I
A
E
S
I
n
t
e
r
n
at
io
n
al
Jou
r
n
al
of
A
r
t
if
ic
ia
l
I
n
t
e
ll
ig
e
n
c
e
(
I
J
-
A
I
)
V
ol
. 13, No. 4,
D
e
c
e
m
be
r
2024, pp.
4138
~
4146
I
S
S
N
:
2252
-
8938
,
D
O
I
:
10.11591/
ij
a
i.
v
13
.i
4
.pp
4138
-
4146
4138
Jou
r
n
al
h
om
e
page
:
ht
tp
:
//
ij
ai
.
ia
e
s
c
or
e
.c
om
S
e
e
d
i
n
g
p
r
e
c
i
si
on
:
a
m
as
k
r
e
gi
on
b
as
e
d
c
on
vol
u
t
i
on
al
n
e
u
r
al
n
e
t
w
or
k
s
c
l
ass
i
f
i
c
at
i
on
ap
p
r
oac
h
f
or
t
h
e
c
l
ass
i
f
i
c
at
i
on
of
p
ad
d
y
se
e
d
s
R
aj
as
h
r
e
e
N
a
m
b
ia
r
1,
2
,
R
an
j
it
h
B
h
at
1,
2
,
V
ar
u
n
a
K
u
m
ar
a
2,
3
1
D
e
pa
r
t
m
e
nt
of
R
obot
i
c
s
a
nd
A
r
t
i
f
i
c
i
a
l
I
nt
e
l
l
i
ge
nc
e
E
ngi
ne
e
r
i
ng,
N
M
A
M
I
ns
t
i
t
ut
e
of
T
e
c
hnol
ogy
,
N
I
T
T
E
(
D
e
e
m
e
d
to
be
U
ni
ve
r
s
i
t
y)
,
N
i
t
t
e
,
K
a
r
na
t
a
ka
,
I
ndi
a
2
F
a
c
ul
t
y
of
E
ngi
ne
e
r
i
ng
a
nd
T
e
c
hnol
ogy
,
J
A
I
N
(
D
e
e
m
e
d
to
be
U
ni
ve
r
s
i
t
y)
,
B
e
n
ga
l
ur
u,
I
ndi
a
3
D
e
pa
r
t
m
e
nt
of
E
l
e
c
t
r
oni
c
s
a
nd
C
om
m
uni
c
a
t
i
on
E
ngi
ne
e
r
i
ng,
M
oodl
a
ka
t
t
e
I
ns
t
i
t
ut
e
of
T
e
c
hnol
ogy,
K
unda
pur
a
,
I
ndi
a
A
r
t
ic
le
I
n
f
o
A
B
S
T
R
A
C
T
A
r
ti
c
le
h
is
to
r
y
:
R
e
c
e
iv
e
d
F
e
b 23, 2024
R
e
vi
s
e
d
J
un 25, 2024
A
c
c
e
pt
e
d
J
un 28, 2024
The
generation
of
sufficient
training
data
that
is
accurately
labelled
for
a
deep
neural
network
involves
a
significant
amount
of
effort
and
freq
uently
constit
utes
a
bottleneck
in
the
implementation
process.
For
the
purp
ose
of
this
research,
we
are
training
a
neural
network
model
to
perform
in
stance
segmentation
and
classifi
cation
of
crop
seeds
for
various
rice
cultivars
.
Synthetic
ally
construct
ed
dataset
is
used
here.
The
concept
of
domain
randomizatio
n,
which
offers
a
productive
alternati
ve
to
th
e
laborious
p
rocess
of
data
annotati
on,
serves
as
the
basis
for
our
methodology.
We
make
use
of
the
domain
randomizatio
n
technique
in
order
to
produce
synthetic
da
ta,
and
the
mask
region
-
based
convolut
ional
neural
network
(Mask
R
-
CNN)
architectu
re
is
utilized
in
order
to
train
our
neural
network
mod
els.
A
cultivar
name
is
used
to
designate
the
seeds,
and
they
are
differentiate
d
from
one
another
using
colors
that
are
comparable
to
those
used
in
the
actual
dataset
of
paddy
cultivars
.
Our
mission
focuses
on
the
identifica
ti
on
and
categorizati
on
of
rice
paddy
varieties
within
automati
cally
ge
nerated
photographs.
Farmer
s
are
able
to
accurately
sort
crop
se
eds
from
a
var
iety
of
rice
cultivars
with
the
use
of
this
approach,
which
is
particularly
use
ful
for
phenotyping
and
optimizing
yields
in
laborator
y
settings.
K
e
y
w
o
r
ds
:
B
ounding
box
M
a
s
k
r
e
gi
on
ba
s
e
d
c
onvolut
io
na
l
ne
ur
a
l
ne
twor
ks
P
a
ddy
c
la
s
s
if
ic
a
ti
on
R
e
gi
on
of
in
te
r
e
s
t
S
ynt
he
ti
c
da
ta
This
is
an
open
access
article
under
the
CC
BY
-
SA
license.
C
or
r
e
s
pon
di
n
g
A
u
th
or
:
R
a
nj
it
h
B
ha
t
D
e
pa
r
tm
e
nt
of
R
obot
ic
s
a
nd
A
r
ti
f
ic
ia
l
I
nt
e
ll
ig
e
nc
e
E
ngi
ne
e
r
in
g,
N
M
A
M
I
n
s
ti
tu
te
of
T
e
c
hnol
ogy
N
I
T
T
E
(
D
e
e
m
e
d
to
be
U
ni
ve
r
s
it
y)
N
it
te
,
K
a
r
ka
la
T
a
lu
k,
U
dupi
,
K
a
r
na
t
a
ka
574110,
I
ndi
a
E
m
a
il
:
r
a
nj
it
hbha
t@gma
il
.c
om
1.
I
N
T
R
O
D
U
C
T
I
O
N
D
e
e
p
le
a
r
ni
ng
ha
s
ga
in
e
d
popula
r
it
y
in
bot
h
th
e
s
c
ie
nt
if
ic
a
nd
in
dus
tr
ia
l
s
phe
r
e
s
.
D
e
e
p
-
le
a
r
ni
ng
m
e
th
ods
,
s
uc
h
as
c
onvolut
io
na
l
ne
ur
a
l
ne
twor
ks
(
C
N
N
s
)
[
1]
,
a
r
e
e
xt
e
ns
iv
e
ly
e
m
pl
oye
d
in
c
om
put
e
r
vi
s
io
n
f
or
ta
s
ks
li
ke
im
a
ge
c
la
s
s
if
ic
a
ti
on,
obj
e
c
t
de
te
c
ti
on,
a
nd
s
e
m
a
nt
ic
as
w
e
ll
as
in
s
ta
nc
e
s
e
gm
e
nt
a
ti
on
[
2]
–
[
4]
.
U
s
in
g
th
e
s
e
m
e
th
od
s
ha
s
a
l
s
o
a
f
f
e
c
te
d
a
gr
ic
ul
tu
r
e
.
A
c
c
or
di
ng
to
K
a
m
i
la
r
is
a
nd
B
ol
dú
[
5]
,
im
a
ge
-
ba
s
e
d
phe
not
ypi
ng
de
te
c
ts
w
e
e
ds
,
a
gr
ic
ul
tu
r
a
l
di
s
e
a
s
e
s
,
a
nd
f
r
ui
ts
.
D
e
e
p
le
a
r
ni
ng
c
om
pl
e
m
e
nt
s
th
e
s
e
c
to
r
'
s
[
6]
a
bunda
nt
hi
gh
-
c
ont
e
xt
da
ta
.
H
ow
e
ve
r
,
de
e
p
le
a
r
ni
ng
r
e
qui
r
e
s
c
ons
id
e
r
a
bl
e
la
be
ll
e
d
da
ta
pr
e
pa
r
a
ti
on.
As
of
2012,
I
m
a
ge
N
e
t
ha
s
1.2
m
il
li
on
tr
a
in
in
g
im
a
ge
s
a
nd
150,000
va
li
da
ti
on/
te
s
t
im
a
ge
s
w
it
h
ha
nd
c
a
te
gor
iz
a
ti
on
[
7]
.
328,000
pi
c
tu
r
e
s
w
it
h
2.5
m
il
li
on
ta
gge
d
obj
e
c
t
s
f
r
om
91
c
a
te
g
or
ie
s
w
e
r
e
u
s
e
d
f
or
th
e
2014
c
om
m
on
obj
e
c
t
s
in
c
ont
e
xt
(
C
O
C
O
)
obj
e
c
t
de
te
c
ti
on
ta
s
k
[
8]
.
T
hi
s
a
nnot
a
ti
n
g
th
e
da
ta
s
e
t
or
de
r
m
a
y
be
c
ha
ll
e
ngi
ng
f
o
r
a
r
e
s
e
a
r
c
he
r
.
A
gr
ic
ul
tu
r
e
r
e
s
e
a
r
c
h
r
e
v
e
a
ls
th
a
t
a
gr
a
in
he
a
d
de
te
c
ti
on
ne
twor
k
m
a
y
be
tr
a
in
e
d
w
it
h
52
phot
os
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
ti
f
I
nt
e
ll
I
S
S
N
:
2252
-
8938
Se
e
di
ng pr
e
c
i
s
io
n:
a m
as
k
r
e
gi
on bas
e
d c
on
v
ol
ut
io
nal
ne
ur
al
n
e
tw
or
k
s
... (
R
aj
as
hr
e
e
N
am
bi
a
r
)
4139
a
ve
r
a
gi
ng
400
obj
e
c
ts
pe
r
im
a
ge
[
9
]
a
nd
a
c
r
op
s
te
m
de
te
c
ti
o
n
ne
twor
k
w
it
h
822
im
a
ge
s
[
10]
.
T
he
s
e
c
a
s
e
s
tu
di
e
s
d
e
m
ons
tr
a
te
th
a
t
I
m
a
ge
N
e
t
c
la
s
s
if
ic
a
ti
on
a
nd
C
O
C
O
de
te
c
ti
on
r
e
qui
r
e
m
or
e
d
a
ta
th
a
n
s
p
e
c
ia
li
z
e
d
w
or
k.
W
hi
le
dom
a
in
a
da
pt
a
ti
on
a
nd
a
c
ti
ve
le
a
r
ni
ng
a
r
e
us
e
d
in
pl
a
nt
/b
io
s
c
ie
nc
e
a
ppl
ic
a
ti
ons
to
c
ut
la
bor
c
os
ts
,
r
e
s
e
a
r
c
he
r
s
f
in
d
a
nnot
a
ti
ng
unpl
e
a
s
a
nt
be
c
a
us
e
it
'
s
li
ke
r
u
nni
ng
a
m
a
r
a
th
on
w
it
hout
a
ta
r
ge
t
[
11]
–
[
13]
.
T
he
s
im
2r
e
a
l
tr
a
ns
f
e
r
,
or
le
a
r
ni
ng
f
r
om
s
ynt
he
ti
c
im
a
ge
s
,
r
e
duc
e
s
m
a
nua
l
a
nnot
a
ti
ons
.
T
r
a
in
in
g
da
ta
f
or
pl
a
nt
im
a
ge
a
na
ly
s
is
w
a
s
pr
e
pa
r
e
d
s
im
il
a
r
ly
.
U
s
in
g
s
ynt
h
e
ti
c
pl
a
nt
m
ode
ls
,
I
s
oka
ne
et
al
.
[
14]
pr
e
di
c
te
d
br
a
nc
hi
ng
pa
tt
e
r
n,
w
hi
le
s
e
ve
r
a
l
r
e
s
e
a
r
c
he
r
s
[
15]
,
[
16]
ge
ne
r
a
te
d
r
e
a
li
s
ti
c
im
a
ge
s
f
r
om
ge
ne
r
a
te
d
da
ta
s
e
ts
us
in
g
ge
ne
r
a
ti
ve
a
dve
r
s
a
r
ia
l
ne
twor
k
(
G
A
N
)
.
G
A
N
-
ge
ne
r
a
te
d
i
m
a
ge
s
w
e
r
e
us
e
d
to
tr
a
in
a
ne
ur
a
l
ne
twor
k
f
or
A
r
a
bi
dops
is
le
a
f
c
ount
in
g
by
G
iu
f
f
r
id
a
et
al
.
[
17]
.
S
i
m
il
a
r
to
A
r
s
e
novi
c
et
al
.
[
18]
S
ty
le
G
A
N
28
c
r
e
a
te
d
pl
a
nt
di
s
e
a
s
e
c
la
s
s
if
ic
a
ti
on
tr
a
in
in
g
pi
c
tu
r
e
s
.
H
ow
e
ve
r
,
s
im
2r
e
a
l
g
e
n
e
r
a
te
s
ne
a
r
ly
li
m
it
le
s
s
tr
a
in
in
g
da
ta
.
To
br
id
ge
th
e
s
im
2r
e
a
l
ga
p,
dom
a
in
r
a
ndomi
z
a
ti
on
tr
a
in
s
de
e
p
n
e
twor
ks
w
it
h
e
nor
m
ous
va
r
ia
nt
s
of
s
ynt
he
ti
c
im
a
ge
s
w
it
h
r
a
ndoml
y
s
e
le
c
te
d
phys
ic
a
l
a
tt
r
ib
ut
e
s
.
D
om
a
in
r
a
ndomi
z
a
ti
on
is
r
e
la
te
d
to
da
ta
a
ugm
e
nt
a
ti
on
(
e
.g.,
r
a
ndoml
y
f
li
ppi
ng
a
nd
r
ot
a
ti
ng
pho
to
gr
a
phs
)
,
but
th
e
s
ynt
he
ti
c
e
nvi
r
onm
e
nt
can
r
e
f
le
c
t
va
r
ie
ty
unde
r
num
e
r
ous
s
c
e
na
r
io
s
,
unl
ik
e
ge
nui
ne
im
a
ge
s
.
T
he
c
onve
nt
io
n
a
l
a
ppr
oa
c
h,
as
s
how
n
in
F
ig
ur
e
1,
in
vol
ve
s
m
a
nua
ll
y
la
be
li
ng
phot
os
to
c
r
e
a
te
th
e
tr
a
in
in
g
da
ta
s
e
t.
In
c
ont
r
a
s
t,
our
s
ugge
s
t
e
d
m
e
th
od
e
li
m
in
a
te
s
th
is
s
te
p
by
ut
il
iz
in
g
a
s
ynt
he
ti
c
da
ta
s
e
t
f
or
th
e
c
r
op
s
e
e
d
in
s
ta
nc
e
s
e
gm
e
nt
a
ti
on
m
ode
l.
F
ig
ur
e
1.
O
ve
r
vi
e
w
of
th
e
s
ugge
s
te
d
tr
a
in
in
g
pr
oc
e
dur
e
f
or
s
e
e
d
in
s
ta
nc
e
s
e
gm
e
nt
a
ti
on
T
hi
s
a
ppr
oa
c
h
in
vol
ve
s
tr
a
in
in
g
de
e
p
ne
ur
a
l
ne
twor
k
m
ode
ls
to
pe
r
f
or
m
th
e
in
tr
ic
a
te
ta
s
k
of
in
s
ta
nc
e
s
e
gm
e
nt
a
ti
on,
w
he
r
e
in
in
di
vi
dua
l
s
e
e
d
s
a
r
e
c
la
s
s
if
ie
d
a
nd
pr
e
c
is
e
ly
lo
c
a
li
z
e
d
w
it
hi
n
im
a
ge
s
.
By
le
ve
r
a
gi
ng
s
ynt
he
ti
c
a
ll
y
ge
ne
r
a
te
d
d
a
ta
s
e
t
s
a
nd
r
a
ndomi
z
a
ti
on
te
c
hni
qu
e
s
,
we
can
c
r
e
a
t
e
a
r
obus
t
a
nd
v
e
r
s
a
ti
le
tr
a
in
in
g
e
nvi
r
onm
e
nt
f
or
th
e
s
e
m
ode
ls
.
T
he
be
n
e
f
it
s
of
pa
ddy
s
e
e
d
c
la
s
s
if
ic
a
ti
on
us
in
g
de
e
p
le
a
r
ni
ng
a
r
e
m
a
ni
f
ol
ds
.
It
not
onl
y
s
ig
ni
f
ic
a
nt
ly
r
e
duc
e
s
th
e
la
bor
a
nd
time
r
e
qui
r
e
d
f
or
s
e
e
d
s
or
ti
ng
but
a
ls
o
e
ns
ur
e
s
c
on
s
is
te
n
c
y
a
nd
pr
e
c
is
io
n
in
th
e
c
la
s
s
if
ic
a
ti
on
pr
oc
e
s
s
.
M
or
e
ove
r
,
it
ha
s
th
e
pot
e
nt
ia
l
to
im
pr
ove
c
r
op
m
a
na
ge
m
e
nt
pr
a
c
ti
c
e
s
,
as
a
c
c
ur
a
te
c
ul
ti
va
r
-
le
ve
l
s
e
e
d
da
ta
can
in
f
or
m
de
c
is
io
ns
r
e
la
t
e
d
to
pl
a
nt
in
g,
f
e
r
ti
li
z
a
ti
on,
a
nd
pe
s
t
c
ont
r
ol
.
M
a
ny
s
tu
di
e
s
h
a
ve
f
ound
th
a
t
us
in
g
s
e
e
d
w
id
th
as
a
pr
im
a
r
y
pa
r
a
m
e
te
r
in
c
r
e
a
s
e
s
r
ic
e
out
put
.
T
he
f
oc
u
s
on
m
or
phol
ogi
c
a
l
s
e
e
d
tr
a
it
s
s
how
s
pr
om
is
e
f
or
im
pr
ovi
ng
a
gr
ic
ul
tu
r
a
l
pr
oduc
ti
vi
ty
a
nd
p
r
om
ot
in
g
bi
ol
ogi
c
a
l
r
e
s
e
a
r
c
h.
It
is
im
por
ta
nt
to
r
e
m
e
m
be
r
,
ne
v
e
r
th
e
le
s
s
,
th
a
t
m
a
ny
e
a
r
li
e
r
r
e
s
e
a
r
c
he
s
e
va
lu
a
te
d
s
e
e
d
f
or
m
us
in
g
qua
li
ta
ti
ve
m
e
a
s
ur
e
s
,
V
e
r
ni
e
r
c
a
ll
ip
e
r
s
,
or
m
a
nua
ll
y
a
nnot
a
ti
n
g
im
a
ge
s
us
in
g
im
a
ge
-
pr
oc
e
s
s
in
g
to
ol
s
.
T
hi
s
phe
not
ypi
ng
pr
oc
e
dur
e
m
a
y
le
a
d
to
qua
nt
if
ic
a
ti
on
m
is
ta
ke
s
th
a
t
di
f
f
e
r
a
m
ongs
t
a
nnot
a
to
r
s
a
nd
is
of
te
n
la
bor
-
in
te
ns
iv
e
.
2.
R
E
L
A
T
E
D
WO
R
K
S
W
id
ia
s
tu
ti
et
al
.
[
19]
s
ugge
s
ts
th
a
t
r
ic
e
s
e
e
d
qua
li
ty
is
tr
a
di
ti
ona
ll
y
de
te
r
m
in
e
d
by
hum
a
n
vi
s
ua
l
a
s
s
e
s
s
m
e
nt
.
T
hi
s
m
e
th
od
is
hi
ghl
y
s
ubj
e
c
ti
ve
w
he
n
c
om
p
a
r
in
g
r
ic
e
va
r
ie
ti
e
s
w
it
h
s
im
il
a
r
phys
ic
a
l
f
e
a
tu
r
e
s
.
T
he
r
e
s
e
a
r
c
h
r
e
c
om
m
e
nd
s
f
la
tb
e
d
s
c
a
nni
ng
a
nd
di
gi
ta
l
im
a
ge
pr
oc
e
s
s
in
g
to
a
s
s
e
s
s
r
ic
e
s
e
e
d
pur
it
y
to
ove
r
c
om
e
th
is
ba
r
r
ie
r
.
A
f
ie
ld
-
ba
s
e
d
gr
ow
out
te
s
t
(
G
O
T
)
va
li
da
te
s
r
ic
e
s
e
e
d
s
ha
pe
a
na
ly
s
i
s
in
th
is
m
e
th
od.
An
a
na
ly
s
is
of
th
e
14
m
or
phol
ogi
c
a
l
qua
li
ti
e
s
f
ound
r
e
la
ti
on
s
hi
ps
in
onl
y
s
ix
a
r
e
a
,
f
e
r
e
t,
m
in
im
um
f
e
r
e
t,
a
s
pe
c
t
r
a
ti
o,
r
oundne
s
s
,
a
nd
s
ol
id
it
y.
G
r
ow
in
g
m
e
th
ods
,
ha
r
ve
s
ti
ng,
s
hi
ppi
ng,
a
nd
pos
t
-
ha
r
ve
s
t
pr
oc
e
s
s
in
g
can
a
f
f
e
c
t
s
e
e
d
pur
it
y.
In
a
ddi
ti
on
to
qua
li
ty
,
s
e
e
d
c
e
r
ti
f
ic
a
te
la
be
l
s
m
us
t
c
le
a
r
ly
di
s
pl
a
y
s
e
e
d
pur
it
y
va
lu
e
s
.
T
he
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
, V
ol
.
13
, N
o.
4
,
D
e
c
e
m
be
r
2024
:
4138
-
4146
4140
pr
opos
e
d
m
e
th
od
[
20]
im
pr
ove
s
r
ic
e
s
e
e
d
pur
it
y
te
s
ti
ng
due
to
i
ts
s
pe
e
d
a
nd
c
o
s
t,
gr
ow
-
out
te
s
t
de
pe
nda
bi
li
ty
.
It
can
be
di
f
f
ic
ul
t
to
di
s
ti
ngui
s
h
b
e
twe
e
n
s
e
e
d
s
w
it
h
th
e
s
a
m
e
m
or
phol
ogy
dur
in
g
pur
it
y
te
s
ti
ng.
M
ol
e
c
ul
a
r
a
ppr
oa
c
he
s
a
r
e
be
in
g
s
tu
di
e
d
to
di
f
f
e
r
e
nt
ia
te
s
uc
h
s
e
e
d
s
as
a
tr
e
a
tm
e
nt
.
T
he
m
e
th
od
in
A
dj
e
m
out
et
al
.
[
21]
,
e
m
pl
oys
m
a
c
hi
ne
le
a
r
ni
ng
a
nd
im
a
g
e
pr
oc
e
s
s
in
g
a
lg
or
it
hm
s
to
c
a
te
gor
ie
s
w
hol
e
a
nd
br
oke
n
r
ic
e
by
how
w
e
ll
th
e
y
m
e
e
t
na
ti
ona
l
r
ic
e
qua
li
ty
s
ta
nda
r
ds
.
T
he
obj
e
c
ts
a
r
e
c
la
s
s
if
ie
d
us
in
g
C
N
N
te
c
hnol
ogy.
T
h
e
im
a
ge
da
ta
ba
s
e
us
e
d
in
th
i
s
s
tu
dy
c
ont
a
in
s
s
e
lf
-
c
ol
le
c
te
d
phot
os
of
L
oc
T
r
oi
20
br
e
e
d
r
ic
e
f
or
m
s
.
T
he
phot
os
w
e
r
e
ta
ke
n
w
it
h
a
S
ony
Z1
s
m
a
r
tp
hone
'
s
20.7
MP
c
a
m
e
r
a
.
T
he
e
xpe
r
im
e
nt
s
r
e
ve
a
l
th
a
t
c
onvolut
io
na
l
ne
ur
a
l
ne
twor
ks
ha
ve
99.16%
pr
e
c
is
io
n.
S
on
et
al
.
[
22]
in
tr
oduc
e
d
de
e
p
-
r
ic
e
,
a
ne
w
r
ic
e
e
va
lu
a
ti
on
m
e
th
od.
It
e
xt
r
a
c
ts
di
s
ti
ngui
s
hi
ng
a
tt
r
ib
ut
e
s
f
r
om
r
ic
e
phot
o
pe
r
s
pe
c
ti
ve
s
us
in
g
a
m
ul
ti
-
vi
e
w
C
N
N
a
r
c
hi
te
c
tu
r
e
.
A
ddi
ti
ona
ll
y,
it
us
e
s
a
r
e
de
s
ig
ne
d
S
of
tM
a
x
lo
s
s
f
unc
ti
on
to
o
pt
im
iz
e
C
N
N
pa
r
a
m
e
te
r
s
.
T
hi
s
c
r
e
a
te
d
a
ne
w
r
ic
e
-
r
a
ti
ng
a
lg
or
it
hm
unde
r
de
e
p
-
r
ic
e
,
th
is
s
ol
v
e
s
r
ic
e
gr
a
di
ng
pr
obl
e
m
s
us
in
g
d
e
e
p
r
e
s
id
ua
l
ne
twor
ks
a
nd
de
e
p
le
a
r
ni
ng.
W
ij
e
r
a
th
na
a
nd
R
a
na
th
ung
a
[
23]
de
s
c
r
ib
e
s
a
c
o
m
put
e
r
vi
s
io
n
a
nd
im
a
g
e
pr
oc
e
s
s
in
g
s
ys
te
m
f
or
r
ic
e
s
e
e
d
pr
oduc
ti
on
th
a
t
a
ut
om
a
ti
c
a
ll
y
c
la
s
s
if
ie
s
r
ic
e
ty
pe
s
.
S
in
c
e
r
ic
e
s
e
e
ds
f
r
om
di
f
f
e
r
e
nt
va
r
ie
ti
e
s
m
ig
ht
lo
ok
id
e
nt
ic
a
l
in
c
ol
or
,
s
ha
pe
,
a
nd
te
xt
ur
e
,
c
a
te
gor
iz
in
g
th
e
m
c
or
r
e
c
tl
y
is
di
f
f
ic
ul
t.
T
he
s
tu
dy
e
va
lu
a
te
d
f
e
a
tu
r
e
e
xt
r
a
c
ti
on
m
e
th
ods
to
por
tr
a
y
r
ic
e
s
e
e
ds
[
24]
.
T
he
y
a
ls
o
te
s
te
d
pow
e
r
f
ul
c
la
s
s
if
ie
r
s
'
pe
r
f
or
m
a
nc
e
w
it
h
th
e
s
e
e
xt
r
a
c
t
e
d
a
tt
r
ib
ut
e
s
to
s
e
le
c
t
th
e
m
o
s
t
tr
us
twor
th
y
c
la
s
s
if
ie
r
.
T
he
r
e
s
e
a
r
c
h
s
ho
w
e
d
th
a
t
th
e
ir
r
a
ndom
f
or
e
s
t
(
R
F
)
c
a
te
gor
iz
a
ti
on
te
c
hni
que
ha
d
an
a
ve
r
a
ge
a
c
c
ur
a
c
y
r
a
te
of
90.54
[
25]
,
[
26]
.
T
he
a
va
il
a
bi
li
ty
of
di
ve
r
s
e
c
ul
ti
va
r
s
in
di
f
f
e
r
e
nt
pl
a
c
e
s
m
a
ke
s
da
ta
c
ol
le
c
ti
ng
f
or
th
is
s
tu
dy
di
f
f
ic
ul
t.
3.
M
E
T
H
O
D
F
our
s
te
ps
a
r
e
s
ugge
s
te
d
in
th
e
m
ode
l
f
lo
w
c
ont
r
ib
ut
in
g
to
th
e
de
ve
lo
pm
e
nt
of
a
de
pe
nda
bl
e
m
e
c
ha
ni
s
m
f
or
c
la
s
s
if
yi
ng
s
e
e
d
s
as
s
how
n
in
F
ig
ur
e
2.
T
he
in
it
ia
l
pa
ddy
s
e
e
d
da
ta
s
e
t
c
om
pr
is
e
s
G
id
da
,
J
a
ya
,
J
yot
hi
,
a
nd
M4
pa
ddy
s
e
e
ds
.
T
he
di
ve
r
s
e
r
a
nge
of
da
ta
in
th
is
da
ta
s
e
t
e
na
bl
e
s
our
pr
ogr
a
m
m
e
to
a
c
c
ur
a
te
ly
di
s
ti
ngui
s
h
be
twe
e
n
di
f
f
e
r
e
nt
ty
pe
s
of
s
e
e
ds
.
F
i
g
u
r
e
2.
P
r
o
p
o
s
e
d
a
r
c
h
i
t
e
c
t
u
r
e
of
p
a
d
d
y
s
e
e
d
c
l
a
s
s
i
f
i
c
a
t
i
o
n
C
r
e
a
ti
ng
a
c
om
pr
e
he
ns
iv
e
da
ta
ba
s
e
of
s
e
e
d
im
a
ge
s
is
c
r
uc
ia
l
f
or
doi
ng
f
ur
th
e
r
da
ta
c
ol
le
c
ti
ng.
T
hi
s
pool
s
e
r
ve
s
as
th
e
f
r
a
m
e
w
or
k
f
or
s
ynt
he
ti
c
im
a
ge
s
,
w
hi
c
h
a
r
e
an
e
s
s
e
nt
ia
l
to
ol
f
or
r
e
s
e
a
r
c
h
pur
pos
e
s
.
We
e
m
pl
oy
dom
a
in
r
a
ndomi
z
a
ti
on
to
ge
ne
r
a
te
a
s
e
t
of
2,000
s
ynt
he
ti
c
i
m
a
ge
s
,
w
it
h
1,400
im
a
ge
s
de
s
ig
na
t
e
d
f
or
tr
a
in
in
g
pur
pos
e
s
a
nd
600
im
a
ge
s
r
e
s
e
r
ve
d
f
or
te
s
ti
ng.
S
ubs
e
que
nt
ly
,
th
e
a
r
ti
f
ic
ia
l
da
ta
s
e
t
is
e
m
pl
oye
d
to
tr
a
in
th
e
m
ode
l
us
in
g
th
e
m
a
s
k
r
e
gi
on
-
ba
s
e
d
c
onvolut
io
na
l
ne
ur
a
l
ne
twor
k
(
M
a
s
k
R
-
C
N
N
)
m
e
th
odol
ogy.
T
hi
s
s
ta
ge
e
na
bl
e
s
our
m
ode
l
to
r
e
c
ogni
z
e
a
nd
c
la
s
s
if
y
s
e
e
ds
,
pr
ovi
di
ng
pr
e
di
c
ti
ons
th
a
t
in
c
lu
de
th
e
s
e
e
d
na
m
e
,
as
w
e
ll
as
th
e
bounding
box
a
nd
ove
r
la
y
c
ol
or
.
U
lt
im
a
te
ly
,
th
e
m
ode
l
und
e
r
goe
s
r
ig
or
ous
te
s
ti
ng
to
a
s
s
e
s
s
its
e
f
f
ic
a
c
y
a
nd
s
ui
ta
bi
li
ty
in
r
e
a
l
-
w
or
ld
s
c
e
na
r
io
s
.
T
h
e
pe
r
f
or
m
a
nc
e
of
th
e
s
y
s
te
m
can
be
e
va
lu
a
te
d
in
m
a
ny
c
ont
e
xt
s
u
s
in
g
a
s
s
e
s
s
m
e
nt
te
c
hni
que
s
th
a
t
c
on
s
id
e
r
bot
h
s
ynt
he
ti
c
a
nd
r
e
a
l
-
w
or
ld
da
ta
s
e
t
s
.
T
he
a
r
c
hi
te
c
tu
r
e
of
th
e
M
a
s
k
R
-
C
N
N
m
ode
l
is
il
lu
s
tr
a
te
d
in
F
ig
ur
e
3.
R
e
gi
on
of
in
te
r
e
s
t
a
li
gn
(
R
oI
A
li
gn)
a
im
s
to
e
xt
r
a
c
t
a
s
m
a
ll
,
f
ix
e
d
-
s
iz
e
f
e
a
tu
r
e
m
a
p
(
li
ke
H
×W
)
f
r
om
each
r
e
gi
on
of
in
te
r
e
s
t
w
it
h
s
ub
-
pi
xe
l
a
c
c
ur
a
c
y,
im
pr
ovi
ng
u
pon
th
e
ol
de
r
R
oI
pool
in
g
m
e
th
od
by
a
voi
di
ng
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
ti
f
I
nt
e
ll
I
S
S
N
:
2252
-
8938
Se
e
di
ng pr
e
c
i
s
io
n:
a m
as
k
r
e
gi
on bas
e
d c
on
v
ol
ut
io
nal
ne
ur
al
n
e
tw
or
k
s
... (
R
aj
as
hr
e
e
N
am
bi
a
r
)
4141
qua
nt
iz
a
ti
on
e
r
r
or
s
.
In
(
1
)
is
t
he
r
e
pr
e
s
e
nt
a
ti
on
of
in
te
r
po
la
te
d
f
e
a
tu
r
e
va
lu
e
at
a
s
pe
c
if
ic
lo
c
a
ti
on
(
,
)
w
it
hi
n
th
e
out
put
f
e
a
tu
r
e
m
a
p
of
th
e
R
oI
.
(
,
)
=
∑
(
,
)
.
(
0
,
1
−
|
−
|
)
.
,
(
0
,
1
−
|
−
|
)
(
1)
W
he
r
e
∑
,
is
a
s
um
m
a
ti
on
ove
r
th
e
ne
ig
hbor
hood
of
th
e
poi
nt
(
,
)
in
th
e
in
put
f
e
a
tu
r
e
m
a
p.
A
nd
we
c
ons
id
e
r
th
e
va
lu
e
s
of
ne
ig
hbor
in
g
poi
nt
s
(
,
)
in
th
e
or
ig
in
a
l
f
e
a
tu
r
e
m
a
p.
(
,
)
is
th
e
f
e
a
tu
r
e
va
lu
e
lo
c
a
te
d
at
(
,
)
in
th
e
in
put
f
e
a
tu
r
e
m
a
p
f
r
om
w
hi
c
h
we
a
r
e
tr
yi
ng
to
e
xt
r
a
c
t
th
e
R
oI
.
m
a
x
(
0
,
1
−
|
−
|
)
a
nd
m
a
x
(
0
,
1
−
|
−
|
)
c
a
lc
ul
a
te
th
e
bi
li
ne
a
r
in
te
r
pol
a
ti
on
w
e
ig
ht
s
.
To
de
te
r
m
in
e
s
th
e
c
la
s
s
as
m
e
nt
io
ne
d
in
(
2
)
,
we
us
e
th
e
S
of
tM
a
x
a
c
ti
va
ti
on
f
unc
ti
on
w
it
h
w
e
ig
ht
W
a
nd
bi
a
s
b
.
H
e
r
e
,
∆
is
th
e
pr
e
di
c
te
d
of
f
s
e
ts
in
(
3
)
.
=
(
.
+
)
(
2)
∆
=
′
.
+
′
(
3)
H
e
r
e
,
in
(
4
)
out
li
ne
s
a
c
om
m
on
pa
tt
e
r
n
in
d
e
e
p
le
a
r
ni
ng,
e
s
pe
c
ia
ll
y
in
ta
s
ks
r
e
la
te
d
to
c
om
put
e
r
vi
s
io
n
a
nd
pa
tt
e
r
n
r
e
c
ogni
ti
on,
w
he
r
e
x
w
oul
d
be
a
m
ul
ti
-
di
m
e
ns
io
na
l
a
r
r
a
y
(a
te
ns
or
)
r
e
pr
e
s
e
nt
in
g
th
e
im
a
ge
da
ta
a
nd
M
is
th
e
c
onvolut
e
d
out
put
th
r
ough
a
s
e
r
ie
s
of
C
N
N
la
y
e
r
s
w
it
h
a
s
ig
m
oi
d
a
c
ti
va
ti
on
f
unc
ti
on.
=
(
(
)
)
(
4)
F
ig
ur
e
3.
M
a
s
k
R
-
C
N
N
m
ode
l
s
tr
uc
tu
r
e
3.1.
C
ol
le
c
t
in
g
p
ad
d
y
s
e
e
d
s
f
or
d
at
as
e
t
We
c
a
r
e
f
ul
ly
c
ol
le
c
te
d
a
da
ta
s
e
t
of
f
our
pa
ddy
s
e
e
d
c
la
s
s
e
s
to
s
e
gm
e
nt
c
r
ops
.
T
he
s
e
c
la
s
s
e
s
r
e
pr
e
s
e
nt
K
a
r
na
ta
ka
pa
ddy
s
e
e
d
va
r
ie
ti
e
s
G
id
da
,
J
a
ya
,
J
yot
h
i,
a
nd
M
4.
O
ur
s
e
gm
e
nt
a
ti
on
m
ode
l
w
il
l
be
tr
a
in
e
d
on
th
is
c
a
r
e
f
ul
ly
c
ur
a
te
d
da
ta
s
e
t
to
r
e
li
a
bl
y
id
e
nt
if
y
a
n
d
c
a
te
gor
ie
s
pa
ddy
s
e
e
d
c
la
s
s
e
s
in
a
gr
ic
ul
tu
r
a
l
phot
ogr
a
phy.
3.2.
S
yn
t
h
e
t
ic
im
age
ge
n
e
r
at
io
n
,
p
r
e
p
r
oc
e
s
s
in
g
an
d
t
r
ai
n
in
g
We
a
ppl
ie
d
c
ut
ti
ng
-
e
dge
dom
a
in
r
a
ndomi
z
a
ti
on
to
opt
i
m
iz
e
our
M
a
s
k
R
-
C
N
N
m
ode
l
f
or
pa
ddy
s
e
e
d
c
la
s
s
if
ic
a
ti
on
vi
a
s
ynt
he
ti
c
pi
c
tu
r
e
s
ynt
he
s
is
.
T
hi
s
m
e
th
od
us
e
s
f
our
r
ic
e
s
e
e
d
ty
pe
s
,
a
va
r
ie
d
s
e
e
d
pool
,
a
nd
r
e
s
iz
in
g
th
e
phot
ogr
a
phs
to
1024
×
1024
pi
xe
l
s
.
S
ta
r
ti
ng
w
it
h
th
is
s
e
e
d
pool
,
we
c
r
e
a
te
d
a
hug
e
da
ta
s
e
t
of
2,000
m
e
ti
c
ul
ous
ly
c
r
e
a
te
d
s
ynt
he
ti
c
phot
os
f
or
tr
a
in
in
g
a
nd
te
s
ti
ng
our
m
ode
l.
D
om
a
in
r
a
ndomi
z
a
ti
on
is
us
e
d
to
tr
a
in
a
ne
ur
a
l
ne
twor
k
c
la
s
s
if
ie
r
th
a
t
e
qua
ls
th
e
pe
r
f
or
m
a
nc
e
of
c
ur
r
e
nt
m
ode
ls
tr
a
in
e
d
ju
s
t
on
a
c
tu
a
l
da
ta
s
e
ts
,
de
m
on
s
tr
a
ti
ng
its
ve
r
s
a
ti
li
ty
a
nd
e
f
f
ic
a
c
y.
O
ur
a
r
e
a
of
r
a
ndomi
z
a
ti
on
e
xpe
r
im
e
nt
s
how
e
d
th
a
t
s
ubj
e
c
t
va
r
ie
ty
is
m
or
e
r
e
le
va
nt
th
a
n
s
e
c
ond
a
r
y
c
r
it
e
r
ia
li
ke
il
lu
m
in
a
ti
on
a
nd
te
xt
ur
in
g
in
de
te
r
m
in
in
g
m
ode
l
c
or
r
e
c
tn
e
s
s
.
M
a
s
k
R
-
C
N
N
w
it
h
K
e
r
a
s
or
T
e
ns
or
F
lo
w
w
a
s
e
m
pl
oye
d
f
or
s
e
e
d
c
la
s
s
if
ic
a
ti
on.
T
h
e
r
e
pos
it
or
y
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
, V
ol
.
13
, N
o.
4
,
D
e
c
e
m
be
r
2024
:
4138
-
4146
4142
s
e
tu
p
ne
twor
k
de
s
ig
ns
a
nd
lo
s
s
f
unc
ti
ons
w
e
r
e
e
m
pl
oye
d.
F
e
a
tu
r
e
s
w
e
r
e
e
xt
r
a
c
te
d
u
s
in
g
R
e
s
N
e
t1
01,
a
r
e
s
id
ua
l
ne
twor
k
in
it
ia
li
z
e
d
us
in
g
MS
C
O
C
O
d
a
ta
s
e
t
w
e
ig
ht
s
[
27]
.
N
e
xt
,
we
f
in
e
-
tu
ne
d
our
c
ount
e
r
f
e
it
s
e
e
d
pi
c
tu
r
e
da
ta
s
e
t
us
in
g
10
tr
a
in
in
g
e
po
c
hs
w
it
h
100
s
te
ps
pe
r
e
po
c
h
a
nd
0.001
le
a
r
ni
ng
r
a
te
.
1,400
im
a
ge
s
w
e
r
e
us
e
d
f
or
tr
a
in
in
g
a
nd
a
m
ong
th
e
400
im
a
ge
s
f
r
om
th
e
600
in
t
he
s
ynt
he
ti
c
da
ta
s
e
t,
w
e
r
e
us
e
d
f
or
va
li
da
ti
on
a
nd
200
f
or
te
s
ti
ng
pur
pos
e
s
.
It
is
not
e
w
or
th
y
th
a
t
we
a
voi
de
d
us
in
g
pi
c
tu
r
e
e
nha
nc
e
m
e
nt
w
h
e
n
tr
a
in
in
g.
T
he
a
r
ti
f
ic
ia
l
tr
a
in
in
g
da
ta
m
a
in
ta
in
e
d
a
1024
×
1024
pi
c
tu
r
e
s
iz
e
c
on
s
ta
nt
ly
.
3.3.
R
e
al
t
im
e
d
at
as
e
t
f
or
m
od
e
l
e
val
u
at
io
n
We
put
th
e
M
a
s
k
R
-
C
N
N
m
od
e
l
in
in
f
e
r
e
nc
e
m
ode
a
nd
va
li
da
te
d
it
us
in
g
our
va
li
da
ti
on
da
ta
s
e
t
to
a
ppr
opr
ia
te
ly
a
s
s
e
s
s
its
pe
r
f
or
m
a
nc
e
.
A
c
om
pr
e
he
ns
iv
e
va
li
da
ti
on
a
ppr
oa
c
h
le
t
s
us
a
s
s
e
s
s
th
e
m
ode
l'
s
a
c
c
ur
a
c
y
a
nd
dur
a
bi
li
ty
in
r
e
a
l
-
w
or
ld
s
it
ua
ti
ons
.
We
s
e
le
c
te
d
a
uni
que
da
ta
s
e
t
of
10
im
a
ge
s
of
s
e
e
ds
f
r
om
4
pa
ddy
r
ic
e
ki
nds
f
or
r
e
a
l
-
w
or
ld
te
s
ti
ng.
R
e
a
l
-
w
or
ld
pi
c
tu
r
e
s
a
r
e
a
lwa
ys
1024
×
1204
pi
xe
l
s
a
nd
f
ol
lo
w
s
ta
nda
r
d
pr
opor
ti
ons
.
O
ur
r
e
a
l
-
w
o
r
ld
da
ta
s
e
t
ha
s
20
im
a
ge
s
w
it
h
10
s
e
e
ds
e
a
c
h.
O
ur
s
ys
te
m
a
c
c
ur
a
te
ly
pr
e
di
c
ts
a
nd
la
be
ls
e
a
c
h
s
e
e
d
w
it
h
its
c
ul
ti
va
r
na
m
e
a
nd
c
ol
or
-
c
ode
s
each
s
e
e
d
va
r
ie
ty
in
th
e
phot
o.
O
ur
m
ode
l'
s
f
in
a
l
te
s
t
is
th
is
r
e
a
l
-
time
da
ta
s
e
t,
w
hi
c
h
pr
ove
s
its
e
f
f
ic
a
c
y
a
nd
r
e
li
a
bi
li
t
y
in
r
e
a
l
-
w
or
ld
s
it
ua
ti
ons
.
4.
R
E
S
U
L
T
S
AND
D
I
S
C
U
S
S
I
O
N
S
U
nde
r
s
ta
ndi
ng
th
e
f
e
a
tu
r
e
s
ne
e
de
d
to
s
uc
c
e
s
s
f
ul
ly
r
e
pl
ic
a
te
r
e
a
l
-
w
or
ld
da
ta
s
e
ts
is
e
s
s
e
nt
ia
l
to
unde
r
s
ta
nd
s
ynt
he
ti
c
da
ta
'
s
va
lu
e
in
de
e
p
le
a
r
ni
ng.
O
ur
m
a
jo
r
f
ounda
ti
on
w
a
s
th
a
t
th
e
ne
ur
a
l
n
e
twor
k
m
us
t
le
a
r
n
to
de
te
c
t
a
nd
s
e
pa
r
a
te
r
a
ndoml
y
in
s
e
r
te
d
or
ove
r
la
pp
in
g
s
e
e
d
s
in
to
obj
e
c
t
s
dur
in
g
s
e
e
d
in
s
ta
nc
e
s
e
gm
e
nt
a
ti
on.
W
hi
le
d
e
s
ig
ni
ng
our
s
ynt
h
e
ti
c
pi
c
tu
r
e
c
ol
le
c
ti
on,
we
pr
io
r
it
iz
e
d
s
e
e
d
or
ie
nt
a
ti
ons
ove
r
s
e
e
d
te
xt
ur
e
s
.
T
h
e
num
be
r
of
im
a
g
e
s
in
th
e
tr
a
in
in
g
da
t
a
s
e
t
a
nd
th
e
r
e
s
ol
ut
io
n
a
nd
va
r
ia
nc
e
of
th
e
s
e
e
d
im
a
ge
s
us
e
d
to
pr
oduc
e
s
ynt
he
ti
c
im
a
ge
s
w
e
r
e
e
xpe
c
t
e
d
to
s
ig
ni
f
ic
a
nt
ly
a
f
f
e
c
t
m
ode
l
pe
r
f
or
m
a
nc
e
.
P
r
ovi
di
ng
e
xa
c
t
bounding
boxe
s
a
nd
m
a
s
k
s
f
or
each
s
e
e
d
it
e
m
a
ll
ow
e
d
our
m
od
e
l
to
c
or
r
e
c
tl
y
de
te
c
t
in
s
ta
nc
e
s
in
th
e
s
uppl
ie
d
phot
ogr
a
phs
a
nd
s
e
gm
e
nt
each
s
e
e
d.
To
tr
a
in
m
a
c
hi
ne
le
a
r
ni
ng
m
ode
ls
f
or
c
om
put
e
r
vi
s
io
n
a
ppl
ic
a
ti
ons
li
ke
im
a
ge
c
a
te
gor
iz
a
ti
on,
obj
e
c
t
r
e
c
ogni
ti
on,
a
nd
pi
c
tu
r
e
s
ynt
he
s
is
,
m
a
ny
s
ynt
he
ti
c
im
a
ge
s
a
r
e
ne
e
de
d.
S
ynt
he
ti
c
im
a
ge
s
ge
ne
r
a
te
d
as
in
F
ig
ur
e
4
a
r
e
c
r
e
a
te
d
by
a
m
ode
l
or
ot
he
r
m
e
a
ns
r
a
th
e
r
th
a
n
us
in
g
r
e
a
l
-
w
or
ld
da
ta
.
F
ig
ur
e
4.
S
ynt
he
ti
c
im
a
ge
ge
ne
r
a
ti
on
us
in
g
s
e
e
d
im
a
ge
pool
M
a
s
k
R
-
C
N
N
s
e
gm
e
nt
s
pa
ddy
s
e
e
ds
pr
e
c
is
e
ly
.
T
h
e
m
a
s
ks
c
le
a
r
ly
id
e
nt
if
ie
d
phot
o
s
e
e
d
r
e
gi
on
s
.
T
hi
s
s
how
s
how
th
e
m
ode
l
a
c
c
ur
a
te
ly
di
s
pl
a
ys
a
ll
4
ty
pe
s
of
s
e
e
ds
.
A
c
c
ur
a
cy
a
r
ound
99
%,
f
or
a
ll
s
e
e
d
va
r
ie
ta
ls
as
s
how
n
in
F
ig
ur
e
5.
F
or
m
a
nd
s
iz
e
of
s
e
e
ds
(
gr
a
in
s
)
a
f
f
e
c
t
c
r
op
qua
li
ty
a
nd
pr
oduc
ti
on.
O
u
r
w
or
kf
lo
w
a
ll
ow
s
us
to
phe
not
ype
m
a
ny
s
e
e
ds
w
it
hout
c
on
s
id
e
r
in
g
or
ie
nt
a
ti
on
dur
in
g
im
a
ge
a
c
qui
s
it
io
n
.
F
ig
ur
e
5.
R
e
a
lt
im
e
s
a
m
pl
e
s
a
nd
th
e
vi
s
ua
li
z
e
d
r
a
w
out
put
s
how
in
g
th
e
a
c
c
ur
a
c
y
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
ti
f
I
nt
e
ll
I
S
S
N
:
2252
-
8938
Se
e
di
ng pr
e
c
i
s
io
n:
a m
as
k
r
e
gi
on bas
e
d c
on
v
ol
ut
io
nal
ne
ur
al
n
e
tw
or
k
s
... (
R
aj
as
hr
e
e
N
am
bi
a
r
)
4143
A
c
om
pr
e
he
ns
iv
e
a
na
ly
s
is
of
tr
a
in
in
g
a
nd
va
li
da
ti
on
lo
s
s
e
s
w
a
s
pe
r
f
or
m
e
d
in
th
e
pa
ddy
c
la
s
s
if
ic
a
ti
on
s
tu
dy
f
or
J
a
ya
,
G
id
da
,
J
yot
hi
,
a
nd
M
4,
us
in
g
1
,
17
6,
1
,
159,
1
,
157,
a
nd
1
,
152
s
a
m
pl
e
s
di
s
tr
ib
ut
e
d
a
c
r
os
s
an
80:
20%
tr
a
in
-
te
s
t
s
pl
it
.
T
r
a
in
/b
ox_l
os
s
,
tr
a
in
/
s
e
g_l
os
s
,
tr
a
in
/d
f
l_
lo
s
s
,
tr
a
in
/c
ls
_l
os
s
,
a
nd
va
l/
box_los
s
,
va
l/
s
e
g_l
os
s
,
va
l/
df
l_
lo
s
s
,
a
nd
va
l/
c
l
s
_l
os
s
w
a
s
e
va
lu
a
t
e
d.
T
h
e
r
e
s
ul
t
s
pr
ovi
de
d
in
tr
ig
ui
ng
m
ode
l
pe
r
f
or
m
a
nc
e
in
s
ig
ht
s
.
O
ur
e
xpe
r
im
e
nt
a
l
in
ve
s
ti
ga
ti
on
us
e
d
M
a
s
k
R
-
C
N
N
as
th
e
f
unda
m
e
nt
a
l
m
e
th
od
f
or
pi
c
tu
r
e
s
e
gm
e
nt
a
ti
on,
be
nc
hm
a
r
ki
ng
it
a
g
a
in
s
t
a
va
r
ie
ty
of
s
e
gm
e
nt
a
ti
on
m
ode
ls
in
T
a
bl
e
1.
To
e
va
lu
a
te
each
m
ode
l'
s
a
bi
li
ty
to
s
e
gm
e
nt
c
om
pl
ic
a
te
d
im
a
ge
s
,
th
e
s
tr
uc
tu
r
a
l
s
im
il
a
r
it
y
in
de
x
m
e
a
s
ur
e
(
S
S
I
M
)
,
a
c
c
ur
a
c
y,
pr
e
c
is
io
n,
r
e
c
a
ll
,
a
nd
F1
-
s
c
or
e
w
e
r
e
a
s
s
e
s
s
e
d.
M
a
s
k
R
-
C
N
N
a
c
hi
e
ve
d
an
S
S
I
M
s
c
or
e
of
0.90,
de
m
ons
tr
a
ti
ng
its
a
bi
li
ty
to
m
a
in
ta
in
s
tr
uc
tu
r
a
l
s
im
il
a
r
it
y
be
tw
e
e
n
s
e
gm
e
nt
e
d
pi
c
tu
r
e
s
a
nd
gr
ound
tr
ut
h.
M
a
s
k
R
-
C
N
N
s
ur
pa
s
s
e
d
its
c
om
pe
ti
to
r
s
w
it
h
0.95
a
c
c
ur
a
c
y,
0.94
pr
e
c
is
io
n,
0.94
r
e
c
a
ll
,
a
nd
0.94
F1
-
s
c
or
e
,
de
m
ons
tr
a
ti
ng
its
r
e
s
il
ie
nc
e
in
de
te
c
ti
ng
a
nd
out
li
ni
ng
obj
e
c
ts
in
im
a
ge
s
.
T
a
bl
e
1.
C
om
pa
r
a
ti
ve
a
na
ly
s
is
of
im
a
ge
s
e
gm
e
nt
a
ti
on
m
ode
l
s
b
a
s
e
d
on
S
S
I
M
a
nd
ot
he
r
pe
r
f
or
m
a
nc
e
m
e
tr
ic
s
pr
e
s
e
nt
in
g
an
ove
r
vi
e
w
of
th
e
pe
r
f
or
m
a
nc
e
of
va
r
io
us
s
e
gm
e
nt
a
ti
on
m
ode
ls
a
c
r
os
s
m
ul
ti
pl
e
m
e
tr
ic
s
s
uc
h
as
S
S
I
M
,
a
c
c
ur
a
c
y,
pr
e
c
is
io
n,
r
e
c
a
ll
,
a
nd
F1
-
s
c
or
e
M
ode
l
S
S
I
M
A
c
c
ur
a
c
y
P
r
e
c
i
s
i
on
R
e
c
a
l
l
F1
-
S
c
or
e
R
e
m
a
r
ks
U
-
N
e
t
[
28]
0.85
0.92
0.90
0.89
0.89
H
i
gh
pr
e
c
i
s
i
on
in
bi
om
e
di
c
a
l
i
m
a
ge
s
e
gm
e
nt
a
t
i
on.
F
C
N
[
29]
0.83
0.90
0.88
0.87
0.87
G
ood
f
or
ge
ne
r
a
l
pur
pos
e
s
,
ve
r
s
a
t
i
l
e
.
D
e
e
pL
a
b
(
v3+)
[
30]
0.88
0.93
0.91
0.92
0.91
C
a
pt
ur
e
s
m
ul
t
i
s
c
a
l
e
i
nf
or
m
a
t
i
on
e
f
f
e
c
t
i
ve
l
y.
P
S
P
N
e
t
[
31]
0.86
0.91
0.89
0.90
0.89
E
f
f
e
c
t
i
ve
gl
oba
l
c
ont
e
xt
i
nf
or
m
a
t
i
on.
S
e
gN
e
t
[
32]
0.82
0.89
0.87
0.86
0.86
E
f
f
i
c
i
e
nt
,
s
ui
t
a
bl
e
f
or
r
e
a
l
-
time
a
ppl
i
c
a
t
i
ons
.
R
e
f
i
ne
N
e
t
[
33]
0.87
0.92
0.90
0.91
0.90
H
i
gh
-
r
e
s
ol
ut
i
on
i
m
a
ge
r
y,
f
i
ne
-
gr
a
i
ne
d
s
e
gm
e
nt
a
t
i
on.
E
ne
t
[
34]
0.80
0.88
0.85
0.84
0.84
O
pt
i
m
i
z
e
d
f
or
s
pe
e
d,
r
e
a
l
-
time
pr
oc
e
s
s
i
ng.
H
R
N
e
t
[
35]
0.89
0.94
0.92
0.93
0.92
M
a
i
nt
a
i
ns
hi
gh
-
r
e
s
ol
ut
i
on
r
e
pr
e
s
e
nt
a
t
i
ons
M
a
s
k
R
-
C
N
N
[
36]
0.90
0.95
0.94
0.94
0.94
S
upe
r
i
or
f
or
i
ns
t
a
nc
e
s
e
gm
e
nt
a
t
i
on
w
i
t
h
hi
gh
de
t
a
i
l
.
H
e
r
e
,
th
e
T
a
bl
e
2
s
how
s
c
la
s
s
c
or
r
e
c
tn
e
s
s
a
nd
F
ig
ur
e
6
il
lu
s
tr
a
te
s
c
onf
us
io
n
m
a
tr
ix
.
T
he
s
e
r
e
s
ul
ts
de
m
ons
tr
a
te
M
a
s
k
R
-
C
N
N
'
s
r
e
m
a
r
ka
bl
e
in
s
ta
nc
e
s
e
gm
e
nt
a
ti
on
c
a
pa
bi
li
ti
e
s
,
e
s
pe
c
ia
ll
y
in
hi
gh
-
pr
e
c
is
io
n
a
nd
de
ta
il
s
e
tt
in
gs
.
O
ur
f
in
di
ngs
d
e
m
ons
tr
a
te
M
a
s
k
R
-
C
N
N
'
s
c
r
uc
ia
l
r
ol
e
in
im
a
g
e
s
e
gm
e
nt
a
ti
on
te
c
hnol
ogi
e
s
,
gi
vi
ng
ne
w
in
s
ig
ht
s
f
or
r
e
s
e
a
r
c
he
r
s
a
nd
pr
a
c
ti
ti
one
r
s
us
in
g
de
e
p
le
a
r
ni
ng
f
or
c
om
pl
ic
a
te
d
im
a
ge
pr
oc
e
s
s
in
g
a
ppl
ic
a
ti
ons
.
T
a
bl
e
2.
A
c
c
ur
a
c
y
pr
e
di
c
ti
on
f
or
th
e
s
e
pa
r
a
te
4
c
la
s
s
e
s
G
id
da
,
J
a
ya
,
J
yot
hi
,
a
nd
M4
G
r
ound
t
r
ut
h
M
a
s
k
C
ol
or
P
r
e
di
c
t
e
d
N
a
m
e
A
c
c
ur
a
c
y
J
a
ya
Y
e
l
l
ow
J
a
ya
0.983
J
yot
hi
P
i
nk
J
yot
hi
0.998
G
i
dda
C
ya
n
G
i
dda
1.00
J
a
ya
V
i
ol
e
t
J
a
ya
0.992
G
i
dda
B
l
ue
G
i
dda
0.997
M4
Y
e
l
l
ow
M4
0.999
J
a
ya
O
r
a
nge
J
a
ya
0.985
F
ig
ur
e
6.
V
is
ua
li
z
in
g
th
e
a
c
c
ur
a
c
y
of
c
la
s
s
if
yi
ng
J
a
ya
,
G
id
da
,
J
yot
hi
,
a
nd
M4
us
in
g
c
onf
us
io
n
m
a
tr
ix
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
, V
ol
.
13
, N
o.
4
,
D
e
c
e
m
be
r
2024
:
4138
-
4146
4144
A
c
r
os
s
th
e
tr
a
in
in
g
pha
s
e
,
th
e
m
ode
l
de
m
ons
tr
a
te
d
a
c
ons
is
te
nt
de
c
r
e
a
s
e
in
bot
h
s
e
gm
e
nt
a
ti
on
(
s
e
g_l
os
s
)
a
nd
c
la
s
s
if
ic
a
ti
on
(
c
ls
_l
o
s
s
)
lo
s
s
e
s
.
T
hi
s
dow
n
w
a
r
d
tr
e
nd
in
lo
s
s
e
s
in
di
c
a
te
s
th
a
t
th
e
m
ode
l
e
f
f
e
c
ti
ve
ly
le
a
r
ne
d
to
di
f
f
e
r
e
nt
ia
te
be
twe
e
n
th
e
c
la
s
s
e
s
a
nd
s
e
gm
e
nt
th
e
pa
ddy
im
a
ge
s
a
c
c
ur
a
te
ly
.
N
ot
a
bl
y,
th
e
box
lo
s
s
(
box_los
s
)
a
ls
o
e
xhi
bi
te
d
a
s
im
il
a
r
d
e
c
r
e
a
s
in
g
t
r
e
nd,
hi
ghl
ig
ht
in
g
th
e
m
ode
l'
s
pr
of
ic
ie
nc
y
in
lo
c
a
li
z
in
g
a
nd
pr
e
c
is
e
ly
de
li
ne
a
ti
ng
th
e
pa
ddy
a
r
e
a
s
w
it
hi
n
th
e
im
a
ge
s
.
D
ur
in
g
va
li
da
ti
on,
th
e
obs
e
r
ve
d
tr
e
nds
in
lo
s
s
e
s
w
e
r
e
r
e
la
ti
ve
ly
s
ta
bl
e
,
a
lb
e
it
w
it
h
m
in
or
f
lu
c
tu
a
ti
ons
.
T
he
va
li
da
ti
on
lo
s
s
e
s
c
lo
s
e
ly
m
ir
r
or
e
d
th
e
tr
a
in
in
g
lo
s
s
e
s
,
a
f
f
ir
m
in
g
th
e
m
ode
l'
s
ge
ne
r
a
li
z
a
ti
on
a
bi
li
ty
a
nd
r
obus
tn
e
s
s
in
r
e
c
ogni
z
in
g
a
nd
c
la
s
s
if
yi
ng
pa
ddy
c
la
s
s
e
s
uns
e
e
n
dur
in
g
tr
a
in
in
g.
T
he
m
a
r
gi
na
l
f
lu
c
tu
a
ti
ons
in
va
li
da
ti
on
lo
s
s
e
s
m
ig
ht
in
di
c
a
te
a
s
li
ght
ove
r
f
it
ti
ng
te
nde
nc
y
or
th
e
c
om
pl
e
xi
ty
of
d
is
ti
ngui
s
hi
ng
c
e
r
ta
in
c
la
s
s
e
s
w
it
hi
n
th
e
va
li
da
ti
on
s
e
t.
O
ve
r
a
ll
,
th
e
m
ode
l'
s
pe
r
f
or
m
a
nc
e
s
how
c
a
s
e
s
pr
om
is
in
g
c
a
pa
bi
li
ti
e
s
in
a
c
c
ur
a
te
ly
s
e
gm
e
nt
in
g
a
nd
c
la
s
s
if
yi
ng
di
f
f
e
r
e
nt
pa
ddy
va
r
ie
ti
e
s
.
T
he
c
ons
is
t
e
nt
r
e
duc
ti
on
in
lo
s
s
e
s
dur
in
g
tr
a
in
in
g,
c
oupl
e
d
w
it
h
va
li
da
ti
on
lo
s
s
e
s
a
li
gni
ng
c
lo
s
e
ly
w
it
h
tr
a
in
in
g
lo
s
s
e
s
,
s
ig
ni
f
ie
s
th
e
m
ode
l'
s
c
om
pe
te
nc
y
in
le
a
r
ni
ng
th
e
di
s
ti
nc
ti
ve
f
e
a
tu
r
e
s
of
each
c
la
s
s
.
4.1.
M
e
t
r
ic
s
e
val
u
at
io
n
4.1.1.
B
in
ar
y
c
la
s
s
i
f
ic
at
io
n
m
e
t
r
ic
s
P
r
e
c
is
io
n
(
B
)
a
nd
r
e
c
a
ll
(
B
)
m
e
tr
ic
s
w
e
r
e
a
s
s
e
s
s
e
d
to
m
e
a
s
ur
e
th
e
m
ode
l'
s
p
e
r
f
or
m
a
nc
e
in
di
f
f
e
r
e
nt
ia
ti
ng
be
twe
e
n
bi
na
r
y
c
la
s
s
e
s
.
P
r
e
c
is
io
n
(
B
)
s
ig
ni
f
ie
s
th
e
a
c
c
ur
a
c
y
of
pos
it
iv
e
c
la
s
s
pr
e
di
c
ti
ons
,
w
hi
le
r
e
c
a
ll
(
B
)
ga
uge
s
th
e
m
ode
l'
s
a
bi
li
ty
to
c
a
pt
ur
e
a
ll
pos
it
iv
e
in
s
ta
nc
e
s
w
it
hi
n
th
e
da
ta
s
e
t.
A
ll
th
e
pl
ot
s
a
r
e
s
how
n
in
th
e
F
ig
ur
e
7.
F
ig
ur
e
7.
P
lo
t
of
lo
s
s
,
pr
e
c
is
io
n
a
nd
r
e
c
a
ll
dur
in
g
tr
a
in
in
g
a
nd
va
li
da
ti
on
f
or
our
da
ta
s
e
t
4.1.2. M
e
an
ave
r
age
p
r
e
c
i
s
io
n
m
e
t
r
ic
s
T
he
e
va
lu
a
ti
on
m
e
a
s
ur
e
d
th
e
m
e
a
n
a
ve
r
a
g
e
pr
e
c
is
io
n
(
m
A
P
)
a
t
50%
in
te
r
s
e
c
ti
on
ove
r
uni
on
(
m
A
P
50)
f
or
bo
th
bi
na
r
y
(
B
)
a
nd
m
ul
ti
c
la
s
s
(
M
)
s
it
ua
ti
ons
.
T
he
s
e
m
e
tr
ic
s
e
va
lu
a
te
th
e
m
ode
l'
s
pr
e
c
is
io
n
in
id
e
nt
if
yi
ng
a
nd
c
a
te
gor
is
in
g
obj
e
c
ts
a
t
di
f
f
e
r
e
nt
in
te
r
s
e
c
ti
on
ove
r
uni
on
th
r
e
s
hol
ds
.
T
he
a
c
hi
e
v
e
d
m
A
P
50
s
c
or
e
s
f
or
bot
h
bi
na
r
y
a
nd
m
ul
ti
c
la
s
s
s
c
e
na
r
io
s
de
m
on
s
tr
a
te
d
c
ons
is
te
nt
a
nd
hi
gh
va
lu
e
s
,
in
di
c
a
ti
ng
th
e
m
ode
l'
s
a
c
c
ur
a
c
y
in
lo
c
a
li
s
in
g
a
nd
c
l
a
s
s
if
yi
ng
obj
e
c
t
s
a
t
va
r
io
u
s
th
r
e
s
hol
ds
.
T
he
pl
ot
a
xe
s
of
a
r
e
r
e
pr
e
s
e
nt
e
d
on t
he
t
op e
a
c
h gr
a
ph obta
in
e
d.
5.
C
O
N
C
L
U
S
I
O
N
T
he
m
ode
l'
s
r
obus
t
pe
r
f
or
m
a
nc
e
in
di
f
f
e
r
e
nt
ia
ti
ng
pa
ddy
ty
pe
s
is
de
m
ons
tr
a
te
d
by
bi
na
r
y
a
nd
m
ul
ti
c
la
s
s
c
la
s
s
if
ic
a
ti
on
m
e
tr
ic
s
in
th
e
pr
opos
e
d
w
or
k.
T
he
m
ode
l'
s
hi
gh
pr
e
c
is
io
n
a
nd
r
e
c
a
ll
r
a
ti
ngs
f
o
r
bi
na
r
y
a
nd
m
ul
ti
c
la
s
s
c
la
s
s
if
ic
a
ti
on
s
s
how
its
a
bi
li
ty
to
a
c
c
ur
a
te
ly
id
e
nt
if
y
s
pe
c
if
ic
c
la
s
s
e
s
w
hi
le
ba
la
nc
in
g
pos
it
iv
e
c
a
s
e
s
a
c
r
os
s
th
e
da
t
a
s
e
t.
To
s
ol
ve
th
is
c
ha
ll
e
nge
,
we
c
r
e
a
te
d
s
ynt
he
ti
c
da
ta
s
e
ts
to
tr
a
in
th
e
m
ode
l
a
nd
te
s
t
it
us
in
g
a
va
li
da
ti
on
da
ta
s
e
t
us
in
g
dom
a
in
r
a
ndomi
z
a
ti
on.
T
he
m
ode
l
can
s
e
gm
e
nt
th
e
s
e
s
e
e
ds
in
to
in
s
ta
nc
e
s
e
gm
e
nt
s
f
r
om
th
e
v
a
li
da
ti
on
da
ta
s
e
t,
w
hi
c
h
c
om
pr
is
e
s
s
ynt
he
ti
c
a
ll
y
c
r
e
a
t
e
d
s
e
e
ds
w
it
h
a
ppr
opr
ia
te
pr
e
c
is
io
n
a
nd
lo
w
e
r
r
or
.
A
ddi
ti
ona
ll
y,
th
e
m
ode
l'
s
s
tr
ong
m
A
P
m
e
tr
ic
s
at
va
r
ie
d
in
te
r
s
e
c
ti
on
ove
r
uni
on
th
r
e
s
hol
ds
de
m
ons
tr
a
te
its
a
bi
li
ty
to
lo
c
a
li
s
e
a
nd
c
a
te
gor
is
e
pa
d
dy
da
ta
a
c
r
o
s
s
c
h
a
ngi
ng
obj
e
c
t
ov
e
r
la
p.
T
he
s
e
c
om
pr
e
he
ns
iv
e
e
v
a
lu
a
ti
ons
a
nd
hi
gh
-
pe
r
f
or
m
a
nc
e
m
e
tr
ic
s
d
e
m
ons
tr
a
te
th
e
m
ode
l'
s
pa
ddy
c
la
s
s
if
ic
a
ti
on
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
ti
f
I
nt
e
ll
I
S
S
N
:
2252
-
8938
Se
e
di
ng pr
e
c
i
s
io
n:
a m
as
k
r
e
gi
on bas
e
d c
on
v
ol
ut
io
nal
ne
ur
al
n
e
tw
or
k
s
... (
R
aj
as
hr
e
e
N
am
bi
a
r
)
4145
e
f
f
ic
a
c
y,
de
m
ons
tr
a
ti
ng
its
pot
e
nt
ia
l
f
or
r
e
a
l
-
w
or
ld
a
ppl
ic
a
ti
ons
in
r
e
li
a
bl
y
r
e
c
ogni
s
in
g
a
nd
c
a
te
gor
is
in
g
va
r
ie
d
r
ic
e
ki
nds
.
R
e
f
in
e
m
e
nt
a
nd
opt
im
is
a
ti
on
c
oul
d
im
pr
ove
th
e
m
ode
l'
s
pe
r
f
or
m
a
nc
e
a
nd
us
e
f
ul
ne
s
s
in
a
gr
ic
ul
tu
r
e
or
a
ut
om
a
te
d
c
r
op
m
oni
to
r
in
g
s
ys
te
m
s
.
R
E
F
E
R
E
N
C
E
S
[
1]
J.
H
e
a
t
on,
“
I
a
n
G
oodf
e
l
l
ow
,
Y
os
hua
B
e
ngi
o,
a
nd
A
a
r
on
C
our
vi
l
l
e
:
de
e
p
l
e
a
r
ni
ng,”
G
e
ne
t
i
c
P
r
ogr
am
m
i
ng
and
E
v
ol
v
abl
e
M
ac
hi
ne
s
,
vol
.
19,
no.
1
–
2,
pp.
305
–
307,
J
un.
2018,
doi
:
10.1007/
s
10710
-
017
-
9
314
-
z.
[
2]
O.
R
onne
be
r
ge
r
,
P.
F
i
s
c
he
r
,
a
nd
T.
B
r
ox,
“U
-
N
e
t
:
c
onvol
ut
i
ona
l
ne
t
w
o
r
ks
f
or
bi
om
e
di
c
a
l
i
m
a
ge
s
e
gm
e
nt
a
t
i
on,”
2015,
pp.
234
–
241
, doi
:
10.1007/
978
-
3
-
319
-
24574
-
4_28.
[
3]
E.
S
he
l
ha
m
e
r
,
J.
L
ong,
a
nd
T.
D
a
r
r
e
l
l
,
“
F
ul
l
y
c
onvol
ut
i
ona
l
ne
t
w
or
ks
f
or
s
e
m
a
nt
i
c
s
e
gm
e
nt
a
t
i
on,”
I
E
E
E
T
r
ans
ac
t
i
ons
on
P
at
t
e
r
n
A
nal
y
s
i
s
and
M
ac
hi
ne
I
nt
e
l
l
i
ge
nc
e
,
vol
.
39,
no.
4,
pp.
640
–
651,
A
pr
.
2017,
doi
:
10.1109/
T
P
A
M
I
.2016.2572683.
[
4]
K.
H
e
,
G.
G
ki
oxa
r
i
,
P.
D
ol
l
a
r
,
a
nd
R.
G
i
r
s
hi
c
k,
“
M
a
s
k
R
-
C
N
N
,”
in
2017
I
E
E
E
I
nt
e
r
nat
i
onal
C
onf
e
r
e
nc
e
on
C
om
put
e
r
V
i
s
i
on
(
I
C
C
V
)
,
I
E
E
E
,
O
c
t
.
2017,
pp.
2980
–
2988
, doi
:
10.1109/
I
C
C
V
.2017.322.
[
5]
A.
K
a
m
i
l
a
r
i
s
a
nd
F.
X.
P
.
-
B
ol
dú,
“
D
e
e
p
l
e
a
r
ni
ng
in
a
gr
i
c
ul
t
ur
e
:
a
s
ur
ve
y,”
C
om
put
e
r
s
and
E
l
e
c
t
r
oni
c
s
in
A
gr
i
c
ul
t
ur
e
,
vol
.
147,
pp.
70
–
90,
A
pr
.
2018,
doi
:
10.1016/
j
.c
om
pa
g.2018.02.016.
[
6]
Y.
K
a
ne
da
,
S.
S
hi
ba
t
a
,
a
nd
H.
M
i
ne
no,
“
M
ul
t
i
-
m
oda
l
s
l
i
di
ng
w
i
ndow
-
ba
s
e
d
s
uppor
t
ve
c
t
or
r
e
gr
e
s
s
i
on
f
or
pr
e
di
c
t
i
ng
pl
a
nt
w
a
t
e
r
s
t
r
e
s
s
,”
K
now
l
e
dge
-
B
as
e
d
Sy
s
t
e
m
s
,
vol
.
134,
pp.
135
–
148,
O
c
t
.
2017,
doi
:
10.1016/
j
.knos
ys
.2017.07.028.
[
7]
O.
R
us
s
a
kov
s
ky
et
al
.
,
“
I
m
a
ge
N
e
t
l
a
r
ge
s
c
a
l
e
vi
s
u
a
l
r
e
c
ogni
t
i
on
c
ha
l
l
e
nge
,”
I
nt
e
r
nat
i
onal
J
our
nal
of
C
om
put
e
r
V
i
s
i
on
,
vol
.
115,
no.
3,
pp.
211
–
252,
D
e
c
.
2015,
doi
:
10.1007/
s
11263
-
015
-
0816
-
y.
[
8]
Y.
A
yt
a
r
a
nd
A.
Z
i
s
s
e
r
m
a
n,
“
I
m
m
e
di
a
t
e
,
s
c
a
l
a
bl
e
obj
e
c
t
c
a
t
e
gor
y
de
t
e
c
t
i
on,”
in
2014
I
E
E
E
C
onf
e
r
e
nc
e
on
C
om
put
e
r
V
i
s
i
on
and
P
at
t
e
r
n
R
e
c
ogni
t
i
on
,
I
E
E
E
,
J
un.
2014,
pp.
2385
–
2392
, doi
:
10.1109/
C
V
P
R
.2014.305.
[
9]
W.
G
uo
et
al
.
,
“
A
e
r
i
a
l
i
m
a
ge
r
y
a
na
l
ys
i
s
–
qua
nt
i
f
yi
ng
a
ppe
a
r
a
nc
e
a
nd
num
b
e
r
of
s
or
ghum
he
a
ds
f
or
a
ppl
i
c
a
t
i
ons
in
br
e
e
di
ng
a
n
d
a
gr
onom
y,”
F
r
ont
i
e
r
s
in
P
l
ant
Sc
i
e
nc
e
,
vol
.
9,
O
c
t
.
2018,
doi
:
10.3389/
f
pl
s
.201
8.01544.
[
10]
X.
J
i
n,
S.
M
a
de
c
,
D.
D
ut
a
r
t
r
e
,
B.
de
S
ol
a
n,
A.
C
om
a
r
,
a
nd
F.
B
a
r
e
t
,
“
H
i
gh
-
t
hr
oughput
m
e
a
s
ur
e
m
e
nt
s
of
s
t
e
m
c
ha
r
a
c
t
e
r
i
s
t
i
c
s
to
e
s
t
i
m
a
t
e
ear
de
ns
i
t
y
a
nd
a
bove
-
gr
ound
bi
om
a
s
s
,”
P
l
ant
P
he
nom
i
c
s
,
vol
.
2019,
J
a
n.
2019,
doi
:
10.34133/
2019/
4820305.
[
11]
S.
G
hos
a
l
et
al
.
,
“A
w
e
a
kl
y
s
upe
r
vi
s
e
d
de
e
p
l
e
a
r
ni
ng
f
r
a
m
e
w
or
k
f
or
s
or
ghum
he
a
d
de
t
e
c
t
i
on
a
nd
c
ount
i
ng,”
P
l
ant
P
he
nom
i
c
s
,
vol
.
2019,
J
a
n.
2019,
doi
:
10.34133/
2019/
1525874.
[
12]
A.
L.
C
h
a
ndr
a
,
S.
V.
D
e
s
a
i
,
V.
N.
B
a
l
a
s
ubr
a
m
a
ni
a
n,
S.
N
i
nom
i
ya
,
a
nd
W.
G
u
o,
“
A
c
t
i
ve
l
e
a
r
ni
ng
w
i
t
h
poi
nt
s
up
e
r
vi
s
i
on
f
or
c
os
t
-
e
f
f
e
c
t
i
ve
pa
ni
c
l
e
de
t
e
c
t
i
on
in
c
e
r
e
a
l
c
r
ops
,”
P
l
ant
M
e
t
hods
,
vol
.
16,
no.
1,
D
e
c
.
2020,
doi
:
10.1186/
s
13007
-
020
-
00575
-
8.
[
13]
T.
N
a
t
h,
A.
M
a
t
hi
s
,
A.
C.
C
he
n,
A.
P
a
t
e
l
,
M.
B
e
t
hge
,
a
nd
M.
W.
M
a
t
hi
s
,
“
U
s
i
ng
D
e
e
pL
a
b
C
ut
f
or
3D
m
a
r
ke
r
l
e
s
s
pos
e
e
s
t
i
m
a
t
i
on
a
c
r
os
s
s
p
e
c
i
e
s
a
nd
be
ha
vi
or
s
,”
N
at
ur
e
P
r
ot
oc
ol
s
,
vol
.
14,
no.
7,
pp.
2152
–
2176,
J
ul
.
2019,
doi
:
10.1038/
s
41596
-
019
-
0176
-
0.
[
14]
T.
I
s
oka
ne
,
F.
O
kur
a
,
A.
I
de
,
Y.
M
a
t
s
us
hi
t
a
,
a
nd
Y.
Y
a
gi
,
“
P
r
oba
bi
l
i
s
t
i
c
pl
a
nt
m
ode
l
i
ng
vi
a
m
ul
t
i
-
vi
e
w
i
m
a
ge
-
to
-
i
m
a
ge
t
r
a
ns
l
a
t
i
on,”
in
2018
I
E
E
E
/
C
V
F
C
onf
e
r
e
nc
e
on
C
om
put
e
r
V
i
s
i
on
and
P
at
t
e
r
n
R
e
c
ogni
t
i
on
,
I
E
E
E
,
J
un.
2018,
pp.
2906
–
2915
,
doi
:
10.1109/
C
V
P
R
.2018.00307.
[
15]
C.
L
a
z
o,
“
S
e
gm
e
nt
a
t
i
on
of
s
ki
n
l
e
s
i
on
s
a
nd
t
he
i
r
a
t
t
r
i
but
e
s
us
i
ng
g
e
ne
r
a
t
i
ve
a
dve
r
s
a
r
i
a
l
n
e
t
w
or
ks
,”
in
L
at
i
nX
in
AI
at
N
e
u
r
al
I
nf
or
m
at
i
on
P
r
oc
e
s
s
i
ng
Sy
s
t
e
m
s
C
onf
e
r
e
nc
e
2019
,
D
e
c
.
2019
, doi
:
10.52591/
l
x
a
i
201912083.
[
16]
A.
S
hr
i
va
s
t
a
va
,
T.
P
f
i
s
t
e
r
,
O.
T
uz
e
l
,
J.
S
us
s
ki
nd,
W.
W
a
ng,
a
nd
R.
W
e
bb,
“
L
e
a
r
ni
ng
f
r
om
s
i
m
ul
a
t
e
d
a
nd
uns
upe
r
vi
s
e
d
i
m
a
ge
s
t
hr
ough
a
dve
r
s
a
r
i
a
l
t
r
a
i
ni
ng,”
in
2017
I
E
E
E
C
onf
e
r
e
nc
e
on
C
om
put
e
r
V
i
s
i
on
and
P
at
t
e
r
n
R
e
c
ogni
t
i
on
(
C
V
P
R
)
,
I
E
E
E
,
J
ul
.
2017,
pp.
2242
–
2251
, doi
:
10.1109/
C
V
P
R
.2017.241.
[
17]
M.
V.
G
i
uf
f
r
i
da
,
H.
S
c
ha
r
r
,
a
nd
S.
A.
T
s
a
f
t
a
r
i
s
,
“
A
R
I
G
A
N
:
s
ynt
he
t
i
c
a
r
a
bi
dop
s
i
s
pl
a
nt
s
u
s
i
ng
ge
ne
r
a
t
i
ve
a
dve
r
s
a
r
i
a
l
ne
t
w
or
k,”
in
2017
I
E
E
E
I
nt
e
r
nat
i
onal
C
onf
e
r
e
nc
e
on
C
om
put
e
r
V
i
s
i
on
W
or
k
s
hops
(
I
C
C
V
W
)
,
I
E
E
E
,
O
c
t
.
2017,
pp.
2064
–
2071
,
doi
:
10.1109/
I
C
C
V
W
.2017.242.
[
18]
M.
A
r
s
e
novi
c
,
M.
K
a
r
a
novi
c
,
S.
S
l
a
doj
e
vi
c
,
A.
A
nde
r
l
a
,
a
nd
D.
S
t
e
f
a
novi
c
,
“
S
ol
vi
ng
c
ur
r
e
nt
l
i
m
i
t
a
t
i
ons
of
de
e
p
l
e
a
r
ni
ng
ba
s
e
d
a
ppr
oa
c
he
s
f
or
pl
a
nt
di
s
e
a
s
e
de
t
e
c
t
i
on,”
Sy
m
m
e
t
r
y
,
vol
.
11,
no.
7,
J
ul
.
2019,
doi
:
10.3390/
s
ym
11070939.
[
19]
M.
L.
W
i
di
a
s
t
ut
i
,
A.
H
a
i
r
m
a
ns
i
s
,
E.
R.
P
a
l
upi
,
a
nd
S.
I
l
ya
s
,
“
D
i
gi
t
a
l
i
m
a
ge
a
na
l
ys
i
s
u
s
i
ng
f
l
a
t
be
d
s
c
a
nni
ng
s
ys
t
e
m
f
or
pur
i
t
y
t
e
s
t
i
ng
of
r
i
c
e
s
e
e
d
a
nd
c
onf
i
r
m
a
t
i
on
by
gr
ow
out
t
e
s
t
,”
I
ndone
s
i
an
J
our
nal
of
A
gr
i
c
ul
t
ur
al
Sc
i
e
nc
e
,
vol
.
19,
no.
2,
p
p.
49
-
56
,
D
e
c
.
2018,
doi
:
10.21082/
i
j
a
s
.v19n2.2018.p49
-
56.
[
20]
K.
S.
J
a
m
una
,
S.
K
a
r
pa
ga
va
l
l
i
,
M.
S.
V
i
j
a
ya
,
P.
R
e
va
t
hi
,
S.
G
oki
l
a
va
ni
,
a
nd
E.
M
a
dhi
ya
,
“
C
l
a
s
s
i
f
i
c
a
t
i
on
of
s
e
e
d
c
ot
t
on
yi
e
l
d
ba
s
e
d
on
t
he
gr
ow
t
h
s
t
a
ge
s
of
c
ot
t
on
c
r
op
us
i
ng
m
a
c
hi
ne
l
e
a
r
ni
ng
t
e
c
hni
que
s
,
”
in
2010
I
nt
e
r
nat
i
onal
C
onf
e
r
e
nc
e
on
A
dv
anc
e
s
in
C
om
put
e
r
E
ngi
ne
e
r
i
ng
,
I
E
E
E
,
J
un.
2010,
pp.
312
–
315
, doi
:
10.1109/
A
C
E
.2010
.71.
[
21]
O.
A
dj
e
m
out
,
K.
H
a
m
m
ouc
he
,
a
nd
M.
D
i
a
f
,
“
A
ut
om
a
t
i
c
s
e
e
ds
r
e
c
ogni
t
i
on
by
s
i
z
e
,
f
or
m
a
nd
t
e
xt
ur
e
f
e
a
t
ur
e
s
,”
in
2007
9t
h
I
nt
e
r
nat
i
onal
Sy
m
pos
i
um
on
Si
gnal
P
r
oc
e
s
s
i
ng
and
I
t
s
A
ppl
i
c
a
t
i
ons
,
I
E
E
E
,
F
e
b.
2007,
pp.
1
–
4
,
doi
:
10.1109/
I
S
S
P
A
.2007.4555428.
[
22]
N.
H
.
S
on
a
nd
N.
T
ha
i
-
N
ghe
,
“
D
e
e
p
l
e
a
r
ni
ng
f
or
r
i
c
e
qua
l
i
t
y
c
l
a
s
s
i
f
i
c
a
t
i
on,
”
in
2019
I
nt
e
r
nat
i
onal
C
onf
e
r
e
nc
e
on
A
dv
anc
e
d
C
om
put
i
ng
and
A
ppl
i
c
at
i
ons
(
A
C
O
M
P
)
,
I
E
E
E
,
N
ov.
2019,
pp.
92
–
96
, doi
:
10.1
109/
A
C
O
M
P
.2019.00021.
[
23]
P.
W
i
j
e
r
a
t
hna
a
nd
L.
R
a
na
t
hunga
,
“
R
i
c
e
c
a
t
e
gor
y
i
de
nt
i
f
i
c
a
t
i
on
us
i
ng
he
ur
i
s
t
i
c
f
e
a
t
ur
e
gui
de
d
m
a
c
hi
ne
vi
s
i
on
a
ppr
oa
c
h,
”
in
2018
I
E
E
E
13t
h
I
nt
e
r
nat
i
onal
C
onf
e
r
e
nc
e
on
I
ndus
t
r
i
al
and
I
nf
or
m
at
i
on
Sy
s
t
e
m
s
(
I
C
I
I
S)
,
I
E
E
E
,
D
e
c
.
2018,
pp.
185
–
190
,
doi
:
10.1109/
I
C
I
I
N
F
S
.2018.8721396.
[
24]
S.
K
hunkhe
t
t
a
nd
T.
R
e
m
s
ungn
e
n,
“
N
on
-
de
s
t
r
uc
t
i
ve
i
de
nt
i
f
i
c
a
t
i
on
of
pur
e
br
e
e
di
ng
r
i
c
e
s
e
e
d
u
s
i
ng
di
gi
t
a
l
i
m
a
ge
a
na
l
y
s
i
s
,”
in
T
h
e
4t
h
J
oi
nt
I
nt
e
r
nat
i
onal
C
onf
e
r
e
nc
e
on
I
nf
or
m
at
i
on
and
C
om
m
uni
c
at
i
on
T
e
c
hnol
ogy
,
E
l
e
c
t
r
oni
c
and
E
l
e
c
t
r
i
c
al
E
ngi
ne
e
r
i
ng
(
J
I
C
T
E
E
)
,
I
E
E
E
,
M
a
r
.
2014,
pp.
1
–
4
,
doi
:
10.1109/
J
I
C
T
E
E
.2014.6804096.
[
25]
H.
-
T.
D
uong
a
nd
V.
T.
H
oa
ng,
“
D
i
m
e
n
s
i
ona
l
i
t
y
r
e
duc
t
i
on
ba
s
e
d
on
f
e
a
t
ur
e
s
e
l
e
c
t
i
on
f
or
r
i
c
e
va
r
i
e
t
i
e
s
r
e
c
ogni
t
i
on,”
in
2019
4t
h
I
nt
e
r
nat
i
onal
C
onf
e
r
e
nc
e
on
I
nf
or
m
at
i
on
T
e
c
hnol
og
y
(
I
nC
I
T
)
,
I
E
E
E
,
O
c
t
.
2019
,
pp.
199
–
202
, doi
:
10.1109/
I
N
C
I
T
.2019.8912121.
[
26]
Y.
W
u,
Z.
Y
a
ng,
W.
W
u,
X.
L
i
,
a
nd
D.
T
a
o,
“
D
e
e
p
-
R
i
c
e
:
de
e
p
m
ul
t
i
-
s
e
n
s
or
i
m
a
ge
r
e
c
ogni
t
i
on
f
or
gr
a
di
ng
r
i
c
e
,”
in
2018
I
E
E
E
I
nt
e
r
nat
i
onal
C
onf
e
r
e
n
c
e
on
I
nf
or
m
at
i
on
and
A
ut
om
at
i
on
(
I
C
I
A
)
,
I
E
E
E
,
A
ug.
2018,
pp.
116
–
120
,
doi
:
10.1109/
I
C
I
nf
A
.2018.8812590.
[
27]
T.
-
Y.
L
i
n
et
al
.
,
“
M
i
c
r
os
of
t
C
O
C
O
:
c
om
m
on
obj
e
c
t
s
in
c
ont
e
xt
,”
C
om
put
e
r
V
i
s
i
on
–
E
C
C
V
2014:
13
t
h
E
ur
ope
an
C
onf
e
r
e
nc
e
,
Z
ur
i
c
h, S
w
i
t
z
e
r
l
a
nd,
2014,
pp.
740
–
755
, doi
:
10.1007/
978
-
3
-
319
-
10602
-
1_48.
[
28]
O.
R
onne
be
r
ge
r
,
“
I
nvi
t
e
d
t
a
l
k:
U
-
N
e
t
c
onvol
ut
i
ona
l
ne
t
w
or
ks
f
or
bi
om
e
di
c
a
l
i
m
a
ge
s
e
gm
e
nt
a
t
i
on,”
B
i
l
dv
e
r
ar
be
i
t
ung
f
ü
r
di
e
M
e
di
z
i
n
, B
e
r
l
i
n, H
e
i
de
l
be
r
g:
S
pr
i
nge
r
,
2017,
doi
:
10.1007/
978
-
3
-
662
-
54345
-
0_
3.
[
29]
M.
G
oya
l
,
M.
Y
a
p,
a
nd
S.
H
a
s
s
a
npour
,
“
M
ul
t
i
-
c
l
a
s
s
s
e
m
a
nt
i
c
s
e
gm
e
nt
a
t
i
on
of
s
ki
n
l
e
s
i
ons
vi
a
f
ul
l
y
c
onvol
ut
i
ona
l
ne
t
w
or
ks
,”
in
P
r
oc
e
e
di
ngs
of
t
he
13t
h
I
nt
e
r
nat
i
onal
J
oi
nt
C
onf
e
r
e
nc
e
on
B
i
om
e
di
c
al
E
ngi
ne
e
r
i
ng
Sy
s
t
e
m
s
and
T
e
c
hnol
ogi
e
s
,
S
C
I
T
E
P
R
E
S
S
-
S
c
i
e
nc
e
a
nd
T
e
c
hnol
ogy
P
ubl
i
c
a
t
i
ons
,
2020,
pp.
290
–
295
, doi
:
10.5220/
0009380300002513.
[
30]
L.
-
C.
C
he
n,
Y.
Z
hu,
G.
P
a
pa
ndr
e
ou,
F.
S
c
hr
of
f
,
a
nd
H.
A
da
m
,
“
E
nc
ode
r
-
de
c
o
de
r
w
i
t
h
a
t
r
ous
s
e
pa
r
a
bl
e
c
onvol
ut
i
on
f
or
s
e
m
a
nt
i
c
i
m
a
ge
s
e
gm
e
nt
a
t
i
on,”
in
P
r
oc
e
e
di
ngs
of
t
he
E
ur
ope
an
c
onf
e
r
e
nc
e
on
c
om
put
e
r
v
i
s
i
on
(
E
C
C
V
)
,
2018,
pp.
801
–
818.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
, V
ol
.
13
, N
o.
4
,
D
e
c
e
m
be
r
2024
:
4138
-
4146
4146
[
31]
H.
Z
ha
o,
J.
S
hi
,
X.
Q
i
,
X.
W
a
ng,
a
nd
J.
J
i
a
,
“
P
yr
a
m
i
d
s
c
e
ne
pa
r
s
i
ng
ne
t
w
or
k,”
in
2017
I
E
E
E
C
onf
e
r
e
nc
e
on
C
om
put
e
r
V
i
s
i
on
and
P
at
t
e
r
n
R
e
c
ogni
t
i
on
(
C
V
P
R
)
,
I
E
E
E
,
J
ul
.
2017,
pp.
6230
–
6239
, do
i
:
10.1109/
C
V
P
R
.2017.660.
[
32]
V.
B
a
dr
i
na
r
a
ya
na
n,
A.
K
e
nda
l
l
,
a
nd
R.
C
i
pol
l
a
,
“
S
e
gN
e
t
:
a
de
e
p
c
onvo
l
ut
i
ona
l
e
nc
ode
r
-
de
c
ode
r
a
r
c
hi
t
e
c
t
ur
e
f
or
i
m
a
ge
s
e
gm
e
nt
a
t
i
on,”
I
E
E
E
T
r
ans
ac
t
i
ons
on
P
at
t
e
r
n
A
na
l
y
s
i
s
and
M
ac
hi
ne
I
nt
e
l
l
i
ge
n
c
e
,
vol
.
39,
no.
12,
pp.
2481
–
2495,
D
e
c
.
2017,
doi
:
10.1109/
T
P
A
M
I
.2016.2644615.
[
33]
G.
L
i
n,
A.
M
i
l
a
n,
C.
S
he
n,
a
nd
I.
R
e
i
d,
“
R
e
f
i
ne
N
e
t
:
m
ul
t
i
-
pa
t
h
r
e
f
i
ne
m
e
nt
ne
t
w
or
ks
f
or
hi
gh
-
r
e
s
ol
ut
i
on
s
e
m
a
nt
i
c
s
e
gm
e
nt
a
t
i
on,”
in
2017
I
E
E
E
C
onf
e
r
e
nc
e
on
C
om
put
e
r
V
i
s
i
on
and
P
at
t
e
r
n
R
e
c
ogni
t
i
on
(
C
V
P
R
)
,
I
E
E
E
,
J
ul
.
2017,
pp.
5168
–
5177
,
doi
:
10.1109/
C
V
P
R
.2017.549.
[
34]
W
.
B
a
i
,
“
E
ne
t
s
e
m
a
nt
i
c
s
e
gm
e
nt
a
t
i
on
c
om
bi
ne
d
w
i
t
h
a
t
t
e
nt
i
on
m
e
c
ha
ni
s
m
,”
R
e
s
e
ar
c
h
Squar
e
,
2021,
doi
:
10.21203/
r
s
.3.r
s
-
425438/
v1.
[
35]
J.
W
a
ng
et
al
.
,
“
D
e
e
p
hi
gh
-
r
e
s
ol
ut
i
on
r
e
pr
e
s
e
nt
a
t
i
on
l
e
a
r
ni
ng
f
or
vi
s
ua
l
r
e
c
ogn
i
t
i
on,”
I
E
E
E
T
r
ans
ac
t
i
ons
on
P
at
t
e
r
n
A
nal
y
s
i
s
an
d
M
ac
hi
ne
I
nt
e
l
l
i
ge
nc
e
,
vol
.
43,
no.
10,
pp.
3349
–
3364,
O
c
t
.
2021,
doi
:
10.1109/
T
P
A
M
I
.2020.2983686.
[
36]
M.
G
a
j
j
a
,
“
B
r
a
i
n
t
um
or
de
t
e
c
t
i
on
us
i
ng
m
a
s
k
R
-
C
N
N
,”
J
our
nal
of
A
dv
anc
e
d
R
e
s
e
ar
c
h
in
D
y
nam
i
c
al
and
C
ont
r
ol
Sy
s
t
e
m
s
,
vol
.
12,
no.
8,
pp.
101
–
108,
J
ul
.
2020,
doi
:
10.5373/
J
A
R
D
C
S
/
V
12S
P
8/
20202506.
B
I
O
G
R
A
P
H
I
E
S
OF
A
U
T
H
O
R
S
Rajashree
Nambiar
holds
a
Masters
of
Technology
degree
from
Nitte
University,
India
in
2014.
She
also
received
his
Bachelo
r
of
Enginee
ring
from
Visvesvaraya
Technological
University,
Belagav
i,
India.
She
is
currently
an
Assistant
Profes
sor
at
Department
of
Robotics
and
Artificial
Engineering
at
NMAM
Institut
e
of
Technology,
NITTE
(Deemed
to
be
University),
Nitte,
India.
She
is
currently
a
resear
ch
scholar
at
the
JAIN
(Deemed
to
be
University),
Bengalur
u.
Her
research
includes
artifici
al
intelligence
,
machine
learning,
deep
learning,
imag
e,
and
signal
processing.
She
can
be
contacted
at
em
ail:
raji24oct@
gmail.com
or
rajashree.n@
nitte.edu.
Ranjith
Bhat
holds
a
Masters
of
Technology
degree
from
Nitte
University,
India
in
2011.
He
also
received
his
Bachelo
r
of
Engineering
from
Visv
esvaraya
Technological
University,
Belagav
i,
India.
He
is
currently
an
Assistant
Profes
sor
at
Department
of
Robotics
and
Artificial
Engineering
at
NMAM
Institut
e
of
Technology,
NITTE
(Deemed
to
be
University),
Nitte,
India.
He
is
currently
a
research
scholar
at
the
JAIN
(deemed
to
be)
university,
Bengalur
u.
His
research
includes
artificial
intelligence
,
machine
learning,
deep
learning,
network
security,
and
computer
networks.
He
can
be
contacted
at
email:
ranjithbhat
@
gmail.com
or
ranjith.bh
at@
nitte.edu.
Varuna
Kumara
is
a
Resear
ch
Scholar
in
the
Department
of
Electronics
Engineering
at
JAIN
Deemed
to
be
University,
Bengalur
u,
India.
He
also
received
his
B.E
.
and
M.Tech
.
from
Visvesvaraya
Technological
University,
Belagav
i,
India
in
2009
and
2012
respectively.
He
is
currently
Assistant
Profes
sor
at
Electronic
s
and
Communicat
ion
Engineering
in
Moodlakatte
Institut
e
of
Technology,
Kundapura,
Indi
a.
His
research
interests
are
in
artificial
intelligence
,
signal
processing,
and
control
systems.
He
can
be
contacted
at
email:
vkumarg.24@
gmail.com.
Evaluation Warning : The document was created with Spire.PDF for Python.