T
E
L
KO
MNIK
A
, V
ol
.
17
,
No.
6,
Dec
e
mb
er
20
1
9,
p
p.3
07
3~
30
85
IS
S
N: 1
69
3
-
6
93
0
,
accr
ed
ited
F
irst
Gr
ad
e b
y K
em
en
r
istekdikti,
Decr
ee
No: 2
1/E/
K
P
T
/20
18
DOI:
10.12928/TE
LK
OM
N
IK
A
.v
1
7
i
6
.
12689
◼
30
73
Rec
ei
v
ed
Ma
r
c
h
15
,
20
1
9
;
Rev
i
s
ed
J
u
l
y
2
,
20
1
9
;
A
c
c
ep
ted
J
u
l
y
18
, 2
0
1
9
Com
pu
ter
visi
o
n
for
pu
ri
ty, p
he
no
l, a
nd
pH
d
et
ectio
n
of
L
u
w
ak
C
of
f
e
e
green
bea
n
Y
u
suf
Hend
r
aw
an*
1
,
S
h
int
a W
id
ya
n
ing
t
ya
s
2
,
S
u
cipt
o
S
u
cipt
o
3
1
Dep
a
rt
m
e
n
t
o
f
Ag
r
i
c
u
l
tu
r
a
l
En
g
i
n
e
e
r
i
n
g
,
Ag
ri
c
u
l
tu
ra
l
T
e
c
h
n
o
l
o
g
y
F
a
c
u
l
ty
,
Uni
v
e
r
s
i
ta
s
Bra
wi
j
a
y
a
,
Ve
te
ra
n
S
t.
,
M
a
l
a
n
g
,
In
d
o
n
e
s
i
a
,
te
l
p
/f
a
x
:
+
6
2
-
3
4
1
-
5
8
0
1
0
6
/+
6
2
-
341
-
5
6
8
9
1
7
2
Dep
a
rt
m
e
n
t
o
f
Ag
r
o
i
n
d
u
s
t
ri
a
l
Te
c
h
n
o
l
o
g
y
,
Ag
ri
c
u
l
t
u
ra
l
T
e
c
h
n
o
l
o
g
y
F
a
c
u
l
ty
,
Uni
v
e
rs
i
ta
s
Br
a
wij
a
y
a
,
Ve
te
ra
n
S
t.
,
M
a
l
a
n
g
,
In
d
o
n
e
s
i
a
3
Hal
a
l
a
n
d
Qu
a
l
i
fi
e
d
I
n
d
u
s
tr
y
Dev
e
l
o
p
m
e
n
t
(HAL
-
Q
I
D),
Un
i
v
e
rs
i
ta
s
Bra
wi
j
a
y
a
,
Ve
te
ra
n
S
t.
,
M
a
l
a
n
g
,
In
d
o
n
e
s
i
a
*C
o
rre
s
p
o
n
d
i
n
g
a
u
th
o
r,
e
-
m
a
i
l
:
y
u
s
u
fh
e
n
d
ra
wa
n
@
g
m
a
i
l
.c
o
m
1
,
s
h
i
n
ta
wid
y
a
1
4
3
5
@g
m
a
i
l
.
c
o
m
2
,
c
i
p
to
ti
p
@ub
.
a
c
.
i
d
3
Ab
strac
t
Com
p
u
te
r v
i
s
i
o
n
a
s
a
n
o
n
-
i
n
v
a
s
i
v
e
b
i
o
-
s
e
n
s
i
n
g
m
e
th
o
d
p
r
o
v
i
d
e
d
o
p
p
o
rtu
n
i
t
y
t
o
d
e
te
c
t
p
u
r
i
t
y
,
t
o
ta
l
p
h
e
n
o
l
,
a
n
d
p
H
i
n
L
u
wak
c
o
ff
e
e
g
re
e
n
b
e
a
n
.
Th
i
s
s
tu
d
y
a
i
m
e
d
to
o
b
ta
i
n
th
e
b
e
s
t
Arti
f
i
c
i
a
l
Neu
ra
l
Net
wor
k
(ANN
)
m
o
d
e
l
t
o
d
e
te
c
t
th
e
p
e
r
c
e
n
t
a
g
e
o
f
p
u
ri
t
y
,
to
ta
l
p
h
e
n
o
l
,
a
n
d
p
H
o
n
L
u
wak
c
o
ff
e
e
g
re
e
n
b
e
a
n
b
y
u
s
i
n
g
c
o
l
o
r
f
e
a
tu
r
e
s
(re
d
-
g
re
e
n
-
b
l
u
e
,
g
ra
y
,
h
u
e
-
s
a
t
u
ra
t
i
o
n
-
v
a
l
u
e
,
h
u
e
-
s
a
tu
ra
t
i
o
n
-
l
i
g
h
tn
e
s
s
,
L
*a*b
*
),
a
n
d
Har
a
l
i
c
k
te
x
tu
r
a
l
fe
a
tu
re
s
wi
th
c
o
l
o
r
c
o
-
o
c
c
u
rre
n
c
e
m
a
tri
x
i
n
c
l
u
d
i
n
g
e
n
tro
p
y
,
e
n
e
r
g
y
,
c
o
n
tr
a
s
t,
h
o
m
o
g
e
n
e
i
t
y
,
s
u
m
m
e
a
n
,
v
a
r
i
a
n
c
e
,
c
o
rre
l
a
ti
o
n
,
m
a
x
i
m
u
m
p
ro
b
a
b
i
l
i
ty
,
i
n
v
e
rs
e
d
i
ff
e
r
e
n
c
e
m
o
m
e
n
t,
a
n
d
c
l
u
s
te
r
te
n
d
e
n
c
y
.
Th
e
b
e
s
t
ANN
s
tru
c
tu
re
wa
s
(5
i
n
p
u
t
s
;
3
0
n
o
d
e
s
i
n
h
i
d
d
e
n
l
a
y
e
r
1
;
4
0
n
o
d
e
s
i
n
h
i
d
d
e
n
l
a
y
e
r
2
;
a
n
d
3
o
u
tp
u
t
s
) wh
i
c
h
had
tra
i
n
i
n
g
m
e
a
n
s
q
u
a
re
e
rro
r (M
SE)
o
f
0
.0
0
8
5
a
n
d
v
a
l
i
d
a
ti
o
n
M
SE
o
f
0
.0
4
4
2
.
Key
w
ords
:
a
rt
i
fi
c
i
a
l
n
e
u
ra
l
n
e
t
work
,
c
o
m
p
u
te
r
v
i
s
i
o
n
,
L
u
wak
c
o
ff
e
e
Copy
righ
t
©
2
0
1
9
Uni
v
e
rsi
t
a
s
Ahm
a
d
D
a
hl
a
n.
All
rig
ht
s
r
e
s
e
rve
d
.
1.
Int
r
o
d
u
ctio
n
In
th
e
l
as
t
tw
o
d
ec
ad
es
,
g
l
ob
a
l
c
off
ee
c
o
ns
um
p
ti
o
n
g
r
owth
h
as
c
on
t
i
n
ue
d
to
gro
w
,
as
driv
en
by
c
of
fee
-
b
as
ed
pr
od
uc
ts
an
d
b
ev
erage
for
mu
l
ati
on
s
an
d
th
e
i
nc
r
ea
s
i
ng
n
um
b
er
o
f
c
off
ee
s
h
op
s
[1
]
.
O
ne
ty
pe
of
c
off
ee
,
k
n
own
to
be
ex
pe
ns
i
v
e
an
d
r
are
i
n
t
he
w
orld
i
s
Lu
wak
(
c
i
v
et)
c
off
ee
[2]
.
A
s
a
h
i
g
h
-
pric
ed
c
om
mo
d
i
ty
,
Lu
w
a
k
c
off
ee
i
s
prone
to
b
e
mi
x
ed
wi
t
h
r
eg
u
l
ar
c
off
ee
be
a
ns
.
A
t
pres
en
t
,
an
i
nte
r
n
ati
on
a
l
l
y
r
ec
o
gn
i
z
ed
m
eth
o
d
of
di
s
t
i
ng
ui
s
h
i
n
g
Lu
w
ak
an
d
r
eg
ul
ar
c
off
ee
,
r
e
ma
i
ns
un
n
oti
c
ed
.
T
hi
s
,
th
erefore
g
i
v
e
s
the
op
p
ortun
i
ty
to
de
s
i
gn
a
s
i
m
p
l
e,
fas
t,
ac
c
urate,
an
d
n
on
-
de
s
tr
uc
t
i
v
e
eq
ui
p
me
nt
,
c
a
pa
b
l
e
of
d
ete
c
t
i
ng
the
pe
r
c
e
nta
g
e
of
mi
x
ed
p
orti
o
n
be
twe
en
Lu
w
ak
c
off
ee
an
d
r
eg
ul
ar
c
off
e
e.
T
he
s
tud
y
r
es
ul
ts
of
J
u
mh
aw
an
[3]
fou
nd
ou
t
th
at
the
tas
tes
of
r
oa
s
te
d
Lu
w
ak
c
off
e
e
an
d
r
eg
ul
ar
r
oa
s
ted
c
off
e
e
ar
e
c
i
tr
i
c
ac
i
d
a
nd
ma
l
i
c
ac
i
d.
Res
ea
r
c
h
o
n
th
e
Lu
wak
c
off
ee
gree
n
be
an
ha
s
n
ev
er
be
e
n
c
on
d
uc
ted
,
a
l
be
i
t
ab
ou
t
75%
of
Ind
o
ne
s
i
a
n
c
of
fee
ex
po
r
ts
are
i
n
t
he
form
of
gree
n
b
e
an
.
I
n
ad
d
i
ti
on
to
de
t
ec
ti
n
g
Lu
w
ak
c
off
e
e
mi
x
tures
i
n
r
eg
u
l
ar
c
off
e
e,
thi
s
s
tu
dy
a
l
s
o
me
as
ure
s
t
ota
l
p
he
n
ol
as
a
n
an
t
i
ox
i
d
an
t
an
d
pH
to
me
as
ure
the
c
off
e
e
ac
i
di
ty
.
Coffe
e
b
ec
om
es
a
s
ou
r
c
e
of
a
nti
ox
i
d
an
ts
to
war
d
off
fr
ee
r
a
di
c
a
l
s
tha
t
ar
e
be
ne
f
i
c
i
a
l
for
h
ea
l
th.
T
he
l
arg
es
t
an
t
i
ox
i
da
nt
c
om
po
ne
nt
i
n
c
o
ffe
e
i
s
p
he
no
l
[4
,
5]
.
A
t
pres
en
t,
c
on
s
um
pti
on
of
green
be
an
ex
tr
ac
t
b
ec
o
me
s
a
ne
w
tr
e
nd
du
e
to
i
ts
l
ow
c
al
orie
c
on
ten
t
[6]
.
Me
as
urin
g
tot
al
p
he
n
ol
i
n
gree
n
b
ea
n
,
he
l
ps
me
as
ure
th
e
a
nti
ox
i
da
nt
ac
ti
v
i
ty
.
In
ad
d
i
ti
on
,
c
of
fee
ha
s
an
ac
i
di
c
tas
te
th
at
i
s
i
d
en
t
i
c
al
to
i
ts
pH
c
on
te
nt
.
T
he
tr
e
nd
of
c
on
s
u
mi
n
g
green
be
a
ns
ex
tr
ac
t
r
eq
ui
r
es
a
s
tud
y
of
pH
,
du
e
t
o
c
on
s
um
er
s
en
s
i
t
i
v
i
ty
of
c
off
ee
ac
i
d
i
ty
,
es
pe
c
i
a
l
l
y
i
n
arab
i
c
a
c
off
ee
.
T
hi
s
r
es
e
arc
h
i
s
uti
l
i
z
e
d
a
s
on
e
of
t
he
s
t
ag
es
i
n
de
s
i
gn
i
ng
t
oo
l
s
for
c
off
ee
i
ns
pe
c
t
i
on
.
Com
pu
t
er
v
i
s
i
on
t
ec
hn
o
l
o
gy
ha
s
be
en
wi
d
el
y
a
pp
l
i
e
d
i
n
i
d
en
ti
fy
i
ng
an
d
c
op
y
i
ng
c
off
e
e
as
an
ex
a
mp
l
e
o
f
r
es
ea
r
c
h
as
c
on
d
uc
ted
by
O
l
i
v
ei
r
a
[7]
,
ap
pl
y
i
ng
c
o
mp
u
ter
v
i
s
i
on
an
d
c
om
p
uta
t
i
o
na
l
i
nte
l
l
i
g
en
c
e
t
o
c
l
as
s
i
fy
gre
en
b
ea
n
c
off
e
e.
T
h
e
r
es
u
l
ts
s
ho
w
the
pe
r
forma
nc
e
of
c
o
mp
u
ter
v
i
s
i
o
n
w
hi
c
h
ac
hi
ev
es
c
l
as
s
i
fi
c
at
i
on
ac
c
urac
y
of
up
to
10
0%.
Nans
en
[8]
ap
p
l
i
ed
c
o
mp
ute
r
v
i
s
i
on
by
us
i
n
g
hy
p
e
r
s
pe
c
tr
al
i
ma
g
i
n
g
to
i
d
en
t
i
fy
c
om
me
r
c
i
al
r
oa
s
ted
c
off
ee
bran
ds
ba
s
ed
o
n
th
ei
r
q
ua
l
i
ty
.
C
ap
oras
o
[9]
de
t
ec
t
mo
i
s
ture
c
o
nte
nt
i
n
s
i
ng
l
e
green
be
a
n
c
off
e
e
by
u
s
i
ng
c
om
pu
ter
v
i
s
i
on
.
T
he
r
es
ul
ts
s
ho
w
op
t
i
m
al
r
es
u
l
t
s
for
mo
i
s
t
ure
Evaluation Warning : The document was created with Spire.PDF for Python.
◼
IS
S
N: 16
93
-
6
93
0
T
E
L
KO
MNIK
A
V
ol
.
17
,
No
.
6,
D
ec
em
b
er 20
19
:
30
7
3
-
3085
3074
c
on
ten
t
d
ete
c
ti
on
i
n
s
i
ng
l
e
gree
n
be
a
n
c
off
ee
an
d
s
uc
c
es
s
ful
l
y
c
l
as
s
i
fy
th
e
ty
pe
s
of
c
off
ee
(
A
r
ab
i
c
a
an
d Ro
bu
s
ta
)
.
Na
v
arr
o
[10]
em
pl
oy
e
d
d
i
g
i
ta
l
i
ma
g
i
ng
te
c
hn
ol
og
y
t
o m
od
el
t
he
qu
a
l
i
ty
of
c
off
ee
du
r
i
ng
t
he
r
oa
s
t
i
n
g
proc
es
s
.
T
he
r
es
ul
ts
pres
en
t
go
od
p
erfor
ma
nc
e
i
n
us
i
ng
a
c
om
bi
na
t
i
on
of
di
gi
t
al
i
ma
gi
ng
wi
t
h
ad
ap
t
i
v
e
ne
tw
ork
ba
s
ed
fuz
z
y
i
nfe
r
e
nc
e
s
y
s
tem
s
(
A
NF
IS
)
to
mo
n
i
tor
c
off
ee
c
o
l
or
d
urin
g
the
r
o
as
ti
n
g
proc
es
s
.
T
he
us
e
of
arti
f
i
c
i
a
l
i
nte
l
l
i
g
en
t
mo
de
l
i
ng
s
uc
h
as
arti
fi
c
i
al
ne
ura
l
n
etwo
r
k
(
A
NN)
ha
s
be
en
s
uc
c
es
s
ful
l
y
ap
pl
i
ed
i
n
v
ari
ou
s
c
off
e
e
i
de
nti
fi
c
at
i
o
n
s
tud
i
es
[1
1,
1
2]
.
Ho
wev
er,
t
he
r
e
ha
v
e
be
e
n
no
s
tu
di
es
tha
t
h
av
e
ob
s
erv
ed
l
i
gh
t
c
o
mp
ut
er
v
i
s
i
on
an
d
art
i
f
i
c
i
a
l
i
n
tel
l
i
ge
nt
m
o
de
l
i
ng
pe
r
for
m
an
c
e
to
i
de
nti
fy
th
e
pu
r
i
ty
of
gr
ee
n
b
ea
n
c
off
ee
f
or
Lu
wak
c
off
ee
ty
p
e.
Ima
g
e
an
al
y
s
i
s
i
s
i
de
n
ti
f
i
ed
as
a
fas
t,
no
n
-
de
s
tr
u
c
ti
v
e
an
d
l
ow
-
c
os
t
m
eth
o
d
for
as
s
es
s
i
ng
t
he
qu
a
l
i
ty
of
foo
d
prod
uc
ts
[1
3,
14
]
.
A
c
c
ordi
ng
to
P
at
el
[1
5],
m
ac
hi
n
e
v
i
s
i
on
de
v
el
op
m
en
t
i
s
ba
s
ed
o
n
the
i
ns
pe
c
t
i
on
of
th
e
foo
d
qu
a
l
i
ty
a
nd
a
gric
u
l
t
ural
pro
du
c
ts
,
un
fortu
na
t
el
y
fac
e
d
s
ev
era
l
ob
s
tac
l
es
whi
c
h
l
at
er
r
eq
u
i
r
es
s
uc
h
an
ac
c
urate,
f
as
t
an
d
ob
j
ec
ti
v
e
tec
hn
i
qu
e
i
n
d
ete
r
m
i
n
i
ng
t
he
qu
al
i
ty
of
the
m
ea
s
ure
d
ma
ter
i
a
l
.
T
hi
s
tec
hn
ol
og
y
ap
pe
ars
i
n
the
d
ev
el
o
pm
en
t
o
f
au
tom
ate
d
ma
c
h
i
ne
r
y
i
n
the
ag
r
i
c
ul
ture
an
d
f
oo
d
i
n
du
s
tr
i
es
[16]
.
S
ev
era
l
s
tud
i
es
[17
-
2
0]
de
p
i
c
t
op
t
i
ma
l
r
es
ul
ts
i
n
m
ac
hi
ne
v
i
s
i
on
a
pp
l
i
c
at
i
o
n
w
he
n
us
i
ng
a
c
om
bi
n
ati
on
of
A
NN
mo
d
el
i
ng
wi
th
c
o
l
or
fea
t
ures
(
RG
B
,
grey
,
H
S
L,
HS
V
,
L*
a
*
b
*
)
an
d
H
aral
i
c
k
t
ex
tural
fea
ture
[2
1
].
In
thi
s
s
tud
y
,
th
e
gree
n
be
a
n
i
ma
ge
d
ata
as
de
r
i
v
ed
fr
o
m
a
mi
x
tur
e
of
Lu
w
ak
c
off
e
e
an
d
r
eg
ul
ar
c
off
ee
are
i
de
nti
f
i
e
d
by
us
i
ng
c
ol
or
fea
t
ures
,
s
uc
h
as
:
Red
(RGB),
G
r
ee
n
(
RGB)
,
B
l
ue
(RG
B)
,
grey
,
Hu
e,
S
a
turat
i
on
(
HSL)
,
Li
gh
t
ne
s
s
(HSL)
,
S
atu
r
at
i
on
(
HS
V
)
,
V
a
l
ue
(H
S
V
)
,
L*
,
a*
,
b
*
,
an
d
tex
tura
l
fea
tures
i
n
ea
c
h
ty
pe
o
f
c
ol
or
(
i
nc
l
u
di
ng
e
ntropy
,
en
ergy
,
c
on
tr
as
t,
ho
m
og
e
ne
i
t
y
,
s
um
me
an
,
v
aria
nc
e,
c
orr
e
l
at
i
on
,
ma
x
i
mu
m
proba
bi
l
i
ty
,
i
nv
ers
e
di
ff
erent
mo
m
en
t
a
nd
c
l
us
ter
ten
d
en
c
y
)
.
In
ad
d
i
ti
on
,
a
l
l
t
he
c
ol
or
an
d
tex
tura
l
fea
t
ures
i
n
th
i
s
s
tud
y
to
s
el
ec
t
the
b
es
t
fea
ture
-
s
ub
s
e
t
c
om
bi
na
t
i
on
are
c
l
as
s
i
f
i
e
d
by
us
i
ng
t
he
f
ea
t
ure
s
el
ec
ti
on
me
t
ho
d
(
f
i
l
ter
me
tho
d)
be
fore
be
i
ng
us
ed
as
i
n
pu
t
i
n
A
NN
mo
d
el
i
ng
.
T
he
s
el
ec
t
ed
c
o
l
or
a
nd
t
ex
tural
fe
atu
r
es
are
t
h
en
m
od
el
e
d
by
us
i
ng
A
NN
to
es
ti
m
ate
th
e
pe
r
c
en
tag
e
of
pH,
t
ota
l
ph
e
no
l
an
d
p
urit
y
of
Lu
w
ak
c
off
ee
wi
t
h
the
l
owes
t
pa
r
a
me
ter v
a
l
u
e
of
M
ea
n
S
q
ua
r
e
E
r
r
or
(
MS
E
)
.
2.
Re
se
a
r
ch
Me
t
h
o
d
T
hi
s
s
tud
y
uti
l
i
z
es
gre
en
be
an
of
arab
i
c
a
L
uwak
(
c
i
v
et
)
c
off
ee
an
d
r
e
gu
l
ar
a
r
ab
i
c
a
c
off
ee
fr
om
In
do
n
es
i
a
n
P
l
an
tat
i
o
n
C
om
p
an
y
(
P
T
P
erk
eb
un
an
Nus
an
t
ara
X
II)
,
B
an
y
uw
an
g
i
,
Ind
o
ne
s
i
a
.
A
r
ab
i
c
a
L
uwak
c
off
ee
us
ed
i
n
th
e
r
es
e
arc
h
i
s
L
on
g
an
Lu
w
ak
c
off
ee
.
R
eg
u
l
ar
ar
ab
i
c
a
c
off
ee
i
s
proc
es
s
ed
by
us
i
ng
we
t
proc
es
s
i
ng
m
eth
od
.
T
he
t
oo
l
for
c
a
ptu
r
i
ng
pi
c
tures
is
d
i
g
i
ta
l
c
am
era
(
wi
t
h
s
pe
c
i
f
i
c
at
i
on
of:
Ni
k
o
n
Coo
l
p
i
x
A
1
0,
1
6
me
ga
pi
x
e
l
s
,
J
ap
a
n)
pl
ac
ed
i
n
a
b
l
ac
k
bo
x
,
wi
th
the
ba
c
k
grou
nd
o
f
b
l
a
c
k
s
urf
ac
e,
wi
th
c
on
s
tan
t
fl
uo
r
es
c
en
t
l
i
gh
t
i
ng
an
d
ev
e
nl
y
d
i
s
tr
i
bu
ted
throug
ho
u
t
the
gree
n
be
a
n
c
off
ee
s
urfac
e,
a
nd
d
i
r
ec
tl
y
pl
ac
ed
u
nd
er
a
v
erti
c
al
l
y
m
ou
nt
ed
c
am
era.
T
he
i
ma
g
e
da
t
a
proc
es
s
i
ng
too
l
ap
pl
i
es
a
n
In
tel
(
R)
C
ore
(
T
M)
i
3
of
3
2
bi
t
C
P
U
c
om
pu
t
er
2.1
0
G
hz
.
S
oft
w
are
us
ed
i
s
by
W
i
nd
ows
7
32
b
i
t
O
pe
r
a
ti
n
g
S
y
s
te
m,
wi
th
a
s
el
f
-
b
ui
l
t
v
i
s
ua
l
ba
s
i
c
6
.0
ba
s
ed
c
ol
o
r
an
d
t
ex
tura
l
a
na
l
y
s
i
s
s
oft
war
e,
e
qu
i
pp
ed
w
i
th
W
ai
k
a
to
E
nv
i
r
on
me
n
t
for
K
no
w
l
ed
ge
A
n
al
y
s
i
s
(
W
E
K
A
)
3.8
[2
2]
,
an
d
wi
th
Ma
t
l
a
b
R2
01
2a
[
23
]
.
G
r
ee
n
be
a
n
w
i
th
a
pred
ete
r
m
i
n
ed
pe
r
c
en
t
ag
e,
i
s
pl
ac
e
d
o
n
a
p
l
atf
orm
w
i
th
an
area
of
2
56
c
m
2
.
T
h
e
i
m
ag
e
form
at
us
ed
i
s
a
b
i
tm
ap
.
T
he
i
m
ag
e
ac
qu
i
s
i
ti
on
de
s
i
g
n
i
s
de
p
i
c
ted
i
n
F
i
gu
r
e
1.
T
hi
s
s
tud
y
ut
i
l
i
z
es
the
gre
en
be
a
n
of
arab
i
c
a
Lu
wak
c
of
fee
an
d
r
eg
u
l
ar
arabi
c
a
c
off
e
e
as
th
e
r
es
ea
r
c
h
ob
j
ec
t.
E
ac
h
da
ta
c
ol
l
ec
ti
on
i
s
ga
t
he
r
ed
by
us
i
ng
1
60
c
o
ffe
e
be
an
s
,
wh
i
l
e
c
al
c
u
l
at
i
ng
th
e
pe
r
c
en
tag
e
of
the
m
i
x
ture
i
s
pe
r
f
orme
d
i
n
un
i
t
o
f
s
ee
ds
.
Mi
x
e
d
pro
po
r
ti
o
ns
c
on
s
i
s
t
of
:
0%
,
1
0
%,
30
%,
40
%
,
50
%,
70
%,
90
%,
an
d
1
00
%
of
Lu
wak
c
off
ee
as
s
h
own
i
n
F
i
gu
r
e
2
.
T
ota
l
ph
en
o
l
tes
t
was
me
as
ure
d
us
i
ng
th
e
F
o
l
i
n
Ci
oc
a
l
teu
me
t
ho
d
[24
].
T
he
pH
me
as
ur
em
e
nt
was
c
arr
i
ed
ou
t
on
c
off
ee
ex
tr
ac
t u
s
i
n
g a
pH m
ete
r
.
T
he
i
ma
ge
i
s
c
on
v
erte
d
fr
o
m
RG
B
c
ol
ou
r
s
pa
c
e
to
gr
ey
,
HS
L,
H
S
V
an
d
L
*
a
*
b*
c
ol
ou
r
s
pa
c
es
[25
].
T
he
r
es
ul
t
of
fea
ture
ex
tr
ac
ti
on
i
s
th
e
c
o
l
or
c
o
-
oc
c
urr
en
c
e
ma
tr
i
x
(
CCM)
i
n
e
ac
h
c
ol
or
grou
p
(
Red
(
RGB)
,
G
r
ee
n
(RGB)
,
B
l
ue
(RGB)
,
gr
ey
,
Hue,
S
at
urati
on
(HSL
)
,
Li
g
htn
es
s
(HSL)
,
S
atu
r
ati
on
(HSV)
,
V
a
l
u
e
(HSV)
,
L*
,
a
*
,
a
nd
b
*
)
.
A
l
Q
ai
s
i
[2
6]
de
v
e
l
op
ed
di
ff
erent
me
t
ho
ds
us
ed
to
ex
tr
ac
t te
x
ture
fea
t
ures
fro
m a
c
o
l
or
i
m
ag
e
.
T
ex
ture
v
al
u
es
ex
tr
ac
ted
i
n
ea
c
h
ty
p
e o
f c
o
l
or
ba
s
ed
on
Har
al
i
c
k
’
s
tex
ture
an
a
l
y
s
i
s
.
T
he
r
es
u
l
ts
of
i
ma
g
e
d
ata
ac
q
ui
s
i
ti
on
pr
od
uc
e
the
12
0
c
ol
or
an
d
tex
tural
fea
t
ures
.
Har
al
i
c
k
’
s
te
x
tural
eq
ua
t
i
o
ns
are as
f
ol
l
ows
:
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KO
MNIK
A
IS
S
N: 1
69
3
-
6
93
0
◼
Comp
ute
r
v
i
s
i
on
f
or pur
i
ty
,
ph
en
ol
,
an
d p
H
de
t
ec
ti
o
n o
f
Lu
wak
C
off
e
e
... (
Y
us
uf
He
nd
r
awa
n
)
3075
=
∑
∑
2
[
,
]
(
1
)
=
−
∑
∑
2
[
,
]
(
2
)
=
∑
∑
(
−
)
2
[
,
]
(
3
)
=
∑
∑
[
,
]
1
+
|
−
|
(
4
)
=
∑
∑
[
,
]
|
−
|
≠
(
5
)
=
∑
∑
(
−
)
(
−
)
[
,
]
2
(
6
)
=
1
2
∑
∑
(
[
,
]
+
[
,
]
)
(
7
)
=
1
2
∑
∑
(
(
−
)
2
[
,
]
+
(
−
)
2
[
,
]
)
(
8
)
=
∑
∑
(
+
−
2
)
[
,
]
(
9
)
=
N
M
j
i
M
a
x
,
,
[
,
]
(
1
0
)
where:
P
(
i
,
j
)
i
s
the
(
i
,
j
)
th
el
e
me
nt
of
a
no
r
m
al
i
z
ed
c
o
-
oc
c
urr
en
c
e
ma
tr
i
x
,
an
d
μ
an
d
σ
are
the
m
ea
n
an
d s
tan
da
r
d
de
v
i
at
i
on
of
t
he
p
i
x
el
el
em
en
t
g
i
v
en
by
t
he
f
ol
l
ow
i
n
g rel
ati
o
ns
hi
ps
:
[
,
]
=
(
,
)
(
11
)
=
∑
∑
[
,
]
(
12
)
=
∑
(
−
)
2
∑
[
,
]
(
13
)
where:
N(
i
,
j
)
i
s
th
e
n
um
be
r
c
ou
nts
i
n
t
he
i
m
ag
e
wi
t
h
pi
x
el
i
nte
ns
i
ty
i
fol
l
ow
ed
by
pi
x
e
l
i
nte
ns
i
ty
j
at
on
e
pi
x
e
l
d
i
s
pl
ac
em
en
t to
the
l
ef
t, a
nd
M
i
s
th
e t
ot
al
nu
mb
er
of
p
i
x
el
s
.
A
NN
t
op
o
l
og
y
op
t
i
mi
z
a
ti
on
i
s
c
on
du
c
te
d
by
us
i
ng
Ma
t
l
ab
R2
01
2
a
s
oft
w
are.
T
h
e
r
es
ul
ts
of
da
t
a
ac
qu
i
s
i
ti
on
o
f
di
gi
t
al
i
ma
g
e
proc
es
s
i
n
g
me
t
ho
d
s
ob
tai
ne
52
8
i
ma
g
es
at
a
predet
ermi
ne
d
pe
r
c
en
ta
ge
.
Im
ag
e
d
ata
i
s
di
v
i
d
ed
i
nto
66
.
67
%
as
tr
ai
ni
n
g
d
ata
a
nd
33
.
33
%
as
v
al
i
da
t
i
on
d
ata
.
T
he
d
i
s
tr
i
b
uti
on
of
tr
a
i
n
i
ng
da
ta
an
d
v
al
i
d
ati
on
da
ta
i
s
the
i
n
i
ti
al
s
t
ag
e
i
n
A
N
N
[2
7]
.
T
r
ai
n
i
n
g
d
ata
i
s
ap
pl
i
ed
to
up
da
te
we
i
g
ht
s
,
bi
as
es
an
d
s
tud
y
da
t
a
p
att
erns
.
T
he
ac
c
urac
y
of
th
e
mo
de
l
us
es
v
al
i
d
ati
on
da
ta
to
fi
nd
ou
t
the
a
bi
l
i
ty
of
th
e
ne
tw
ork
to
i
de
nt
i
fy
ne
w
d
ata
pa
tt
erns
[2
8]
.
A
NN
Evaluation Warning : The document was created with Spire.PDF for Python.
◼
IS
S
N: 16
93
-
6
93
0
T
E
L
KO
MNIK
A
V
ol
.
17
,
No
.
6,
D
ec
em
b
er 20
19
:
30
7
3
-
3085
3076
mo
de
l
i
ng
a
pp
l
i
es
th
e
ba
c
k
propa
ga
t
i
on
n
eu
r
a
l
ne
tw
or
k
(
B
P
NN)
al
g
orit
hm
,
wh
i
c
h
i
s
a
po
p
ul
ar
al
g
orit
h
m
,
w
i
d
el
y
us
e
d
i
n
A
NN
[29
,
3
0]
.
B
ef
ore
p
erfor
mi
n
g
A
NN
mo
d
el
i
ng
,
i
np
ut
an
d
ou
t
pu
t
da
ta
are no
r
m
al
i
z
ed
to
r
an
ge
of
-
1 a
nd
1.
Inp
ut
l
ay
ers
i
nc
l
ud
e
c
ol
ors
a
nd
tex
tura
l
fe
atu
r
e
s
.
T
he
ou
tp
ut
l
ay
er
ex
pres
s
es
the
pe
r
c
en
tag
e
of
pu
r
i
ty
,
tot
al
ph
en
o
l
an
d
pH
i
n
Lu
w
ak
c
off
ee
.
D
es
i
gn
i
n
g
the
b
es
t
A
NN
to
po
l
og
y
i
s
ac
c
om
pl
i
s
he
d
thro
ug
h
s
en
s
i
ti
v
i
ty
an
a
l
y
s
i
s
wi
th
a
v
arie
ty
of
l
ea
r
n
i
ng
f
un
c
ti
on
s
;
ac
ti
v
at
i
o
n
fun
c
ti
on
;
l
ea
r
n
i
n
g
r
ate
an
d
mo
m
en
tu
m
(
0.
1,
0.
5,
0.
9);
hi
d
de
n
l
ay
er
(
1,
2);
h
i
dd
en
l
ay
er
no
d
e
(
10
,
20
,
30
,
40
)
wi
th
the
l
owes
t
v
al
i
da
t
i
o
n
of
MS
E
pa
r
a
me
t
er.
T
h
i
s
s
tud
y
form
ul
a
tes
t
he
3
ac
t
i
v
at
i
on
fun
c
ti
on
s
i
.
e.
p
urel
i
n,
ta
ns
i
g
, a
nd
l
o
gs
i
g
[31
]
.
F
i
gu
r
e
1.
D
es
i
gn
of
i
m
ag
e
ac
qu
i
s
i
ti
o
n s
y
s
tem
(
a)
(
b)
(
c
)
(
d)
(
e)
(
f)
(
g)
(
h)
F
i
gu
r
e
2.
M
i
x
ture
of
L
uwak
(
c
i
v
et)
c
off
ee
an
d r
eg
u
l
ar c
off
ee
:
(
a) 0%; (
b) 10
%; (
c
)
30
%; (
d) 40%; (
e)
50
%;
(
f)
70
%; (
g) 90%; (
h)
10
0
%
3.
Re
sult
s
a
nd
An
aly
s
is
T
he
fe
atu
r
e
ex
tr
ac
ti
on
r
e
s
ul
ts
i
n
12
0
c
o
l
or
s
an
d
t
ex
tural
fea
t
ures
wh
i
c
h
r
e
pres
en
t
i
nfo
r
ma
t
i
on
r
e
l
at
ed
t
o
th
e
i
ma
ge
(
a
m
i
x
ture
of
Lu
w
ak
c
off
ee
a
nd
r
e
gu
l
ar
gree
n
b
ea
n
i
n
v
ari
ou
s
pe
r
c
en
ta
ge
s
)
.
T
h
e
ma
i
n
pr
ob
l
em
em
erges
tha
t
n
ot
al
l
c
ol
or
an
d
tex
t
ural
fe
atu
r
es
are
c
ap
ab
l
e
of
predi
c
t
i
ng
de
p
en
d
en
t
v
ar
i
ab
l
e
or
ob
j
ec
ti
v
e
fun
c
t
i
o
n.
T
hi
s
s
tag
e
i
s
i
nte
nd
e
d
to
fi
nd
o
ut
the
fea
tures
aff
ec
t
i
ng
ei
t
h
er
d
ep
e
nd
e
nt
v
aria
b
l
e
or
ob
j
ec
ti
v
e
fun
c
t
i
on
.
F
e
atu
r
e
s
el
ec
ti
o
n
i
s
c
on
du
c
te
d
by
prepr
oc
es
s
i
n
g
da
ta
i
n
da
t
a
m
i
n
i
ng
.
S
el
ec
ti
on
of
fea
tures
b
ec
om
e
s
an
i
mp
ortant
s
tag
e
t
o
s
pe
ed
up
the
mo
d
el
i
ng
proc
es
s
an
d
to
fac
i
l
i
ta
te
th
e
d
es
i
g
n
of
to
ol
s
.
T
h
e
ma
i
n
p
urpos
e
of
fe
atu
r
e
s
e
l
ec
t
i
on
i
s
t
o
pr
ev
en
t
ov
erf
i
tt
i
ng
,
as
c
h
arac
teri
z
ed
by
h
i
gh
M
S
E
v
a
l
i
da
ti
on
;
to
r
ed
uc
e
tr
ai
n
i
ng
t
i
me
a
nd
to
i
mp
r
ov
e
m
od
e
l
ac
c
urac
y
[32
-
34]
.
T
he
r
es
e
arc
h
r
es
u
l
ts
o
f
t
he
K
arabu
l
ut
[3
5]
pres
en
te
d
th
at
f
ea
tur
e
s
el
e
c
ti
on
mi
gh
t
i
nc
r
ea
s
e
ac
c
urac
y
by
15
.5
5%
i
n
A
NN,
Na
i
v
e
B
ay
es
,
an
d
J
48
Dec
i
s
i
on
T
r
ee
mo
d
el
i
n
g.
T
hi
s
s
tud
y
em
p
l
oy
s
6
at
tr
i
bu
t
e
ev
a
l
ua
t
ors
,
s
uc
h
as
:
Cfs
S
ub
s
e
t,
Cor
r
el
at
i
o
n
A
ttri
bu
te
,
O
ne
R
A
ttr
i
bu
te,
Re
l
i
e
fF,
G
a
i
n
R
ati
o
A
t
tr
i
b
ute
,
an
d
G
a
i
n
Inf
o
A
ttr
i
bu
t
e.
T
hi
s
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KO
MNIK
A
IS
S
N: 1
69
3
-
6
93
0
◼
Comp
ute
r
v
i
s
i
on
f
or pur
i
ty
,
ph
en
ol
,
an
d p
H
de
t
ec
ti
o
n o
f
Lu
wak
C
off
e
e
... (
Y
us
uf
He
nd
r
awa
n
)
3077
s
tud
y
ap
p
l
i
es
fi
l
ter
m
od
e
l
t
o
fi
n
d
ou
t
t
he
fea
t
ure
s
el
e
c
ti
on
.
T
h
e
fi
l
t
er
mo
d
el
w
hi
c
h
i
s
fas
t
an
d
s
i
mp
l
e
,
as
s
es
s
es
r
el
ev
an
t
fea
t
ures
by
k
no
w
i
ng
th
e
i
ntri
ns
i
c
n
atu
r
e
of
d
ata
.
F
i
l
ter
me
t
ho
d
al
g
orit
h
ms
r
an
k
f
ea
t
ures
ba
s
ed
on
th
ei
r
pr
ox
i
m
i
t
y
to
th
e
c
l
as
s
.
T
he
f
i
l
te
r
me
th
od
ha
s
the
a
dv
an
t
ag
es
o
f
be
i
n
g a
f
as
t a
nd
s
i
mp
l
e
c
om
p
uti
ng
me
th
od
[
36
,
37
]
.
T
he
fea
ture
s
el
ec
ti
o
n
ou
tpu
t
for
di
g
i
ta
l
i
ma
ge
da
ta
i
s
i
n
the
to
p
10
r
a
nk
for
eff
i
c
i
en
c
y
an
d
s
i
mp
l
i
f
i
c
ati
on
of
A
NN
i
n
pu
t.
A
fte
r
ob
t
ai
n
i
n
g
the
to
p
10
r
an
k
,
the
i
np
ut
da
t
a
i
s
t
he
n
m
od
e
l
e
d
by
us
i
ng
A
NN
to
s
e
l
ec
t
th
e
i
n
pu
t
whi
c
h
pr
od
uc
es
the
l
o
wes
t
M
S
E
v
a
l
i
da
t
i
on
.
A
NN
s
tr
uc
tures
us
ed
are:
the
4
0
no
de
s
i
n
1
st
hi
d
de
n
l
ay
er
an
d
40
no
de
s
i
n
2
nd
hi
dd
e
n
l
ay
er
;
ac
ti
v
at
i
on
fun
c
ti
o
n
us
ed
i
n
hi
d
de
n
l
ay
er
a
nd
ou
t
pu
t
l
ay
er
was
t
an
s
i
g
;
tr
a
i
n
l
m
as
l
ea
r
n
i
n
g
fu
nc
ti
o
n
;
l
e
arni
ng
r
ate
of
0.1
an
d
mo
m
en
tu
m
of
0.9
.
T
ab
l
e
1
s
ho
ws
t
he
R
ed
(RGB)
s
um
m
ea
n
,
wh
i
c
h
ha
s
a
s
tr
on
g
c
orr
el
at
i
o
n
wi
th
the
pe
r
c
en
t
ag
e
o
f
r
e
gu
l
ar
c
off
ee
m
i
x
tures
i
n
L
uwak
c
o
ffe
e
we
i
gh
i
n
g
of
0.2
35
9
9.
A
fte
r
ob
t
ai
ni
ng
the
we
i
g
hts
an
d
r
ati
ng
s
,
t
he
da
ta
i
n
T
a
bl
e
1
are
m
od
e
l
ed
by
us
i
n
g
A
N
N
to
f
i
nd
ou
t
fe
atu
r
es
whi
c
h
c
a
n
pre
di
c
t
tot
a
l
ph
e
no
l
,
pH,
an
d
th
e
p
erc
en
ta
g
e
of
r
eg
u
l
ar
c
off
ee
mi
x
es
i
n
Lu
w
ak
c
off
ee
wi
th
th
e
l
ow
es
t
MS
E
v
a
l
i
da
t
i
on
p
arame
ter.
A
NN
o
utp
ut
for
d
i
g
i
ta
l
i
m
ag
e
d
ata
fe
atu
r
e
s
el
ec
ti
on
i
n
T
ab
l
e
2
s
ho
ws
tha
t
w
he
n
t
he
12
0
c
ol
or
an
d
tex
tura
l
f
ea
tures
are
us
e
d
as
A
NN
i
np
uts
,
th
ere
i
s
no
v
a
l
ue
d
ue
t
o
ne
tw
ork
err
ors
.
T
hi
s
i
s
d
ue
to
i
nc
o
mp
a
ti
b
i
l
i
ty
of
th
e
tr
a
i
n
l
m
l
e
arni
n
g
fu
nc
ti
o
n
w
i
th
the
a
mo
u
nt
o
f i
np
u
t d
a
ta.
T
hu
s
,
fe
atu
r
e s
e
l
ec
t
i
on
r
e
ma
i
ns
ne
c
es
s
ary
to
be
pe
r
for
me
d
.
T
he
r
es
ul
ts
of
f
ea
ture
s
el
ec
ti
on
,
pres
en
t
5
da
t
a
i
np
uts
c
orr
el
ati
ng
w
i
th
the
pe
r
c
en
t
ag
e
of
r
eg
ul
ar
c
off
ee
mi
x
es
i
n
Lu
w
ak
c
off
ee
.
T
he
f
i
v
e
d
ata
r
es
ul
ts
fr
om
f
ea
ture
s
e
l
ec
t
i
on
are
l
ab
el
ed
as
tex
ture
f
ea
tur
es
.
E
x
tr
ac
ti
o
n
of
i
m
ag
e
fea
tures
i
s
b
as
ed
o
n
CC
M.
T
h
e
fr
e
q
ue
nt
l
y
ap
pl
i
ed
c
on
v
en
ti
on
a
l
m
eth
o
d
i
n
t
ex
ture
an
a
l
y
s
i
s
is
the
gray
l
ev
el
c
o
-
oc
c
urr
en
c
e
ma
tr
i
x
(
G
LCM)
,
wh
i
c
h
i
s
a p
o
pu
l
ar me
tho
d f
or r
e
pre
s
en
ti
n
g t
ex
t
ure fe
atu
r
es
as
de
v
el
op
e
d b
y
Harr
a
l
i
c
k
.
F
i
gu
r
e
3
de
pi
c
ts
t
he
r
el
a
ti
o
ns
hi
p
of
r
eg
ul
ar
c
off
e
e
mi
x
ture
p
erc
en
ta
ge
i
n
L
uwak
c
off
ee
wi
th
Re
d
(RGB)
s
um
me
an
.
T
he
r
es
u
l
ts
s
ho
w
t
ha
t
the
Red
(RGB)
s
um
m
ea
n
de
c
r
e
as
e
s
al
on
g
w
i
th
the
i
nc
r
ea
s
i
ng
pe
r
c
en
t
ag
e
of
Lu
wak
c
off
ee
.
T
he
v
a
l
ue
of
th
e
t
ex
tural
fea
ture
R
ed
(
RGB)
s
um
me
a
n
s
tat
es
the
av
erag
e
nu
mb
e
r
of
r
ed
v
a
l
ue
s
i
n
th
e
i
ma
ge
(
the
hi
gh
er
t
he
v
a
l
ue
o
f
Red
(RG
B)
s
um
mean
,
the
av
er
ag
e
n
um
b
er
of
r
ed
s
i
n
th
e
tex
tura
l
fea
ture
w
i
l
l
b
e
great
er)
.
F
i
gu
r
e
3
s
ho
ws
the
t
ex
tural
fea
t
ure R
ed
(RGB
)
s
um
m
ea
n
of
1
00
%
Lu
w
ak
c
off
ee
was
l
ower t
ha
n
0%
Lu
wak
c
off
ee
.
F
i
gu
r
e
4
s
h
ows
V
al
ue
(HS
V)
s
um
m
ea
n
i
n
wh
i
c
h
the
v
al
ue
d
ec
r
ea
s
es
w
i
th
the
i
nc
r
ea
s
e
i
n
the
p
erc
en
ta
ge
o
f
Lu
wak
c
off
ee
.
T
ex
tura
l
fe
atu
r
e
V
a
l
u
e
(HSV)
s
um
me
an
v
a
l
ue
s
t
at
es
the
av
er
ag
e
nu
mb
er
of
v
al
u
e
i
n
t
he
i
ma
ge
(
th
e
hi
gh
er
th
e
v
al
ue
,
the
t
ex
tural
f
ea
t
ure
w
i
l
l
be
gre
ate
r
)
.
T
he
tex
ture
v
al
u
e
s
tat
es
th
e
am
ou
nt
of
l
i
g
ht
r
ec
e
i
v
ed
by
the
ey
e
r
eg
ard
l
es
s
of
the
c
ol
or.
T
hi
s
i
s
i
n
ac
c
ord
an
c
e
w
i
th
th
e
c
ol
o
r
of
Lu
wak
c
off
e
e
a
nd
r
e
gu
l
ar
c
off
ee
wh
i
c
h
c
an
be
ob
s
erv
ed
v
i
s
ua
l
l
y
.
Lu
wak
c
off
ee
us
ed
i
n
th
e
s
tud
y
ha
s
a d
ark
er c
ol
or t
ha
n
i
n
r
e
gu
l
ar c
off
ee
,
aff
ec
t
i
n
g
the
v
a
l
ue
s
um
mean
to
d
ec
r
ea
s
e
a
l
on
g
w
i
th
t
he
i
nc
r
ea
s
e
i
n
th
e
pe
r
c
en
ta
ge
of
Lu
w
ak
c
off
ee
.
F
i
gu
r
e
5
s
ho
ws
S
atu
r
ati
on
(HSL)
s
um
me
a
n
i
n
w
hi
c
h
the
v
al
ue
de
c
r
ea
s
es
a
l
on
g
w
i
th
th
e
i
nc
r
ea
s
e
i
n
the
pe
r
c
en
tag
e
of
Lu
wak
c
off
ee
.
S
at
urat
i
on
(H
SL)
s
um
me
an
v
a
l
ue
s
ta
tes
the
av
er
ag
e
n
um
b
er
of
s
atu
r
ati
on
v
al
ue
i
n
the
i
m
ag
e
(
the
hi
g
he
r
th
e
v
al
u
e
,
the
grea
ter
the
a
mo
un
t
of
s
atu
r
at
i
on
)
.
F
i
gu
r
e
6
s
ho
ws
B
l
ue
(RGB)
v
aria
nc
e
w
hi
c
h
de
c
r
ea
s
es
wi
th
the
i
nc
r
ea
s
e
i
n
the
pe
r
c
en
ta
ge
of
Lu
wak
c
off
ee
.
B
l
ue
(R
GB)
v
a
r
i
an
c
e
s
h
ows
v
aria
t
i
o
ns
i
n
c
o
-
oc
c
urr
en
c
e
ma
tr
i
x
el
e
me
nts
.
I
ma
ge
s
wi
th
s
ma
l
l
c
o
l
or
de
gree
t
r
an
s
i
ti
on
s
wi
l
l
h
av
e
l
i
ttl
e
v
aria
nc
e.
If
t
he
v
aria
nc
e
v
al
ue
i
s
hi
g
h,
the
d
eg
r
ee
of
c
ol
or
of
th
e
i
ma
ge
wi
l
l
s
pre
ad
.
V
ari
an
c
e
i
s
the
s
u
m
of
s
q
ua
r
es
of
di
ff
erenc
es
i
n
i
nte
ns
i
ty
a
mo
ng
t
he
ne
i
gh
bo
r
i
n
g
p
i
x
el
s
.
F
i
g
ure
6
s
ho
ws
tha
t
Lu
w
ak
(
c
i
v
et)
c
off
ee
gre
en
be
an
(
10
0%)
ha
v
e
a
l
es
s
di
ffu
s
e
d
bl
u
e
c
ol
or
,
wh
i
l
e
green
b
ea
n
i
n
Lu
wak
c
off
e
e
(
0%)
ha
s
a
di
ffu
s
e
d
bl
u
e
c
o
l
or.
F
i
gu
r
e
7
s
ho
ws
the
Hue
v
ari
an
c
e
whi
c
h
i
nc
r
ea
s
es
al
on
g
w
i
t
h
t
he
i
nc
r
ea
s
e
i
n
the
pe
r
c
en
tag
e
of
Lu
wak
c
off
ee
.
F
r
om
the
gr
ap
h,
i
t
i
s
ob
v
i
ou
s
th
at
Lu
w
ak
c
off
ee
(
10
0%)
ha
s
a
mo
r
e d
i
ff
us
e
d
H
ue
c
o
l
or th
an
th
e
L
uwak
c
off
ee
(
0%)
.
A
NN
m
od
e
l
i
ng
pr
od
uc
es
predi
c
t
i
v
e
ou
t
pu
t,
wei
gh
t
an
d
b
i
as
wh
i
c
h
i
s
o
pti
ma
l
i
n
es
ti
ma
ti
n
g
t
he
p
erc
en
ta
ge
of
pu
r
i
ty
,
to
tal
ph
en
o
l
an
d
pH.
T
h
e
m
os
t
i
mp
orta
nt
s
t
ep
i
n
d
es
i
g
ni
n
g
A
NN
s
tr
uc
ture
i
s
th
e
s
el
ec
ti
on
of
op
t
i
ma
l
w
ei
g
hts
an
d
bi
as
es
am
on
g
ne
ur
on
s
wi
t
h
h
i
gh
ge
ne
r
al
i
z
ati
on
s
[38
]
.
T
he
s
el
ec
te
d
A
N
N
s
tr
ue
c
ture
i
s
pres
en
te
d
i
n
F
i
gu
r
e
8.
T
h
e
i
ni
t
i
a
l
s
tag
e
i
n
de
s
i
g
ni
n
g
A
NN
s
tr
uc
ture
i
s
a
tr
a
i
ni
ng
err
or
of
th
e
l
ea
r
ni
ng
fun
c
t
i
on
.
L
ea
r
n
i
ng
fun
c
t
i
on
p
l
ay
s
a
r
ol
e
i
n
c
ha
ng
i
ng
we
i
gh
ts
an
d
bi
as
es
du
r
i
ng
tr
ai
n
i
n
g.
A
NN
mo
d
e
l
i
ng
r
es
ul
ts
c
o
ns
i
s
t
of
wei
gh
ts
an
d
bi
as
es
tha
t
af
fec
t
M
S
E
v
al
i
da
t
i
on
.
F
or
t
hi
s
r
e
as
on
,
a
tr
a
i
n
i
ng
err
or
of
the
l
ea
r
n
i
ng
fun
c
t
i
on
i
s
c
arr
i
ed
ou
t
.
T
he
r
es
ea
r
c
h
r
es
ul
ts
of
S
ha
r
m
a
an
d
V
en
ug
o
pu
l
an
[3
9]
a
nd
A
gg
arw
al
an
d
Raj
en
dra
[40]
po
i
nt
ou
t
tha
t
th
e
l
ea
r
n
i
n
g f
u
nc
ti
o
n i
nfl
ue
nc
es
A
NN
pe
r
form
an
c
e.
Evaluation Warning : The document was created with Spire.PDF for Python.
◼
IS
S
N: 16
93
-
6
93
0
T
E
L
KO
MNIK
A
V
ol
.
17
,
No
.
6,
D
ec
em
b
er 20
19
:
30
7
3
-
3085
3078
T
ab
l
e
1.
F
ea
ture
S
el
ec
t
i
on
of
Di
gi
t
al
Ima
ge
D
ata
N
o
.
A
t
t
r
ibu
t
e
E
v
a
lua
t
o
r
S
e
a
r
c
h
Met
h
o
d
I
mag
e
Fea
t
u
r
e
s
W
e
igh
t
R
a
n
k
1.
C
f
s
S
u
b
s
e
t
E
v
a
lua
t
o
r
B
e
s
t
F
irs
t
R
e
d
(R
G
B)
E
n
t
r
o
p
y
-
1
H
u
e
C
o
n
t
r
a
s
t
-
2
H
u
e
I
n
v
e
r
s
e
-
3
H
u
e
C
o
r
r
e
lat
ion
-
4
S
a
t
u
r
a
t
ion
(H
SL
)
C
o
r
r
e
lat
ion
-
5
S
a
t
u
r
a
t
ion
(H
S
V)
C
o
r
r
e
lat
ion
-
6
R
e
d
(R
G
B)
S
u
m
Mea
n
-
7
H
u
e
S
u
m
Mea
n
-
8
S
a
t
u
r
a
t
ion
(H
SL
)
S
u
m
Mea
n
-
9
S
a
t
u
r
a
t
ion
(H
S
V)
S
u
m
Mea
n
-
10
Gr
e
e
d
y
S
t
e
p
w
is
e
R
e
d
(R
G
B)
E
n
t
r
o
p
y
-
1
H
u
e
C
o
n
t
r
a
s
t
-
2
H
u
e
I
n
v
e
r
s
e
-
3
S
a
t
u
r
a
t
ion
(H
SL
)
C
o
r
r
e
lat
ion
-
4
S
a
t
u
r
a
t
ion
(H
S
V)
C
o
r
r
e
lat
ion
-
5
R
e
d
(R
G
B)
S
u
m
Mea
n
-
6
H
u
e
S
u
m
Mea
n
-
7
S
a
t
u
r
a
t
ion
(H
SL
)
S
u
m
Mea
n
-
8
S
a
t
u
r
a
t
ion
(H
S
V)
S
u
m
Mea
n
-
9
B
lue
(R
G
B
)
V
a
r
ian
c
e
-
10
2.
C
o
r
r
e
lat
ion
A
t
t
r
ibu
t
e
E
v
a
lua
t
o
r
R
a
n
k
e
r
Gr
e
e
n
(R
G
B)
S
u
m
Mea
n
0
.
3
2
4
1
Gr
e
y
S
u
m
Mea
n
0
.
3
2
4
2
R
e
d
(R
G
B)
S
u
m
Mea
n
0
.
3
2
3
3
V
a
lue
(H
SV)
S
u
m
Mea
n
0
.
3
2
3
4
L
igh
t
n
e
s
s
(H
SL
)
S
u
m
Mea
n
0
.
3
2
3
5
L
(L
a
b
)
Mea
n
0
.
3
1
8
6
B
lue
(R
G
B
)
V
a
r
ian
c
e
0
.
3
1
7
7
L
igh
t
n
e
s
s
(H
SL
)
V
a
r
ian
c
e
0
.
3
1
7
8
B
lue
(R
G
B
)
C
lu
s
t
e
r
0
.
3
1
6
9
Gr
e
y
V
a
r
ian
c
e
0
.
3
1
6
10
3.
On
e
R
A
t
t
r
ibu
t
e
R
a
n
k
e
r
S
a
t
u
r
a
t
ion
(H
SL
)
S
u
m
Mea
n
6
5
.
5
3
0
1
H
u
e
H
o
mog
e
n
e
it
y
6
1
.
9
3
2
2
H
u
e
I
n
v
e
r
s
e
6
1
.
9
3
2
3
H
u
e
E
n
e
r
g
y
6
1
.
9
3
2
4
H
u
e
C
o
n
t
r
a
s
t
6
0
.
7
9
5
5
H
u
e
S
u
m
Mea
n
6
0
.
6
0
6
6
H
u
e
E
n
t
r
o
p
y
5
9
.
8
4
8
7
H
u
e
V
a
r
ian
c
e
5
7
.
5
7
6
8
B
lue
(R
G
B
)
V
a
r
ian
c
e
5
6
.
0
6
1
9
H
u
e
C
lus
t
e
r
5
5
.
3
0
3
10
4.
R
e
li
e
f
F
R
a
n
k
e
r
R
e
d
(R
G
B)
S
u
m
Mea
n
0
.
2
3
6
1
V
a
lue
(H
SV)
S
u
m
Mea
n
0
.
2
3
5
2
S
a
t
u
r
a
t
ion
(H
SL
)
S
u
m
Mea
n
0
.
2
3
4
3
B
lue
(R
G
B
)
V
a
r
ian
c
e
0
.
2
3
1
4
H
u
e
V
a
r
ian
c
e
0
.
2
3
1
5
Gr
e
y
S
u
m
Mea
n
0
.
2
2
7
6
L
igh
t
n
e
s
s
(H
SL
)
V
a
r
ian
c
e
0
.
2
2
7
7
Gr
e
y
V
a
r
ian
c
e
0
.
2
2
5
8
Gr
e
e
n
(R
G
B)
V
a
r
ian
c
e
0
.
2
2
5
9
Gr
e
e
n
(R
G
B)
S
u
m
Mea
n
0
.
2
2
3
10
5.
Ga
in
R
a
t
io
A
t
t
r
ibu
t
e
E
v
a
lua
t
o
r
R
a
n
k
e
r
Gr
e
y
S
u
m
Mea
n
0
.
7
1
6
1
R
e
d
(R
G
B)
S
u
m
Mea
n
0
.
6
9
8
2
L
igh
t
n
e
s
s
(H
SL
)
S
u
m
Mea
n
0
.
6
8
8
3
V
a
lue
(H
SV)
S
u
m
Mea
n
0
.
6
8
3
4
Gr
e
e
n
(R
G
B)
S
u
m
Mea
n
0
.
6
8
2
5
V
a
lue
(H
SV)
V
a
r
ian
c
e
0
.
6
6
5
6
S
a
t
u
r
a
t
ion
(H
SL
)
S
u
m
Mea
n
0
.
6
6
3
7
L
igh
t
n
e
s
s
(H
SL
)
V
a
r
ian
c
e
0
.
6
6
2
8
Gr
e
y
V
a
r
ian
c
e
0
.
6
4
3
9
Gr
e
y
C
lu
s
t
e
r
0
.
6
4
0
10
6.
I
n
f
o
Ga
in
A
t
t
r
ibu
t
e
E
v
a
lua
t
o
r
R
a
n
k
e
r
S
a
t
u
r
a
t
ion
(H
SL
)
S
u
m
Mea
n
2
.
0
8
2
1
H
u
e
I
n
v
e
r
s
e
2
.
0
1
0
2
H
u
e
C
o
n
t
r
a
s
t
1
.
9
1
9
3
H
u
e
E
n
e
r
g
y
1
.
9
1
9
4
H
u
e
E
n
t
r
o
p
y
1
.
9
0
8
5
H
u
e
H
o
mog
e
n
e
it
y
1
.
8
8
3
6
H
u
e
C
o
r
r
e
lat
ion
1
.
8
7
2
7
H
u
e
S
u
m
Mea
n
1
.
8
5
2
8
H
u
e
V
a
r
ian
c
e
1
.
8
3
8
9
H
u
e
C
lus
t
e
r
1
.
8
0
8
10
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KO
MNIK
A
IS
S
N: 1
69
3
-
6
93
0
◼
Comp
ute
r
v
i
s
i
on
f
or pur
i
ty
,
ph
en
ol
,
an
d p
H
de
t
ec
ti
o
n o
f
Lu
wak
C
off
e
e
... (
Y
us
uf
He
nd
r
awa
n
)
3079
T
ab
l
e
2.
A
NN
P
erf
ormanc
e
u
s
i
ng
Co
l
ou
r
an
d T
ex
tural
F
ea
ture
S
e
l
ec
t
i
on
N
o
.
A
t
t
r
ibu
t
e
E
v
a
lua
t
o
r
S
e
a
r
c
h
Met
h
o
d
I
n
p
u
t
MS
E
Tr
a
inin
g
MS
E
V
a
li
d
a
t
ion
1.
-
-
A
ll
Fea
t
u
r
e
(
1
2
0
I
n
p
u
t
s
)
-
-
2.
C
f
s
S
u
b
s
e
t
E
v
a
lua
t
o
r
B
e
s
t
F
irs
t
Fea
t
u
r
e
r
a
n
k
1
~2
0
.
0
0
9
9
0
.
1
0
9
4
Fea
t
u
r
e
r
a
n
k
1
~3
0
.
0
0
9
8
0
.
2
4
8
5
Fea
t
u
r
e
r
a
n
k
1
~4
0
.
0
0
9
0
0
.
7
9
2
1
Fea
t
u
r
e
r
a
n
k
1
~5
0
.
0
0
9
8
0
.
1
7
4
0
Fea
t
u
r
e
r
a
n
k
1
~6
0
.
0
0
9
9
0
.
0
9
7
3
Fea
t
u
r
e
r
a
n
k
1
~7
0
.
0
0
8
3
0
.
0
8
0
2
Fea
t
u
r
e
r
a
n
k
1
~8
0
.
0
1
0
0
0
.
1
2
3
9
Fea
t
u
r
e
r
a
n
k
1
~9
0
.
0
0
7
9
0
.
3
2
2
0
Fea
t
u
r
e
r
a
n
k
1
~1
0
0
.
3
2
0
4
0
.
6
6
0
5
Gr
e
e
d
y
S
t
e
p
w
i
s
e
Fea
t
u
r
e
r
a
n
k
1
~2
0
.
0
0
9
9
0
.
1
0
9
4
Fea
t
u
r
e
r
a
n
k
1
~3
0
.
0
0
9
8
0
.
2
4
8
5
Fea
t
u
r
e
r
a
n
k
1
~4
0
.
0
0
9
8
0
.
1
1
6
7
Fea
t
u
r
e
r
a
n
k
1
~5
0
.
0
0
8
4
0
.
2
6
4
2
Fea
t
u
r
e
r
a
n
k
1
~6
0
.
0
0
8
8
0
.
3
5
8
2
Fea
t
u
r
e
r
a
n
k
1
~7
0
.
0
0
9
8
0
.
1
2
4
5
Fea
t
u
r
e
r
a
n
k
1
~8
0
.
0
0
6
0
0
.
3
2
1
3
Fea
t
u
r
e
r
a
n
k
1
~9
0
.
0
0
9
3
0
.
1
6
9
9
Fea
t
u
r
e
r
a
n
k
1
~1
0
0
.
0
0
6
2
0
.
1
8
5
3
3.
C
o
r
r
e
lat
ion
A
t
t
r
ibu
t
e
E
v
a
lua
t
o
r
R
a
n
k
e
r
Fea
t
u
r
e
r
a
n
k
1
~2
0
.
0
0
9
9
0
.
2
8
7
8
Fea
t
u
r
e
r
a
n
k
1
~3
0
.
0
1
0
0
0
.
4
0
9
0
Fea
t
u
r
e
r
a
n
k
1
~4
0
.
0
0
9
7
0
.
1
7
8
9
Fea
t
u
r
e
r
a
n
k
1
~5
0
.
0
1
0
0
0
.
2
6
7
4
Fea
t
u
r
e
r
a
n
k
1
~6
0
.
0
0
9
7
0
.
0
6
2
1
Fea
t
u
r
e
r
a
n
k
1
~7
0
.
0
0
9
7
0
.
0
9
1
8
Fea
t
u
r
e
r
a
n
k
1
~8
0
.
0
1
0
0
0
.
0
6
8
4
Fea
t
u
r
e
r
a
n
k
1
~9
0
.
0
0
9
9
0
.
0
6
7
2
Fea
t
u
r
e
r
a
n
k
1
~1
0
0
.
0
8
0
3
0
.
3
2
4
6
4.
On
e
R
A
t
t
r
ibu
t
e
R
a
n
k
e
r
Fea
t
u
r
e
r
a
n
k
1
~2
0
.
0
0
8
8
0
.
3
9
1
9
Fea
t
u
r
e
r
a
n
k
1
~3
0
.
0
0
9
7
0
.
5
3
2
5
Fea
t
u
r
e
r
a
n
k
1
~4
0
.
0
0
8
2
0
.
7
7
6
0
Fea
t
u
r
e
r
a
n
k
1
~5
0
.
0
0
9
1
0
.
3
9
6
8
Fea
t
u
r
e
r
a
n
k
1
~6
0
.
0
0
4
8
0
.
9
0
2
1
Fea
t
u
r
e
r
a
n
k
1
~7
0
.
0
0
9
8
0
.
2
6
4
2
Fea
t
u
r
e
r
a
n
k
1
~8
0
.
0
0
8
6
0
.
1
0
9
8
Fea
t
u
r
e
r
a
n
k
1
~9
0
.
0
0
9
6
0
.
1
2
4
9
Fea
t
u
r
e
r
a
n
k
1
~1
0
0
.
0
0
9
7
0
.
1
6
2
2
5.
R
e
li
e
F
R
a
n
k
e
r
Fea
t
u
r
e
r
a
n
k
1
~2
0
.
0
1
3
5
0
.
0
6
8
9
Fea
t
u
r
e
r
a
n
k
1
~3
0
.
0
1
0
0
0
.
0
8
7
2
Fea
t
u
r
e
r
a
n
k
1
~4
0
.
0
0
9
8
0
.
0
8
2
6
Fea
t
u
r
e
r
a
n
k
1
~5
0
.
0
0
9
7
0
.
0
4
9
1
Fea
t
u
r
e
r
a
n
k
1
~6
0
.
0
0
9
4
0
.
0
5
5
2
Fea
t
u
r
e
r
a
n
k
1
~7
0
.
0
0
9
5
0
.
0
5
6
9
Fea
t
u
r
e
r
a
n
k
1
~8
0
.
0
0
9
8
0
.
0
5
4
2
Fea
t
u
r
e
r
a
n
k
1
~9
0
.
0
0
9
7
0
.
2
6
1
2
Fea
t
u
r
e
r
a
n
k
1
~1
0
0
.
3
0
9
9
0
.
3
9
3
7
6.
Ga
in
R
a
t
io
A
t
t
r
ibu
t
e
E
v
a
lua
t
o
r
R
a
n
k
e
r
Fea
t
u
r
e
r
a
n
k
1
~2
0
.
0
0
9
9
0
.
1
5
2
8
Fea
t
u
r
e
r
a
n
k
1
~3
0
.
0
0
9
9
0
.
1
0
1
4
Fea
t
u
r
e
r
a
n
k
1
~4
0
.
0
1
0
0
0
.
1
7
4
1
Fea
t
u
r
e
r
a
n
k
1
~5
0
.
3
9
7
0
0
.
9
2
8
7
Fea
t
u
r
e
r
a
n
k
1
~6
0
.
0
1
0
0
0
.
0
9
0
0
Fea
t
u
r
e
r
a
n
k
1
~7
0
.
0
0
9
8
0
.
0
5
8
8
Fea
t
u
r
e
r
a
n
k
1
~8
0
.
0
0
9
9
0
.
0
5
3
1
Fea
t
u
r
e
r
a
n
k
1
~9
0
.
0
0
9
9
0
.
0
5
6
8
Fea
t
u
r
e
r
a
n
k
1
~1
0
0
.
0
0
9
5
0
.
1
4
3
2
7.
I
n
f
o
Ga
in
R
a
t
io
A
t
t
r
ibu
t
e
E
v
a
lua
t
o
r
R
a
n
k
e
r
Fea
t
u
r
e
r
a
n
k
1
~2
0
.
0
0
9
5
0
.
5
3
4
8
Fea
t
u
r
e
r
a
n
k
1
~3
0
.
0
0
9
5
0
.
1
4
5
2
Fea
t
u
r
e
r
a
n
k
1
~4
0
.
0
0
9
9
0
.
4
1
1
6
Fea
t
u
r
e
r
a
n
k
1
~5
0
.
0
0
9
7
0
.
6
4
3
7
Fea
t
u
r
e
r
a
n
k
1
~6
0
.
0
0
9
9
0
.
1
1
6
1
Fea
t
u
r
e
r
a
n
k
1
~7
0
.
0
0
9
8
0
.
1
7
8
7
Fea
t
u
r
e
r
a
n
k
1
~8
0
.
0
1
0
0
0
.
1
5
6
1
Fea
t
u
r
e
r
a
n
k
1
~9
0
.
6
2
3
1
0
.
2
7
9
7
Fea
t
u
r
e
r
a
n
k
1
~1
0
0
.
0
0
9
7
0
.
2
6
1
8
Evaluation Warning : The document was created with Spire.PDF for Python.
◼
IS
S
N: 16
93
-
6
93
0
T
E
L
KO
MNIK
A
V
ol
.
17
,
No
.
6,
D
ec
em
b
er 20
19
:
30
7
3
-
3085
3080
F
i
gu
r
e
3.
T
h
e rel
ati
on
s
h
i
p
o
f
Lu
w
ak
c
off
ee
pu
r
i
ty
to
r
e
d
(
RGB)
s
um
me
a
n
F
i
gu
r
e
4.
R
el
a
ti
o
ns
hi
p o
f
Lu
wak
c
off
ee
pu
r
i
ty
to
v
a
l
ue
(
H
SV)
s
um
m
ea
n
F
i
gu
r
e
5.
T
h
e rel
ati
on
s
h
i
p
o
f
Lu
w
ak
c
off
ee
pu
r
i
ty
to
s
at
urati
o
n
(HSL)
s
u
m m
ea
n
F
i
gu
r
e
6.
R
el
a
ti
o
ns
hi
p o
f
Lu
wak
c
off
ee
pu
r
i
ty
to
b
l
u
e
(RG
B)
v
aria
nc
e
1
3
7
,
8
6
1
3
5
,
2
5
1
3
4
,
2
4
1
3
3
,
7
2
1
3
1
,
9
5
1
2
8
,
0
9
1
2
3
,
2
2
1
2
2
,
0
7
0%
10%
30%
40%
50%
70%
90%
100%
R
e
d(
R
GB
)
S
um
M
e
a
n
P
e
rc
e
nt
a
ge
of
C
i
v
e
t
C
of
f
e
e
1
3
7
,
5
4
1
3
4
,
9
8
1
3
3
,
9
8
1
3
3
,
4
9
1
3
1
,
7
5
127,
93
1
2
3
,
1
1
1
2
1
,
9
7
0%
10%
30%
40%
50%
70%
90%
100%
V
a
l
ue(
H
S
V
)
S
um
M
e
a
n
P
e
rc
e
nt
a
ge
of
C
i
v
e
t
C
of
f
e
e
5
1
,
7
4
4
9
,
7
7
49,
01
4
7
,
8
8
4
7
,
0
9
4
4
,
7
0
4
2
,
4
7
4
0
,
9
5
0%
10%
30%
40%
50%
70%
90%
100%
Sat
ura
t
i
on(
H
SL)
Sum
M
e
a
n
P
e
rc
e
nt
a
ge
of
C
i
v
e
t
C
of
f
e
e
8
8
9
,
9
9
8
7
6
,
2
8
8
6
1
,
7
0
8
5
7
,
6
5
831,
06
7
8
9
,
2
6
7
1
7
,
5
6
6
8
2
,
4
1
0%
10%
30%
40%
50%
70%
90%
100%
B
l
ue(
R
GB
)
V
a
ri
a
nce
P
e
rc
e
nt
a
ge
of
C
i
v
e
t
C
of
f
e
e
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KO
MNIK
A
IS
S
N: 1
69
3
-
6
93
0
◼
Comp
ute
r
v
i
s
i
on
f
or pur
i
ty
,
ph
en
ol
,
an
d p
H
de
t
ec
ti
o
n o
f
Lu
wak
C
off
e
e
... (
Y
us
uf
He
nd
r
awa
n
)
3081
F
i
gu
r
e
7.
R
el
a
ti
o
ns
hi
p o
f
l
u
wak
(
c
i
v
et
)
c
off
ee
pu
r
i
ty
to
hue
v
ari
an
c
e
F
i
gu
r
e
8.
A
NN s
tr
uc
ture
wi
t
h 5
s
e
l
ec
ted
i
n
pu
t
pa
r
a
me
t
er i
m
ag
es
f
or es
ti
ma
t
i
ng
th
e p
erc
en
t
ag
e o
f
Lu
wak
(
c
i
v
et) c
off
ee
mi
x
tur
e,
tot
al
ph
e
no
l
an
d p
H
T
ab
l
e
3
pres
en
ts
tr
a
i
n
l
m
a
s
the
s
el
ec
ted
l
e
arni
ng
f
un
c
ti
on
w
hi
c
h
prod
uc
es
the
l
owes
t
MS
E
v
a
l
i
da
t
i
on
.
T
r
ai
nl
m
i
s
a
l
e
arni
ng
fun
c
t
i
on
th
at
u
pd
at
es
th
e
w
ei
gh
ts
an
d
b
i
as
es
ba
s
e
d
on
La
v
en
b
erg
M
arqua
dt
o
pti
mi
z
at
i
on
.
T
r
a
i
n
l
m
i
s
c
ate
go
r
i
z
ed
as
the
f
as
tes
t
al
go
r
i
th
m
a
nd
i
s
r
ec
om
me
nd
e
d
as
the
fi
r
s
t
s
up
erv
i
s
i
on
al
go
r
i
t
hm
de
s
pi
te
en
t
ai
l
i
ng
mo
r
e
me
m
ory
tha
n
ot
he
r
al
g
orit
h
ms
.
T
r
ai
n
l
m
i
s
proc
ee
de
d
by
us
i
n
g
J
ac
ob
i
an
M
atri
x
c
al
c
u
l
at
i
o
ns
,
whi
l
e
ne
twork
pe
r
forma
nc
e
i
s
me
as
ured
fr
om
MS
E
.
T
r
ai
nl
m
i
s
de
s
i
gn
ed
t
o
ha
v
e
a
tw
o
-
l
ev
e
l
t
r
ai
ni
ng
s
pe
ed
,
whi
c
h
i
s
fas
ter
wi
t
ho
u
t
c
al
c
ul
at
i
n
g
the
H
es
s
i
an
ma
tr
i
x
.
A
fte
r
o
bta
i
ni
ng
t
he
b
es
t
l
e
arni
n
g
fun
c
t
i
on
tha
t
g
i
v
es
the
l
owes
t
M
S
E
v
al
i
d
ati
on
,
t
he
n
tr
a
i
n
i
ng
err
or
i
s
ma
na
ge
d
i
n
th
e
ac
ti
v
a
ti
on
f
un
c
ti
on
as
i
l
l
us
tr
ate
d i
n Tab
l
e
4.
T
he
r
es
ul
ts
of
t
he
tr
a
i
ni
ng
err
or
s
ho
w
tha
t
t
he
ta
ns
i
g
fun
c
ti
on
i
n
the
h
i
d
de
n
l
ay
e
r
an
d
pu
r
el
i
n
i
n
t
he
ou
t
pu
t
l
ay
er
gi
v
es
t
he
l
o
wes
t
MS
E
v
a
l
i
da
ti
on
.
P
urel
i
n
ac
ti
v
a
ti
o
n
f
un
c
ti
o
n
i
s
o
nl
y
us
ed
i
n
th
e
o
utp
ut
l
ay
er.
P
ure
l
i
n
pro
du
c
es
y
=
x
,
w
hi
c
h
c
an
n
ot
s
o
l
v
e
no
n
-
l
i
ne
a
r
prob
l
em
s
on
hi
d
de
n
l
ay
er
no
de
s
.
A
NN
mo
de
l
i
s
i
de
nti
c
a
l
to
f
i
n
di
n
g
the
r
e
l
at
i
o
ns
hi
p
of
n
on
-
l
i
ne
ar
da
ta
be
twe
en
i
np
ut
an
d
ou
t
pu
t.
F
or
thi
s
r
ea
s
on
,
ac
t
i
v
ati
on
fun
c
ti
o
n
s
s
uc
h
as
tan
s
i
g
an
d
l
og
s
i
g
are
r
ec
om
me
nd
e
d
i
n
t
he
hi
dd
en
l
ay
er.
T
r
a
i
n
i
ng
err
or
ac
ti
v
ati
on
fun
c
t
i
on
i
s
pe
r
form
ed
b
ec
au
s
e
i
t
aff
ec
ts
MS
E
v
al
i
da
t
i
o
n.
Res
ea
r
c
h
by
Chan
g
an
d
Chun
g
[
41
]
s
ho
ws
th
at
the
r
es
ul
ts
of
the
s
en
s
i
ti
v
i
ty
an
a
l
y
s
i
s
in
t
he
ac
ti
v
a
ti
o
n
fu
nc
ti
o
n
,
a
ffe
c
t
the
pe
r
f
orma
nc
e
of
A
N
N
i
n
produc
i
ng
2
6
6
,
5
6
3
4
0
,
9
9
3
6
5
,
2
0
4
1
2
,
9
5
4
5
3
,
3
0
5
2
1
,
9
6
6
1
4
,
5
5
6
3
9
,
2
9
0%
10%
30%
40%
50%
70%
90%
100%
H
ue
V
a
ri
a
nce
P
e
rc
e
nt
a
ge
of
C
i
v
e
t
C
of
f
e
e
Evaluation Warning : The document was created with Spire.PDF for Python.
◼
IS
S
N: 16
93
-
6
93
0
T
E
L
KO
MNIK
A
V
ol
.
17
,
No
.
6,
D
ec
em
b
er 20
19
:
30
7
3
-
3085
3082
the
l
owes
t
MS
E
v
al
i
d
ati
on
.
T
he
u
ns
ui
t
ab
l
e
ac
t
i
v
ati
o
n
fun
c
t
i
on
c
a
us
es
an
i
nc
r
ea
s
e
i
n
M
S
E
v
al
i
d
ati
on
.
A
f
ter
o
bta
i
n
i
ng
t
he
b
es
t
ac
t
i
v
ati
on
fun
c
t
i
on
,
A
NN
s
tr
uc
ture
i
s
furt
he
r
d
es
i
gn
ed
w
i
th
a
v
arie
ty
of
l
ea
r
ni
n
g rat
es
, h
i
dd
en
l
ay
er
no
d
es
, a
n
d t
h
e
nu
mb
er of
hi
dd
e
n l
ay
ers
.
T
ab
l
e
5
s
ho
ws
th
e
be
s
t
s
tr
uc
ture
wh
i
c
h
i
s
th
e
5
-
30
-
40
-
3
wi
t
h
l
e
arni
ng
r
at
e
of
0.
1
an
d
mo
m
en
tu
m
of
0.5
,
w
hi
c
h
p
r
od
uc
es
MS
E
v
al
i
d
ati
on
of
0.0
44
2.
De
term
i
na
t
i
o
n
of
t
he
nu
mb
er
o
f
hi
d
de
n
l
ay
er
no
de
s
an
d
t
he
nu
mb
er
o
f
hi
dd
e
n
l
ay
e
r
s
pl
ay
s
as
the
mo
s
t
i
mp
ortant
s
ta
ge
i
n
de
s
i
g
ni
n
g
A
N
N
s
tr
uc
ture.
T
he
r
es
ul
ts
s
ho
w
t
ha
t
tw
o
hi
d
de
n
l
ay
ers
c
an
pred
i
c
t
the
o
utp
u
t
v
aria
b
l
e
wh
i
c
h
i
s
be
tte
r
th
a
n
1
hi
d
de
n
l
ay
er,
du
e
t
o
th
e
2
-
hi
d
de
n
-
l
ay
er
ab
i
l
i
ty
to
s
ol
v
e
no
n
-
l
i
ne
ar
probl
em
s
whi
c
h
i
s
b
ett
er
th
an
1
hi
d
de
n
l
ay
er.
How
ev
er,
m
ore
hi
d
de
n
l
ay
ers
h
i
n
de
r
the
c
o
mp
ut
er
r
un
ni
ng
.
T
he
r
efo
r
e,
th
e
h
i
d
de
n
l
ay
er
s
en
s
i
ti
v
i
ty
a
na
l
y
s
i
s
i
s
ne
ed
e
d.
In
thi
s
s
tu
dy
,
the
m
ax
i
m
um
nu
mb
er
of
h
i
dd
en
l
ay
ers
i
s
de
term
i
n
ed
,
as
K
ars
ol
i
y
a
[42
]
s
tat
ed
t
ha
t
th
e
2
-
h
i
d
de
n
-
l
ay
ers
c
an
s
ol
v
e
no
n
-
l
i
ne
ar
prob
l
em
s
.
S
en
s
i
ti
v
i
ty
an
a
l
y
s
i
s
of
l
e
a
r
ni
n
g
r
ate
an
d
mo
me
n
tu
m
i
s
r
eq
ui
r
e
d
be
c
au
s
e
bo
th
of
t
he
s
e
i
n
di
c
ato
r
s
pl
ay
a
r
o
l
e
i
n c
ha
ng
es
of
we
i
g
ht
a
nd
b
i
as
d
urin
g t
r
ai
ni
ng
.
T
ab
l
e
3.
Tr
a
i
n
i
ng
E
r
r
or
b
as
ed
o
n
Le
arni
ng
Fu
nc
ti
on
N
o
.
L
e
a
r
n
ing
Fun
c
t
ion
R
t
r
a
inin
g
R
v
a
li
d
a
t
ion
MS
E
t
r
a
inin
g
MS
E
v
a
li
d
a
t
ion
1.
Tr
a
inc
g
b
(
C
o
n
jug
a
t
e
Gr
a
d
ien
t
B
P
w
it
h
P
o
w
e
l
l
–
B
e
a
le
R
e
s
t
a
r
t
)
0
.
9
8
8
9
1
0
.
9
8
2
3
1
0
.
0
1
0
0
0
.
0
4
8
2
2.
Tr
a
inc
g
f
(
C
o
n
jug
a
t
e
Gr
a
d
ien
t
B
P
w
it
h
Fle
t
c
h
e
r
R
e
e
v
e
s
U
p
d
a
t
e
)
0
.
9
8
8
8
2
0
.
9
8
8
4
9
0
.
0
1
0
0
0
.
0
5
1
7
3.
Tr
a
inc
g
p
(
C
o
n
jug
a
t
e
Gr
a
d
ien
t
B
P
w
it
h
P
o
lak
R
ibie
r
e
U
p
d
a
t
e
)
0
.
9
8
8
8
5
0
.
9
8
7
1
3
0
.
0
1
0
0
0
.
0
4
8
8
4.
Tr
a
ing
d
(
Gr
a
d
ien
t
D
e
s
c
e
n
t
B
P
)
0
.
9
8
3
0
6
0
.
9
8
8
8
4
0
.
0
1
5
1
0
.
0
5
0
9
5.
Tr
a
ing
d
a
(
Gr
a
d
ien
t
D
e
s
c
e
n
t
w
it
h
A
d
a
p
t
iv
e
L
e
a
r
n
ing
R
a
t
e
B
P
)
0
.
9
8
3
0
2
0
.
9
8
7
9
2
0
.
0
1
5
2
0
.
0
5
1
8
6.
Tr
a
ing
d
m
(
Gr
a
d
ien
t
D
e
s
c
e
n
t
w
i
t
h
Momen
t
u
m
B
P
)
0
.
9
8
8
8
3
0
.
9
8
8
3
6
0
.
0
0
9
6
0
.
0
5
3
4
7.
Tr
a
ing
d
x
(
Gr
a
d
ien
t
D
e
s
c
e
n
t
w
it
h
Momen
t
u
m
A
d
a
p
t
i
v
e
L
e
a
r
n
ing
R
a
t
e
B
P
)
0
.
9
8
6
8
0
0
.
9
8
8
7
6
0
.
0
1
1
8
0
.
0
5
0
8
8.
Tr
a
inlm
(
L
a
v
e
n
b
e
r
g
Mar
q
u
a
d
t
B
P
)
0
.
9
9
0
8
5
0
.
9
8
9
3
2
0
.
0
0
9
2
0
.
0
4
6
4
9.
Tr
a
ino
s
s
(
On
e
S
t
e
p
S
e
c
a
n
t
B
P
)
0
.
9
8
8
8
7
0
.
9
8
7
7
2
0
.
0
1
0
0
0
.
0
5
4
2
10.
Tr
a
inr
p
(
R
e
s
il
ien
t
B
P
)
0
.
9
8
8
8
5
0
.
9
8
7
2
9
0
.
0
1
0
0
0
.
0
4
7
4
11.
Tr
a
ins
c
g
(
S
c
a
led
C
o
n
jug
a
t
e
Gr
a
d
i
e
n
t
B
P
)
0
.
9
8
8
8
7
0
.
9
8
6
5
0
0
.
0
1
0
0
0
.
0
5
3
9
T
ab
l
e
4.
Tr
a
i
n
i
ng
E
r
r
or
b
as
ed
o
n
A
c
t
i
v
ati
on
Fu
nc
ti
o
n
L
e
a
r
n
ing
f
u
n
c
t
ion
A
c
t
i
v
a
t
ion
f
u
n
c
t
ion
R
t
r
a
inin
g
R
v
a
li
d
a
t
ion
MS
E
t
r
a
inin
g
MS
E
v
a
li
d
a
t
ion
H
idd
e
n
L
a
y
e
r
1
H
idd
e
n
L
a
y
e
r
2
Ou
t
p
u
t
L
a
y
e
r
TR
A
I
N
L
M
Tan
s
ig
Tan
s
ig
P
u
r
e
li
n
0
.
9
9
0
8
5
0
.
9
8
9
3
2
0
.
0
0
9
2
0
.
0
4
6
4
Tan
s
ig
Tan
s
ig
Tan
s
ig
0
.
9
9
0
8
5
0
.
9
8
9
3
2
0
.
0
0
9
2
0
.
0
5
7
9
Tan
s
ig
Tan
s
ig
L
o
g
s
ig
0
.
8
8
2
2
3
0
.
8
6
9
2
6
0
.
2
3
0
0
0
.
2
6
7
1
L
o
g
s
ig
L
o
g
s
ig
P
u
r
e
li
n
0
.
9
8
9
1
5
0
.
9
7
4
2
2
0
.
0
0
9
7
0
.
0
5
9
6
L
o
g
s
ig
L
o
g
s
ig
Tan
s
ig
0
.
9
8
9
2
5
0
.
8
5
3
8
1
0
.
0
1
0
0
0
.
2
1
3
6
L
o
g
s
ig
L
o
g
s
ig
L
o
g
s
ig
0
.
4
2
8
8
2
0
.
8
6
9
1
4
0
.
3
7
4
7
0
.
3
7
9
1
T
ab
l
e
5.
A
c
c
urac
y
an
d
MS
E
of
A
NN
P
erf
orma
nc
e
b
as
ed
o
n
Le
arni
ng
R
ate
an
d
M
om
en
tum
L
e
a
r
n
ing
R
a
t
e
Momen
t
u
m
A
N
N
S
t
r
u
c
t
u
r
e
R
t
r
a
inin
g
R
v
a
li
d
a
t
ion
MS
E
t
r
a
inin
g
MS
E
v
a
li
d
a
t
ion
0
.
1
0
.
5
5
-
30
-
3
0
.
9
8
8
9
5
0
.
9
7
2
8
1
0
.
0
0
9
9
0
.
0
6
8
4
5
-
40
-
3
0
.
9
8
9
00
0
.
9
8
3
8
6
0
.
0
0
9
8
0
.
0
4
4
5
5
-
30
-
30
-
3
0
.
9
9
1
7
8
0
.
9
7
8
0
2
0
.
0
0
7
5
0
.
0
5
4
0
5
-
30
-
40
-
3
0
.
99052
0
.
9
7
9
3
3
0
.
0
0
8
5
0
.
0
4
4
2
5
-
40
-
40
-
3
0
.
9
8
8
9
4
0
.
9
7
3
9
8
0
.
0
0
9
9
0
.
0
5
1
6
0
.
1
0
.
9
5
-
30
-
3
0
.
9
8
9
1
0
0
.
9
6
8
3
9
0
.
0
0
9
8
0
.
0
6
8
4
5
-
40
-
3
0
.
9
8
8
9
6
0
.
9
6
9
1
4
0
.
0
0
9
9
0
.
0
5
4
6
5
-
30
-
30
-
3
0
.
9
8
8
8
7
0
.
9
7
8
9
0
0
.
0
1
00
0
.
0
5
0
9
5
-
30
-
40
-
3
0
.
9
8
9
2
9
0
.
9
8
1
9
1
0
.
0
0
9
6
0
.
0
4
7
6
5
-
40
-
40
-
3
0
.
9
8
8
9
4
0
.
9
7
3
9
8
0
.
0
0
9
9
0
.
0
5
1
6
F
i
gu
r
e
9
s
ho
ws
t
hree
de
s
c
r
i
pti
o
ns
,
wh
i
c
h
are:
the
bl
ue
l
i
n
e
r
ep
r
es
e
nti
ng
t
he
tr
ai
n
i
ng
,
the
bl
ue
da
s
he
d
l
i
ne
r
ep
r
e
s
en
t
i
n
g
t
he
be
s
t,
an
d
t
he
bl
ac
k
d
as
he
d
l
i
n
e
r
e
pres
e
nt
i
n
g
t
he
go
al
.
T
he
tr
ai
ni
ng
s
ho
ws
an
i
ter
ati
v
e
r
e
l
at
i
on
s
h
i
p
t
o
M
S
E
d
urin
g
tr
ai
n
i
n
g.
F
i
gu
r
e
9
s
h
o
ws
de
c
r
ea
s
i
n
g
err
or
al
on
g
w
i
th
i
nc
r
ea
s
i
n
g
i
terat
i
on
d
ue
to
th
e
s
ta
bl
e
n
etwo
r
k
a
bi
l
i
ty
to
r
ec
og
ni
z
e
da
ta
pa
tte
r
ns
.
A
NN
i
s
a
"
bl
ac
k
bo
x
"
mo
de
l
i
n
g
tha
t
i
s
oft
en
us
e
d
i
n
d
e
al
i
ng
wi
th
no
n
-
l
i
ne
ar
probl
e
ms
.
A
NN
ha
s
the
ab
i
l
i
ty
to
l
e
arn
fr
o
m
i
tera
ti
on
s
,
w
hi
c
h
i
s
w
i
d
el
y
us
e
d
i
n
v
ari
ou
s
f
i
e
l
d
s
of
s
c
i
en
c
e
.
T
he
a
dv
an
t
ag
es
of
A
NN
ar
e
b
ei
n
g
ab
l
e
t
o
ad
o
pt,
s
t
ud
y
an
d
ge
ne
r
a
l
i
z
e
[4
3,
44]
.
T
he
ad
v
an
t
ag
es
of
A
N
N
are
al
s
o
to
q
ui
c
k
l
y
an
d
ac
c
urate
l
y
s
tud
y
d
ata
p
att
erns
c
o
mp
ar
ed
t
o
c
on
v
en
t
i
on
al
Evaluation Warning : The document was created with Spire.PDF for Python.