
ete
r
s
.
T
h
es
e
p
a
r
am
ete
r
s
a
r
e
c
o
m
m
o
n
l
y
u
s
ed
m
et
r
i
cs
th
a
t
elab
o
r
ate
s
p
e
ci
f
ic
all
y
a
b
o
u
t
t
h
e
s
y
s
t
em
p
e
r
f
o
r
m
a
n
c
e
a
n
d
b
e
h
a
v
i
o
r
f
o
r
a
p
p
lie
d
c
o
n
d
iti
o
n
s
.
T
h
e
i
n
cl
u
d
e
d
p
e
r
f
o
r
m
a
n
c
e
p
a
r
a
m
e
te
r
s
a
r
e
ac
c
u
r
a
cy
,
s
p
ec
i
f
i
cit
y
,
r
e
ca
l
l,
an
d
p
r
e
cisi
o
n
.
T
h
is
s
ec
t
io
n
h
el
p
s
t
o
u
n
d
e
r
s
ta
n
d
v
a
r
i
o
u
s
m
et
r
i
cs
,
f
ir
s
tl
y
,
a
n
d
t
h
en
t
h
e
ac
c
u
r
ac
y
,
l
o
s
s
cu
r
v
es
,
an
d
c
o
n
f
u
s
i
o
n
m
at
r
i
x
ar
e
d
is
c
u
s
s
e
d
.
F
u
r
th
e
r
,
t
h
e
p
a
r
a
m
e
tr
i
c
p
l
o
ts
a
r
e
an
al
y
z
ed
t
o
u
n
d
e
r
s
ta
n
d
th
e
o
v
e
r
a
ll
b
eh
a
v
io
r
o
f
th
e
esta
b
l
is
h
e
d
s
y
s
t
em
.
4
.
1
.
E
v
a
lua
t
i
o
n ind
ica
t
o
rs
T
h
e
f
o
llo
win
g
ar
e
t
h
e
ass
ess
m
en
t
m
et
r
i
cs t
o
a
n
a
ly
ze
t
h
e
e
f
f
e
cti
v
e
n
ess
o
f
t
h
e
i
m
p
le
m
e
n
t
ed
m
o
d
el:
Acc
u
r
ac
y
:
I
t
is
a
p
a
r
a
m
e
te
r
u
s
e
d
t
o
m
ea
s
u
r
e
t
h
e
p
r
o
p
o
r
ti
o
n
o
f
co
r
r
ec
tly
id
e
n
tifie
d
a
n
o
m
alies
an
d
n
o
r
m
al
in
s
tan
ce
s
am
o
n
g
th
e
t
o
tal
n
u
m
b
er
o
f
i
n
s
tan
ce
s
.
A
c
c
ura
c
y
=
T
r
ue
Po
s
i
t
i
v
e
+
T
r
ue
N
eg
at
i
v
e
T
o
t
al
S
am
p
l
es
(
6)
Pre
cisi
o
n
:
it
is
r
e
f
er
r
e
d
t
o
as
t
h
e
p
r
o
p
o
r
ti
o
n
o
f
t
r
u
e
a
n
o
m
alies
am
o
n
g
t
h
e
i
n
s
t
an
ce
s
f
la
g
g
e
d
a
s
a
n
o
m
al
o
u
s
b
y
t
h
e
m
o
d
el.
Pr
e
c
ision
=
T
r
ue
Po
s
i
t
i
v
e
T
r
ue
Po
s
i
t
i
v
e
+
T
r
ue
N
eg
at
i
v
e
(
7
)
R
ec
al
l:
th
is
in
d
i
ca
t
o
r
m
e
as
u
r
es
t
h
e
a
b
il
it
y
o
f
t
h
e
s
y
s
t
em
t
o
c
o
r
r
e
ctl
y
id
e
n
ti
f
y
ac
t
u
a
l
an
o
m
al
ie
s
.
R
e
c
a
l
l
=
T
r
ue
Po
s
i
t
i
v
e
T
r
ue
Po
s
i
t
i
v
e
+
F
al
s
e
N
eg
at
i
v
e
(
8
)
F1
-
s
c
o
r
e:
T
h
is
o
n
e
is
t
h
e
h
ar
m
o
n
i
c
m
e
an
o
f
p
r
ec
is
io
n
a
n
d
r
e
ca
ll
,
w
h
ic
h
p
r
o
v
id
es
a
s
t
a
b
l
e
m
et
r
i
c
t
o
m
ea
s
u
r
e
wh
e
n
cl
ass
es
a
r
e
n
o
t
b
a
la
n
c
ed
.
F1
s
c
or
e
=
2
∗
Pr
ec
i
s
i
o
n
∗
Recal
l
Pr
ec
i
s
i
o
n
+
Recal
l
(
9
)
Fals
e
p
o
s
i
ti
v
e
r
at
e
(
FP
R
)
:
th
is
is
p
r
o
p
o
r
t
io
n
o
f
n
o
r
m
al
a
cti
o
n
s
cl
ass
i
f
ie
d
i
n
c
o
r
r
ec
t
ly
as
a
b
n
o
r
m
al
a
cti
o
n
s
.
F
PR
=
F
al
s
e
Po
s
i
t
i
v
e
T
r
ue
N
eg
at
i
v
e
+
F
al
s
e
Po
s
i
t
i
v
e
(
1
0
)
Fals
e
n
e
g
at
iv
e
r
a
te
(
F
NR
)
:
it is
th
e
p
r
o
p
o
r
tio
n
o
f
ac
tu
al
ab
n
o
r
m
alities
in
co
r
r
ec
tly
id
en
tifie
d
as n
o
r
m
al
ac
tio
n
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
52
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
,
Vo
l.
41
,
No
.
3
,
Ma
r
ch
20
26
:
1
1
0
5
-
1
1
1
6
1112
F
N
R
=
T
r
ue
N
eg
at
i
v
e
T
r
ue
N
eg
at
i
v
e
+
F
al
s
e
N
eg
at
i
v
e
(
1
1
)
Ar
e
a
u
n
d
e
r
t
h
e
r
ec
e
i
v
e
r
o
p
e
r
ati
n
g
c
h
a
r
a
cte
r
is
t
ic
cu
r
v
e
(
A
UC
-
R
OC
)
:
T
h
e
ch
a
n
g
es
o
cc
u
r
wit
h
t
r
u
e
p
o
s
iti
v
e
r
at
e
a
n
d
f
a
ls
e
p
o
s
it
iv
e
r
at
e
d
u
e
t
o
v
ar
io
u
s
th
r
esh
o
l
d
v
a
lu
es
a
r
e
g
r
ap
h
ica
ll
y
r
e
p
r
e
s
en
t
e
d
b
y
t
h
e
AUC
-
R
OC
c
u
r
v
es.
T
h
e
v
alu
e
o
f
AU
C
in
d
icate
s
th
e
p
e
r
f
o
r
m
an
ce
o
f
th
e
m
o
d
el.
Hig
h
er
AUC
is
c
o
n
s
id
er
ed
b
etter
as
it d
if
f
er
en
tiates th
e
n
o
r
m
al
an
d
ab
n
o
r
m
al
ac
tio
n
s
s
ig
n
if
ican
t
ly
.
C
o
n
f
u
s
io
n
Ma
tr
ix
:
I
t
is
a
tab
l
e
th
at
p
r
esen
ts
th
e
tr
u
e
p
o
s
itiv
es,
f
alse
p
o
s
itiv
es,
tr
u
e
n
e
g
a
tiv
es,
an
d
f
alse n
eg
ativ
es,
o
f
f
e
r
in
g
a
d
eta
iled
in
s
ig
h
t in
to
th
e
m
o
d
el'
s
p
er
f
o
r
m
a
n
ce
.
4
.
2
.
T
ra
ini
ng
a
nd
v
a
lid
a
t
io
n
a
cc
ura
cy
a
nd
lo
s
s
curv
es
T
h
e
C
NN
-
L
STM
m
o
d
el'
s
p
er
f
o
r
m
a
n
ce
ca
n
b
e
ass
ess
ed
u
s
in
g
v
ar
io
u
s
ty
p
es
o
f
cu
r
v
es;
h
er
e
,
th
e
s
y
s
tem
was
v
er
if
ied
with
tw
o
o
f
th
em
,
tr
ain
in
g
an
d
v
alid
atio
n
ac
cu
r
ac
y
as
well
as
lo
s
s
cu
r
v
es.
B
o
th
th
e
cu
r
v
es
ar
e
r
e
f
er
r
ed
h
e
r
e
t
o
ass
is
t
in
id
en
tify
in
g
an
y
p
r
o
b
le
m
s
s
u
ch
as
o
v
er
f
itti
n
g
o
r
u
n
d
er
f
itti
n
g
b
y
v
is
u
ally
r
ep
r
esen
tin
g
h
o
w
th
e
m
o
d
el
le
ar
n
s
o
v
e
r
tim
e.
At
ev
er
y
e
p
o
c
h
s
(
iter
atio
n
)
in
th
e
tr
ain
in
g
p
r
o
ce
s
s
,
th
e
tr
ai
n
in
g
ac
cu
r
ac
y
c
u
r
v
e
illu
s
tr
ates
th
e
m
o
d
el'
s
ac
cu
r
ac
y
o
n
th
e
tr
ain
i
n
g
d
ata.
T
h
e
m
o
d
el'
s
ac
cu
r
ac
y
o
n
th
e
v
alid
atio
n
d
ataset
—
wh
ich
it
d
o
es
n
o
t
o
b
s
er
v
e
d
u
r
i
n
g
tr
ain
in
g
—
is
r
ep
r
esen
ted
b
y
t
h
e
Valid
atio
n
a
cc
u
r
ac
y
cu
r
v
e
.
I
t
is
em
p
lo
y
ed
to
o
b
s
er
v
e
th
e
m
o
d
el's
ca
p
ac
ity
f
o
r
a
d
ap
tatio
n
.
I
d
ea
lly
,
b
o
th
th
e
tr
ai
n
in
g
an
d
v
alid
atio
n
ac
c
u
r
ac
y
s
h
o
u
ld
in
cr
ea
s
e
o
v
er
tim
e.
T
h
e
tr
ain
in
g
d
ataset'
s
lo
s
s
(
er
r
o
r
)
is
r
ep
r
esen
ted
b
y
t
h
e
tr
ain
in
g
lo
s
s
cu
r
v
e
at
ea
ch
ep
o
ch
s
.
Sy
s
tem
aim
s
to
r
ed
u
ce
th
e
lo
s
s
.
T
h
e
lo
s
s
o
n
th
e
v
alid
atio
n
d
ataset
is
d
is
p
lay
ed
b
y
th
e
v
alid
atio
n
lo
s
s
cu
r
v
e.
B
etter
m
o
d
el
p
er
f
o
r
m
a
n
ce
is
in
d
icate
d
b
y
a
lo
wer
lo
s
s
.
T
h
e
in
f
o
r
m
a
ti
o
n
p
r
o
v
id
e
d
b
y
t
h
e
tr
ai
n
i
n
g
a
n
d
v
al
id
ati
o
n
l
o
s
s
v
al
u
es
is
c
r
u
c
ia
l b
e
ca
u
s
e
i
t h
el
p
s
u
s
i
d
e
n
t
if
y
a
n
y
lea
r
n
i
n
g
is
s
u
es
t
h
at
m
a
y
r
es
u
lt
i
n
a
n
u
n
d
er
f
it
o
r
an
o
v
e
r
f
it
m
o
d
el
b
y
al
lo
wi
n
g
u
s
t
o
s
ee
h
o
w
s
y
s
te
m
s
t
y
p
i
ca
l
b
e
h
a
v
i
o
r
v
e
r
i
f
i
ed
wit
h
t
h
e
ite
r
ati
o
n
o
f
s
u
cc
ee
d
i
n
g
ep
o
c
h
s
.
A
t t
h
e
p
r
ed
icti
n
g
s
ta
g
e
,
t
h
e
y
wil
l a
ls
o
te
ll
u
s
wh
ic
h
e
p
o
c
h
s
t
o
u
t
ili
ze
th
e
t
r
ai
n
i
n
g
m
o
d
el
wei
g
h
ts
.
Fi
g
u
r
e
6
s
h
o
ws
t
h
e
ac
c
u
r
ac
y
cu
r
v
es
f
o
r
t
r
ai
n
i
n
g
a
n
d
v
a
li
d
at
io
n
af
te
r
2
a
n
d
1
0
e
p
o
c
h
s
.
I
t
c
an
b
e
e
asil
y
o
b
s
er
v
ed
t
h
at
th
e
ac
cu
r
a
c
y
g
o
t
i
m
p
r
o
v
e
d
as
th
e
n
u
m
b
er
o
f
ep
o
ch
s
i
n
cr
ea
s
e
s
.
F
ig
u
r
e
7
s
h
o
ws
t
h
e
tr
ai
n
i
n
g
an
d
v
ali
d
ati
o
n
lo
s
s
cu
r
v
es
f
o
r
2
a
n
d
1
0
ep
o
c
h
s
.
T
h
e
l
o
s
s
f
u
n
ct
io
n
d
ec
r
e
ases
wi
th
an
i
n
c
r
e
ase
i
n
ep
o
ch
s
.
Fig
u
r
e
6
.
T
r
ain
in
g
ac
c
u
r
ac
y
a
n
d
v
alid
atio
n
ac
cu
r
ac
y
cu
r
v
es f
o
r
2
a
n
d
1
0
ep
o
ch
s
Fig
u
r
e
7
.
T
r
ain
in
g
l
o
s
s
an
d
v
al
id
atio
n
lo
s
s
cu
r
v
es f
o
r
2
a
n
d
1
0
ep
o
c
h
s
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:
2
5
0
2
-
4
7
52
A
n
in
n
o
v
a
tive
d
ee
p
le
a
r
n
in
g
b
a
s
ed
a
p
p
r
o
a
ch
fo
r
a
n
o
ma
ly
d
e
tectio
n
in
in
tellig
en
t
…
(
Meg
h
a
G.
P
a
llew
a
r
)
1113
4
.
3
.
Co
nfusi
o
n
m
a
t
rix
A
co
n
f
u
s
io
n
m
atr
ix
ass
is
ts
in
th
e
v
is
u
aliza
tio
n
o
f
th
e
o
u
tc
o
m
e
o
f
a
class
if
icatio
n
task
b
y
g
iv
in
g
a
tab
le
s
tr
u
ctu
r
e
o
f
th
e
v
ar
i
o
u
s
p
r
ed
ictio
n
s
a
n
d
o
u
tc
o
m
es.
T
h
is
co
n
f
u
s
io
n
m
atr
i
x
s
h
o
ws
y
o
u
th
e
p
r
o
p
o
r
tio
n
o
f
tim
es
ea
ch
ac
tiv
ity
was
m
i
s
i
d
en
tifie
d
(
f
al
s
e
p
o
s
itiv
es
an
d
f
alse
n
eg
ativ
es)
co
m
p
ar
ed
t
o
th
e
p
r
o
p
o
r
tio
n
o
f
tim
es
it
wa
s
co
r
r
ec
tly
d
etec
ted
(
tr
u
e
p
o
s
itiv
es).
T
h
is
h
as
th
e
p
o
ten
tial
to
b
r
in
g
em
p
h
asis
o
n
p
ar
ticu
lar
ac
tio
n
s
th
at
th
e
m
o
d
el
s
tr
u
g
g
les
to
r
ec
o
g
n
ize.
Fig
u
r
e
8
s
h
o
ws
th
e
co
n
f
u
s
io
n
m
atr
ix
with
in
d
iv
id
u
al
class
-
wis
e
ac
cu
r
ac
y
.
T
h
r
o
u
g
h
w
h
ich
th
e
p
er
ce
n
tag
e
ac
c
u
r
ac
y
o
f
all
1
2
class
es c
an
b
e
id
en
tifie
d
.
T
h
e
m
atr
ix
r
ev
ea
ls
th
at
th
e
h
i
g
h
est
m
is
class
if
icatio
n
o
cc
u
r
s
with
th
e
ass
au
l
t
ac
tio
n
,
wh
ich
is
o
f
ten
m
is
tak
en
f
o
r
ab
u
s
e
o
r
a
f
ig
h
t
.
Similar
ly
,
ab
u
s
e
is
f
r
eq
u
en
tly
m
is
class
if
ied
as
ass
au
lt o
r
f
ig
h
t
.
T
h
is
is
lik
ely
d
u
e
to
th
e
s
im
ilar
ity
in
th
e
ac
tio
n
s
in
v
o
lv
ed
.
Ho
wev
er
,
th
e
p
er
ce
n
tag
e
o
f
ac
cu
r
atel
y
m
atch
in
g
th
e
ac
tu
al
an
d
p
r
ed
icted
ac
tio
n
s
r
em
ain
s
r
elativ
ely
h
ig
h
,
p
r
o
v
id
in
g
a
clea
r
in
d
icatio
n
o
f
th
e
i
m
p
lem
en
ted
m
o
d
el'
s
p
er
f
o
r
m
an
ce
ac
r
o
s
s
th
e
v
ar
io
u
s
ac
tiv
ities
.
Fig
u
r
e
8
.
C
o
n
f
u
s
io
n
m
atr
i
x
f
o
r
1
2
ac
tiv
ity
r
ec
o
g
n
itio
n
4
.
4
.
P
er
f
o
r
m
a
nce
pa
ra
m
et
e
rs a
nd
t
heir
plo
t
s
Pre
cisi
o
n
,
r
ec
all,
F1
s
co
r
e,
an
d
ac
c
u
r
ac
y
ar
e
f
r
eq
u
e
n
t
ly
r
ep
r
esen
ted
g
r
a
p
h
ically
i
n
o
r
d
er
to
d
em
o
n
s
tr
ate
h
o
w
well
th
e
cla
s
s
if
icatio
n
m
o
d
els
ar
e
p
er
f
o
r
m
in
g
.
I
t
is
s
im
p
ler
to
u
n
d
er
s
t
an
d
an
d
ev
alu
ate
t
h
e
p
er
f
o
r
m
an
ce
o
f
class
if
icatio
n
m
o
d
els
ac
r
o
s
s
s
ev
er
al
class
es
wh
en
p
r
ec
is
io
n
,
r
ec
all,
F1
s
co
r
e,
an
d
ac
c
u
r
ac
y
ar
e
s
h
o
wn
g
r
ap
h
ically
.
T
h
ey
p
r
o
v
id
e
an
illu
s
tr
ativ
e
o
v
er
v
iew
o
f
a
m
o
d
el'
s
s
tr
en
g
th
s
an
d
p
o
ten
tial
ar
ea
s
f
o
r
im
p
r
o
v
em
e
n
t.
T
h
e
class
-
w
is
e
co
m
p
a
r
is
o
n
is
s
h
o
w
n
i
n
Fi
g
u
r
es
9
a
n
d
1
0
.
A
co
m
p
ar
ativ
e
a
n
aly
s
is
with
ex
is
tin
g
s
y
s
tem
s
is
p
r
esen
ted
,
as illu
s
tr
ated
in
Fig
u
r
e
1
1
.
Fig
u
r
e
9
.
C
lass
-
wis
e
p
er
f
o
r
m
a
n
ce
p
ar
am
eter
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
52
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
,
Vo
l.
41
,
No
.
3
,
Ma
r
ch
20
26
:
1
1
0
5
-
1
1
1
6
1114
Fig
u
r
e
1
0
.
C
lass
wi
s
e
lin
e
g
r
ap
h
f
o
r
p
r
ec
is
io
n
,
r
ec
all,
ac
cu
r
a
cy
an
d
F1
s
co
r
e
Fig
u
r
e
1
1
.
C
o
m
p
ar
is
o
n
with
o
th
er
an
o
m
al
y
d
etec
tio
n
s
y
s
tem
s
T
h
e
UC
F
C
r
im
e
d
ataset
i
s
m
o
r
e
co
m
p
lex
s
in
ce
it
g
iv
es
a
v
ar
iety
o
f
ac
tio
n
s
an
d
v
id
eo
k
in
d
s
f
o
r
ea
ch
ac
tio
n
,
in
co
n
tr
ast
to
o
th
er
p
u
b
licly
av
ailab
le
d
atasets
th
at
o
n
l
y
in
cl
u
d
e
v
id
eo
s
with
f
ix
e
d
s
tan
d
p
o
in
ts
,
u
n
if
o
r
m
b
ac
k
d
r
o
p
s
,
an
d
u
s
u
al
an
g
les.
T
h
is
ch
allen
g
i
n
g
UC
F
C
r
im
e
d
ataset
is
u
s
ed
to
ass
ess
th
e
C
NN
-
L
ST
M
m
o
d
el'
s
ex
p
er
im
en
tal
o
u
tc
o
m
es.
T
h
e
UC
F
C
r
im
e
d
ataset
r
ec
o
r
d
s
cr
im
e
ac
tiv
ity
in
r
ea
l
tim
e,
ca
p
tu
r
ed
b
y
C
C
T
V
ca
m
er
as
in
u
n
s
tr
u
ctu
r
e
d
s
u
r
r
o
u
n
d
in
g
s
,
w
h
ile
o
th
e
r
d
atasets
co
n
ce
n
tr
ate
o
n
f
r
eq
u
e
n
t
ac
tio
n
s
.
B
ec
au
s
e
o
f
its
d
iv
er
s
ity
,
it
s
er
v
es
a
s
a
u
s
ef
u
l
b
aselin
e
f
o
r
ev
alu
atin
g
an
o
m
aly
d
etec
tio
n
m
eth
o
d
s
'
p
er
f
o
r
m
an
ce
in
p
r
ac
tical
s
ettin
g
s
.
T
h
e
p
r
ese
n
t
m
o
d
el
w
o
r
k
s
e
x
c
ell
en
tl
y
f
o
r
m
ax
im
u
m
class
es
f
o
r
t
h
e
t
ask
o
f
a
n
o
m
al
y
d
et
ec
ti
o
n
.
T
h
e
p
e
r
f
o
r
m
an
ce
is
im
p
r
o
v
e
d
wit
h
d
if
f
i
cu
lt
ac
ti
v
it
y
d
ata
b
as
e.
Evaluation Warning : The document was created with Spire.PDF for Python.