I
AE
S In
t
er
na
t
io
na
l J
o
urna
l o
f
Art
if
icia
l In
t
ellig
ence
(
I
J
-
AI
)
Vo
l.
1
5
,
No
.
1
,
Feb
r
u
ar
y
2
0
2
6
,
p
p
.
642
~
654
I
SS
N:
2
2
5
2
-
8
9
3
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijai.v
15
.i
1
.
p
p
6
4
2
-
6
5
4
642
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
a
i
.
ia
esco
r
e.
co
m
Deforma
ble spa
ti
a
l pyra
mid po
o
ling
-
enha
nced
E
ff
ici
entNe
t
wi
th
weig
hted
featur
e
fusio
n f
o
r pom
eg
r
a
na
te
frui
t
disea
s
e diag
no
sis
H
a
rish
B
o
m
m
ena
ha
lli M
a
llik
a
rj
un
a
ia
h
1
,
B
a
la
j
i P
r
a
bh
u B
a
luv
a
nera
lu Ve
er
a
nn
a
2
1
D
e
p
a
r
t
me
n
t
o
f
I
n
f
o
r
mat
i
o
n
S
c
i
e
n
c
e
a
n
d
E
n
g
i
n
e
e
r
i
n
g
,
M
a
l
n
a
d
C
o
l
l
e
g
e
o
f
E
n
g
i
n
e
e
r
i
n
g
-
H
a
ssa
n
,
A
f
f
i
l
i
a
t
e
d
t
o
V
i
sv
e
sv
a
r
a
y
a
T
e
c
h
n
o
l
o
g
i
c
a
l
U
n
i
v
e
r
si
t
y
,
B
e
l
a
g
a
v
i
,
I
n
d
i
a
2
D
e
p
a
r
t
me
n
t
o
f
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
a
n
d
E
n
g
i
n
e
e
r
i
n
g
,
M
a
l
n
a
d
C
o
l
l
e
g
e
o
f
En
g
i
n
e
e
r
i
n
g
-
H
a
ss
a
n
,
A
f
f
i
l
i
a
t
e
d
t
o
V
i
s
v
e
sv
a
r
a
y
a
Te
c
h
n
o
l
o
g
i
c
a
l
U
n
i
v
e
r
si
t
y
,
B
e
l
a
g
a
v
i
,
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
J
u
l 9
,
2
0
2
5
R
ev
is
ed
J
an
1
,
2
0
2
6
Acc
ep
ted
J
an
2
2
,
2
0
2
6
P
o
m
e
g
ra
n
a
te
is
a
fru
it
o
f
h
ig
h
n
u
tri
ti
o
n
a
l
a
n
d
e
c
o
n
o
m
ic
imp
o
rtan
c
e
.
S
ti
ll
,
it
is
h
ig
h
ly
s
u
sc
e
p
ti
b
le
t
o
d
i
ffe
re
n
t
d
ise
a
se
s
d
u
rin
g
it
s
g
r
o
win
g
sta
g
e
s,
lea
d
in
g
to
sig
n
ifi
c
a
n
t
y
iel
d
lo
ss
e
s
a
n
d
f
in
a
n
c
ial
se
tb
a
c
k
s
fo
r
fa
rm
e
rs.
Th
is
a
rti
c
le
p
ro
p
o
se
s
a
n
o
v
e
l
d
ise
a
se
d
e
tec
ti
o
n
m
o
d
e
l
t
h
a
t
i
n
teg
ra
tes
h
a
n
d
c
ra
fte
d
fe
a
tu
re
s
with
d
e
e
p
fe
a
tu
re
s
e
x
trac
ted
u
sin
g
a
d
e
v
e
lo
p
e
d
d
e
fo
rm
a
b
le
sp
a
ti
a
l
p
y
ra
m
id
p
o
o
li
n
g
(DSP
P
)
-
Eff
icie
n
tNe
t
a
r
c
h
it
e
c
tu
re
.
Ha
n
d
c
ra
fted
fe
a
tu
re
s
su
c
h
a
s
c
o
lo
r
(
RG
B
a
n
d
HSV
h
isto
g
ra
m
s
)
,
tex
tu
re
fe
a
tu
re
s
fr
o
m
gr
a
y
lev
e
l
c
o
-
o
c
c
u
rre
n
c
e
m
a
tri
x
(G
LCM
),
a
n
d
s
h
a
p
e
a
tt
ri
b
u
tes
e
x
trac
ted
fro
m
c
o
n
t
o
u
r
d
e
sc
rip
to
rs
a
n
d
H
u
m
o
m
e
n
ts
a
re
c
a
p
tu
re
d
a
n
d
f
u
se
d
wit
h
d
e
e
p
f
e
a
tu
re
s
b
y
we
ig
h
ted
f
u
sio
n
stra
teg
y
,
re
su
lt
e
d
in
th
e
m
o
st
d
isc
rimin
a
ti
v
e
in
f
o
rm
a
ti
o
n
.
Th
e
fu
se
d
fe
a
tu
re
s
a
re
c
a
teg
o
rize
d
u
si
n
g
a
su
p
p
o
rt
v
e
c
to
r
m
a
c
h
in
e
(S
VM)
in
a
c
las
sifica
ti
o
n
p
h
a
se
,
w
h
ich
e
ffe
c
ti
v
e
ly
c
las
sifies
d
iffere
n
t
c
las
se
s
o
f
p
o
m
e
g
ra
n
a
te
fru
it
d
ise
a
se
s.
Th
e
c
o
m
b
in
e
d
d
e
e
p
a
n
d
h
a
n
d
c
ra
fted
fe
a
tu
re
s
o
b
tai
n
e
d
9
6
.
6
6
%
a
c
c
u
ra
c
y
,
9
6
.
2
6
%
p
re
c
isio
n
,
9
6
.
5
0
%
re
c
a
ll
,
9
6
.
3
7
%
F
1
-
sc
o
re
,
a
n
d
9
5
.
6
4
%
sp
e
c
ifi
c
it
y
o
n
t
h
e
p
o
m
e
g
ra
n
a
te
fru
it
d
ise
a
se
d
a
tas
e
t
wh
ich
c
o
m
p
a
re
d
to
e
x
isti
n
g
tec
h
n
iq
u
e
s.
K
ey
w
o
r
d
s
:
C
L
AHE
Dee
p
f
ea
tu
r
es
Def
o
r
m
ab
le
s
p
atial
p
y
r
am
id
p
o
o
lin
g
Han
d
cr
af
ted
f
ea
tu
r
es
Po
m
eg
r
an
ate
f
r
u
it d
is
ea
s
e
Su
p
p
o
r
t
v
ec
to
r
m
ac
h
in
e
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
:
Har
is
h
B
o
m
m
en
ah
alli M
allik
ar
ju
n
aiah
Dep
ar
tm
en
t o
f
I
n
f
o
r
m
atio
n
Scien
ce
an
d
E
n
g
in
ee
r
i
n
g
,
Ma
ln
a
d
C
o
lleg
e
o
f
E
n
g
in
ee
r
i
n
g
-
Has
s
an
Af
f
iliated
to
Vis
v
esv
ar
ay
a
T
e
ch
n
o
lo
g
ical
Un
iv
er
s
ity
B
elag
av
i
,
I
n
d
ia
E
m
ail:
h
ar
is
h
b
m
1
3
7
@
g
m
ail.
co
m
1.
I
NT
RO
D
UCT
I
O
N
Po
m
eg
r
an
ate,
k
n
o
wn
f
o
r
t
h
eir
h
ig
h
n
u
tr
ie
n
t
co
n
tex
t
an
d
u
n
i
q
u
e
b
le
n
d
o
f
s
wee
t
an
d
tar
t
f
l
av
o
r
s
,
is
a
m
ajo
r
f
r
u
it
cr
o
p
p
r
im
ar
ily
c
u
ltiv
ated
in
ar
id
an
d
s
em
i
-
ar
id
ar
ea
s
[
1
]
,
[
2
]
.
T
h
is
h
as
d
if
f
er
en
t
q
u
alities
an
d
is
n
ativ
e
o
n
Af
g
h
an
is
tan
an
d
I
r
a
n
.
I
n
r
ec
e
n
t
y
ea
r
s
,
p
o
m
e
g
r
an
a
te
cu
ltiv
atio
n
h
as
ex
ten
d
ed
to
ar
ea
s
o
v
er
Af
r
ica,
Au
s
tr
alia,
Am
er
ica,
E
u
r
o
p
e
,
a
n
d
Mid
d
le
E
ast
[
3
]
–
[
5
]
.
C
o
n
s
u
m
er
s
en
jo
y
p
o
m
eg
r
a
n
ates
in
d
if
f
er
e
n
t
f
o
r
m
ats,
in
v
o
lv
es
f
r
esh
f
r
u
it,
ju
ice,
o
i
l,
jam
,
an
d
in
f
u
s
io
n
[
6
]
.
Fru
its
ar
e
p
r
ed
is
p
o
s
ed
to
d
is
ea
s
es
an
d
p
ar
ticu
lar
ly
p
o
m
eg
r
a
n
ates
ar
e
g
r
ea
tly
p
r
o
n
e
to
in
f
ec
tio
n
s
in
d
if
f
e
r
en
t
d
ev
elo
p
m
en
t
p
h
ases
[
7
]
.
T
h
e
v
u
ln
er
a
b
ilit
y
o
f
p
o
m
eg
r
a
n
ates
to
d
is
ea
s
es
ca
u
s
es
ef
f
ec
tiv
e
ec
o
n
o
m
ic
ch
alle
n
g
es
to
f
ar
m
er
s
,
h
i
g
h
lig
h
tin
g
th
e
r
eq
u
ir
em
e
n
t
to
id
en
tify
d
ev
elo
p
m
en
t
p
h
ases
f
o
r
s
u
itab
le
p
lan
t
m
o
n
ito
r
in
g
an
d
f
u
lf
ilm
en
t
o
n
c
r
itical
r
eq
u
ir
em
e
n
ts
[
8
]
.
Ob
tain
in
g
k
n
o
wled
g
e
ab
o
u
t
v
ar
ian
t
s
eg
m
en
ts
o
n
p
o
m
eg
r
an
a
te
f
r
u
it
g
r
o
win
g
en
s
u
r
es
a
c
u
ltiv
ato
r
s
m
ak
e
b
etter
ch
o
ices
ab
o
u
t
f
er
tili
za
tio
n
,
c
o
n
tr
o
llin
g
p
ests
an
d
o
p
tim
u
m
tim
in
g
o
f
h
ar
v
est,
th
r
o
u
g
h
o
p
tim
izin
g
th
e
f
r
u
it
q
u
ality
an
d
y
ield
[
9
]
.
T
h
is
is
s
ig
n
if
ican
t
f
o
r
ac
k
n
o
wled
g
in
g
th
e
tim
in
g
an
d
attr
ib
u
tes
in
t
eg
r
ated
with
ev
e
r
y
d
ev
elo
p
m
e
n
t
p
h
ase
th
at
v
ar
y
b
ased
o
n
p
ar
am
ete
r
s
lik
e
t
h
e
v
ar
iety
o
f
p
o
m
eg
r
a
n
ate,
t
h
e
co
n
d
itio
n
s
o
f
th
e
en
v
ir
o
n
m
en
t
an
d
th
e
c
r
iter
ia
f
o
r
cu
ltiv
atio
n
[
1
0
]
.
T
h
e
cu
ltiv
atio
n
o
f
p
o
m
e
g
r
an
ate
h
as
ess
en
tial
s
ig
n
if
ican
ce
in
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
Defo
r
ma
b
le
s
p
a
tia
l p
yra
mid
p
o
o
lin
g
-
e
n
h
a
n
ce
d
E
fficien
tN
et
w
ith
…
(
Ha
r
is
h
B
o
mme
n
a
h
a
ll
i Ma
llika
r
ju
n
a
ia
h
)
643
th
e
ag
r
icu
ltu
r
al
f
ield
b
ec
au
s
e
o
f
th
e
g
r
o
win
g
d
em
an
d
f
o
r
th
eir
n
u
tr
itio
u
s
f
r
u
its
an
d
th
eir
d
if
f
er
en
t
a
p
p
licatio
n
s
in
b
ev
er
a
g
es,
f
o
o
d
an
d
c
o
s
m
etics
[
1
1
]
.
T
h
o
u
g
h
,
th
e
in
d
u
s
tr
y
s
tr
u
g
g
les
with
s
ev
er
al
ch
allen
g
es,
in
clu
d
in
g
d
is
ea
s
e
m
an
ag
em
en
t,
alig
n
in
g
q
u
ality
ev
alu
atio
n
with
d
ev
elo
p
m
en
t
s
tag
es,
f
lu
ctu
ati
o
n
o
f
clim
ate
an
d
m
an
ag
in
g
s
o
il
n
u
tr
ien
t
b
alan
ce
[
1
2
]
–
[
1
4
]
.
Far
m
e
r
s
o
f
ten
s
tr
u
g
g
le
to
p
r
ec
is
ely
id
en
tify
th
e
p
o
m
eg
r
an
at
e
d
ev
elo
p
m
e
n
t
p
h
ases
,
lead
in
g
to
in
ef
f
icien
t
r
eso
u
r
ce
allo
c
atio
n
an
d
a
f
f
ec
tin
g
d
ec
is
io
n
s
r
elate
d
to
h
ar
v
est
tim
in
g
an
d
e
n
tire
cr
o
p
m
an
ag
in
g
[
1
5
]
–
[
1
8
]
.
I
m
p
r
ec
is
io
n
in
id
en
tific
atio
n
r
esu
lted
i
n
s
o
u
r
ce
wastag
e,
lo
s
s
es
in
f
in
an
ce
an
d
d
r
awb
a
ck
s
o
n
o
p
p
o
r
tu
n
ities
to
ad
v
a
n
ce
d
ef
f
ec
tiv
e
n
ess
an
d
s
u
s
tain
ab
ilit
y
o
n
f
ar
m
in
g
p
r
ac
tices
[
1
9
]
.
Fo
r
ad
d
r
ess
in
g
th
ese
ch
allen
g
es,
ad
v
an
ce
d
te
ch
n
iq
u
es
lik
e
tr
an
s
f
er
an
d
d
ee
p
lear
n
in
g
(
DL
)
ar
e
ass
ig
n
ed
f
o
r
m
itig
atin
g
is
s
u
es
f
ac
ed
th
r
o
u
g
h
f
ar
m
er
s
an
d
ex
p
lo
r
in
g
p
o
te
n
tial a
v
en
u
es o
n
s
im
p
le
an
d
im
p
r
o
v
e
d
ag
r
icu
ltu
r
al
p
r
ac
tices.
W
an
g
et
a
l.
[
2
0
]
d
ev
elo
p
e
d
to
m
ato
leaf
d
is
ea
s
e
d
etec
tio
n
al
g
o
r
ith
m
-
b
ased
atten
tio
n
m
ec
h
an
is
m
s
an
d
m
u
lti
-
s
ca
le
f
ea
tu
r
e
f
u
s
io
n
.
I
n
i
tially
,
co
n
v
o
l
u
tio
n
al
b
lo
c
k
att
en
tio
n
m
o
d
u
le
(
C
B
AM
)
was
im
p
lem
en
ted
in
a
b
ac
k
b
o
n
e
f
ea
tu
r
e
e
x
tr
ac
tio
n
n
etwo
r
k
f
o
r
im
p
r
o
v
in
g
a
ca
p
a
b
ilit
y
to
ca
p
tu
r
e
af
f
ec
ted
f
ea
t
u
r
es.
Nex
t,
s
h
allo
w
f
ea
tu
r
e
m
ap
s
wer
e
d
ev
elo
p
e
d
to
r
e
-
p
ar
am
eter
ized
g
en
er
alize
d
f
ea
tu
r
e
p
y
r
am
id
n
etw
o
r
k
(
R
ep
GPFN)
to
en
h
an
ce
lo
ca
lizatio
n
ca
p
ab
ilit
y
f
o
r
s
m
all
lesi
o
n
f
ea
tu
r
es.
A
t
last
,
R
ep
GFPN
r
ep
lace
d
p
ath
ag
g
r
eg
atio
n
f
ea
t
u
r
e
p
y
r
am
id
n
etwo
r
k
(
PAFP
N)
in
a
y
o
u
o
n
ly
lo
o
k
o
n
ce
v
e
r
s
io
n
6
(
YOL
Ov
6
)
m
eth
o
d
f
o
r
o
b
tain
in
g
ef
f
icien
t
in
teg
r
atio
n
o
f
d
ee
p
s
em
an
tic
an
d
s
h
allo
w
s
p
atial
d
ata.
Mo
d
el
s
u
f
f
er
ed
f
r
o
m
n
o
is
e
an
d
c
o
m
p
r
ess
io
n
ar
tifa
cts,
wh
ich
m
in
im
izes
q
u
ality
o
f
d
is
ea
s
e
d
etec
tio
n
.
Kh
an
et
a
l.
[
2
1
]
p
r
esen
ted
a
m
eth
o
d
wh
ic
h
em
p
lo
y
ed
r
o
b
u
s
t
f
ea
tu
r
e
ex
tr
ac
tio
n
,
in
clu
d
in
g
g
r
ay
lev
el
co
-
o
cc
u
r
r
en
ce
m
atr
ix
(
GL
C
M
)
an
d
s
ca
le
in
v
ar
ia
n
t
f
ea
tu
r
e
tr
an
s
f
o
r
m
(
SIFT
)
,
in
teg
r
ated
with
s
u
p
p
o
r
t
v
ec
to
r
m
ac
h
in
e
(
SVM
)
f
o
r
ef
f
ec
tiv
e
class
if
icatio
n
.
T
h
e
e
x
ten
d
ed
d
ataset
o
f
2
,
7
0
0
to
m
at
o
leaf
im
ag
es
wit
h
a
m
in
im
u
m
o
f
3
0
0
im
ag
es
f
o
r
ev
e
r
y
n
in
e
d
if
f
e
r
en
t
d
is
ea
s
e
class
es.
T
h
is
d
ata
f
ac
ilit
ated
th
e
tr
ain
in
g
an
d
t
esti
n
g
o
f
v
ar
io
u
s
m
ac
h
in
e
le
ar
n
in
g
(
ML
)
a
n
d
DL
-
b
ased
alg
o
r
ith
m
s
.
T
h
o
u
g
h
u
n
ev
en
lig
h
tin
g
a
n
d
l
o
w
co
n
tr
ast
m
ak
e
d
is
ea
s
e
s
y
m
p
to
m
s
co
m
p
lex
f
o
r
d
etec
tio
n
,
em
p
lo
y
i
n
g
co
n
tr
ast
lim
ited
ad
ap
tiv
e
h
is
to
g
r
am
eq
u
aliza
tio
n
(
C
L
AHE
)
im
p
r
o
v
es
lo
ca
l
co
n
tr
ast
an
d
en
h
an
ce
s
v
is
ib
ilit
y
o
f
th
ese
s
y
m
p
to
m
s
.
O
ad
et
a
l.
[
2
2
]
s
u
g
g
ested
th
e
ar
tific
ial
in
tellig
en
ce
(
AI
)
m
eth
o
d
,
wh
ic
h
d
ete
cted
an
d
ex
p
lain
ed
p
lan
t
d
is
ea
s
es
b
y
im
ag
e
an
aly
s
is
.
T
h
e
s
u
g
g
ested
m
eth
o
d
id
en
tifie
d
s
ev
er
al
d
is
ea
s
es
in
f
r
u
its
an
d
v
eg
etab
les
th
r
o
u
g
h
ass
ig
n
in
g
an
en
s
em
b
le
lear
n
in
g
class
if
ier
with
f
o
u
r
DL
alg
o
r
ith
m
s
v
is
u
al
g
eo
m
etr
y
g
r
o
u
p
1
6
(
VGG1
6
)
,
VGG1
9
,
R
esNet1
0
1
v
2
,
an
d
I
n
ce
p
tio
n
-
V3
.
Ad
d
it
io
n
ally
,
p
r
o
v
id
e
d
ex
p
lan
atio
n
s
f
o
r
p
r
ed
ictio
n
s
b
y
lo
ca
l
in
ter
p
r
etab
le
m
o
d
el
-
ag
n
o
s
tic
ex
p
lan
atio
n
s
(
L
I
ME
)
em
p
lo
y
ed
f
o
r
i
n
ter
p
r
etin
g
p
r
ed
ictio
n
s
o
f
DL
alg
o
r
ith
m
s
.
Vis
u
aliza
tio
n
s
p
r
o
d
u
ce
d
f
r
o
m
s
ev
e
r
al
alg
o
r
ith
m
s
f
o
r
p
ix
els
in
f
lu
en
ce
o
n
p
r
ec
is
e
an
d
in
c
o
r
r
ec
t
p
r
ed
ictio
n
s
.
T
h
e
s
u
g
g
ested
m
eth
o
d
m
is
s
ed
s
u
b
tle
co
lo
r
,
te
x
tu
r
e
,
a
n
d
s
h
ap
e
f
ea
tu
r
es;
in
co
n
tr
ast,
th
is
ar
ticle
ca
p
tu
r
ed
d
etailed
h
a
n
d
cr
af
te
d
f
ea
tu
r
es
f
o
r
p
r
ec
is
ely
ex
tr
ac
tin
g
v
is
u
al
v
ar
iatio
n
s
.
Naseer
et
a
l.
[
2
3
]
in
tr
o
d
u
ce
d
tr
an
s
f
er
lear
n
in
g
-
en
ab
led
C
R
Net
alg
o
r
ith
m
to
ca
p
tu
r
i
n
g
s
p
a
tial f
ea
tu
r
es f
r
o
m
p
o
m
e
g
r
an
at
e
im
ag
es in
5
p
h
ases
o
f
p
o
m
eg
r
a
n
ate
g
r
o
wth
.
T
h
e
ca
p
tu
r
ed
s
p
atial
f
ea
tu
r
es
wer
e
f
ed
in
to
r
an
d
o
m
f
o
r
est
(
R
F)
alg
o
r
ith
m
,
r
esu
ltin
g
in
a
d
ev
elo
p
m
en
t
o
f
n
ew
p
r
o
b
a
b
ilis
tic
f
ea
tu
r
e
s
et.
T
h
ese
f
ea
tu
r
es
ass
is
ted
in
p
r
e
cisely
id
en
tify
in
g
p
o
m
eg
r
a
n
ate
d
e
v
elo
p
m
e
n
tal
p
h
ases
.
Fo
r
e
v
alu
atin
g
p
er
f
o
r
m
a
n
ce
,
ex
is
tin
g
class
if
icatio
n
alg
o
r
ith
m
s
co
n
s
id
er
ed
wer
e
c
o
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
(
C
NN)
,
k
-
n
eig
h
b
o
r
s
class
if
ier
(
KNC),
lo
g
is
tic
r
eg
r
ess
io
n
(
L
R
)
,
an
d
Gau
s
s
ian
n
aïv
e
B
ay
es
(
GNB).
T
r
ad
itio
n
al
ce
ll
-
lev
el
lear
n
in
g
(
C
L
L
)
alg
o
r
it
h
m
s
h
av
e
d
if
f
icu
lt
y
h
an
d
lin
g
m
u
lti
-
s
ca
le
an
d
ir
r
e
g
u
lar
d
is
ea
s
e
p
atter
n
s
.
C
o
m
b
in
in
g
d
ef
o
r
m
ab
le
s
p
atial
p
y
r
a
m
id
p
o
o
lin
g
(
DSPP
)
with
E
f
f
icien
tNet
en
ab
les ad
a
p
tiv
e,
m
u
lti
-
s
ca
le
p
o
o
lin
g
co
n
ce
n
tr
ated
o
n
d
is
ea
s
e
-
r
elev
an
t r
eg
io
n
s
.
Hu
et
a
l.
[
2
4
]
im
p
lem
e
n
ted
th
e
lig
h
tweig
h
t
d
etec
tio
n
tech
n
i
q
u
e
th
at
d
e
p
en
d
s
o
n
en
h
a
n
ce
d
YOL
Ov
5
.
I
n
itially
,
Fas
ter
-
C
3
m
o
d
u
le
was
d
ev
elo
p
ed
to
r
e
p
lace
th
e
ac
tu
al
cr
o
s
s
s
tag
e
p
ar
tial
(
C
S
P
)
m
o
d
u
le
in
YOL
Ov
5
f
o
r
ef
f
ec
tiv
ely
m
i
n
im
izin
g
n
u
m
b
er
o
f
p
a
r
am
eter
s
in
th
e
f
ea
tu
r
e
ex
tr
ac
tio
n
p
r
o
ce
s
s
.
T
h
en
,
C
o
o
r
d
C
o
n
v
a
n
d
en
h
an
ce
d
c
o
n
ten
t
-
awa
r
e
r
ea
s
s
em
b
ly
o
f
f
ea
tu
r
es
(
C
AR
AF
E
)
wer
e
im
p
lem
en
ted
i
n
th
e
n
ec
k
n
etwo
r
k
f
o
r
en
h
an
cin
g
th
e
r
e
f
in
em
en
t
o
f
p
o
s
itio
n
d
ata
in
f
e
atu
r
e
f
u
s
io
n
an
d
r
ef
i
n
in
g
s
em
an
tic
d
ata
in
th
e
d
o
wn
-
s
am
p
lin
g
p
r
o
ce
s
s
.
At
last
,
ch
an
n
el
-
wis
e
k
n
o
wled
g
e
d
is
till
atio
n
alg
o
r
ith
m
was
u
tili
ze
d
in
m
o
d
el
t
r
ain
in
g
f
o
r
e
n
h
an
cin
g
d
etec
tio
n
ac
cu
r
ac
y
with
o
u
t
m
ax
im
izin
g
th
e
n
u
m
b
er
o
f
m
o
d
el
p
ar
am
eter
s
.
C
o
n
s
id
er
in
g
h
a
n
d
cr
af
ted
an
d
d
ee
p
f
ea
tu
r
es
s
ep
ar
ately
ca
u
s
es
in
co
m
p
lete
d
ata
r
ep
r
esen
tatio
n
;
a
weig
h
ted
f
u
s
io
n
s
tr
ateg
y
was
d
ev
elo
p
ed
t
o
ef
f
ec
tiv
ely
in
te
g
r
ate
b
o
th
f
ea
t
u
r
e
ty
p
es.
J
ian
g
et
a
l.
[
2
5
]
d
ev
elo
p
ed
th
e
m
eth
o
d
b
ased
o
n
ef
f
icien
t
f
ea
tu
r
e
s
eg
m
en
tatio
n
tr
an
s
f
o
r
m
er
(
E
FS
-
F
o
r
m
er
)
.
T
h
e
e
x
ten
d
e
d
lo
ca
l
d
etail
(
E
L
D)
m
o
d
u
le
ex
t
en
d
ed
th
e
r
ec
ep
tiv
e
f
ield
o
f
th
e
m
o
d
el
th
r
o
u
g
h
e
x
t
en
d
in
g
th
e
c
o
n
v
o
lu
tio
n
,
g
o
o
d
h
an
d
lin
g
o
f
f
in
e
s
p
o
ts
an
d
ef
f
i
cien
tly
m
in
im
izin
g
d
ata
lo
s
s
.
H
-
atten
tio
n
m
in
im
ized
co
m
p
u
tatio
n
al
r
e
d
u
n
d
an
cy
th
r
o
u
g
h
im
p
o
s
in
g
m
u
lti
-
lay
er
co
n
v
o
lu
tio
n
s
,
en
h
an
cin
g
f
ea
tu
r
e
f
ilter
in
g
.
Par
allel
f
u
s
io
n
f
r
am
ewo
r
k
e
f
f
icien
tly
u
tili
ze
d
v
ar
io
u
s
in
ter
v
als
o
f
f
ea
tu
r
e
ex
tr
ac
tio
n
o
f
C
NN
an
d
tr
an
s
f
o
r
m
er
s
en
c
o
d
er
s
,
o
b
tain
in
g
in
tr
in
s
ic
f
ea
tu
r
e
e
x
tr
ac
tio
n
a
n
d
in
teg
r
atin
g
s
em
an
tic
d
ata
in
ch
an
n
el
an
d
s
p
atial
d
im
en
s
io
n
s
in
f
ea
tu
r
e
f
u
s
io
n
m
o
d
u
le
(
FF
M)
.
T
h
e
in
itial
class
if
ier
s
f
ailed
to
d
if
f
er
en
tiate
b
etwe
en
clo
s
ely
r
esem
b
lin
g
d
is
ea
s
e
s
y
m
p
to
m
s
;
in
co
r
p
o
r
atin
g
a
n
SVM
class
if
ier
im
p
r
o
v
ed
b
o
u
n
d
ar
y
-
e
n
ab
led
s
ep
ar
atio
n
,
im
p
r
o
v
in
g
o
v
er
all
class
if
icatio
n
ac
cu
r
ac
y
.
T
r
ad
itio
n
al
d
is
ea
s
e
d
etec
tio
n
an
d
class
if
icatio
n
alg
o
r
ith
m
s
ar
e
m
an
u
al,
tim
e
-
co
n
s
u
m
in
g
an
d
lead
to
h
u
m
an
e
r
r
o
r
,
m
ak
in
g
tim
ely
an
d
p
r
ec
is
e
d
iag
n
o
s
is
ch
allen
g
in
g
.
Vis
u
al
s
y
m
p
to
m
s
lik
e
d
is
co
lo
r
atio
n
,
t
ex
tu
r
e
ch
an
g
es
an
d
s
h
ap
e
d
ef
o
r
m
atio
n
ar
e
s
u
b
tle
an
d
v
a
r
y
in
in
ten
s
ity
,
ca
u
s
in
g
ch
allen
g
es
f
o
r
tr
ad
itio
n
al
m
eth
o
d
s
.
Ad
d
itio
n
ally
,
i
n
co
n
s
is
ten
t
lig
h
tin
g
an
d
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2252
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
1
5
,
No
.
1
,
Feb
r
u
ar
y
2
0
2
6
:
642
-
6
5
4
644
im
ag
e
n
o
is
e
r
ed
u
ce
d
r
eliab
le
f
ea
tu
r
e
ex
tr
ac
tio
n
.
T
h
e
p
r
im
ar
y
o
b
jectiv
e
o
f
th
is
p
a
p
er
is
to
d
ev
elo
p
h
y
b
r
i
d
im
ag
e
-
b
ased
d
is
ea
s
e
d
etec
tio
n
alg
o
r
ith
m
f
o
r
p
o
m
e
g
r
an
ate
f
r
u
its
th
r
o
u
g
h
co
m
b
in
in
g
h
an
d
c
r
af
ted
with
DL
-
b
ased
f
ea
tu
r
es.
T
h
e
h
an
d
cr
af
ted
f
ea
t
u
r
es
lik
e
c
o
lo
r
,
te
x
tu
r
e
a
n
d
s
h
a
p
e
ar
e
in
teg
r
ate
d
with
d
ee
p
f
e
atu
r
es
ca
p
tu
r
ed
b
y
DSPP
-
E
f
f
icien
tN
et.
T
h
en
,
th
e
weig
h
ted
f
ea
tu
r
e
f
u
s
io
n
s
tr
ateg
y
is
em
p
lo
y
ed
f
o
r
in
teg
r
atin
g
th
ese
f
ea
tu
r
es a
n
d
class
if
icatio
n
is
p
er
f
o
r
m
e
d
b
y
SVM.
I
n
th
is
m
an
u
s
cr
ip
t,
p
r
o
p
o
s
ed
DSPP
-
en
h
an
ce
d
E
f
f
icien
tNet
with
weig
h
ted
f
ea
tu
r
e
f
u
s
io
n
i
s
d
ev
elo
p
ed
in
p
o
m
eg
r
a
n
ate
d
is
ea
s
e
class
i
f
icatio
n
f
o
r
ch
allen
g
es
lik
e
v
ar
y
in
g
lesi
o
n
s
h
ap
es,
s
u
b
tle
tex
tu
r
e
d
if
f
er
e
n
ce
an
d
co
m
p
lex
b
ac
k
g
r
o
u
n
d
s
.
E
f
f
icien
tNet
ac
t
as
lig
h
tweig
h
t
m
o
d
el
f
o
r
d
ee
p
f
ea
tu
r
e
ex
tr
ac
tio
n
,
p
r
o
v
id
es
o
p
tim
al
b
alan
ce
b
etwe
en
ac
cu
r
ac
y
a
n
d
co
m
p
u
tatio
n
al
co
s
t.
T
h
e
en
h
an
ce
m
en
t
o
f
m
o
d
el’
s
c
ap
ab
ilit
y
to
ex
tr
ac
t
m
u
lti
-
s
ca
le
co
n
tex
tu
al
d
ata,
t
h
e
DSPP
i
s
in
co
r
p
o
r
ated
,
all
o
ws
th
e
m
o
d
el
to
ca
p
tu
r
e
f
e
atu
r
es
in
d
if
f
er
en
t
r
ec
ep
tiv
e
f
ield
s
.
T
o
ad
d
r
ess
th
e
ch
allen
g
es
o
n
d
ee
p
f
ea
tu
r
e
s
,
p
r
o
p
o
s
ed
m
o
d
el
in
co
r
p
o
r
at
ed
weig
h
ted
f
u
s
io
n
s
tr
ateg
y
wh
ich
in
teg
r
ated
d
ee
p
f
ea
tu
r
es
with
h
an
d
cr
a
f
ter
f
ea
tu
r
es.
T
h
is
f
u
s
io
n
p
r
o
ce
s
s
en
s
u
r
es
th
e
f
ea
tu
r
e
s
p
ac
e
th
r
o
u
g
h
s
em
an
tic
d
ata
with
d
o
m
ain
-
r
elev
a
n
t lo
w
-
lev
el
p
atter
n
s
,
en
h
an
cin
g
class
s
e
p
ar
ab
ilit
y
.
T
h
e
u
s
ag
e
o
f
lear
n
i
n
g
weig
h
ts
in
f
u
s
io
n
en
s
u
r
es
f
ea
tu
r
e
t
y
p
e
p
r
o
p
o
r
tio
n
ally
f
o
r
f
in
al
d
ec
is
io
n
,
im
p
r
o
v
in
g
r
o
b
u
s
tn
ess
to
im
ag
e
v
ar
iab
ilit
y
.
T
h
e
s
ig
n
if
ic
an
t c
o
n
tr
ib
u
tio
n
s
o
f
t
h
e
r
esear
ch
ar
e
d
escr
ib
e
d
as f
o
llo
ws
.
i)
Han
d
cr
a
f
ter
f
ea
t
u
r
es
lik
e
c
o
lo
r
h
is
to
g
r
am
s
,
GL
C
M
-
b
ased
te
x
tu
r
e
d
escr
ip
to
r
s
an
d
Hu
m
o
m
en
ts
an
d
d
ee
p
f
ea
tu
r
es
f
r
o
m
d
e
v
elo
p
e
d
DSPP
-
E
f
f
icien
tNet
ar
e
ca
p
tu
r
ed
t
o
d
if
f
er
e
n
tiate
th
e
lo
w
-
lev
el
an
d
h
ig
h
-
lev
el
ch
ar
ac
ter
is
tics
o
f
d
is
ea
s
es.
ii)
I
n
teg
r
atio
n
o
f
DSPP
with
E
f
f
i
cien
tNet
f
o
r
ef
f
ec
tiv
ely
f
o
c
u
s
in
g
o
n
d
is
ea
s
e
r
elev
an
ce
r
eg
io
n
s
,
b
y
u
tili
zin
g
ad
ju
s
tab
le
p
o
o
lin
g
g
r
i
d
s
an
d
e
n
h
an
cin
g
m
u
lti
-
s
ca
le
f
ea
tu
r
e
r
ep
r
esen
tatio
n
.
iii)
T
h
e
weig
h
ted
f
ea
tu
r
e
f
u
s
io
n
s
tr
ateg
y
in
te
g
r
ates
h
an
d
cr
af
ted
an
d
d
ee
p
f
ea
t
u
r
es,
wh
ich
b
ala
n
ce
s
f
o
r
m
u
ch
p
r
ec
is
e
an
d
r
o
b
u
s
t d
is
ea
s
e
class
if
icatio
n
.
T
h
is
r
esear
ch
p
ap
er
is
o
r
g
a
n
ized
as
f
o
llo
ws:
s
ec
tio
n
2
p
r
o
v
id
es
d
etails
o
f
a
p
r
o
p
o
s
ed
DSPP
-
E
f
f
icien
tNet
alg
o
r
ith
m
.
Sectio
n
3
v
alid
ates
a
p
er
f
o
r
m
an
ce
o
f
th
e
DSPP
-
E
f
f
icien
tNet
alg
o
r
ith
m
.
Fin
ally
,
th
e
co
n
clu
s
io
n
is
g
iv
e
n
in
s
ec
tio
n
4
.
2.
P
RO
P
O
SE
D
S
E
CT
I
O
N
T
h
is
p
ap
er
p
r
o
p
o
s
es
a
n
o
v
el
d
is
ea
s
e
d
etec
tio
n
m
o
d
el
th
at
in
teg
r
ates
h
an
d
cr
af
ted
f
ea
tu
r
es
with
d
ee
p
f
ea
tu
r
es
ex
tr
ac
ted
u
s
in
g
a
DSPP
en
h
an
ce
d
E
f
f
icien
tNet
ar
c
h
itectu
r
e.
T
h
e
p
o
m
eg
r
an
ate
f
r
u
it
d
is
ea
s
e
d
ataset
is
u
s
ed
in
th
is
ar
ticle
an
d
th
e
im
ag
es
ar
e
p
r
e
-
p
r
o
ce
s
s
ed
b
y
u
s
in
g
a
m
ed
ian
f
ilter
an
d
C
L
AHE
.
Nex
t,
th
e
h
an
d
cr
a
f
ted
an
d
d
ee
p
f
e
atu
r
es
ar
e
ex
tr
ac
ted
b
y
DSPP
-
E
f
f
icien
tNet
an
d
th
ese
f
ea
tu
r
e
s
ar
e
f
u
s
ed
u
s
in
g
weig
h
ted
f
ea
tu
r
e
f
u
s
io
n
.
I
n
th
e
class
if
icatio
n
p
h
ase,
we
u
s
ed
th
e
SVM
f
o
r
m
u
lti
-
class
class
if
icatio
n
.
Fig
u
r
e
1
r
ep
r
esen
ts
th
e
p
r
o
ce
s
s
o
f
p
o
m
eg
r
an
ate
f
r
u
it d
is
ea
s
e
class
if
ic
atio
n
.
Fig
u
r
e
1
.
Pro
ce
s
s
o
f
p
o
m
e
g
r
an
ate
f
r
u
it d
is
ea
s
e
class
if
icatio
n
2
.
1
.
Da
t
a
s
et
T
h
is
p
ap
er
u
s
es
th
e
p
u
b
licly
a
v
ailab
le
p
o
m
eg
r
a
n
ate
f
r
u
it
d
is
ea
s
e
d
ataset,
a
s
tan
d
ar
d
ized
r
e
s
o
u
r
ce
f
o
r
ca
teg
o
r
izin
g
p
o
m
e
g
r
an
ate
f
r
u
it
d
is
ea
s
es
[
2
6
]
.
T
h
e
d
ataset
co
n
tain
s
5
,
0
9
9
lab
elled
an
d
c
lass
if
ied
im
ag
es
o
f
h
ea
lth
y
an
d
d
is
ea
s
ed
p
o
m
eg
r
a
n
ate
f
r
u
it,
d
iv
i
d
ed
in
to
f
i
v
e
v
a
r
io
u
s
p
o
m
eg
r
an
ate
f
r
u
it
d
is
ea
s
es
class
e
s
s
u
ch
as
h
ea
lth
y
,
b
ac
ter
ial
b
lig
h
t,
an
th
r
ac
n
o
s
e,
ce
r
co
s
p
o
r
a
f
r
u
it
s
p
o
t
,
an
d
alter
n
ar
ia
f
r
u
it
s
p
o
t.
T
h
e
r
aw
im
ag
es
ar
e
g
iv
en
as in
p
u
t to
p
r
e
-
p
r
o
ce
s
s
in
g
p
h
ase
to
im
p
r
o
v
e
th
e
q
u
ality
o
f
im
ag
es.
2
.
2
.
P
re
-
pro
ce
s
s
ing
2
.
2
.
1
.
M
edia
n f
ilte
r
I
t
is
a
n
o
n
-
lin
ea
r
s
m
o
o
th
i
n
g
m
eth
o
d
u
tili
ze
d
f
o
r
elim
in
atin
g
th
e
im
p
u
ls
e
n
o
is
e
wh
en
p
r
e
s
er
v
in
g
th
e
ed
g
es.
T
h
e
win
d
o
w
s
ize
5
×
5
s
lid
es
o
v
er
th
e
im
ag
e.
Fo
r
ev
er
y
p
i
x
el,
th
e
win
d
o
w
ca
p
tu
r
e
s
n
eig
h
b
o
r
in
g
p
ix
el
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
Defo
r
ma
b
le
s
p
a
tia
l p
yra
mid
p
o
o
lin
g
-
e
n
h
a
n
ce
d
E
fficien
tN
et
w
ith
…
(
Ha
r
is
h
B
o
mme
n
a
h
a
ll
i Ma
llika
r
ju
n
a
ia
h
)
645
v
alu
es.
T
h
e
v
alu
es
ar
e
s
o
r
ted
an
d
m
ed
ian
is
ch
o
s
en
.
T
h
en
,
t
h
e
ce
n
ter
p
ix
el
is
r
ep
lace
d
wit
h
th
e
m
ed
ian
v
al
u
es
an
d
its
m
ath
em
atica
l
f
o
r
m
u
la
is
g
iv
en
as
(
1
)
.
I
t
r
em
o
v
es
b
a
ck
g
r
o
u
n
d
n
o
is
e
ca
u
s
ed
b
y
co
m
p
r
ess
io
n
ar
tifa
cts.
Un
lik
e
m
ea
n
f
ilter
,
m
ed
ian
f
i
lter
p
r
eser
v
es
ed
g
es,
th
at
ar
e
ess
en
tial
f
o
r
id
en
tify
in
g
d
is
ea
s
e
b
o
u
n
d
ar
ies
an
d
r
etain
s
f
in
e
p
atter
n
s
lik
e
s
p
o
ts
o
r
d
is
ea
s
ed
r
eg
io
n
s
o
n
a
lea
f
.
(
,
)
=
{
(
,
)
|
(
,
)
∈
(
,
)
}
(
1
)
2
.
2
.
2
.
CL
AH
E
C
L
AHE
is
a
s
tr
o
n
g
m
et
h
o
d
f
o
r
co
n
t
r
as
t e
n
h
a
n
c
em
en
t,
p
ar
tic
u
lar
ly
in
n
o
n
-
u
n
i
f
o
r
m
l
ig
h
t
in
g
c
o
n
d
iti
o
n
s
.
T
h
is
is
v
a
r
ia
n
t
o
f
a
d
a
p
ti
v
e
h
is
t
o
g
r
am
eq
u
aliz
ati
o
n
(
AH
E
)
w
h
ich
li
m
its
t
h
e
am
p
l
if
ic
ati
o
n
o
f
n
o
is
e
.
T
h
e
im
ag
e
is
s
ep
ar
ate
d
to
s
m
all
r
e
g
i
o
n
s
ca
l
l
ed
til
es.
Fo
r
ev
er
y
til
e,
e
x
e
cu
t
e
a
h
is
to
g
r
a
m
a
n
d
r
e
d
is
tr
i
b
u
t
e
th
e
p
ix
el
in
te
n
s
it
ies
to
f
la
tte
n
t
h
e
h
is
to
g
r
a
m
.
C
li
p
t
h
e
h
is
t
o
g
r
a
m
t
o
t
h
e
t
h
r
es
h
o
l
d
t
o
a
v
o
i
d
n
o
is
e
o
v
er
-
a
m
p
li
f
i
ca
t
i
o
n
.
Ass
ig
n
b
il
in
ea
r
in
t
er
p
o
l
ati
o
n
f
o
r
s
m
o
o
t
h
i
n
g
t
r
a
n
s
it
io
n
s
a
m
o
n
g
til
es,
a
n
d
its
m
ath
em
ati
ca
l
f
o
r
m
u
la
is
g
iv
e
n
a
s
(
2
)
.
(
,
)
=
(
,
)
(
2
)
T
h
is
im
p
r
o
v
es
th
e
lo
ca
l
c
o
n
tr
ast,
u
s
ed
to
h
ig
h
lig
h
t
th
e
l
o
c
al
d
is
ea
s
e
f
ea
tu
r
es
lik
e
d
is
co
l
o
r
atio
n
o
r
p
atch
es.
Als
o
d
ea
ls
with
u
n
e
v
en
lig
h
tin
g
,
co
m
p
en
s
atin
g
to
n
o
n
-
u
n
if
o
r
m
illu
m
in
atio
n
in
ca
p
tu
r
e
d
im
ag
es
.
E
n
h
an
ce
s
th
e
v
is
ib
ilit
y
o
f
b
o
t
h
s
u
b
tle
an
d
s
tr
o
n
g
d
is
ea
s
e
-
r
ele
v
an
t te
x
tu
r
es a
n
d
co
lo
r
v
ar
iati
o
n
s
.
2
.
3
.
F
e
a
t
ure
ex
t
r
a
ct
io
n
T
h
e
p
r
e
-
p
r
o
ce
s
s
ed
im
ag
es
ar
e
f
ed
as
in
p
u
t
to
a
f
ea
t
u
r
e
ex
tr
ac
tio
n
p
h
ase
to
ca
p
tu
r
e
h
an
d
c
r
af
ted
an
d
d
ee
p
f
ea
tu
r
es
to
d
if
f
e
r
en
tiatin
g
th
e
v
ar
io
u
s
class
es
o
f
d
is
ea
s
e
r
eg
io
n
s
in
p
o
m
eg
r
a
n
ate
f
r
u
its
.
T
h
e
ca
p
tu
r
ed
,
h
an
d
cr
a
f
ted
an
d
d
ee
p
f
ea
tu
r
e
s
to
d
if
f
er
e
n
tiate
d
is
ea
s
ed
r
eg
io
n
s
ar
e
d
escr
ib
e
d
in
t
h
is
s
ec
tio
n
.
T
h
e
ex
tr
ac
ted
h
an
d
cr
a
f
ted
an
d
d
ee
p
f
ea
tu
r
es
ar
e
ex
p
lain
ed
in
th
is
s
u
b
s
ec
tio
n
.
2
.
3
.
1
.
Co
lo
r
f
e
a
t
ures
I
n
th
is
s
ec
tio
n
,
th
e
h
an
d
cr
af
te
d
f
ea
tu
r
es
s
u
ch
as
co
lo
r
f
ea
tu
r
es,
tex
tu
r
e
f
ea
tu
r
es
an
d
s
h
ap
e
f
ea
tu
r
es
ar
e
ca
p
tu
r
ed
a
n
d
ex
p
lain
e
d
as f
o
llo
ws.
Nu
m
er
o
u
s
p
lan
ts
h
av
e
co
lo
r
ch
an
g
es
lik
e
y
ello
win
g
,
b
r
o
wn
s
p
o
ts
,
p
ale
p
atch
es,
ca
p
tu
r
in
g
th
e
c
o
lo
r
d
is
tr
ib
u
tio
n
h
elp
s
to
id
en
tify
th
ese
s
y
m
p
to
m
s
.
I
n
itially
co
n
v
er
ts
th
e
im
a
g
e
to
R
GB
an
d
HSV
co
lo
r
s
p
ac
es.
T
h
en
d
iv
id
e
ev
er
y
ch
an
n
el
in
to
b
in
s
an
d
c
o
u
n
t
th
e
n
u
m
b
er
o
f
p
ix
els
in
ev
er
y
b
i
n
to
f
o
r
m
a
h
is
to
g
r
am
,
at
last
ly
n
o
r
m
alize
th
e
h
is
to
g
r
am
to
d
e
v
elo
p
a
f
ea
t
u
r
e
v
ec
t
o
r
.
2
.
3
.
2
.
T
y
pes
o
f
co
lo
r
f
e
a
t
ures
Dif
f
er
en
t c
o
lo
r
f
ea
t
u
r
e
r
ep
r
ese
n
tatio
n
s
ar
e
o
f
ten
u
tili
ze
d
in
i
m
ag
e
an
aly
s
is
to
ca
p
tu
r
e
d
if
f
e
r
en
ce
s
in
co
lo
r
in
f
o
r
m
atio
n
.
i)
R
GB
h
is
to
g
r
a
m
:
ca
p
t
u
r
es
d
is
tr
ib
u
tio
n
o
f
r
ed
,
g
r
ee
n
an
d
b
lu
r
in
ten
s
ities
.
I
t
h
elp
s
to
id
en
ti
f
y
r
aw
co
lo
r
ch
an
g
es
ii)
HSV
h
is
to
g
r
am
:
it
ef
f
ec
tiv
ely
s
ep
ar
ates
co
lo
r
(
h
u
e)
f
r
o
m
illu
m
in
atio
n
(
v
alu
e
)
a
n
d
s
atu
r
a
tio
n
,
m
ak
in
g
it
in
v
ar
ian
t
to
lig
h
tin
g
co
n
d
i
tio
n
s
.
T
h
is
is
r
o
b
u
s
t
f
o
r
lig
h
tin
g
c
o
n
d
itio
n
s
an
d
h
ig
h
lig
h
ts
th
e
s
u
b
tle
d
is
ea
s
e
-
s
p
ec
if
ic
co
lo
r
s
h
if
ts
.
2
.
3
.
3
.
T
ex
t
ure
f
ea
t
ures
T
o
ca
p
tu
r
e
t
h
e
tex
tu
r
e
f
ea
tu
r
es,
in
th
is
ar
ticle
th
e
GL
C
M
is
u
s
ed
.
I
t
em
p
lo
y
s
a
c
o
n
ce
p
t
o
f
p
i
x
el
in
ten
s
ity
d
is
tr
ib
u
tio
n
,
th
at
i
n
clu
d
es
b
lack
,
wh
ite
,
an
d
v
ar
io
u
s
g
r
a
y
s
h
a
d
es.
I
n
im
a
g
e
f
o
r
ea
c
h
p
i
x
el,
h
o
m
o
g
en
eity
v
alu
e
is
m
ea
s
u
r
ed
an
d
c
h
an
g
es
a
r
e
ac
q
u
ir
ed
th
en
th
e
r
e
h
as
g
r
ea
test
ch
a
n
c
e
to
g
et
ab
n
o
r
m
al
r
eg
io
n
.
I
n
a
p
r
e
-
p
r
o
c
ess
ed
im
ag
e,
d
is
ea
s
e
ar
ea
s
o
f
f
er
ir
r
eg
u
lar
s
u
r
f
ac
e
p
atter
n
s
lik
e
r
o
u
g
h
n
ess
,
lesi
o
n
s
,
p
o
wd
er
y
tex
tu
r
es,
th
ese
ar
e
ca
p
tu
r
es
b
y
tex
t
u
r
e
d
escr
ip
to
r
s
.
GL
C
M
ca
p
tu
r
es
s
p
atial
r
elatio
n
s
h
ip
s
am
o
n
g
p
ix
el
in
ten
s
ities
,
r
ep
r
esen
tin
g
h
o
w
p
air
s
o
f
g
r
a
y
lev
els
an
d
o
cc
u
r
at
ce
r
tain
an
g
les
an
d
d
is
ta
n
ce
s
.
C
o
n
v
er
t
th
e
im
ag
e
in
to
g
r
ay
s
ca
le
an
d
e
x
ec
u
te
th
e
GL
C
Ms in
m
u
ltip
le
o
r
i
en
tatio
n
s
.
Fin
ally
,
s
tatis
tical
f
e
atu
r
es a
r
e
ca
p
tu
r
es
f
r
o
m
e
v
er
y
o
r
ien
tatio
n
.
T
ex
tu
r
e
f
ea
tu
r
es
ex
tr
ac
ted
f
r
o
m
GL
C
M
in
clu
d
e:
i)
C
o
n
tr
ast
: i
t c
ap
tu
r
es lo
ca
l in
ten
s
ity
v
ar
iatio
n
s
ii)
C
o
r
r
elatio
n
: i
t c
ap
tu
r
es p
ix
el
c
o
r
r
elatio
n
iii)
E
n
er
g
y
: i
t c
a
p
tu
r
es tex
tu
r
al
u
n
if
o
r
m
ity
iv
)
Ho
m
o
g
en
eity
: i
t c
a
p
tu
r
es th
e
c
lo
s
en
ess
o
f
d
iag
o
n
al
d
is
tr
ib
u
tio
n
2
.
3
.
4
.
Sh
a
pe
f
ea
t
ures
T
h
e
d
is
ea
s
es
ca
u
s
e
leaf
d
ef
o
r
m
atio
n
,
cu
r
lin
g
o
r
h
o
les
th
at
alter
s
n
o
r
m
al
s
h
ap
e
g
eo
m
etr
y
.
C
ap
tu
r
ed
th
e
leaf
co
n
to
u
r
b
y
ed
g
e
d
et
ec
tio
n
lik
e
ca
n
n
y
an
d
id
e
n
tif
y
ex
ter
n
al
b
o
u
n
d
ar
ies.
E
x
ec
u
te
th
e
asp
ec
t
r
atio
,
ex
ten
d
s
th
e
leaf
ar
ea
o
r
b
o
u
n
d
in
g
b
o
x
ar
ea
an
d
s
o
lid
ity
.
T
h
ese
ca
p
tu
r
e
th
e
g
eo
m
etr
ic
d
i
s
to
r
tio
n
s
b
ec
au
s
e
o
f
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2252
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
1
5
,
No
.
1
,
Feb
r
u
ar
y
2
0
2
6
:
642
-
6
5
4
646
th
e
d
is
ea
s
e.
Hu
m
o
m
en
ts
ar
e
t
h
e
s
h
ap
e
d
escr
ip
to
r
s
d
r
iv
en
f
r
o
m
im
ag
e
m
o
m
e
n
ts
,
wh
ich
ar
e
r
o
tatio
n
,
s
ca
le
an
d
tr
an
s
latio
n
in
v
ar
ian
t.
I
t
e
n
co
d
es
th
e
s
p
atial
d
is
tr
ib
u
tio
n
o
f
p
i
x
els.
[
]
=
f
u
n
ctio
n
o
f
ce
n
ter
m
o
m
en
ts
u
p
to
3
rd
o
r
d
er
.
Ge
n
er
ally
,
s
ev
en
Hu
m
o
m
en
ts
ar
e
ca
p
t
u
r
ed
p
er
im
ag
e.
Sh
ap
e
d
escr
ip
to
r
s
ca
p
tu
r
e
th
e
g
lo
b
al
d
ef
o
r
m
atio
n
s
lik
e
s
tr
u
ctu
r
al
d
am
ag
e.
Hu
m
o
m
en
ts
p
r
o
v
id
e
a
co
m
p
ac
t
an
d
in
v
ar
ia
n
t
s
h
ap
e
r
ep
r
esen
tatio
n
.
Ma
th
em
atica
l
f
o
r
m
u
la
f
o
r
c
o
m
p
u
tin
g
h
an
d
cr
af
ted
f
ea
tu
r
es
is
g
iv
en
in
(
3
)
.
I
n
(
3
)
,
th
e
r
ep
r
esen
ts
h
an
d
cr
a
f
ted
f
ea
t
u
r
es,
th
e
r
ep
r
esen
ts
ca
p
tu
r
ed
c
o
lo
r
ed
f
ea
tu
r
es,
th
e
r
ep
r
esen
ts
ca
p
tu
r
ed
te
x
tu
r
e
f
ea
t
u
r
es
an
d
r
ep
r
esen
ts
ca
p
tu
r
ed
s
h
ap
e
f
ea
tu
r
es.
=
[
,
,
]
(
3
)
2
.
3
.
5
.
Ca
pturing
deep
f
ea
t
ures us
i
ng
E
f
f
icient
Net
E
f
f
icien
tNet
is
C
NN
ar
ch
itect
u
r
e
c
r
ea
ted
t
h
r
o
u
g
h
Go
o
g
le
B
r
ain
T
ea
m
.
T
h
is
ar
ticle
a
n
aly
z
es
n
etwo
r
k
s
ca
lin
g
an
d
id
en
tifie
s
wh
ich
o
p
tim
izin
g
th
e
d
e
p
th
,
wid
t
h
,
an
d
r
eso
lu
ti
o
n
o
f
t
h
e
n
et
wo
r
k
en
h
a
n
ce
s
its
p
er
f
o
r
m
an
ce
.
Fo
r
d
ev
elo
p
in
g
th
e
n
ew
m
o
d
el,
it
s
ca
les
th
e
n
eu
r
al
n
etwo
r
k
f
o
r
cr
ea
tin
g
D
L
-
b
ased
alg
o
r
ith
m
s
th
at
p
r
o
v
id
e
m
u
ch
h
i
g
h
e
f
f
ec
tiv
en
ess
an
d
ac
c
u
r
ac
y
wh
ile
co
m
p
ar
in
g
with
p
r
e
v
io
u
s
ly
u
tili
ze
d
C
NNs.
T
h
e
E
f
f
icien
tNet
p
r
o
ce
s
s
ed
h
ig
h
-
s
ca
le
v
is
u
al
r
ec
o
g
n
itio
n
with
co
n
s
is
ten
cy
an
d
ac
cu
r
ac
y
.
W
h
ile
co
m
p
a
r
in
g
to
o
th
er
alg
o
r
ith
m
s
lik
e
VGGN
et,
Go
o
g
leNe
t,
Xce
p
tio
n
,
R
esNet
,
an
d
I
n
ce
p
tio
n
Net
,
th
e
E
f
f
icien
tNet
o
u
tp
er
f
o
r
m
ed
th
ese
alg
o
r
ith
m
s
.
E
f
f
icien
tNet
u
tili
ze
s
co
m
p
o
s
ite
s
ca
lin
g
alg
o
r
ith
m
th
at
d
ev
elo
p
s
v
a
r
io
u
s
m
eth
o
d
s
in
C
NN.
Nu
m
b
er
o
f
lay
er
s
in
th
e
n
etwo
r
k
with
r
es
p
ec
t
to
n
etwo
r
k
d
e
p
th
.
C
o
n
v
o
l
u
tio
n
al
lay
e
r
wid
th
is
p
r
o
p
o
r
tio
n
al
to
am
o
u
n
t
o
f
f
ilter
s
in
clu
d
ed
.
Heig
h
t
a
n
d
wid
th
o
f
in
p
u
t
im
ag
e
d
e
f
in
e
a
r
eso
lu
tio
n
o
f
t
h
e
im
ag
e.
Fig
u
r
e
2
r
ep
r
esen
ts
an
ar
ch
itectu
r
e
o
f
E
f
f
icien
tNet
f
o
r
f
ea
tu
r
e
e
x
tr
ac
tio
n
.
Fig
u
r
e
2
.
Ar
c
h
itectu
r
e
o
f
E
f
f
ic
ien
tNet
f
o
r
f
ea
tu
r
e
ex
tr
ac
tio
n
T
h
e
alg
o
r
ith
m
ca
p
tu
r
es
th
e
c
h
ar
ac
ter
is
tics
o
v
er
lay
er
s
b
y
s
e
v
er
al
co
n
v
o
lu
tio
n
al
lay
er
s
with
r
ec
ep
tiv
e
f
ield
o
f
3
×
3
an
d
a
m
o
b
ile
in
v
er
ted
b
o
ttlen
ec
k
C
o
n
v
.
Ma
th
e
m
atica
l
f
r
o
m
(
4
)
to
(
8
)
,
r
e
p
r
es
en
ts
to
s
ca
le
d
ep
th
,
wid
th
,
an
d
r
eso
lu
tio
n
with
r
es
p
ec
t
to
.
I
n
(
4
)
to
(
8
)
,
th
e
,
,
an
d
r
ep
r
esen
t
d
ep
th
,
wid
th
an
d
r
eso
lu
tio
n
o
f
th
e
n
etwo
r
k
an
d
,
,
an
d
r
ep
r
esen
ts
co
n
s
tan
t
ter
m
s
ar
e
d
ef
in
ed
b
y
a
g
r
id
s
ea
r
c
h
h
y
p
er
p
a
r
a
m
eter
s
tu
n
in
g
m
eth
o
d
.
C
o
ef
f
icien
t is th
e
u
s
er
-
d
ef
in
e
d
v
ar
ia
b
le
th
at
h
an
d
les wh
o
le
-
s
ca
le
s
o
u
r
ce
s
o
f
th
e
m
eth
o
d
.
=
(
4
)
=
(
5
)
=
(
6
)
.
.
2
.
2
≈
2
,
(
7
)
≥
1
,
≥
1
,
≥
1
.
(
8
)
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
Defo
r
ma
b
le
s
p
a
tia
l p
yra
mid
p
o
o
lin
g
-
e
n
h
a
n
ce
d
E
fficien
tN
et
w
ith
…
(
Ha
r
is
h
B
o
mme
n
a
h
a
ll
i Ma
llika
r
ju
n
a
ia
h
)
647
T
h
is
m
eth
o
d
a
d
ju
s
ted
th
e
d
ep
th
,
wid
th
an
d
r
eso
lu
tio
n
o
f
th
e
n
etwo
r
k
f
o
r
o
p
tim
izin
g
th
e
n
etwo
r
k
’
s
ac
cu
r
ac
y
an
d
m
em
o
r
y
co
n
s
u
m
p
tio
n
u
s
in
g
a
v
ailab
le
r
eso
u
r
ce
s
.
E
f
f
icien
tNet
ad
ju
s
ts
ev
er
y
d
im
e
n
s
io
n
b
y
p
r
e
-
d
ef
i
n
ed
g
r
o
u
p
o
f
s
ca
lin
g
co
ef
f
icien
ts
,
o
u
tp
er
f
o
r
m
i
n
g
o
th
er
DL
-
b
ased
alg
o
r
ith
m
s
.
T
h
e
m
eth
o
d
is
r
elea
s
ed
with
s
ca
lin
g
lev
els
r
a
n
g
in
g
f
r
o
m
0
to
7
,
wh
er
e
ev
er
y
le
v
e
l
r
ep
r
esen
ts
a
n
in
c
r
ea
s
e
in
ac
cu
r
ac
y
an
d
m
o
d
el
p
ar
am
eter
s
ize.
W
ith
r
ec
en
t
ad
v
an
ce
m
e
n
ts
,
E
f
f
icien
tNet
p
r
o
v
id
es
e
n
h
an
ce
d
u
b
i
q
u
ito
u
s
co
n
n
ec
tiv
ity
an
d
b
r
in
g
s
ca
p
a
b
ilit
ies o
f
DL
to
v
a
r
io
u
s
p
latf
o
r
m
s
,
ef
f
ec
tiv
ely
m
ee
tin
g
d
if
f
er
e
n
t a
p
p
licatio
n
r
e
q
u
ir
em
en
ts
.
i)
Sp
atial
p
y
r
am
id
p
o
o
lin
g
m
o
d
u
le
(
SP
P)
:
th
e
SP
P
m
o
d
u
le
is
d
r
iv
en
f
r
o
m
p
y
r
am
id
s
ce
n
e
p
a
r
s
in
g
n
etwo
r
k
(
PS
PNet
)
,
th
e
p
y
r
am
id
p
o
o
lin
g
m
itig
ates
th
e
d
r
awb
ac
k
s
o
f
f
ix
ed
-
s
ize
r
eq
u
i
r
em
en
t f
o
r
C
N
N
in
p
u
t
im
a
g
e.
T
h
is
is
em
p
lo
y
ed
f
o
r
e
x
tr
ac
tin
g
f
ea
tu
r
es
at
m
u
ltip
le
s
ca
les
an
d
s
y
n
th
esizes
th
e
g
l
o
b
al
d
ata.
Fig
u
r
e
3
r
ep
r
esen
ts
th
e
a
r
ch
itectu
r
e
o
f
DSPP
.
Py
r
am
id
p
o
o
lin
g
m
o
d
u
le
i
n
clu
d
es
4
s
tag
es
s
u
c
h
as
p
y
r
am
i
d
p
o
o
lin
g
,
co
n
v
o
lu
tio
n
,
u
p
-
s
am
p
lin
g
an
d
co
n
ca
ten
atio
n
p
r
o
ce
s
s
.
B
y
p
y
r
am
id
p
o
o
lin
g
,
s
p
atia
l
f
ea
tu
r
es
o
n
4
v
ar
io
u
s
s
p
atial
s
ca
les
ar
e
d
ete
cted
.
Fo
r
im
p
r
o
v
in
g
th
e
ca
p
ab
ilit
y
o
f
n
o
n
-
lin
ea
r
lear
n
i
n
g
o
f
m
u
ltip
le
s
ca
le
f
ea
tu
r
es,
1
×1
c
o
n
v
o
lu
tio
n
is
i
n
co
r
p
o
r
ated
f
o
r
h
an
d
lin
g
f
ea
t
u
r
e
s
ize
an
d
to
m
in
im
ize
n
u
m
b
er
o
f
f
ea
tu
r
e
ch
an
n
el
th
r
o
u
g
h
−
ℎ
am
o
u
n
t
o
f
ch
an
n
els
o
f
an
in
p
u
t
f
ea
tu
r
e
m
ap
,
wh
e
r
e
r
ep
r
esen
ts
am
o
u
n
t
o
f
p
y
r
am
id
p
o
o
lin
g
s
ca
les.
C
o
n
v
o
lu
ted
f
ea
tu
r
e
m
ap
s
a
r
e
in
s
er
t
ed
b
y
a
b
ilin
ea
r
f
ilter
f
o
r
m
atc
h
in
g
in
p
u
t
s
ize
o
f
a
f
ea
tu
r
e
m
a
p
.
I
n
p
u
t
f
ea
tu
r
e
m
ap
s
ar
e
f
u
s
ed
with
4
u
p
-
s
am
p
led
f
ea
t
u
r
e
m
ap
s
,
s
o
th
at
th
e
g
lo
b
al
co
n
tex
t
f
ea
tu
r
es
ar
e
m
ain
tain
e
d
to
m
u
ltip
le
s
ca
le
f
ea
tu
r
es.
F
o
u
r
lev
els
with
s
izes
o
f
1
×1
,
2
×2
,
3
×3
,
an
d
6
×6
ar
e
u
tili
ze
d
in
SP
P m
o
d
u
l
e.
Fig
u
r
e
3
.
Ar
c
h
itectu
r
e
o
f
DSPP
ii)
D
S
PP
m
o
d
i
f
i
es
t
h
e
SP
P
t
h
r
o
u
g
h
m
a
k
i
n
g
t
h
e
p
o
s
i
t
i
o
n
o
f
p
o
o
l
in
g
l
a
y
e
r
s
l
e
a
r
n
a
b
l
e
.
t
h
e
p
o
o
l
i
n
g
a
r
e
a
s
d
e
f
o
r
m
s
p
a
t
i
al
l
y
b
a
s
e
d
o
n
i
n
p
u
t
f
e
a
t
u
r
e
s
,
s
o
t
h
a
t
t
h
e
y
a
l
i
g
n
w
it
h
s
e
m
a
n
t
i
c
o
r
d
i
s
e
as
es
a
r
e
a
s
.
E
v
e
r
y
p
o
o
l
i
n
g
l
a
y
e
r
’
s
s
a
m
p
l
i
n
g
p
o
s
i
ti
o
n
is
a
d
j
u
s
t
e
d
th
r
o
u
g
h
o
f
f
s
e
t
,
l
e
a
r
n
s
d
y
n
a
m
i
ca
l
l
y
t
h
r
o
u
g
h
a
l
i
g
h
t
we
i
g
h
t
s
u
b
-
n
e
t
w
o
r
k
.
T
h
e
m
a
t
h
e
m
a
t
i
ca
l
f
o
r
m
u
l
a
f
o
r
t
h
is
p
r
o
c
e
s
s
is
g
i
v
e
n
i
n
(
9
)
.
′
=
+
∆
(
9
)
I
n
(
9
)
,
th
e
r
ep
r
esen
ts
ac
tu
al
p
o
o
lin
g
p
o
s
itio
n
,
th
e
∆
r
ep
r
esen
ts
lear
n
ed
f
ea
tu
r
e
an
d
th
e
′
r
ep
r
esen
ts
f
in
al
s
am
p
le
p
o
s
itio
n
u
tili
ze
d
in
p
o
o
lin
g
.
T
h
ese
o
f
f
s
ets
∆
ar
e
ac
q
u
ir
ed
f
r
o
m
an
in
d
iv
i
d
u
al
c
o
n
v
o
lu
ti
o
n
lay
er
tr
ain
ed
an
d
in
teg
r
ated
with
th
e
m
eth
o
d
.
DSPP
m
o
d
i
f
ies
wh
er
e
th
e
f
ea
tu
r
es
ar
e
p
o
o
led
b
ased
o
n
in
p
u
t
co
n
tex
t.
T
h
is
en
ab
les
th
e
m
o
d
el
to
f
o
c
u
s
ef
f
ec
tiv
ely
o
n
d
is
ea
s
e
-
af
f
ec
ted
ar
ea
s
,
ev
e
n
wh
en
t
h
ese
r
eg
io
n
s
ar
e
d
is
to
r
ted
an
d
ir
r
e
g
u
lar
.
T
h
e
ex
tr
ac
te
d
h
an
d
c
r
af
ted
an
d
d
ee
p
f
ea
tu
r
es
ar
e
f
u
s
ed
to
g
eth
er
b
y
weig
h
ted
f
ea
tu
r
e
f
u
s
io
n
a
n
d
it
is
ex
p
lain
ed
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2252
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
1
5
,
No
.
1
,
Feb
r
u
ar
y
2
0
2
6
:
642
-
6
5
4
648
3
.
4
.
Weig
hte
d f
ea
t
ure
f
us
io
n
W
eig
h
t
o
f
ea
ch
f
ea
tu
r
e
is
co
n
s
id
er
ed
as
ev
alu
atio
n
o
f
i
ts
f
ea
tu
r
e
s
ig
n
if
ican
ce
.
Sen
s
itiv
ity
o
f
ev
alu
ated
f
ea
tu
r
es
is
d
ef
in
ed
t
h
r
o
u
g
h
em
p
lo
y
in
g
v
alu
es
with
in
r
an
g
e
o
f
[
0
,
1
]
.
T
h
e
e
x
tr
ac
te
d
h
a
n
d
-
cr
a
f
ted
an
d
d
ee
p
f
ea
t
u
r
es
ar
e
co
n
ca
te
n
ated
b
ef
o
r
e
b
ein
g
f
ed
in
to
th
e
S
VM
class
if
ier
f
o
r
class
if
icatio
n
.
T
h
e
r
ep
r
esen
ts
weig
h
t,
wh
ich
r
ep
r
esen
ts
ca
p
ab
ilit
y
to
class
if
y
th
e
d
if
f
er
e
n
t
class
es
o
f
p
o
m
e
g
r
an
ates
a
n
d
its
m
ath
em
atica
l
ex
p
r
ess
io
n
is
,
as g
iv
en
in
(
1
0
)
,
wh
er
e
r
ep
r
esen
ts
n
u
m
b
e
r
o
f
f
ea
tu
r
es.
=
∑
=
1
,
=
1
,
2
,
…
,
(
1
0
)
B
ef
o
r
e
p
er
f
o
r
m
in
g
weig
h
ted
f
ea
tu
r
e
f
u
s
io
n
,
th
at
is
ess
en
tial
f
o
r
s
tan
d
ar
d
ize
f
ea
tu
r
es
to
p
r
ev
en
t
to
ein
g
with
s
m
aller
d
ata
v
alu
es
f
r
o
m
f
ea
tu
r
es
with
h
ig
h
e
r
d
ata
v
alu
es.
T
h
is
en
s
u
r
es
t
h
at
th
e
ca
lcu
latio
n
o
u
tco
m
es
ar
e
n
o
t
d
is
to
r
ted
b
ec
au
s
e
o
f
v
ar
i
o
u
s
d
im
e
n
s
io
n
s
in
f
ea
tu
r
es.
Ma
th
em
atica
l
f
o
r
m
u
la
f
o
r
n
o
r
m
aliza
tio
n
o
f
f
ea
t
u
r
e
v
alu
e
o
f
ℎ
is
g
iv
en
in
(
1
1
)
.
′
=
−
(
)
(
)
−
(
)
,
=
1
,
2
,
…
,
(
1
1
)
T
h
e
f
ea
tu
r
e
t
h
at
d
ef
in
es
a
d
ata
to
h
ig
h
est
ex
ten
t
is
ac
q
u
i
r
ed
b
y
m
u
ltip
ly
in
g
h
a
n
d
cr
af
te
d
f
e
atu
r
es
an
d
d
ee
p
f
ea
tu
r
es
th
r
o
u
g
h
th
eir
co
r
r
esp
o
n
d
i
n
g
weig
h
t
,
th
en
s
u
m
m
in
g
th
e
r
esu
lts
,
with
to
tal
v
alu
e
n
o
r
m
alize
d
to
1
.
Han
d
c
r
af
ted
f
e
atu
r
es
lik
e
co
lo
r
h
is
to
g
r
am
s
,
tex
tu
r
e
p
a
tter
n
s
an
d
s
h
ap
e
d
escr
ip
t
o
r
s
c
ap
tu
r
ed
b
y
d
if
f
er
en
t
tr
ad
itio
n
al
alg
o
r
ith
m
s
an
d
d
ee
p
f
ea
tu
r
es
ar
e
a
u
to
m
atica
lly
le
ar
n
ed
a
n
d
en
co
d
e
h
i
g
h
-
lev
el
s
em
an
tic
an
d
s
p
atial
d
ata
lik
e
co
m
p
lex
tex
tu
r
es
an
d
co
n
tex
tu
al
p
atter
s
ca
p
tu
r
ed
b
y
DSPP
-
E
f
f
icien
tNet
ar
e
f
u
s
ed
.
T
h
e
v
ec
to
r
o
f
h
an
d
cr
a
f
ted
f
ea
tu
r
es
is
r
ep
r
ese
n
ted
as
∈
an
d
v
ec
to
r
o
f
d
ee
p
f
e
atu
r
es
ar
e
r
ep
r
esen
ted
as
∈
.
Her
e,
th
ese
two
f
ea
tu
r
es
ar
e
c
o
m
b
in
ed
an
d
r
ep
r
esen
ted
as
.
T
h
e
m
ath
em
atica
l
f
o
r
m
u
la
f
o
r
wei
g
h
ted
f
ea
t
u
r
e
f
u
s
io
n
is
g
iv
en
in
(
1
2
)
.
I
n
(
1
2
)
,
th
e
,
∈
[
0
,
1
]
r
ep
r
esen
ts
s
ca
lar
we
ig
h
ts
an
d
b
y
ad
d
in
g
th
is
v
alu
e
is
1
,
th
e
(
.
)
r
ep
r
esen
ts
n
o
r
m
alize
d
f
u
n
ctio
n
.
=
∙
(
)
,
∙
(
)
(
1
2
)
3
.
5
.
Su
pp
o
rt
v
ec
t
o
r
ma
chine
SVM
is
cla
s
s
if
icat
io
n
m
o
d
el
d
ep
en
d
e
d
o
n
t
h
e
p
r
in
cip
le
o
f
s
tr
u
ctu
r
al
r
is
k
r
ed
u
ctio
n
.
Ma
in
aim
o
f
SVM
is
to
clas
s
if
y
d
if
f
er
en
t
s
am
p
les
an
d
in
cr
ea
s
e
th
e
s
p
ac
in
g
am
o
n
g
o
p
tim
ally
s
ep
ar
ated
h
y
p
er
p
la
n
e
an
d
th
e
en
tire
tr
ain
in
g
s
am
p
le
.
C
o
n
s
id
er
o
r
ig
in
al
d
ataset
an
d
it is
g
iv
en
as (
1
3
)
.
{
(
,
)
|
∈
,
∈
{
−
1
,
+
1
}
,
=
1
,
2
,
…
,
}
(
1
3
)
I
n
(
1
3
)
,
t
h
e
r
ep
r
esen
ts
n
u
m
b
e
r
o
f
tr
ai
n
in
g
d
ata
s
am
p
les,
th
e
r
ep
r
esen
ts
in
p
u
t
o
f
m
o
d
el,
th
e
is
d
im
en
s
io
n
o
f
tr
ai
n
in
g
s
am
p
le,
th
e
r
ep
r
esen
ts
s
am
p
le
class
,
-
1
an
d
1
r
ep
r
esen
t
ca
teg
o
r
y
l
ab
els.
Fo
r
lin
ea
r
s
ep
ar
ab
le,
eq
u
atio
n
f
o
r
s
ep
ar
atio
n
p
lan
e
is
p
r
esen
ted
as
∙
+
=
0
.
A
m
ath
em
atica
l
ex
p
r
ess
io
n
f
o
r
s
am
p
le
(
,
)
th
at
n
ee
d
s
to
s
atis
f
y
is
g
iv
en
i
n
(
1
4
)
.
[
(
∙
)
+
]
≥
1
,
=
1
,
2
,
…
,
(
1
4
)
I
n
(
1
4
)
,
th
e
r
e
p
r
esen
ts
th
e
n
o
r
m
al
v
ec
t
o
r
p
lan
e
a
n
d
r
e
p
r
e
s
en
ts
co
n
s
tan
t
ter
m
.
Dis
tan
ce
am
o
n
g
clo
s
est
s
am
p
lin
g
p
o
in
t
an
d
s
e
p
ar
atio
n
p
lan
e
is
r
ep
r
esen
ted
as
1
‖
‖
⁄
.
Hen
ce
,
h
ig
h
est
s
p
ac
i
n
g
o
f
1
‖
‖
⁄
i
s
eq
u
al
to
h
ig
h
est
v
alu
e
o
f
‖
‖
2
.
Sep
ar
atio
n
lin
e
d
ef
in
e
d
th
r
o
u
g
h
i
s
o
p
tim
u
m
s
ep
a
r
atio
n
lin
e
an
d
s
am
p
les
in
s
ep
ar
atio
n
lin
e
∙
+
=
±
1
ar
e
k
n
o
wn
as
s
u
p
p
o
r
t
v
ec
to
r
s
.
L
a
g
r
an
g
e
o
p
tim
izatio
n
alg
o
r
ith
m
s
em
p
lo
y
ed
f
o
r
co
n
v
er
tin
g
th
at
in
to
a
d
o
u
b
le
p
r
o
b
lem
.
Ma
th
em
atica
l f
o
r
m
u
la
f
o
r
m
ax
im
is
atio
n
f
u
n
ctio
n
is
g
iv
en
as (
1
5
)
.
(
)
=
∑
−
1
2
(
∙
)
=
1
(
1
5
)
I
n
(
1
5
)
,
th
e
r
ep
r
esen
ts
th
e
L
ag
r
an
g
e
m
u
ltip
lier
a
n
d
≥
0
,
=
1
,
…
,
.
T
h
is
ac
tu
ally
id
en
tifie
s
o
p
tim
u
m
s
o
lu
tio
n
to
q
u
ad
r
atic
f
u
n
ctio
n
with
co
n
s
tr
ain
ts
a
n
d
in
s
tan
ce
s
r
esp
ec
tiv
e
to
n
o
n
-
ze
r
o
in
s
u
p
p
o
r
t
v
ec
to
r
s
.
Ma
th
em
atica
l f
o
r
m
u
l
a
f
o
r
o
b
tain
in
g
o
p
tim
al
class
if
icatio
n
f
u
n
ctio
n
is
g
iv
en
in
(
1
6
)
.
(
)
=
{
(
.
)
+
∗
}
=
{
∑
∗
(
.
)
+
∗
=
1
}
(
1
6
)
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
Defo
r
ma
b
le
s
p
a
tia
l p
yra
mid
p
o
o
lin
g
-
e
n
h
a
n
ce
d
E
fficien
tN
et
w
ith
…
(
Ha
r
is
h
B
o
mme
n
a
h
a
ll
i Ma
llika
r
ju
n
a
ia
h
)
649
I
n
(
1
6
)
,
t
h
e
∗
r
e
p
r
e
s
e
n
t
s
o
p
ti
m
u
m
L
a
g
r
a
n
g
e
m
u
ltip
lier
a
n
d
t
h
e
∗
r
e
p
r
e
s
e
n
ts
c
l
as
s
i
f
i
c
at
i
o
n
t
h
r
esh
o
l
d
;
t
h
e
s
e
a
r
e
p
a
r
a
m
e
t
e
r
s
t
o
d
e
te
r
m
i
n
e
t
h
e
o
p
t
i
m
u
m
h
y
p
e
r
-
p
l
a
n
e
p
a
r
t
i
t
i
o
n
.
P
o
s
i
t
i
v
e
o
r
n
e
g
a
ti
v
e
f
u
n
c
t
i
o
n
r
e
p
r
e
s
e
n
ts
c
l
a
s
s
f
e
a
t
u
r
es
.
F
o
r
l
i
n
e
a
r
l
y
i
n
s
e
p
a
r
a
b
l
e
c
a
s
es
,
a
s
l
ac
k
v
a
r
i
a
b
le
i
s
e
m
p
l
o
y
e
d
,
t
r
a
n
s
f
o
r
m
i
n
g
t
h
e
i
s
s
u
e
o
f
f
i
n
d
i
n
g
t
h
e
h
y
p
e
r
p
l
a
n
e
i
n
t
o
a
q
u
a
d
r
a
t
i
c
p
r
o
g
r
a
m
m
i
n
g
p
r
o
b
l
e
m
a
n
d
i
t
s
m
a
t
h
e
m
a
t
i
ca
l
f
o
r
m
u
l
a
is
g
i
v
e
n
a
s
(
1
7
).
{
(
)
=
1
2
‖
‖
2
+
∑
=
1
.
.
[
(
∙
)
+
]
≥
1
−
,
≥
0
,
=
1
,
…
,
(
1
7
)
I
n
(
1
7
)
,
th
e
r
ep
r
esen
ts
p
o
s
itiv
e
s
lack
v
ar
iab
le
wh
ich
all
o
ws
m
is
class
if
icatio
n
,
d
escr
ib
ed
d
ev
iatio
n
n
u
m
b
er
o
f
r
esp
ec
tiv
e
d
ata
p
o
in
t
f
r
o
m
h
y
p
er
p
la
n
e.
T
h
e
r
ep
r
esen
ts
p
en
alty
f
ac
t
o
r
d
escr
ib
es
d
eg
r
ee
o
f
p
u
n
is
h
m
en
t
to
m
is
class
if
icati
o
n
.
T
h
is
is
u
tili
ze
d
to
co
n
tr
o
l
weig
h
t
am
o
n
g
h
y
p
er
p
lan
es
with
less
s
p
ac
in
g
in
f
itn
ess
f
u
n
ctio
n
an
d
g
u
ar
an
t
ee
s
m
in
im
u
m
d
ev
iatio
n
n
u
m
b
er
in
d
ata
p
o
i
n
t.
Pro
p
o
s
ed
m
eth
o
d
co
m
b
in
e
d
lig
h
tweig
h
t
E
f
f
icen
tNet
with
DSPP
to
im
p
r
v
e
m
u
lti
-
s
cle
f
ea
tu
r
e
e
x
tr
ac
tio
n
.
T
h
i
s
en
s
u
r
es
f
ea
tu
r
e
r
ep
r
esen
tatio
n
,
h
a
n
d
cr
af
te
d
f
e
atu
r
es
ex
tr
ac
t
co
lo
r
,
tex
tu
r
e
an
d
s
h
ap
e
d
ata
ar
e
in
teg
r
ated
with
d
ee
p
f
ea
tu
r
es
b
y
weig
h
ted
f
ea
tu
r
e
f
u
s
io
n
m
ec
h
an
is
m
.
T
h
is
h
y
b
r
id
m
o
d
el
u
s
es
lo
w
-
lev
el
an
d
h
ig
h
-
lev
el
d
es
cr
ip
to
r
s
,
en
ab
les
th
e
m
o
d
el
to
d
if
f
e
r
en
tiate
d
is
ea
s
e
class
es.
T
h
e
f
u
s
ed
f
ea
tu
r
e
s
ar
e
th
en
class
if
ied
b
y
d
en
s
e
n
eu
r
al
lay
e
r
b
y
So
f
tMa
x
f
u
n
ctio
n
as in
Alg
o
r
ith
m
1
.
Alg
o
r
ith
m
1
.
DSPP
-
E
f
f
icien
tNet
with
weig
h
ted
f
ea
tu
r
e
f
u
s
io
n
f
o
r
p
o
m
eg
r
a
n
ate
d
is
ea
s
e
class
if
icatio
n
I
n
p
u
t
:
p
o
m
eg
r
an
ate
f
r
u
it d
is
ea
s
e
d
ataset
Ou
tp
u
t
:
p
r
ed
icted
d
is
ea
s
e
class
f
o
r
ev
er
y
im
ag
e,
ev
alu
atio
n
m
etr
ics.
B
eg
in
Pre
-
p
r
o
ce
s
s
in
g
s
tag
e
Fo
r
ev
er
y
i
m
ag
e
in
:
R
esize
im
ag
e
to
u
n
if
o
r
m
s
ize
(
224
×
224
)
No
r
m
alize
p
ix
el
v
alu
es
Ap
p
ly
d
ata
a
u
g
m
en
tatio
n
Featu
r
e
E
x
tr
ac
tio
n
Fo
r
ev
er
y
i
m
ag
e
:
E
x
tr
ac
t h
an
d
cr
af
ted
f
ea
tu
r
es
=
(
)
Pas
s
b
y
E
f
f
icien
tNet
to
o
b
tain
d
ee
p
f
ea
tu
r
es
Ap
p
ly
DSPP
o
n
to
ac
q
u
ir
e
m
u
lti
-
s
ca
le
en
s
r
ich
ed
f
ea
tu
r
es
Featu
r
e
Fu
s
io
n
Fo
r
ev
er
y
s
am
p
le
Fu
s
e
f
ea
tu
r
es u
s
in
g
weig
h
ted
s
u
m
=
×
+
×
C
las
s
if
icatio
n
Pas
s
to
f
u
lly
c
o
n
n
ec
ted
lay
er
s
Ap
p
ly
So
f
tMa
x
ac
tiv
atio
n
to
o
b
tain
class
p
r
o
b
ab
ilit
ies
E
m
p
lo
y
class
lab
el
with
h
ig
h
e
s
t p
r
o
b
ab
ilit
y
Mo
d
el
T
r
ain
in
g
Def
in
e
lo
s
s
f
u
n
ctio
n
(
C
r
o
s
s
E
n
tr
o
p
y
L
o
s
s
)
Op
tim
ize
Usi
n
g
Ad
am
T
r
ain
m
o
d
el
o
v
er
ep
o
c
h
s
with
b
atch
s
ize
E
v
alu
atio
n
Use te
s
t/v
alid
atio
n
d
ata
to
co
m
p
u
te
Acc
u
r
ac
y
,
Pre
cisi
o
n
,
R
ec
all,
F1
-
s
co
r
e,
Sp
ec
if
icity
Ap
p
ly
K
-
f
o
l
d
cr
o
s
s
-
v
alid
atio
n
f
o
r
s
tab
ilit
y
.
E
n
d
3.
E
XP
E
R
I
M
E
N
T
A
L
RE
SUL
T
S
T
h
e
p
e
r
f
o
r
m
an
ce
o
f
DSP
-
E
f
f
i
cien
tNet
with
weig
h
ted
f
u
s
io
n
f
o
r
th
e
SVM
m
eth
o
d
is
s
im
u
lated
with
Py
th
o
n
en
v
i
r
o
n
m
e
n
t
an
d
s
y
s
tem
co
n
f
ig
u
r
atio
n
s
ar
e
8
GB
R
AM
,
W
in
d
o
ws
1
0
(
6
4
-
b
it)
an
d
i5
p
r
o
ce
s
s
o
r
.
Acc
u
r
ac
y
,
p
r
ec
is
io
n
,
r
e
ca
ll,
F
1
-
s
co
r
e,
an
d
s
p
ec
if
icity
ar
e
m
etr
ics
co
n
s
id
er
ed
in
th
is
r
esea
r
ch
to
v
alid
ate
th
e
p
er
f
o
r
m
an
ce
o
f
d
ev
elo
p
ed
tech
n
iq
u
e.
E
r
r
o
r
r
ates
lik
e
m
ea
n
s
q
u
ar
e
d
er
r
o
r
(
MSE
)
,
r
o
o
t
m
ea
n
s
q
u
a
r
e
er
r
o
r
(
R
MSE
)
,
m
ea
n
ab
s
o
lu
te
e
r
r
o
r
(
MA
E
)
,
m
ea
n
ab
s
o
lu
te
p
r
ec
is
io
n
er
r
o
r
(
MA
PR
)
,
an
d
co
e
f
f
icien
t
o
f
d
eter
m
in
atio
n
(R
2
)
ar
e
co
n
s
id
er
ed
to
ev
al
u
ate
th
e
p
er
f
o
r
m
a
n
ce
.
Ma
th
em
atica
l
f
o
r
m
u
la
f
o
r
th
ese
m
etr
ics
is
g
iv
en
f
r
o
m
(
1
8
)
t
o
(
2
2
)
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2252
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
1
5
,
No
.
1
,
Feb
r
u
ar
y
2
0
2
6
:
642
-
6
5
4
650
=
+
+
+
+
×
100
(
1
8
)
=
+
×
100
(
1
9
)
=
+
×
100
(
2
0
)
1
−
=
2
×
×
+
×
100
(
2
1
)
=
+
×
100
(
2
2
)
Fig
u
r
e
4
r
ep
r
esen
ts
th
e
clas
s
-
wis
e
r
esu
lts
f
o
r
th
e
p
o
m
e
g
r
an
ate
f
r
u
it
d
is
ea
s
e
d
ataset.
T
h
e
f
iv
e
d
if
f
er
en
t
class
es
s
u
ch
as
alter
n
ar
ia,
an
t
h
r
ac
n
o
s
e,
b
ac
ter
ial_
b
lig
h
t,
ce
r
c
o
s
p
o
r
a,
an
d
h
ea
lth
y
ar
e
ev
alu
ated
t
o
s
h
o
w
th
e
p
er
f
o
r
m
an
ce
o
f
th
e
d
ev
elo
p
e
d
tech
n
iq
u
e.
T
h
is
e
v
a
lu
atio
n
h
i
g
h
lig
h
ts
h
o
w
ea
ch
class
p
er
f
o
r
m
s
u
n
d
er
th
e
p
r
o
p
o
s
ed
DSPP
-
E
f
f
icien
tNet
m
eth
o
d
with
weig
h
te
d
f
ea
tu
r
e
f
u
s
io
n
.
Fig
u
r
e
4
.
Per
f
o
r
m
an
c
e
o
f
class
wis
e
r
esu
lts
f
o
r
p
o
m
eg
r
an
ate
f
r
u
it d
is
ea
s
e
d
ataset
I
n
T
a
b
l
e
1
,
t
h
e
p
e
r
f
o
r
m
a
n
c
e
o
f
c
o
m
b
i
n
e
d
h
a
n
d
c
r
a
f
t
e
d
a
n
d
d
e
e
p
f
e
a
t
u
r
e
s
w
i
t
h
d
i
f
f
e
r
e
n
t
p
e
r
f
o
r
m
a
n
c
e
m
e
t
r
i
c
s
a
r
e
e
v
al
u
a
t
e
d
.
H
e
r
e
,
h
a
n
d
c
r
a
f
t
e
d
f
e
a
t
u
r
es
s
u
c
h
a
s
c
o
l
o
r
,
t
e
x
t
u
r
e
,
a
n
d
s
h
a
p
e
f
ea
tu
r
e
s
a
r
e
e
v
a
l
u
a
t
e
d
i
n
d
i
v
i
d
u
a
l
l
y
a
n
d
c
o
m
b
i
n
e
d
.
T
h
e
n
,
t
h
e
d
e
e
p
a
t
t
r
i
b
u
te
s
a
r
e
e
v
alu
a
t
e
d
a
n
d
b
o
t
h
h
a
n
d
c
r
a
f
t
e
d
a
n
d
d
e
e
p
f
e
a
t
u
r
es
a
r
e
e
v
a
l
u
a
t
e
d
i
n
c
o
m
b
i
n
a
t
i
o
n
.
T
h
e
c
o
m
b
i
n
e
d
d
e
e
p
a
n
d
h
a
n
d
c
r
a
f
t
e
d
f
e
a
t
u
r
e
s
o
b
t
a
i
n
e
d
9
6
.
6
6
%
a
c
c
u
r
a
c
y
,
9
6
.
2
6
%
p
r
e
c
i
s
i
o
n
,
9
6
.
5
0
%
r
e
c
al
l
,
9
6
.
3
7
%
F
1
-
s
c
o
r
e
,
a
n
d
9
5
.
6
4
%
s
p
e
c
i
f
i
c
i
t
y
o
n
t
h
e
p
o
m
e
g
r
a
n
a
t
e
f
r
u
i
t
d
i
s
e
as
e
d
a
t
as
e
t
.
T
ab
le
1
.
Per
f
o
r
m
an
ce
o
f
c
o
m
b
in
ed
h
an
d
cr
af
ted
an
d
d
ee
p
f
ea
t
u
r
es
F
e
a
t
u
r
e
s
A
c
c
u
r
a
c
y
(
%)
P
r
e
c
i
s
i
o
n
(
%)
R
e
c
a
l
l
(
%)
F1
-
sc
o
r
e
(
%)
S
p
e
c
i
f
i
c
i
t
y
(
%)
C
o
l
o
r
f
e
a
t
u
r
e
s
9
1
.
2
3
9
0
.
9
2
9
0
.
5
1
9
0
.
3
7
9
0
.
0
3
Te
x
t
u
r
e
f
e
a
t
u
r
e
s
9
1
.
1
5
9
0
.
7
9
9
0
.
3
3
9
0
.
0
4
8
9
.
8
4
S
h
a
p
e
f
e
a
t
u
r
e
s
9
1
.
3
4
9
1
.
0
2
9
0
.
8
7
9
0
.
6
6
9
0
.
3
7
C
o
m
b
i
n
e
d
h
a
n
d
c
r
a
f
t
e
d
f
e
a
t
u
r
e
s
9
4
.
7
8
9
4
.
3
2
9
4
.
0
5
9
3
.
7
6
9
3
.
4
1
D
e
e
p
f
e
a
t
u
r
e
s
9
3
.
2
6
9
3
.
0
5
9
2
.
8
5
9
2
.
5
8
9
2
.
1
7
B
o
t
h
h
a
n
d
c
r
a
f
t
e
d
a
n
d
d
e
e
p
f
e
a
t
u
r
e
s
9
6
.
6
6
9
6
.
2
6
9
6
.
5
0
9
6
.
3
7
9
5
.
6
4
T
ab
le
2
r
ep
r
esen
ts
th
e
p
er
f
o
r
m
an
ce
o
f
d
if
f
er
e
n
t
f
ea
t
u
r
e
e
x
t
r
ac
tio
n
alg
o
r
ith
m
s
with
m
etr
ic
s
an
d
er
r
o
r
r
ates.
T
h
e
ex
is
tin
g
f
ea
tu
r
e
ex
tr
ac
tio
n
alg
o
r
ith
m
s
lik
e
I
n
ce
p
tio
n
Net,
Mo
b
ileNet,
R
esNet,
E
f
f
icien
tNet
ar
e
co
n
s
id
er
ed
to
v
alid
ate
th
e
p
er
f
o
r
m
an
ce
o
f
th
e
d
ev
el
o
p
ed
DSP
-
E
f
f
icien
tNet
with
weig
h
ted
f
u
s
io
n
.
T
h
e
DSP
-
E
f
f
icien
tNet
with
weig
h
t
ed
f
u
s
io
n
o
b
tain
e
d
9
6
.
6
6
%
ac
cu
r
ac
y
,
9
6
.
2
6
%
p
r
ec
is
io
n
,
9
6
.
5
0
%
r
ec
all,
9
6
.
3
7
%
F
1
-
s
co
r
e
,
an
d
9
5
.
6
4
% sp
ec
if
ic
ity
o
n
th
e
p
o
m
e
g
r
an
ate
f
r
u
it d
i
s
ea
s
e
d
ataset.
T
ab
le
3
r
ep
r
esen
ts
th
e
p
er
f
o
r
m
an
ce
d
if
f
e
r
en
t
class
if
ier
s
wi
th
m
etr
ics
an
d
er
r
o
r
r
ates.
T
h
e
ex
is
tin
g
class
if
ier
s
lik
e
ex
tr
em
e
g
r
ad
ien
t
b
o
o
s
tin
g
(
XGBo
o
s
t)
,
d
ec
is
io
n
tr
ee
(
DT
)
,
R
F
,
an
d
tr
ad
itio
n
al
SVM
ar
e
co
n
s
id
er
ed
to
v
alid
ate
th
e
p
er
f
o
r
m
an
ce
o
f
th
e
d
ev
el
o
p
ed
DSP
-
E
f
f
icien
tNet
with
weig
h
ted
f
u
s
io
n
.
T
h
e
DSP
-
E
f
f
icien
tNet
with
weig
h
t
ed
f
u
s
io
n
o
b
tain
e
d
9
6
.
6
6
%
ac
cu
r
ac
y
,
9
6
.
2
6
%
p
r
ec
is
io
n
,
9
6
.
5
0
%
r
ec
all,
9
6
.
3
7
%
F
1
-
s
co
r
e
,
an
d
9
5
.
6
4
%
s
p
ec
if
icity
o
n
th
e
p
o
m
eg
r
a
n
ate
f
r
u
i
t
d
is
ea
s
e
d
ata
s
et.
Fig
u
r
e
s
5
an
d
6
r
ep
r
esen
ts
a
co
n
f
u
s
io
n
m
atr
ix
an
d
R
OC
C
u
r
v
e
f
o
r
p
r
o
p
o
s
ed
m
eth
o
d
u
s
in
g
Po
m
eg
r
an
ate
f
r
u
it d
is
ea
s
e
d
at
aset.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
Defo
r
ma
b
le
s
p
a
tia
l p
yra
mid
p
o
o
lin
g
-
e
n
h
a
n
ce
d
E
fficien
tN
et
w
ith
…
(
Ha
r
is
h
B
o
mme
n
a
h
a
ll
i Ma
llika
r
ju
n
a
ia
h
)
651
T
ab
le
2
.
Per
f
o
r
m
an
ce
o
f
d
if
f
er
en
t f
ea
tu
r
e
e
x
tr
ac
tio
n
alg
o
r
ith
m
s
M
e
t
h
o
d
s
C
l
a
s
si
f
i
c
a
t
i
o
n
p
e
r
f
o
r
ma
n
c
e
Er
r
o
r
r
a
t
e
A
c
c
u
r
a
c
y
(
%)
P
r
e
c
i
s
i
o
n
(
%)
R
e
c
a
l
l
(
%)
F1
-
sc
o
r
e
(
%)
S
p
e
c
i
f
i
c
i
t
y
(
%)
M
S
E
R
M
S
E
M
A
E
M
A
P
E
R
2
I
n
c
e
p
t
i
o
n
N
e
t
9
4
.
5
2
9
4
.
3
1
9
4
.
0
5
9
4
.
2
5
9
3
.
7
5
0
.
5
1
0
.
4
9
0
.
6
4
0
.
5
9
0
.
5
1
M
o
b
i
l
e
N
e
t
9
4
.
7
8
9
4
.
5
3
9
4
.
1
7
9
4
.
4
2
9
4
.
2
8
0
.
4
4
0
.
4
2
0
.
5
9
0
.
5
3
0
.
5
6
R
e
sN
e
t
9
5
.
0
2
9
4
.
8
3
9
4
.
4
2
9
4
.
5
5
9
4
.
6
3
0
.
3
9
0
.
3
4
0
.
5
2
0
.
4
6
0
.
6
0
Ef
f
i
c
i
e
n
t
N
e
t
9
5
.
4
3
9
5
.
2
1
9
5
.
0
7
9
5
.
1
3
9
5
.
0
5
0
.
3
2
0
.
2
8
0
.
4
5
0
.
3
7
0
.
6
8
D
S
P
-
Ef
f
i
c
i
e
n
t
N
e
t
w
i
t
h
w
e
i
g
h
t
e
d
f
u
s
i
o
n
9
6
.
6
6
9
6
.
2
6
9
6
.
5
0
9
6
.
3
7
9
5
.
6
4
0
.
2
5
0
.
2
2
0
.
3
7
0
.
3
1
0
.
7
6
T
ab
le
3
.
Per
f
o
r
m
an
ce
o
f
d
if
f
er
en
t c
lass
if
ier
s
with
m
etr
ics an
d
er
r
o
r
s
r
ates
M
e
t
h
o
d
s
C
l
a
s
si
f
i
c
a
t
i
o
n
p
e
r
f
o
r
ma
n
c
e
Er
r
o
r
r
a
t
e
A
c
c
u
r
a
c
y
(
%)
P
r
e
c
i
s
i
o
n
(
%)
R
e
c
a
l
l
(
%)
F1
-
sc
o
r
e
(
%)
S
p
e
c
i
f
i
c
i
t
y
(
%)
M
S
E
R
M
S
E
M
A
E
M
A
P
E
R
2
X
G
B
o
o
st
9
4
.
5
2
9
4
.
3
1
9
4
.
0
5
9
4
.
2
5
9
3
.
7
5
0
.
5
1
0
.
4
9
0
.
6
4
0
.
5
9
0
.
5
1
DT
9
4
.
7
8
9
4
.
5
3
9
4
.
1
7
9
4
.
4
2
9
4
.
2
8
0
.
4
4
0
.
4
2
0
.
5
9
0
.
5
3
0
.
5
6
RF
9
5
.
0
2
9
4
.
8
3
9
4
.
4
2
9
4
.
5
5
9
4
.
6
3
0
.
3
9
0
.
3
4
0
.
5
2
0
.
4
6
0
.
6
0
S
V
M
9
5
.
4
3
9
5
.
2
1
9
5
.
0
7
9
5
.
1
3
9
5
.
0
5
0
.
3
2
0
.
2
8
0
.
4
5
0
.
3
7
0
.
6
8
D
S
P
-
Ef
f
i
c
i
e
n
t
N
e
t
w
i
t
h
w
e
i
g
h
t
e
d
f
u
s
i
o
n
f
o
r
S
V
M
9
6
.
6
6
9
6
.
2
6
9
6
.
5
0
9
6
.
3
7
9
5
.
6
4
0
.
2
5
0
.
2
2
0
.
3
7
0
.
3
1
0
.
7
6
Fig
u
r
e
5
.
C
o
n
f
u
s
io
n
m
atr
i
x
Fig
u
r
e
6
.
R
OC
cu
r
v
e
T
ab
le
4
p
r
esen
ts
th
e
r
esu
lts
o
f
5
-
f
o
ld
cr
o
s
s
-
v
alid
atio
n
to
ev
a
lu
ate
g
en
e
r
aliza
tio
n
a
b
ilit
y
o
f
p
r
o
p
o
s
ed
DSPP
-
E
f
f
icien
tNet
with
wei
g
h
ted
f
ea
tu
r
e
f
u
s
io
n
m
eth
o
d
f
o
r
p
o
m
eg
r
a
n
ate
d
is
ea
s
e
class
if
icatio
n
.
T
h
e
p
r
o
p
o
s
ed
m
o
d
e
is
tr
ain
ed
a
n
d
test
ac
r
o
s
s
f
iv
e
d
if
f
er
en
t
d
at
a
s
p
lits
,
en
s
u
r
es
r
o
b
u
s
t
m
o
d
el
p
er
f
o
r
m
an
ce
.
T
h
e
r
esu
lts
f
r
o
m
T
a
b
le
4
,
s
h
o
ws
th
at
th
e
p
r
o
p
o
s
ed
m
o
d
el
p
e
r
f
o
r
m
ed
well
o
n
i
n
d
iv
id
u
al
s
p
l
its
an
d
g
en
e
r
alize
d
ef
f
icien
tly
ac
r
o
s
s
v
ar
io
u
s
s
u
b
s
ets o
f
d
ata.
T
ab
le
4
.
K
-
f
o
ld
cr
o
s
s
-
v
alid
a
t
io
n
r
esu
lts
f
o
r
p
r
o
p
o
s
ed
DSPP
-
E
f
f
icien
tNet
with
weig
h
ted
f
e
atu
r
e
f
u
s
io
n
m
o
d
el
ac
r
o
s
s
d
if
f
er
en
t m
etr
ics
K
-
F
o
l
d
A
c
c
u
r
a
c
y
(
%)
P
r
e
c
i
s
i
o
n
(
%)
R
e
c
a
l
l
(
%)
F1
-
sc
o
r
e
(
%)
S
p
e
c
i
f
i
c
i
t
y
(
%)
K
=
1
9
6
.
3
0
9
6
.
1
0
9
6
.
3
0
9
6
.
2
0
9
5
.
5
0
K
=
2
9
6
.
8
0
9
6
.
5
0
9
6
.
6
0
9
6
.
5
5
9
5
.
8
0
K
=
3
9
6
.
9
0
9
6
.
6
0
9
6
.
7
0
9
6
.
6
5
9
5
.
9
0
K
=
4
9
6
.
5
0
9
6
.
3
0
9
6
.
4
0
9
6
.
3
5
9
5
.
6
0
K
=
5
9
6
.
6
0
9
6
.
3
0
9
6
.
5
0
9
6
.
4
0
9
5
.
4
0
M
e
a
n
9
6
.
6
6
9
6
.
3
6
9
6
.
5
0
9
6
.
4
3
9
5
.
6
4
S
t
a
n
d
a
r
d
d
e
v
i
a
t
i
o
n
0
.
2
0
0
.
1
9
0
.
1
4
0
.
1
7
0
.
2
0
T
ab
le
5
p
r
esen
ts
th
e
c
o
m
p
r
e
h
en
s
iv
e
s
tatis
tical
an
d
co
m
p
u
tatio
n
ev
alu
atio
n
f
o
r
DSPP
-
E
f
f
icien
tNet
with
weig
h
ted
f
ea
tu
r
e
f
u
s
io
n
m
o
d
e
f
o
r
p
o
m
eg
r
an
ate
d
is
e
ase
class
if
icat
io
n
.
Me
tr
ics
lik
e
m
ea
n
ac
c
u
r
ac
y
,
co
n
f
id
en
ce
in
te
r
v
al,
s
tan
d
ar
d
d
ev
iatio
n
,
p
-
v
alu
e
f
r
o
m
t
-
te
s
t,
tr
ain
in
g
tim
e,
in
f
er
en
ce
tim
e
p
er
im
ag
e
,
an
d
Evaluation Warning : The document was created with Spire.PDF for Python.