I
nte
rna
t
io
na
l J
o
urna
l o
f
E
lect
rica
l a
nd
Co
m
pu
t
er
E
ng
ineering
(
I
J
E
CE
)
Vo
l.
1
6
,
No
.
3
,
J
u
n
e
20
2
6
,
p
p
.
1307
~
1
3
1
8
I
SS
N:
2088
-
8
7
0
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijece.
v
1
6
i
3
.
pp
1
3
0
7
-
1
3
1
8
1307
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
ec
e.
ia
esco
r
e.
co
m
G
AN
-
a
ug
mented
v
isio
n t
ra
nsfo
rme
r wit
h
bala
nced s
y
nthet
ic
da
ta g
eneratio
n f
o
r robus
t
ric
e leaf
disea
se det
ection
Sa
if
ul I
s
la
m
1,
2
,
M
d.
Na
s
im
Ak
hta
r
2
,
M
.
M
a
ha
di H
a
s
s
a
n
2
,
A.
N.
M
.
Re
z
a
ul K
a
rim
2
,
I
s
ra
t
B
inte
h
H
a
bib
2
1
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
,
D
h
a
k
a
U
n
i
v
e
r
s
i
t
y
o
f
E
n
g
i
n
e
e
r
i
n
g
a
n
d
Te
c
h
n
o
l
o
g
y
,
G
a
z
i
p
u
r
,
B
a
n
g
l
a
d
e
s
h
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
,
I
n
t
e
r
n
a
t
i
o
n
a
l
I
sl
a
mi
c
U
n
i
v
e
r
si
t
y
o
f
C
h
i
t
t
a
g
o
n
g
,
C
h
i
t
t
a
g
o
n
g
,
B
a
n
g
l
a
d
e
sh
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
No
v
7
,
2
0
2
5
R
ev
is
ed
Feb
1
0
,
2
0
2
6
Acc
ep
ted
Ma
r
1
6
,
2
0
2
6
Early
a
n
d
a
c
c
u
ra
te
id
e
n
t
ifi
c
a
ti
o
n
o
f
rice
lea
f
d
ise
a
se
s
is
e
ss
e
n
ti
a
l
fo
r
su
sta
in
a
b
le
c
ro
p
m
a
n
a
g
e
m
e
n
t;
h
o
we
v
e
r,
m
a
n
y
e
x
isti
n
g
c
o
n
v
o
l
u
ti
o
n
a
l
n
e
u
ra
l
n
e
two
rk
s
(CNN
s
)
b
a
se
d
so
l
u
ti
o
n
s
stru
g
g
le
wit
h
c
las
s
imb
a
lan
c
e
a
n
d
li
m
it
e
d
ro
b
u
stn
e
ss
wh
e
n
a
p
p
li
e
d
t
o
re
a
l
-
field
d
a
ta.
In
t
h
is
wo
r
k
,
a
g
e
n
e
ra
ti
v
e
a
d
v
e
rsa
rial
n
e
two
rk
(
G
AN
)
a
u
g
m
e
n
ted
v
isio
n
tra
n
sfo
rm
e
r
(ViT)
fra
m
e
wo
rk
is
in
tr
o
d
u
c
e
d
t
o
o
v
e
rc
o
m
e
t
h
e
se
li
m
it
a
ti
o
n
s
.
A
d
e
e
p
siz
e
re
p
r
e
se
n
tativ
e
sa
m
p
les
fo
r
u
n
d
e
rre
p
re
se
n
ted
d
ise
a
se
c
a
teg
o
ries
,
re
su
lt
in
g
in
a
m
o
re
b
a
lan
c
e
d
trai
n
in
g
d
a
tas
e
t
a
n
d
a
c
h
iev
i
n
g
a
F
ré
c
h
e
t
i
n
c
e
p
ti
o
n
d
ista
n
c
e
(F
ID)
sc
o
re
o
f
1
8
.
6
.
Th
e
b
a
lan
c
e
d
d
a
tas
e
t
is
th
e
n
u
se
d
to
trai
n
a
v
isio
n
tr
a
n
sfo
rm
e
r
m
o
d
e
l
th
a
t
lev
e
ra
g
e
s
se
lf
-
a
tt
e
n
ti
o
n
t
o
c
a
p
t
u
re
g
l
o
b
a
l
c
o
n
te
x
tu
a
l
f
e
a
tu
re
s
o
f
rice
lea
f
ima
g
e
s.
Ex
p
e
rime
n
tal
e
v
a
lu
a
ti
o
n
a
c
ro
ss
ten
d
ise
a
se
c
las
se
s
sh
o
ws
th
a
t
t
h
e
p
ro
p
o
se
d
a
p
p
ro
a
c
h
a
tt
a
in
s
a
n
o
v
e
ra
ll
c
las
sifica
ti
o
n
a
c
c
u
ra
c
y
o
f
9
6
.
5
%
,
e
x
c
e
e
d
in
g
t
h
e
p
e
rfo
rm
a
n
c
e
o
f
se
v
e
ra
l
e
sta
b
li
sh
e
d
CNN
a
rc
h
it
e
c
tu
re
s.
Ad
d
it
i
o
n
a
ll
y
,
th
e
m
o
d
e
l
d
e
m
o
n
st
ra
tes
stro
n
g
g
e
n
e
ra
li
z
a
ti
o
n
c
a
p
a
b
i
li
ty
o
n
a
n
e
x
tern
a
l
fiel
d
d
a
tas
e
t,
a
c
h
iev
i
n
g
9
4
.
8
%
a
c
c
u
ra
c
y
.
T
o
v
a
li
d
a
te
r
e
a
l
-
wo
rld
a
p
p
li
c
a
b
il
i
ty
,
th
e
train
e
d
m
o
d
e
l
i
s
d
e
p
lo
y
e
d
o
n
a
Je
tso
n
Na
n
o
e
d
g
e
d
e
v
ice
,
wh
e
re
it
d
e
li
v
e
rs
e
fficie
n
t
i
n
fe
re
n
c
e
p
e
rfo
rm
a
n
c
e
su
it
a
b
le
fo
r
p
ra
c
ti
c
a
l
a
g
ricu
lt
u
ra
l
a
p
p
li
c
a
ti
o
n
s.
Th
e
fin
d
in
g
s
i
n
d
ica
te
th
a
t
c
o
m
b
i
n
in
g
GA
N
-
b
a
se
d
d
a
ta
a
u
g
m
e
n
tatio
n
with
tran
sf
o
r
m
e
r
-
b
a
se
d
lea
rn
in
g
p
ro
v
i
d
e
s
a
r
e
li
a
b
le
a
n
d
sc
a
lab
le so
lu
ti
o
n
f
o
r
rice
lea
f
d
ise
a
se
d
e
tec
ti
o
n
.
K
ey
w
o
r
d
s
:
Data
au
g
m
en
tatio
n
a
g
r
icu
ltu
r
e
Dee
p
co
n
v
o
lu
tio
n
al
g
en
er
ativ
e
ad
v
er
s
ar
ial
n
etwo
r
k
G
e
n
e
r
a
t
i
v
e
a
d
v
e
r
s
a
r
i
al
n
e
t
w
o
r
k
s
R
ice
leaf
d
is
ea
s
e
d
etec
tio
n
Vis
io
n
tr
an
s
f
o
r
m
er
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 BY
-
SA
li
c
e
n
se
C
o
r
r
e
s
p
o
nd
ing
A
uth
o
r
:
Saif
u
l I
s
lam
Dep
ar
tm
en
t o
f
C
o
m
p
u
ter
Scie
n
ce
an
d
E
n
g
in
ee
r
in
g
,
I
n
ter
n
ati
o
n
al
I
s
lam
ic
Un
iv
er
s
ity
o
f
C
h
ittag
o
n
g
C
h
ittag
o
n
g
,
B
an
g
lad
esh
E
m
ail: saif
u
lcse@
i
iu
c.
ac
.
b
d
1.
I
NT
RO
D
UCT
I
O
N
Fo
r
ap
p
r
o
x
im
ately
f
if
ty
p
er
ce
n
t
o
f
p
eo
p
le
wo
r
l
d
wid
e,
r
ice
s
er
v
es
as
th
eir
p
r
im
ar
y
n
u
tr
itio
n
al
en
er
g
y
s
o
u
r
ce
an
d
th
e
p
r
in
ci
p
al
f
o
o
d
s
o
u
r
ce
f
o
r
o
v
er
3
.
5
b
illi
o
n
p
eo
p
le
in
Asi
a,
L
atin
Am
er
ica,
an
d
p
o
r
tio
n
s
o
f
Af
r
ica,
m
ak
in
g
r
ice
o
n
e
o
f
th
e
m
ajo
r
s
tap
le
f
o
o
d
s
co
n
s
u
m
ed
ar
o
u
n
d
th
e
wo
r
ld
[
1
]
,
[
2
]
.
Alth
o
u
g
h
r
ice
p
lay
s
a
n
im
p
o
r
tan
t
r
o
le,
v
ar
io
u
s
leaf
d
is
ea
s
es
p
o
s
e
s
er
io
u
s
p
r
o
b
lem
s
f
o
r
r
ice
p
r
o
d
u
ctio
n
.
T
h
ese
d
is
ea
s
es
ca
n
lo
wer
y
ield
s
b
y
as
m
u
c
h
as
7
0
%
an
d
d
am
a
g
e
g
r
ain
q
u
ality
[
3
]
.
Dis
ea
s
e
d
iag
n
o
s
is
h
as
h
is
to
r
ically
d
ep
en
d
e
d
o
n
f
ar
m
er
s
’
o
r
ag
r
o
n
o
m
is
ts
’
m
a
n
u
al
v
is
u
al
i
n
v
esti
g
atio
n
,
wh
ic
h
co
s
ts
m
o
n
ey
,
tak
es
a
lo
n
g
tim
e,
an
d
is
p
r
o
n
e
to
h
u
m
an
m
is
tak
es,
esp
ec
ially
f
o
r
s
m
allh
o
ld
e
r
f
a
r
m
er
s
wh
o
h
av
e
lim
ited
ac
ce
s
s
to
p
r
o
f
ess
io
n
al
ass
is
tan
ce
[
4
]
.
T
h
ese
r
estrictio
n
s
f
r
e
q
u
en
tly
r
esu
lt
in
in
co
r
r
ec
t
d
ia
g
n
o
s
es,
p
o
s
tp
o
n
e
d
tr
ea
tm
e
n
ts
,
a
n
d
th
e
o
v
er
u
s
e
o
f
p
esti
cid
es,
all
o
f
wh
ich
wo
r
s
en
ag
r
icu
ltu
r
al
lo
s
s
es
an
d
en
v
ir
o
n
m
en
tal
d
am
ag
e.
R
ec
en
t
d
ev
elo
p
m
en
ts
in
d
ee
p
lear
n
in
g
(
DL
)
an
d
ar
tific
ial
i
n
tellig
en
ce
(
AI
)
h
av
e
s
h
o
wn
p
r
o
m
is
e
in
a
u
to
m
atin
g
th
e
d
iag
n
o
s
is
o
f
p
lan
t
d
is
ea
s
es.
C
o
n
v
o
lu
tio
n
al
n
eu
r
a
l
n
etwo
r
k
s
(
C
NNs)
h
av
e
d
em
o
n
s
tr
ated
ef
f
icac
y
in
d
etec
tin
g
r
ice
d
is
ea
s
es
f
r
o
m
im
ag
es
o
f
leav
es;
h
o
wev
er
,
m
o
d
els
s
u
ch
as
VGG1
6
o
r
R
esNet
f
r
eq
u
en
tly
s
u
f
f
er
f
r
o
m
o
v
er
f
itti
n
g
o
n
s
m
al
l
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
1
6
,
No
.
3
,
J
u
n
e
20
2
6
:
1
3
0
7
-
1
3
1
8
1308
d
atasets
,
s
tr
u
g
g
le
to
ca
p
tu
r
e
lo
n
g
-
r
a
n
g
e
s
p
atial
f
ea
tu
r
es,
an
d
d
em
an
d
s
ig
n
if
ican
t
co
m
p
u
tatio
n
al
r
eso
u
r
ce
s
.
Gen
er
ativ
e
ad
v
e
r
s
ar
ial
n
etwo
r
k
s
(
GANs)
h
av
e
b
ee
n
d
esig
n
e
d
to
p
r
o
d
u
ce
ar
tific
ially
cr
ea
t
ed
im
ag
es
in
o
r
d
er
to
b
alan
ce
d
atasets
an
d
en
h
a
n
ce
m
o
d
el
g
en
e
r
aliza
tio
n
[
5
]
.
T
o
o
v
e
r
co
m
e
th
ese
lim
itatio
n
s
,
v
is
io
n
tr
an
s
f
o
r
m
er
s
(
ViT
s
)
h
av
e
g
ar
n
er
ed
atten
tio
n
b
ec
au
s
e
o
f
th
eir
ab
ilit
y
to
id
en
tify
g
l
o
b
al
co
n
tex
tu
al
lin
k
ag
es
an
d
lo
n
g
-
r
a
n
g
e
d
ep
en
d
e
n
cies
in
im
a
g
es,
o
u
t
p
er
f
o
r
m
in
g
C
NNs
in
ce
r
tain
task
s
[
6
]
.
Fu
r
th
e
r
m
o
r
e
,
Vi
T
s
o
f
f
er
m
u
lti
-
task
lear
n
in
g
,
w
h
ich
e
n
ab
les
th
e
c
lass
if
icatio
n
o
f
d
is
ea
s
e
ty
p
es,
an
d
t
h
is
f
ea
tu
r
e
ca
n
b
e
n
ef
it
p
r
ec
is
io
n
f
ar
m
in
g
.
T
h
is
r
esear
ch
p
r
o
p
o
s
es
a
n
o
v
el
f
r
am
ewo
r
k
f
o
r
th
e
d
etec
tio
n
an
d
class
if
icatio
n
o
f
r
ice
le
af
d
is
ea
s
es
b
ased
o
n
v
is
io
n
tr
an
s
f
o
r
m
er
s
.
T
h
e
ViT
m
o
d
el
is
co
n
tr
asted
with
co
n
v
en
tio
n
al
C
NN
ar
ch
itectu
r
e
s
,
an
d
GAN
-
b
ased
au
g
m
en
tatio
n
is
in
co
r
p
o
r
ated
to
o
v
er
co
m
e
class
im
b
alan
ce
.
T
h
e
g
o
al
o
f
th
is
co
m
p
ar
a
tiv
e
an
aly
s
is
i
s
to
p
r
o
v
id
e
an
a
u
to
m
ated
r
ice
d
i
s
ea
s
e
m
an
ag
em
en
t
s
o
lu
tio
n
t
h
at
is
m
o
r
e
p
r
ec
is
e,
ef
f
ec
tiv
e
,
an
d
s
ca
lab
le.
T
h
is
r
esear
ch
m
ak
es
th
e
f
o
llo
win
g
co
n
tr
ib
u
tio
n
s
:
i)
A
n
o
v
el
GAN
-
b
ased
(
ViT
f
r
am
ewo
r
k
is
p
r
o
p
o
s
ed
f
o
r
r
ice
leaf
d
is
ea
s
e
d
etec
tio
n
(
R
L
DD)
,
wh
er
e
GAN
-
g
en
er
ated
s
y
n
t
h
etic
im
ag
es
ad
d
r
ess
clas
s
im
b
alan
ce
,
r
ed
u
ce
an
o
m
alies
d
u
r
in
g
o
v
e
r
s
am
p
lin
g
,
an
d
p
r
e
v
en
t
r
ed
u
n
d
a
n
t
d
ata
g
en
er
atio
n
a
n
d
ii)
A
r
e
p
r
o
d
u
cib
le
tr
ai
n
in
g
p
ip
elin
e
is
p
r
esen
ted
,
i
n
clu
d
in
g
e
x
p
licit
ViT
a
n
d
d
ee
p
co
n
v
o
lu
tio
n
al
g
en
e
r
ativ
e
a
d
v
er
s
ar
ial
n
etwo
r
k
(
DC
GAN)
h
y
p
er
p
ar
am
eter
s
,
d
ataset
s
p
lit s
tr
ateg
y
,
an
d
o
p
tim
izatio
n
s
ettin
g
s
.
T
h
e
r
em
ain
d
er
o
f
th
is
p
ap
er
is
o
r
g
an
ized
as
f
o
llo
ws:
s
ec
tio
n
2
r
ev
iews
r
elate
d
wo
r
k
,
s
ec
tio
n
3
d
escr
ib
es
th
e
p
r
o
p
o
s
ed
m
eth
o
d
o
lo
g
y
,
s
ec
tio
n
4
p
r
esen
ts
ex
p
er
im
en
tal
r
esu
lts
an
d
d
is
cu
s
s
io
n
,
an
d
s
ec
tio
n
5
co
n
clu
d
es th
e
p
ap
er
with
f
u
tu
r
e
r
esear
ch
d
ir
ec
tio
n
s
.
2.
L
I
T
E
R
AT
U
RE
R
E
VI
E
W
Fo
r
th
e
p
u
r
p
o
s
e
o
f
id
en
tif
y
in
g
s
im
u
ltan
eo
u
s
o
cc
u
r
r
en
ce
o
f
leaf
d
is
ea
s
es
in
r
ice
an
d
wh
ea
t,
J
ian
g
et
a
l.
[
7
]
s
u
g
g
ested
a
m
u
ltit
ask
d
ee
p
tr
a
n
s
f
er
lear
n
in
g
m
o
d
el
b
u
ilt o
n
a
n
e
n
h
an
ce
d
VGG1
6
ar
ch
itectu
r
e.
Usi
n
g
I
m
ag
eNe
t
-
p
r
etr
ai
n
ed
weig
h
ts
with
f
in
e
-
tu
n
in
g
a
n
d
d
ata
au
g
m
en
tatio
n
,
th
e
m
o
d
el
d
em
o
n
s
tr
ated
an
ac
cu
r
ac
y
o
f
9
7
.
2
2
%
f
o
r
r
ice
an
d
9
8
.
7
5
%
f
o
r
wh
ea
t,
o
u
tp
e
r
f
o
r
m
in
g
s
in
g
le
-
task
m
o
d
els,
R
esNet5
0
,
a
n
d
Den
s
eNe
t1
2
1
.
A
r
ea
l
-
tim
e
p
lan
t
d
is
ea
s
e
d
ataset
was
cr
ea
ted
b
y
J
o
s
ep
h
et
a
l
.
[
8
]
,
wh
o
also
s
u
g
g
ested
a
d
ee
p
lear
n
in
g
(
DL
)
b
ased
d
etec
tio
n
tech
n
i
q
u
e.
T
h
ey
g
ath
er
ed
a
n
d
an
n
o
tated
a
s
izab
le
co
llectio
n
o
f
d
is
ea
s
e
p
h
o
to
s
in
au
th
e
n
tic
f
ield
s
ettin
g
s
with
th
e
aim
o
f
o
v
er
co
m
i
n
g
th
e
d
r
aw
b
ac
k
s
ass
o
ciate
d
with
th
e
cu
r
r
en
t
d
ata
s
ets.
T
h
eir
r
esear
ch
,
wh
ich
u
s
ed
co
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
s
(
C
NNs)
f
o
r
class
if
icatio
n
,
s
h
o
wed
r
esil
ien
ce
in
r
ea
l
-
tim
e
ap
p
licatio
n
s
an
d
ac
co
m
p
lis
h
e
d
ex
ce
llen
t
ac
cu
r
ac
y
in
d
iag
n
o
s
in
g
a
v
ar
iety
o
f
p
lan
t
d
is
ea
s
es.
B
ijo
y
et
a
l.
[
9
]
p
r
esen
ted
a
lig
h
tweig
h
t
d
ee
p
C
NN
f
o
r
r
ice
lea
f
d
is
ea
s
e
d
et
ec
tio
n
(
R
L
DD)
,
ac
h
ie
v
in
g
9
9
.
8
1
%
ac
cu
r
ac
y
with
o
n
ly
0
.
1
8
M
p
ar
a
m
eter
s
.
T
o
f
a
cilitate
p
r
ac
tical
ag
r
icu
ltu
r
al
ap
p
licatio
n
,
th
ey
c
o
n
tin
u
e
d
to
cr
ea
te
a
cr
o
p
h
ea
lth
m
o
n
ito
r
in
g
s
y
s
tem
with
an
o
p
en
ap
p
licatio
n
p
r
o
g
r
am
m
i
n
g
in
ter
f
ac
e
(
API
)
,
web
in
ter
f
ac
e,
an
d
An
d
r
o
id
ap
p
licatio
n
.
R
o
y
an
d
Ku
k
r
ej
a
et
a
l.
[
1
0
]
p
r
esen
ted
a
v
is
i
o
n
tr
a
n
s
f
o
r
m
er
(
ViT
)
-
b
ased
m
o
d
el
f
o
r
r
ice
leaf
d
is
ea
s
e
id
en
tific
atio
n
a
n
d
s
ev
er
ity
esti
m
atio
n
.
Ay
y
ap
p
an
et
a
l.
[
1
1
]
a
p
p
lied
C
NNs
f
o
r
r
ic
e
d
is
ea
s
e
d
etec
tio
n
,
co
m
p
ar
in
g
Den
s
eNe
t1
2
1
,
E
f
f
i
cien
tNetB
4
,
Xce
p
tio
n
,
an
d
M
o
b
ileNetV3
o
n
a
d
ataset
o
f
1
0
,
4
0
7
im
a
g
es
ac
r
o
s
s
1
0
class
es.
T
h
e
s
tu
d
y
o
u
tp
e
r
f
o
r
m
ed
ea
r
lier
m
o
d
els
lik
e
VG
G1
9
an
d
R
esNet1
0
1
b
y
u
s
in
g
a
lar
g
er
d
ataset
an
d
ad
v
an
ce
d
ar
c
h
itectu
r
e
.
S
tu
d
y
[
1
2
]
p
r
esen
ts
an
o
p
tim
ized
lig
h
tweig
h
t
n
eu
r
al
m
o
d
el
f
o
r
R
L
DD,
ad
d
r
ess
in
g
th
e
ch
allen
g
e
o
f
lim
ited
c
o
m
p
u
tatio
n
al
r
eso
u
r
ce
s
i
n
ag
r
icu
ltu
r
al
s
ettin
g
s
.
T
h
e
a
u
th
o
r
s
c
o
m
p
ar
ed
C
NN
-
b
ased
m
o
d
els o
n
a
d
ataset
o
f
r
ice
lea
f
im
ag
es a
n
d
f
o
u
n
d
th
eir
p
r
o
p
o
s
ed
lig
h
tweig
h
t m
o
d
el.
A
DL
-
d
r
iv
e
n
a
p
p
r
o
ac
h
f
o
r
r
ec
o
g
n
izin
g
an
d
ca
teg
o
r
izin
g
f
iv
e
im
p
o
r
tan
t
r
ice
d
is
ea
s
es
was
cr
ea
ted
b
y
Har
id
asan
et
a
l.
[
1
3
]
.
Srin
iv
as
et
a
l.
[
1
4
]
p
r
o
p
o
s
ed
an
i
m
p
r
o
v
e
d
C
NN
f
r
am
ewo
r
k
f
o
r
R
L
DD,
r
eso
lv
in
g
p
r
o
b
lem
s
ass
o
ciate
d
with
u
n
e
v
en
lig
h
tin
g
,
in
tr
icate
b
ac
k
g
r
o
u
n
d
p
atter
n
s
,
an
d
s
im
ilar
ity
in
d
is
ea
s
e
s
y
m
p
to
m
s
.
T
h
e
s
tu
d
y
u
s
ed
an
e
n
h
an
ce
d
C
NN
m
o
d
el
with
o
p
tim
ized
f
ea
tu
r
e
ex
tr
ac
tio
n
a
n
d
class
if
icatio
n
,
ev
alu
ated
o
n
m
u
ltip
le
r
ice
d
is
ea
s
e
d
ataset
s
.
An
ag
r
icu
ltu
r
al
I
n
ter
n
et
o
f
T
h
in
g
s
(
I
o
T
)
s
y
s
tem
u
s
in
g
a
h
y
b
r
id
ized
d
ee
p
lear
n
in
g
tech
n
iq
u
e
f
o
r
r
ice
c
r
o
p
d
is
ea
s
e
d
iag
n
o
s
is
was
p
r
esen
ted
b
y
W
an
g
et
a
l.
[
1
5
]
.
T
h
eir
tech
n
iq
u
e
lev
er
ag
es
I
o
T
-
b
ased
d
ata
g
ath
er
in
g
f
o
r
r
ea
l
-
tim
e
m
o
n
ito
r
in
g
b
y
em
p
lo
y
in
g
C
NNs
to
ca
p
tu
r
e
d
is
cr
im
in
ativ
e
f
ea
tu
r
es
an
d
s
u
p
p
o
r
t
v
ec
to
r
m
ac
h
in
es
(
SVMs)
f
o
r
class
if
icatio
n
.
I
m
ag
es
o
f
r
ice
leav
es
w
er
e
u
s
ed
to
test
t
h
e
s
y
s
tem
,
an
d
it
s
h
o
we
d
ex
c
ellen
t
ac
cu
r
ac
y
in
id
e
n
tify
in
g
an
d
ca
teg
o
r
izin
g
c
o
m
m
o
n
r
ice
d
is
ea
s
es.
A
lig
h
tweig
h
t
R
esNet
-
9
clas
s
if
ier
an
d
an
im
p
r
o
v
ed
v
is
io
n
tr
an
s
f
o
r
m
er
(
I
VT
)
ar
e
u
s
e
d
in
a
h
ier
ar
ch
ical
f
r
am
ewo
r
k
f
o
r
r
ec
o
g
n
izin
g
d
is
ea
s
es
in
th
e
f
o
liag
e
o
f
p
la
n
ts
b
y
Vallab
h
ajo
s
y
u
la
et
a
l.
[
1
6
]
.
T
h
e
m
o
d
el
ef
f
ec
tiv
ely
ca
p
tu
r
es
b
o
th
g
lo
b
al
an
d
l
o
ca
l
f
ea
tu
r
es
with
r
e
d
u
ce
d
co
m
p
u
tatio
n
al
co
s
t
an
d
ac
h
iev
ed
u
p
t
o
9
9
.
7
%
ac
cu
r
ac
y
o
n
Plan
tVillag
e
d
atasets
,
o
u
tp
er
f
o
r
m
i
n
g
I
n
c
ep
tio
n
V3
,
Mo
b
ileNetV2
,
a
n
d
R
esNet5
0
.
A
co
n
v
o
l
u
tio
n
al
n
eu
r
al
n
etwo
r
k
-
b
ased
im
ag
e
class
if
icatio
n
m
o
d
el
co
m
b
in
ed
with
a
ch
at
b
o
t
i
n
ter
f
ac
e
is
u
s
ed
in
th
e
au
to
m
atic
r
ice
d
is
ea
s
e
d
etec
tio
n
s
y
s
tem
d
escr
ib
ed
b
y
T
em
n
ir
an
r
at
et
a
l.
[
1
7
]
.
T
h
e
d
ev
ice
ca
n
id
en
tif
y
co
m
m
o
n
r
ice
illn
ess
es
in
clu
d
in
g
leaf
s
m
u
t
an
d
g
iv
es
f
ar
m
er
s
ea
s
ily
ac
ce
s
s
ib
le
r
ea
l
-
tim
e
f
ee
d
b
ac
k
.
A
twin
C
NN
f
r
am
ewo
r
k
f
o
r
r
ice
leaf
d
is
ea
s
e
id
en
tific
atio
n
was
p
r
o
p
o
s
ed
b
y
Pai
et
a
l.
[
1
8
]
.
I
t
co
m
b
in
es
m
u
lti
-
s
ca
le
f
ea
tu
r
es
f
r
o
m
VGG1
6
,
R
esNet5
0
,
an
d
I
n
ce
p
tio
n
V3
with
p
r
in
cip
al
co
m
p
o
n
en
t
an
aly
s
is
(
PC
A)
f
o
r
d
im
en
s
io
n
ality
r
ed
u
ctio
n
.
R
ad
h
ak
r
is
h
n
a
n
et
a
l.
[
1
9
]
in
t
r
o
d
u
ce
d
an
en
h
an
ce
d
m
a
ch
i
n
e
lear
n
in
g
(
ML
)
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
GA
N
-
a
u
g
men
ted
visi
o
n
tr
a
n
s
f
o
r
mer wi
th
b
a
la
n
ce
d
s
yn
th
etic
d
a
ta
g
en
era
tio
n
…
(
S
a
ifu
l
I
s
la
m
)
1309
f
r
am
ewo
r
k
f
o
r
d
etec
tin
g
r
ice
b
last
d
is
ea
s
e
in
p
ad
d
y
cr
o
p
s
,
lev
er
ag
in
g
th
e
Plan
tVillag
e
d
ataset
co
m
p
r
is
in
g
ap
p
r
o
x
im
ately
a
s
et
o
f
6
0
,
0
0
0
r
ice
leaf
im
ag
es
co
v
er
in
g
b
o
th
h
ea
lth
y
a
n
d
in
f
ec
ted
ca
s
es.
T
h
is
ap
p
r
o
ac
h
ad
d
r
ess
es
lim
itatio
n
s
in
tr
ad
iti
o
n
al
d
iag
n
o
s
tic
m
eth
o
d
s
,
o
f
f
e
r
in
g
p
o
te
n
tial
b
en
ef
its
f
o
r
ea
r
l
y
d
is
ea
s
e
d
etec
tio
n
in
ag
r
icu
ltu
r
e
.
A
lig
h
tweig
h
t
q
u
an
tized
C
NN
ca
lled
R
iceL
ea
f
C
las
s
if
ier
-
v
1
.
0
was
p
r
o
p
o
s
ed
b
y
Ma
r
tin
s
et
a
l.
[
2
0
]
to
id
en
tify
f
iv
e
d
if
f
er
en
t
ty
p
es
o
f
r
ice
leaf
co
n
d
itio
n
s
.
T
h
e
m
o
d
el
was
cr
ea
ted
f
r
o
m
s
cr
atch
u
s
in
g
d
ata
au
g
m
en
tatio
n
an
d
o
p
tim
izatio
n
tech
n
iq
u
es
to
im
p
r
o
v
e
g
e
n
e
r
aliza
tio
n
,
an
d
it
was
tr
ain
ed
u
s
in
g
2
,
8
0
7
p
h
o
to
s
.
T
h
e
p
ap
er
e
m
p
h
asizes
h
o
w
i
t
co
u
ld
b
e
u
s
ed
in
p
r
ec
is
io
n
ag
r
icu
ltu
r
e
f
o
r
lo
w
-
c
o
s
t,
r
ea
l
-
tim
e
r
ice
d
is
ea
s
e
m
o
n
ito
r
in
g
.
R
DR
M
-
YOL
O,
a
lig
h
tweig
h
t
YOL
Ov
5
-
b
ased
m
o
d
el
f
o
r
r
ice
d
is
ea
s
e
d
etec
tio
n
in
ch
allen
g
in
g
f
ield
cir
cu
m
s
tan
ce
s
,
was
cr
ea
ted
b
y
L
i
et
a
l.
[
2
1
]
.
Usi
n
g
Ho
r
-
B
NFA,
SP
D
-
C
o
n
v
,
GsC
o
n
v
,
an
d
weig
h
te
d
in
ter
s
ec
tio
n
o
v
e
r
u
n
io
n
(
W
I
o
U)
lo
s
s
,
it
o
u
t
p
er
f
o
r
m
ed
YOL
Ov
6
–
v
8
an
d
Fas
ter
R
-
C
NN,
ac
h
iev
in
g
9
4
.
3
%
p
r
ec
is
io
n
,
8
9
.
6
%
r
ec
all,
an
d
9
3
.
5
%
m
ea
n
av
e
r
ag
e
p
r
ec
is
io
n
(
m
AP)
o
n
5
,
9
3
0
im
a
g
es.
Usi
n
g
p
h
o
to
s
g
ath
er
e
d
f
r
o
m
s
ev
er
al
p
ar
ts
o
f
I
n
d
ia,
K
u
m
ar
et
a
l.
[
2
2
]
c
r
ea
ted
a
C
N
N
-
b
ased
m
o
d
el
f
o
r
th
e
a
u
to
m
a
tic
d
iag
n
o
s
is
o
f
r
ice
leaf
d
is
ea
s
es.
T
h
r
ee
co
n
v
o
l
u
tio
n
al
f
ea
tu
r
e
e
x
tr
ac
tio
n
lay
e
r
s
alo
n
g
with
two
f
u
lly
co
n
n
ec
te
d
lay
er
s
wer
e
p
ar
t
o
f
th
e
ar
ch
itectu
r
e.
Fo
r
r
ea
l
-
tim
e
f
ield
ap
p
licatio
n
,
th
e
s
tu
d
y
r
ec
o
m
m
en
d
s
in
teg
r
atio
n
with
m
o
b
ile
an
d
I
o
T
d
ev
ices.
W
ith
an
em
p
h
asis
o
n
B
an
g
lad
esh
i
ag
r
icu
ltu
r
e,
C
h
o
wd
h
u
r
y
et
a
l.
[
2
3
]
e
x
am
in
ed
t
h
e
u
s
e
o
f
m
ac
h
in
e
lear
n
in
g
(
ML
)
an
d
d
ee
p
lear
n
in
g
(
DL
)
f
o
r
p
lan
t
leaf
d
is
ea
s
e
id
en
tific
atio
n
.
T
h
ey
d
r
aw
at
ten
tio
n
to
d
is
ea
s
e
r
is
k
s
th
at
af
f
ec
t
f
o
o
d
s
ec
u
r
ity
an
d
g
r
o
s
s
d
o
m
esti
c
p
r
o
d
u
ct
(
GDP)
in
cr
o
p
s
lik
e
r
ice,
to
m
a
to
es,
p
o
tato
es,
an
d
b
ell
p
ep
p
er
s
.
T
h
e
f
o
cu
s
o
f
th
is
s
tu
d
y
is
o
n
ea
r
ly
id
e
n
tific
atio
n
to
r
ed
u
ce
t
h
e
u
s
ag
e
o
f
p
esti
cid
es.
R
ah
m
an
et
a
l
.
[
2
4
]
p
r
o
p
o
s
ed
a
wo
r
k
t
h
at
p
r
e
s
en
ts
a
DL
-
b
ased
r
e
al
-
tim
e
p
la
n
t
d
iag
n
o
s
is
s
y
s
tem
f
o
r
s
u
s
tain
ab
le
ag
r
icu
ltu
r
e.
A
to
tal
o
f
3
0
,
9
4
5
p
h
o
to
s
f
r
o
m
ei
g
h
t
p
lan
t
s
p
ec
ies
an
d
3
5
d
is
ea
s
e
class
e
s
wer
e
co
m
b
in
ed
f
r
o
m
v
ar
io
u
s
s
o
u
r
ce
s
,
in
clu
d
in
g
th
e
Plan
tVillag
e
d
at
aset,
an
d
u
tili
ze
d
in
th
e
s
tu
d
y
.
T
h
e
f
in
d
in
g
s
d
em
o
n
s
tr
ate
th
e
g
r
ea
t
p
o
ten
tial
o
f
d
ee
p
lear
n
in
g
-
b
ased
s
y
s
tem
s
f
o
r
b
etter
cr
o
p
m
a
n
ag
em
e
n
t a
n
d
p
r
ec
is
e,
r
ea
l
-
tim
e
p
lan
t d
is
ea
s
e
d
iag
n
o
s
tics
.
Desp
ite
th
e
r
em
ar
k
a
b
le
p
r
o
g
r
ess
ac
h
iev
ed
th
r
o
u
g
h
v
ar
io
u
s
C
NN,
h
y
b
r
id
,
an
d
tr
an
s
f
o
r
m
er
-
b
ased
ar
ch
itectu
r
es,
th
e
r
ev
iewe
d
s
t
u
d
ies
r
ev
ea
l
s
ev
er
al
p
er
s
is
ten
t
ch
allen
g
es
th
at
ju
s
tify
th
e
d
e
v
elo
p
m
en
t
o
f
a
n
ew
m
o
d
el.
B
ased
o
n
th
e
lim
itatio
n
s
id
en
tifie
d
in
e
x
is
tin
g
C
NN
an
d
tr
an
s
f
o
r
m
er
-
b
ased
ap
p
r
o
ac
h
es,
th
is
s
tu
d
y
in
tr
o
d
u
ce
s
a
GAN
-
au
g
m
en
te
d
v
is
io
n
tr
an
s
f
o
r
m
er
f
r
am
ew
o
r
k
d
esig
n
e
d
to
ac
h
iev
e
b
ala
n
ce
d
,
ac
cu
r
ate,
an
d
ed
g
e
-
co
m
p
atib
le
r
ice
leaf
d
is
ea
s
e
d
etec
tio
n
.
3.
M
E
T
H
O
D
Fig
u
r
e
1
p
r
esen
ts
th
e
o
v
er
all
f
lo
wch
ar
t
o
f
th
e
p
r
o
p
o
s
ed
m
eth
o
d
o
lo
g
y
f
o
r
R
L
DD.
T
h
e
f
r
am
ewo
r
k
in
teg
r
ates
d
ata
c
o
llectio
n
,
p
r
ep
r
o
ce
s
s
in
g
,
GAN
-
b
ased
au
g
m
en
tatio
n
,
m
o
d
el
tr
ain
in
g
,
an
d
p
er
f
o
r
m
a
n
ce
ev
alu
atio
n
.
A
d
ataset
o
f
r
ice
leaf
im
ag
es
was
co
m
p
iled
,
an
n
o
tated
,
an
d
p
r
e
p
r
o
ce
s
s
ed
th
r
o
u
g
h
r
esizin
g
,
n
o
r
m
aliza
tio
n
,
an
d
tr
a
d
itio
n
al
au
g
m
e
n
tatio
n
tech
n
iq
u
es
to
en
s
u
r
e
co
n
s
is
ten
cy
.
T
o
ad
d
r
e
s
s
class
im
b
alan
ce
,
s
y
n
th
etic
s
am
p
les
wer
e
g
en
e
r
ated
u
s
in
g
a
GAN.
T
h
e
ViT
m
o
d
el
was
th
en
tr
ain
ed
an
d
c
o
m
p
ar
ed
with
s
ev
er
al
C
NN
-
b
ased
ar
ch
itectu
r
es u
n
d
e
r
id
en
tical
co
n
d
itio
n
s
.
Fig
u
r
e
1
.
E
n
d
-
to
-
en
d
p
ip
elin
e
o
f
th
e
p
r
o
p
o
s
ed
GAN
-
au
g
m
e
n
ted
v
is
io
n
tr
an
s
f
o
r
m
er
f
r
am
ew
o
r
k
,
in
cl
u
d
in
g
d
ata
ac
q
u
is
itio
n
,
p
r
e
p
r
o
ce
s
s
in
g
,
G
AN
-
b
ased
s
y
n
th
etic
im
ag
e
g
e
n
er
atio
n
,
d
ataset
b
alan
cin
g
,
Vi
T
tr
ain
in
g
,
a
n
d
p
er
f
o
r
m
an
ce
ev
al
u
atio
n
3
.
1
.
Da
t
a
s
et
a
nd
a
nn
o
t
a
t
io
n
A
to
tal
o
f
1
3
,
8
7
6
r
ice
lea
f
i
m
ag
es
wer
e
co
llected
f
r
o
m
Kag
g
le
[
2
5
]
.
E
ac
h
im
a
g
e
wa
s
ca
r
ef
u
lly
an
n
o
tated
with
ex
p
er
t
co
n
s
u
lt
atio
n
to
id
en
tify
o
n
e
o
f
ten
m
ajo
r
r
ice
leaf
co
n
d
itio
n
s
an
d
h
ea
lth
y
leav
es.
T
o
ad
d
r
ess
class
im
b
alan
ce
,
a
d
ee
p
co
n
v
o
lu
tio
n
al
g
en
e
r
ativ
e
a
d
v
er
s
ar
ial
n
etwo
r
k
(
DC
GAN)
was
em
p
lo
y
ed
to
g
en
er
ate
s
y
n
th
etic
im
ag
es
f
o
r
u
n
d
er
r
ep
r
esen
ted
d
is
ea
s
e
ca
teg
o
r
ies.
T
h
e
GAN
p
r
o
d
u
c
ed
3
,
6
7
6
s
y
n
th
etic
s
am
p
les
f
o
r
u
n
d
e
r
r
ep
r
esen
te
d
d
is
ea
s
e
ca
teg
o
r
ies.
T
h
is
au
g
m
en
tatio
n
r
esu
lted
in
a
f
in
al
d
ataset
o
f
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
1
6
,
No
.
3
,
J
u
n
e
20
2
6
:
1
3
0
7
-
1
3
1
8
1310
1
7
,
5
5
2
im
ag
es
(
1
3
,
8
7
6
r
ea
l
a
n
d
3
,
6
7
6
s
y
n
th
etic)
,
t
h
er
eb
y
im
p
r
o
v
in
g
class
d
is
tr
ib
u
tio
n
an
d
d
iv
e
r
s
ity
.
T
h
e
q
u
ality
o
f
g
en
er
ated
im
ag
es
was
v
er
if
ied
t
h
r
o
u
g
h
v
is
u
al
i
n
s
p
ec
tio
n
b
y
a
g
r
icu
ltu
r
al
ex
p
er
ts
an
d
a
Fré
ch
et
in
ce
p
tio
n
d
is
tan
ce
(
FID
)
s
co
r
e
o
f
1
8
.
6
,
in
d
icatin
g
s
tr
o
n
g
s
i
m
ilar
ity
to
r
ea
l
s
am
p
les.
T
h
e
d
ataset
was
d
iv
id
ed
in
to
7
0
%
f
o
r
tr
ai
n
in
g
,
1
5
%
f
o
r
v
alid
atio
n
,
an
d
1
5
%
f
o
r
te
s
tin
g
,
m
ak
in
g
s
u
r
e
th
at
all
ill
n
ess
ty
p
es
ar
e
f
air
ly
r
ep
r
esen
ted
f
o
r
f
air
m
o
d
el
e
v
a
lu
atio
n
.
As
s
h
o
wn
in
T
ab
le
1
,
th
e
o
r
ig
in
al
tr
ain
in
g
d
ataset
ex
h
ib
its
s
u
b
s
tan
tial
class
im
b
alan
ce
,
with
s
ev
er
al
d
is
ea
s
e
ca
teg
o
r
ies
h
av
in
g
s
ig
n
if
ican
tly
f
ewe
r
s
am
p
les.
T
o
m
itig
ate
th
is
is
s
u
e,
GAN
-
b
ased
au
g
m
en
tatio
n
was
ap
p
lied
s
elec
tiv
ely
t
o
u
n
d
er
r
ep
r
esen
ted
class
es.
A
to
tal
o
f
3
,
6
7
6
s
y
n
th
etic
im
a
g
es
wer
e
g
en
er
ated
i
n
p
r
o
p
o
r
tio
n
to
th
e
lev
el
o
f
c
lass
im
b
alan
ce
,
r
esu
ltin
g
in
im
p
r
o
v
ed
class
r
ep
r
esen
tatio
n
wh
ile
av
o
id
in
g
ex
ce
s
s
iv
e
o
v
er
-
s
am
p
lin
g
o
f
m
ajo
r
ity
class
es.
T
ab
le
1
.
C
lass
-
wis
e
d
is
tr
ib
u
tio
n
o
f
p
ad
d
y
leaf
im
ag
es b
e
f
o
r
e
an
d
af
ter
GAN
-
b
ased
a
u
g
m
e
n
tatio
n
C
l
a
s
s
n
a
me
O
r
i
g
i
n
a
l
s
a
mp
l
e
s
A
f
t
e
r
G
A
N
a
u
g
m
e
n
t
a
t
i
o
n
N
o
r
mal
1
7
6
4
1
7
6
4
B
l
a
s
t
1
7
3
8
1
7
5
1
H
i
sp
a
1
5
9
4
1
6
8
0
D
e
a
d
h
e
a
r
t
1
4
4
2
1
6
0
6
Tu
n
g
r
o
1
0
8
8
1
4
3
2
B
r
o
w
n
s
p
o
t
9
6
5
1
3
7
1
D
o
w
n
y
m
i
l
d
e
w
6
2
0
1
2
0
1
B
a
c
t
e
r
i
a
l
l
e
a
f
b
l
i
g
h
t
4
7
9
1
1
3
2
B
a
c
t
e
r
i
a
l
l
e
a
f
s
t
r
e
a
k
3
8
0
1
0
8
3
B
a
c
t
e
r
i
a
l
p
a
n
i
c
l
e
b
l
i
g
h
t
3
3
7
1
0
6
3
To
t
a
l
1
0
,
4
0
7
1
4
,
0
8
3
3
.
2
.
P
re
pro
ce
s
s
ing
E
v
er
y
i
m
ag
e
was
s
u
b
jecte
d
t
o
a
n
u
m
b
e
r
o
f
p
r
ep
r
o
ce
s
s
in
g
p
r
o
ce
d
u
r
es
b
ef
o
r
e
tr
ain
in
g
i
n
o
r
d
er
t
o
g
u
ar
an
tee
u
n
i
f
o
r
m
i
t
y
an
d
en
h
an
ce
m
o
d
el
g
en
e
r
aliza
tio
n
.
T
o
s
tan
d
ar
d
ize
p
ix
el
in
ten
s
ity
v
a
lu
es,
ea
ch
r
ice
lea
f
im
ag
e
was r
esized
to
a
f
ix
e
d
s
p
atial
r
eso
lu
tio
n
o
f
2
2
4
×2
2
4
p
ix
els an
d
n
o
r
m
alize
d
u
s
in
g
:
=
−
µ
(
1
)
wh
er
e
d
en
o
tes
th
e
in
p
u
t
i
m
ag
e,
µ
is
th
e
m
ea
n
,
an
d
σ
is
th
e
s
tan
d
ar
d
d
ev
iatio
n
o
f
th
e
d
ataset.
T
h
is
n
o
r
m
aliza
tio
n
en
s
u
r
ed
th
at
al
l
im
ag
es
h
ad
ze
r
o
-
ce
n
ter
e
d
p
ix
el
d
is
tr
ib
u
tio
n
s
,
im
p
r
o
v
in
g
n
u
m
er
ical
s
tab
ilit
y
d
u
r
in
g
tr
ain
i
n
g
.
T
o
p
r
e
v
en
t
o
v
e
r
f
itti
n
g
a
n
d
in
cr
ea
s
e
d
ataset
d
iv
er
s
ity
,
tr
a
d
itio
n
al
d
ata
au
g
m
en
tatio
n
tech
n
iq
u
es
wer
e
ap
p
lied
.
E
ac
h
im
ag
e
u
n
d
e
r
wen
t
a
s
et
o
f
r
an
d
o
m
tr
an
s
f
o
r
m
atio
n
s
s
u
ch
as
r
o
tatio
n
,
h
o
r
iz
o
n
tal
an
d
v
er
tical
f
lip
p
in
g
,
an
d
s
ca
li
n
g
:
′
=
(
)
(
)
(
2
)
′′
=
ℎ
(
′
)
(
′
)
(
ℎ
)
(
3
)
=
(
′′
)
(
)
(
4
)
w
h
er
e
,
ℎ
,
,
an
d
α
r
ep
r
esen
t
th
e
r
o
tatio
n
,
f
lip
p
in
g
,
an
d
s
ca
lin
g
o
p
er
ato
r
s
r
esp
ec
tiv
ely
.
T
h
is
p
r
ep
r
o
ce
s
s
in
g
p
ip
elin
e
n
o
t
o
n
ly
e
n
h
an
ce
d
t
h
e
r
o
b
u
s
tn
ess
o
f
th
e
ViT
m
o
d
el
b
u
t
also
en
s
u
r
e
d
r
eliab
le
p
er
f
o
r
m
an
ce
ac
r
o
s
s
d
iv
er
s
e
d
is
ea
s
e
class
es b
y
cr
ea
tin
g
v
ar
ied
a
n
d
r
ea
lis
tic
in
p
u
t
co
n
d
itio
n
s
f
o
r
m
o
d
el
tr
ain
in
g
.
3
.
3
.
T
ra
ini
ng
f
ra
m
ewo
r
k
a
n
d t
o
o
ls
T
h
e
v
is
io
n
tr
an
s
f
o
r
m
er
was
s
elec
ted
d
u
e
to
its
ab
ilit
y
to
c
ap
tu
r
e
lo
n
g
-
r
a
n
g
e
s
p
atial
d
ep
en
d
en
cies
th
r
o
u
g
h
s
elf
-
atten
tio
n
,
wh
ich
is
cr
itical
f
o
r
d
is
tin
g
u
is
h
in
g
v
is
u
ally
s
im
ilar
r
ice
leaf
d
is
e
ases
.
D
C
GAN
wa
s
ch
o
s
en
o
v
e
r
m
o
r
e
co
m
p
lex
g
en
er
ativ
e
m
o
d
els
s
u
ch
as
Sty
leGAN
o
r
d
if
f
u
s
io
n
m
o
d
els
b
e
ca
u
s
e
it
p
r
o
v
id
es
a
f
av
o
r
a
b
le
b
alan
ce
b
etwe
en
im
ag
e
q
u
ality
,
tr
ain
in
g
s
tab
ilit
y
,
an
d
co
m
p
u
tatio
n
al
ef
f
icie
n
cy
f
o
r
ag
r
icu
ltu
r
al
d
atasets
.
I
m
ag
es
wer
e
r
esized
to
2
2
4
×2
2
4
to
m
atch
s
tan
d
ar
d
ViT
in
p
u
t
r
eso
lu
tio
n
wh
ile
p
r
eser
v
in
g
d
is
ea
s
e
-
r
elate
d
tex
tu
r
e
d
etails.
T
h
e
Ad
am
o
p
tim
izer
was
ad
o
p
ted
f
o
r
b
o
th
ViT
an
d
DC
GAN
tr
ai
n
in
g
d
u
e
to
its
f
ast
co
n
v
er
g
en
ce
an
d
r
o
b
u
s
tn
ess
in
d
ee
p
v
is
io
n
m
o
d
els.
I
n
o
r
d
er
to
in
v
esti
g
ate
th
e
ef
f
icac
y
o
f
th
e
s
u
g
g
ested
V
iT
m
o
d
el
f
o
r
R
L
DD,
an
ex
p
er
im
en
tal
f
r
am
ewo
r
k
was
cr
ea
ted
.
B
ec
au
s
e
o
f
its
ca
p
ac
ity
to
ca
p
tu
r
e
l
o
n
g
-
r
a
n
g
e
s
p
atial
d
e
p
en
d
e
n
cies,
ViT
was
ch
o
s
en
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
GA
N
-
a
u
g
men
ted
visi
o
n
tr
a
n
s
f
o
r
mer wi
th
b
a
la
n
ce
d
s
yn
th
etic
d
a
ta
g
en
era
tio
n
…
(
S
a
ifu
l
I
s
la
m
)
1311
·
as
th
e
m
ain
ar
ch
itectu
r
e.
A
n
u
m
b
er
o
f
b
aselin
e
C
NN
m
o
d
els,
in
clu
d
in
g
VGG1
6
,
R
esNet5
0
,
Mo
b
ileNetV2
,
Den
s
eNe
t1
2
1
,
Den
s
eNe
t1
6
9
,
Xce
p
tio
n
,
an
d
R
esNet5
0
V2
,
w
er
e
em
p
lo
y
ed
f
o
r
c
o
m
p
ar
ativ
e
s
tu
d
y
.
GANs
wer
e
u
s
ed
to
g
en
er
ate
s
y
n
t
h
etic
im
ag
e
f
o
r
u
n
d
er
r
ep
r
esen
ted
clas
s
es
illn
es
s
clas
s
es
f
o
r
m
itig
ati
n
g
class
d
is
tr
ib
u
tio
n
d
is
p
ar
ities
an
d
im
p
r
o
v
e
g
en
er
aliza
tio
n
,
g
u
ar
a
n
teein
g
a
m
o
r
e
r
o
b
u
s
t
an
d
b
alan
ce
d
tr
ai
n
in
g
d
ataset.
T
h
e
ViT
m
o
d
el
d
iv
id
es
ea
ch
im
ag
e
x
∈
R
H×
W×
C
in
to
N
p
atch
es
o
f
s
i
ze
P
×P
,
wh
er
e
=
2
.
E
ac
h
f
latten
ed
p
atch
is
em
b
ed
d
e
d
as:
0
=
(
)
+
,
=
1
,
2
,
…
,
(
5
)
wh
er
e
(
)
is
a
lin
ea
r
p
r
o
jectio
n
a
n
d
E
pos
r
ep
r
esen
ts
p
o
s
itio
n
al
e
n
co
d
in
g
.
W
ith
in
ea
ch
tr
an
s
f
o
r
m
er
en
c
o
d
er
,
th
e
s
elf
-
atten
tio
n
m
ec
h
a
n
is
m
co
m
p
u
tes:
(
,
,
)
=
(
√
)
(
6
)
wh
er
e
Q
=
z
l
W
Q
,
K
=
z
l
W
K
,
a
nd
V
=
z
l
W
V
ar
e
the
q
u
er
y,
k
e
y,
a
n
d
v
al
u
e
pr
o
je
ct
io
n
s
.
T
h
e
e
n
c
od
er
ou
t
pu
t
i
s
up
d
at
e
d
a
s
:
+
1
=
(
(
+
(
)
)
)
(
7
)
Af
ter
s
ev
er
al
lay
er
s
,
th
e
class
t
o
k
en
r
ep
r
esen
tatio
n
is
p
ass
ed
to
a
class
if
icatio
n
h
e
a
d
:
̂
=
(
+
)
(
8
)
T
h
e
tr
ain
in
g
of
t
h
e
m
o
d
el
u
tili
ze
d
ca
teg
o
r
ical
cr
o
s
s
-
en
tr
o
p
y
l
o
s
s
:
=
−
1
∑
∑
,
(
̂
,
)
=
1
=
1
(
9
)
wh
er
e
,
an
d
̂
,
d
en
o
te
th
e
tr
u
e
an
d
p
r
ed
icted
p
r
o
b
a
b
ilit
ies
f
o
r
class
c
of
s
am
p
le
i
,
r
esp
ec
ti
v
ely
.
T
h
e
Ad
am
o
p
tim
izer
was
u
s
ed
to
tr
ain
each
m
o
d
el
in
th
e
s
am
e
way
,
with
a
b
atch
s
ize
of
32,
a
lear
n
in
g
r
ate
of
0
.
0
0
0
1
,
an
d
5
0
ep
o
c
h
s
.
Ac
cu
r
ac
y
,
p
r
ec
is
io
n
,
r
ec
all,
a
n
d
F1
-
s
co
r
e
wer
e
u
s
ed
to
as
s
ess
th
e
m
o
d
el’
s
p
er
f
o
r
m
an
ce
;
c
o
n
f
u
s
io
n
m
atr
i
x
an
aly
s
is
was
u
s
ed
to
p
r
o
v
id
e
ad
d
itio
n
al
u
n
d
er
s
tan
d
in
g
of
p
er
f
o
r
m
an
ce
p
e
r
class
.
A
Deep
C
o
n
v
o
lu
tio
n
al
GAN
(
DC
GA
N)
ar
ch
itectu
r
e
was
em
p
lo
y
ed
to
g
en
er
ate
s
y
n
th
etic
r
ice
lea
f
im
ag
es
f
o
r
u
n
d
er
r
e
p
r
esen
ted
d
is
ea
s
e
class
es.
T
h
e
GAN
f
r
am
ewo
r
k
co
n
s
is
ts
o
f
a
g
en
er
ato
r
G
a
n
d
a
d
is
cr
im
in
ato
r
D
,
tr
ain
e
d
with
t
h
e
em
p
lo
y
in
g
a
m
in
im
ax
o
p
ti
m
izatio
n
cr
iter
io
n
:
(
,
)
=
∼
(
)
(
)
[
l
o
g
(
)
]
+
∼
(
)
[
l
o
g
(
1
−
(
(
)
)
)
]
(
1
0
)
wh
er
e
∼
(
)
is
th
e
in
p
u
t
n
o
is
e
v
ec
to
r
.
T
h
e
g
en
e
r
ato
r
a
n
d
d
is
cr
im
in
ato
r
lo
s
s
f
u
n
ctio
n
s
a
r
e
:
=
−
∼
(
)
[
lo
g
(
(
)
)
]
(
1
1
)
=
−
∼
(
)
(
)
[
l
o
g
(
)
]
−
∼
(
)
[
l
o
g
(
1
−
(
(
)
)
)
]
(
1
2
)
T
r
ain
in
g
o
f
t
h
e
GAN
s
p
an
n
ed
2
0
0
ep
o
ch
s
,
em
p
lo
y
in
g
th
e
Ad
am
o
p
tim
izer
at
a
lear
n
in
g
r
ate
o
f
0
.
0
0
0
2
,
with
b
atc
h
es
of
64
s
am
p
les
an
d
b
in
ar
y
cr
o
s
s
-
en
tr
o
p
y
lo
s
s
.
B
o
th
th
e
g
en
er
ato
r
a
n
d
d
is
cr
im
in
ato
r
wer
e
tr
ain
ed
s
im
u
ltan
eo
u
s
ly
on
2
2
4
×2
2
4
im
ag
es.
T
o
v
er
if
y
t
h
e
ef
f
ec
t
of
a
u
g
m
e
n
tatio
n
,
we
c
o
n
d
u
cted
a
p
er
-
class
ac
cu
r
ac
y
an
aly
s
i
s
s
h
o
win
g
an
av
e
r
ag
e
4
.
2
%
im
p
r
o
v
e
m
en
t
in
m
in
o
r
ity
d
is
ea
s
e
ca
teg
o
r
i
es,
co
n
f
i
r
m
in
g
th
at
GAN
-
b
ased
au
g
m
en
tatio
n
ef
f
ec
tiv
ely
en
h
a
n
ce
d
d
ata
d
iv
er
s
ity
an
d
class
b
alan
ce
.
T
h
e
d
et
ailed
co
n
f
i
g
u
r
atio
n
an
d
h
y
p
er
p
a
r
am
eter
s
o
f
th
e
Vi
s
io
n
T
r
an
s
f
o
r
m
e
r
u
s
ed
in
th
is
s
tu
d
y
ar
e
p
r
esen
ted
in
T
a
b
le
2.
T
h
e
DC
GAN
ar
ch
itectu
r
e
co
n
s
is
ts
o
f
a
co
n
v
o
lu
tio
n
al
g
en
er
ato
r
an
d
d
is
cr
im
in
at
o
r
.
T
h
e
g
en
er
ato
r
co
m
p
r
is
es
f
o
u
r
tr
a
n
s
p
o
s
ed
c
o
n
v
o
lu
tio
n
al
lay
er
s
with
b
atch
n
o
r
m
aliza
tio
n
an
d
R
eL
U
ac
tiv
atio
n
s
,
f
o
llo
wed
b
y
a
T
an
h
o
u
t
p
u
t
lay
er
.
T
h
e
d
is
cr
im
in
ato
r
co
n
s
is
ts
o
f
f
o
u
r
c
o
n
v
o
lu
tio
n
al
lay
er
s
with
L
ea
k
y
R
e
L
U
ac
tiv
atio
n
s
an
d
a
Sig
m
o
id
o
u
tp
u
t
lay
er
.
B
o
t
h
n
etwo
r
k
s
wer
e
tr
ain
ed
u
s
in
g
th
e
Ad
am
o
p
tim
izer
with
lear
n
in
g
r
ate
o
f
0
.
0
0
0
2
,
β
1
=
0
.
5
,
an
d
β
2
=
0
.
9
9
9
.
T
h
e
g
en
e
r
ato
r
an
d
d
is
cr
im
i
n
ato
r
wer
e
u
p
d
ated
alter
n
ately
with
a
1
:1
u
p
d
ate
r
atio
to
en
s
u
r
e
s
tab
le
a
d
v
er
s
ar
i
al
tr
ain
in
g
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
1
6
,
No
.
3
,
J
u
n
e
20
2
6
:
1
3
0
7
-
1
3
1
8
1312
Σ
T
ab
le
2
.
ViT
co
n
f
ig
u
r
atio
n
u
s
ed
in
th
e
p
r
o
p
o
s
ed
f
r
a
m
ewo
r
k
P
a
r
a
me
t
e
r
V
a
l
u
e
I
n
p
u
t
i
m
a
g
e
s
i
z
e
2
2
4
×
2
2
4
P
a
t
c
h
s
i
z
e
1
6
×
1
6
N
u
mb
e
r
o
f
p
a
t
c
h
e
s
1
9
6
Emb
e
d
d
i
n
g
d
i
m
e
n
s
i
o
n
7
6
8
N
u
mb
e
r
o
f
e
n
c
o
d
e
r
l
a
y
e
r
s
12
N
u
mb
e
r
o
f
a
t
t
e
n
t
i
o
n
h
e
a
d
s
12
M
LP
h
i
d
d
e
n
d
i
m
e
n
si
o
n
3
0
7
2
D
r
o
p
o
u
t
r
a
t
e
0
.
1
O
p
t
i
mi
z
e
r
A
d
a
m
Le
a
r
n
i
n
g
r
a
t
e
0
.
0
0
0
1
B
a
t
c
h
si
z
e
32
N
u
mb
e
r
o
f
e
p
o
c
h
s
50
Lo
ss f
u
n
c
t
i
o
n
C
a
t
e
g
o
r
i
c
a
l
c
r
o
ss
-
e
n
t
r
o
p
y
3
.
4
.
F
e
a
t
ure
s
elec
t
io
n
Featu
r
e
s
elec
tio
n
in
th
is
s
tu
d
y
was
p
er
f
o
r
m
ed
au
to
m
atica
lly
b
y
th
e
d
ee
p
lear
n
in
g
m
o
d
els
r
ath
er
th
an
th
r
o
u
g
h
m
a
n
u
al
ex
t
r
ac
tio
n
.
Vi
T
ca
p
tu
r
e
d
b
o
th
lo
ca
l
d
etails
an
d
g
lo
b
al
co
n
tex
tu
al
r
elatio
n
s
h
ip
s
u
s
in
g
its
s
elf
-
atten
tio
n
m
ec
h
a
n
is
m
,
wh
ile
C
NN
b
aselin
es
p
r
im
ar
ily
f
o
c
u
s
ed
o
n
l
o
ca
lized
s
p
atial
f
ea
tu
r
es.
I
n
th
e
ViT
ar
ch
itectu
r
e,
f
ea
t
u
r
e
ex
tr
ac
tio
n
is
g
o
v
er
n
e
d
b
y
th
e
atten
tio
n
-
weig
h
ted
ag
g
r
eg
atio
n
o
f
p
atch
em
b
ed
d
i
n
g
s
:
=
∑
=
1
(
1
3
)
w
h
er
e
F
i
d
e
no
t
e
s
th
e
i
-
th
ou
t
pu
t
f
e
a
tu
r
e
ve
c
to
r
,
V
j
is
the
v
al
u
e
e
m
b
e
dd
in
g
of
p
at
c
h
j
,
an
d
α
ij
r
e
pr
e
s
e
nt
s
t
h
e
at
t
e
nt
io
n
w
ei
g
ht
c
o
m
p
ut
e
d
a
s
:
=
√
∑
ex
p
(
)
=
1
(
1
4)
Her
e,
Q
i
an
d
K
j
ar
e
th
e
q
u
er
y
an
d
k
ey
v
ec
to
r
s
c
o
r
r
esp
o
n
d
in
g
to
p
atch
es
i
an
d
j
,
r
esp
ec
tiv
ely
,
an
d
d
k
is
th
e
d
im
en
s
io
n
ality
o
f
th
e
k
e
y
s
p
ac
e.
T
h
is
f
o
r
m
u
latio
n
allo
ws
th
e
m
o
d
el
t
o
ca
p
tu
r
e
b
o
th
s
h
o
r
t
an
d
lo
n
g
-
r
a
n
g
e
d
ep
en
d
e
n
cies
am
o
n
g
d
if
f
er
e
n
t
r
eg
io
n
s
of
th
e
r
ice
leaf
im
ag
e
.
To
en
h
a
n
ce
f
ea
tu
r
e
d
iv
er
s
ity
,
GANs
wer
e
u
s
ed
to
p
r
o
d
u
ce
s
y
n
th
etic
s
am
p
les
f
o
r
u
n
d
er
r
ep
r
esen
ted
d
is
ea
s
e
ca
teg
o
r
ies.
T
h
is
au
g
m
en
tatio
n
e
x
p
an
d
e
d
th
e
o
v
er
all
f
ea
tu
r
e
s
p
ac
e:
=
∪
ℎ
(
1
5
)
wh
er
e
an
d
ℎ
d
en
o
te
th
e
f
ea
tu
r
e
s
ets
ex
tr
ac
ted
f
r
o
m
r
ea
l
an
d
GAN
-
g
en
er
ated
im
ag
e
s
,
r
esp
ec
tiv
ely
.
T
h
is
co
m
b
in
ati
o
n
en
ab
led
th
e
m
o
d
els
to
lear
n
more
d
is
cr
im
in
ativ
e
an
d
g
e
n
er
aliza
b
le
r
ep
r
esen
tatio
n
s
.
T
h
r
o
u
g
h
th
is
au
to
m
atic
p
r
o
ce
s
s
,
th
e
m
o
s
t
r
elev
an
t
an
d
h
ig
h
-
lev
el
s
em
an
tic
f
ea
tu
r
es
wer
e
r
etain
ed
,
en
s
u
r
in
g
r
o
b
u
s
t
d
etec
tio
n
o
f
r
ice
leaf
d
is
ea
s
es
d
esp
ite
v
ar
iatio
n
s
in
a
p
p
ea
r
a
n
ce
an
d
en
v
ir
o
n
m
en
tal
f
ac
to
r
s
.
3
.
5
.
Dis
ea
s
e
predict
io
n a
nd
perf
o
rm
a
nce
ev
a
lua
t
io
n
On
ce
tr
ain
in
g
was
co
m
p
leted
,
th
e
o
p
tim
ized
ViT
ar
ch
itectu
r
e
was
em
p
lo
y
ed
f
o
r
i
n
f
er
en
ce
o
n
u
n
s
ee
n
test
im
ag
es
to
ev
alu
ate
its
p
r
ac
tical
ap
p
licab
ilit
y
in
r
ea
l
-
wo
r
ld
s
ce
n
ar
io
s
.
E
ac
h
i
n
p
u
t
i
m
ag
e
was
ca
r
ef
u
lly
p
r
o
ce
s
s
ed
th
r
o
u
g
h
th
e
m
o
d
e
l
to
p
r
ed
ict
th
e
d
is
ea
s
e
ty
p
e.
T
h
e
p
r
ed
icted
o
u
tp
u
ts
̂
wer
e
s
y
s
tem
atica
lly
co
m
p
ar
ed
with
th
e
ass
o
ciate
d
tr
u
e
lab
els
y
f
o
r
t
h
e
p
u
r
p
o
s
e
of
m
ea
s
u
r
in
g
ac
cu
r
ac
y
a
n
d
r
eliab
ilit
y
.
T
h
e
p
er
f
o
r
m
an
ce
of
th
e
m
o
d
el
was
ass
e
s
s
ed
b
ased
on
co
m
m
o
n
ly
u
s
ed
ev
alu
atio
n
c
r
iter
ia,
s
u
ch
as
ac
cu
r
ac
y
,
p
r
ec
is
io
n
,
r
ec
all,
F1
-
s
c
o
r
e
,
an
d
an
an
aly
s
is
o
f
th
e
co
n
f
u
s
io
n
m
atr
ix
,
to
p
r
o
v
id
e
b
o
t
h
o
v
er
all
an
d
class
-
wis
e
in
s
ig
h
ts
.
T
h
ese
m
etr
ics ar
e
f
o
r
m
ally
d
e
f
in
ed
as
(
1
6
)
-
(
1
9
)
:
=
+
+
+
+
(
1
6
)
=
+
(
1
7
)
=
+
(
1
8
)
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
GA
N
-
a
u
g
men
ted
visi
o
n
tr
a
n
s
f
o
r
mer wi
th
b
a
la
n
ce
d
s
yn
th
etic
d
a
ta
g
en
era
tio
n
…
(
S
a
ifu
l
I
s
la
m
)
1313
1
=
2
×
×
+
(
1
9
)
wh
er
e
TP
,
TN
,
FP
,
an
d
FN
d
en
o
te
th
e
n
u
m
b
er
of
tr
u
e
p
o
s
itiv
es,
tr
u
e
n
eg
ativ
es,
f
alse
p
o
s
itiv
es,
an
d
f
alse
n
eg
ativ
es,
r
esp
ec
tiv
ely
.
T
h
ese
q
u
a
n
titativ
e
m
ea
s
u
r
es
en
s
u
r
e
a
b
alan
ce
d
ev
alu
atio
n
o
f
b
o
th
p
o
s
itiv
e
an
d
n
eg
ativ
e
class
if
icatio
n
s
ac
r
o
s
s
d
is
ea
s
e
ca
teg
o
r
ies.
T
o
an
al
y
ze
th
e
tr
ain
in
g
d
y
n
am
ics,
m
o
d
el
co
n
v
er
g
e
n
ce
was m
o
n
ito
r
ed
b
y
tr
ac
k
i
n
g
th
e
class
if
icatio
n
lo
s
s
an
d
ac
cu
r
a
cy
o
v
er
ep
o
c
h
s
t
:
(
+
1
)
<
(
)
,
∀
∈
[
1
,
]
(
2
0
)
(
+
1
)
>
(
)
,
.
(
2
1
)
A
s
m
o
o
th
an
d
m
o
n
o
to
n
ic
d
ec
r
ea
s
e
in
C
LS
wi
th
a
co
r
r
esp
o
n
d
in
g
in
cr
ea
s
e
in
A
cc
co
n
f
ir
m
ed
s
tab
le
lear
n
in
g
an
d
ef
f
ec
tiv
e
p
a
r
am
eter
o
p
tim
izatio
n
o
f
th
e
ViT
m
o
d
el.
C
o
m
p
ar
ed
to
c
o
n
v
e
n
tio
n
al
C
NN
-
b
ased
ar
ch
itectu
r
es,
th
e
ViT
m
o
d
el
ex
h
ib
ited
s
u
p
er
io
r
p
e
r
f
o
r
m
an
c
e
in
b
o
th
ac
cu
r
ac
y
an
d
g
en
e
r
a
lizatio
n
,
ac
h
iev
i
n
g
an
o
v
er
all
ac
c
u
r
ac
y
o
f
9
6
.
5
%
as
p
r
esen
ted
in
T
ab
le
3.
Mo
r
eo
v
er
,
t
h
e
in
teg
r
atio
n
o
f
GA
N
-
au
g
m
en
ted
d
ata
m
itig
ated
s
k
ewe
d
class
f
r
e
q
u
e
n
c
i
e
s
a
n
d
e
n
h
a
n
c
e
d
t
h
e
c
a
p
a
b
i
l
i
t
y
of
t
h
e
m
o
d
e
l
to
ac
c
u
r
a
t
e
l
y
c
l
as
s
i
f
y
u
n
d
e
r
r
e
p
r
e
s
e
n
t
e
d
d
i
s
e
as
e
c
a
t
e
g
o
r
i
e
s
.
T
ab
le
3
.
Per
f
o
r
m
an
ce
co
m
p
a
r
is
o
n
o
f
d
i
f
f
er
en
t
d
ee
p
lear
n
in
g
m
o
d
els
M
o
d
e
l
A
c
c
u
r
a
c
y
(
%)
P
r
e
c
i
s
i
o
n
(
%)
R
e
c
a
l
l
(
%)
F
1
S
c
o
r
e
(
%)
V
G
G
1
6
9
1
.
5
9
0
.
8
9
0
.
2
9
0
.
5
R
e
sN
e
t
5
0
9
2
.
7
9
2
.
0
9
1
.
6
9
1
.
8
M
o
b
i
l
e
N
e
t
V
2
9
3
.
2
9
2
.
5
9
2
.
1
9
2
.
3
D
e
n
seN
e
t
1
2
1
9
2
.
0
9
1
.
4
9
1
.
0
9
1
.
2
D
e
n
seN
e
t
1
6
9
9
2
.
3
9
1
.
7
9
1
.
2
9
1
.
5
X
c
e
p
t
i
o
n
9
1
.
8
9
1
.
0
9
0
.
6
9
0
.
8
R
e
sN
e
t
5
0
V
2
9
2
.
5
9
1
.
8
9
1
.
4
9
1
.
6
V
i
T
9
6
.
5
9
5
.
8
9
5
.
4
9
5
.
6
4.
E
XP
E
R
I
M
E
N
T
AND
RE
SUL
T
4
.
1
.
E
x
perim
ent
s
et
up
All
ex
p
er
im
en
tal
ev
alu
atio
n
s
wer
e
p
er
f
o
r
m
ed
o
n
a
h
i
g
h
-
ca
p
ac
ity
co
m
p
u
tatio
n
al
s
etu
p
eq
u
ip
p
ed
with
an
I
n
tel®
C
o
r
e
™
i9
-
1
3
9
0
0
K
C
PU
@
3
.
0
0
GHz,
NVI
DI
A
GeFo
r
ce
R
T
X
4
0
9
0
GPU
(
2
4
GB
VR
AM
)
,
an
d
6
4
GB
R
AM
,
r
u
n
n
in
g
U
b
u
n
tu
2
2
.
0
4
L
T
S.
T
h
e
im
p
lem
en
tat
io
n
was
ca
r
r
ied
o
u
t
u
s
in
g
Py
t
h
o
n
3
.
1
0
with
d
ee
p
lear
n
in
g
f
r
am
ewo
r
k
s
T
en
s
o
r
Fl
o
w
2
.
1
5
a
n
d
Py
T
o
r
ch
2
.
2
.
A
co
n
s
is
ten
t
co
n
f
ig
u
r
atio
n
was
m
ain
tain
ed
f
o
r
b
o
th
th
e
p
r
o
p
o
s
ed
ViT
an
d
all
b
aselin
e
C
NN
m
o
d
els.
Fo
r
all
m
o
d
els,
a
b
atch
s
ize
o
f
3
2
an
d
a
lear
n
in
g
r
ate
o
f
0
.
0
0
0
1
wer
e
u
s
ed
d
u
r
in
g
tr
ai
n
in
g
,
an
d
5
0
ep
o
ch
s
u
s
in
g
th
e
Ad
am
o
p
tim
izer
.
T
h
e
ca
teg
o
r
ical
cr
o
s
s
-
en
tr
o
p
y
lo
s
s
f
u
n
ctio
n
was
u
s
ed
f
o
r
m
u
lti
-
class
clas
s
if
icatio
n
.
T
o
p
r
e
v
en
t
o
v
er
f
itti
n
g
,
ea
r
ly
s
to
p
p
in
g
an
d
lear
n
in
g
r
ate
s
ch
ed
u
lin
g
wer
e
ap
p
lied
d
u
r
in
g
tr
ain
in
g
.
T
o
ex
a
m
in
e
m
o
d
el
g
en
er
aliza
tio
n
,
th
e
p
r
o
p
o
s
ed
ViT
was
test
ed
o
n
an
ex
ter
n
al
f
ield
d
ataset
o
f
5
0
0
u
n
s
ee
n
im
ag
es,
ac
h
iev
i
n
g
9
4
.
8
%
ac
cu
r
ac
y
,
c
o
n
f
ir
m
in
g
r
o
b
u
s
tn
ess
b
ey
o
n
d
th
e
t
r
a
i
n
i
n
g
d
i
s
t
r
i
b
u
t
i
o
n
.
T
h
e
t
r
a
i
n
e
d
m
o
d
e
l
r
e
q
u
i
r
e
s
65
MB
of
s
to
r
a
g
e
a
n
d
d
e
m
o
n
s
t
r
a
t
es
an
a
v
er
a
g
e
p
r
o
c
e
s
s
i
n
g
t
i
m
e
o
f
3
8
m
s
p
e
r
im
a
g
e
o
n
a
n
NVI
DI
A
R
T
X
4
0
9
0
an
d
2
1
0
m
s
o
n
a
J
etso
n
Nan
o
d
ev
ice,
d
em
o
n
s
tr
atin
g
s
u
itab
ilit
y
f
o
r
r
ea
l
-
tim
e
a
n
d
e
d
g
e
d
e
p
lo
y
m
en
t.
4
.
2
.
E
v
a
lua
t
i
o
n
m
et
ho
d
Mo
d
el
ev
alu
atio
n
was
co
n
d
u
cted
u
s
in
g
th
e
p
er
f
o
r
m
an
ce
m
etr
ics
d
ef
in
ed
in
s
ec
tio
n
3
.
5
.
,
n
am
ely
ac
cu
r
ac
y
,
p
r
ec
is
io
n
,
r
ec
all,
a
n
d
F1
-
s
co
r
e.
T
o
ass
ess
th
e
in
d
iv
id
u
al
co
n
tr
ib
u
tio
n
o
f
GAN
-
b
ased
au
g
m
e
n
tatio
n
,
an
ab
latio
n
s
tu
d
y
was
p
er
f
o
r
m
ed
u
n
d
er
th
r
ee
ex
p
er
im
en
ta
l
co
n
f
ig
u
r
atio
n
s
s
u
ch
as
ViT
with
o
u
t
an
y
d
ata
au
g
m
en
tatio
n
,
ac
h
ie
v
in
g
9
3
.
4
%
ac
cu
r
ac
y
,
ViT
with
tr
a
d
itio
n
al
au
g
m
en
tatio
n
o
n
ly
,
ac
h
iev
in
g
9
4
.
1
%
ac
cu
r
ac
y
,
ViT
with
GAN
-
b
as
ed
au
g
m
en
tatio
n
,
ac
h
iev
in
g
9
6
.
5
%.
T
h
e
q
u
an
titativ
e
im
p
r
o
v
em
en
t
o
b
tain
ed
b
y
th
e
p
r
o
p
o
s
ed
GAN
au
g
m
e
n
ted
ViT
was
s
tatis
tically
v
alid
ated
u
s
in
g
a
p
ai
r
ed
t
-
test
.
L
et
an
d
d
en
o
te
th
e
ac
cu
r
ac
y
v
alu
es
f
o
r
th
e
b
as
elin
e
an
d
GAN
-
au
g
m
en
ted
m
o
d
els
o
v
er
ex
p
er
im
en
tal
f
o
ld
s
.
T
h
e
p
air
ed
t
-
s
tatis
t
ic
is
d
ef
in
ed
as:
=
√
(
2
2
)
T
h
e
n
u
ll
h
y
p
o
th
esis
H
0
ass
u
m
es
n
o
s
ig
n
i
f
ican
t
d
if
f
er
en
ce
b
e
twee
n
co
n
f
ig
u
r
atio
n
s
.
R
esu
lts
s
h
o
wed
p
<
0
.
0
1
,
in
d
icatin
g
th
at
th
e
p
e
r
f
o
r
m
an
ce
g
ain
of
th
e
GAN
-
au
g
m
en
t
ed
ViT
is
s
tati
s
tically
s
ig
n
if
ican
t.
I
n
ad
d
itio
n
to
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
1
6
,
No
.
3
,
J
u
n
e
20
2
6
:
1
3
0
7
-
1
3
1
8
1314
o
v
er
all
ac
cu
r
ac
y
im
p
r
o
v
em
e
n
t
(
2
.
4
%
o
v
e
r
tr
ad
itio
n
al
au
g
m
en
tatio
n
an
d
3
.
1
%
o
v
er
b
aselin
e)
,
p
er
-
class
F1
-
s
co
r
es
in
cr
ea
s
ed
by
an
av
er
ag
e
of
4
.
2
%
f
o
r
m
in
o
r
ity
d
is
ea
s
e
ca
teg
o
r
ies,
co
n
f
ir
m
in
g
th
at
s
y
n
th
etic
d
ata
g
en
er
atio
n
ef
f
ec
tiv
el
y
en
h
a
n
c
ed
f
ea
tu
r
e
d
iv
er
s
ity
an
d
b
alan
ce
.
T
o
f
u
r
th
er
v
er
if
y
g
en
er
aliz
atio
n
ca
p
ab
ilit
y
,
th
e
tr
ain
ed
ViT
m
o
d
el
was
ev
al
u
ated
o
n
a
n
ex
ter
n
al
f
ield
d
a
taset
o
f
5
0
0
u
n
s
ee
n
r
ice
leaf
im
ag
es,
ac
h
ie
v
in
g
9
4
.
8
% a
cc
u
r
ac
y
u
n
d
er
n
atu
r
al
illu
m
in
atio
n
an
d
b
ac
k
g
r
o
u
n
d
v
ar
iab
ilit
y
.
Fin
ally
,
co
m
p
u
tatio
n
al
ef
f
icien
cy
was
an
aly
ze
d
in
ter
m
s
of
av
er
ag
e
in
f
er
e
n
ce
tim
e
,
d
ef
in
ed
a
s
:
=
1
∑
(
−
)
=
1
(
2
3
)
wh
er
e
N
is
th
e
n
u
m
b
er
of
test
im
ag
es,
a
n
d
,
d
en
o
te
th
e
s
tar
t
an
d
e
n
d
tim
estam
p
s
of
p
r
e
d
ictio
n
f
o
r
i
m
a
g
e
.
T
h
e
m
o
d
el
d
em
o
n
s
tr
at
ed
an
av
er
ag
e
in
f
er
en
ce
d
u
r
atio
n
of
38
ms
p
er
im
ag
e
on
an
NVI
DI
A
R
T
X
4090
GPU
an
d
210
ms
on
a
J
et
s
o
n
Nan
o
d
ev
ice,
d
em
o
n
s
tr
atin
g
s
u
itab
ilit
y
f
o
r
r
ea
l
-
tim
e
an
d
e
d
g
e
d
ep
lo
y
m
en
ts
in
p
r
ec
is
io
n
a
g
r
i
c
u
l
t
u
r
e
.
5.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
As
s
h
o
wn
in
Fig
u
r
e
2
,
th
e
ViT
ac
h
iev
ed
a
s
m
o
o
th
a
n
d
co
n
s
is
ten
t
in
cr
ea
s
e
in
tr
ain
in
g
an
d
v
alid
atio
n
ac
cu
r
ac
y
,
ev
en
tu
ally
co
n
v
er
g
in
g
to
an
o
v
e
r
all
ac
cu
r
ac
y
o
f
9
6
.
5
%.
I
n
co
n
tr
ast,
C
NN
ar
ch
itectu
r
es
s
u
ch
as
Mo
b
ileNetV2
an
d
R
esNet5
0
ac
h
iev
ed
lo
wer
p
er
f
o
r
m
an
ce
,
with
ac
cu
r
ac
ies
o
f
9
3
.
2
%
an
d
9
2
.
7
%,
r
esp
ec
tiv
ely
.
GAN
-
b
ased
d
ata
au
g
m
en
tat
io
n
h
ad
a
s
u
b
s
tan
tial
im
p
ac
t
o
n
th
e
r
esu
lts
,
p
ar
ticu
lar
ly
in
b
alan
cin
g
u
n
d
er
r
ep
r
esen
ted
class
es.
T
h
is
en
h
an
ce
m
en
t
led
to
a
n
av
er
ag
e
p
er
f
o
r
m
an
ce
g
ain
o
f
4
%
–
5
%
ac
r
o
s
s
m
in
o
r
ity
d
is
ea
s
e
ca
teg
o
r
ies
an
d
r
e
d
u
c
ed
m
is
class
if
icatio
n
r
ates
f
o
r
s
ev
er
e
ca
s
es.
T
h
e
tr
ain
i
n
g
an
d
v
alid
atio
n
lo
s
s
s
tead
ily
d
ec
r
ea
s
e,
in
d
icatin
g
im
p
r
o
v
e
d
m
o
d
el
o
p
tim
izatio
n
,
wh
ile
b
o
th
t
r
ain
in
g
a
n
d
v
alid
atio
n
ac
cu
r
ac
y
in
cr
ea
s
e,
d
em
o
n
s
tr
atin
g
en
h
a
n
ce
d
g
en
er
aliza
tio
n
af
ter
f
i
n
e
-
t
u
n
in
g
.
Fig
u
r
e
2
.
T
r
ain
v
s
v
alid
atio
n
a
cc
u
r
ac
y
Fo
r
th
e
ea
r
ly
an
d
p
r
ec
is
e
d
et
ec
tio
n
o
f
r
ice
d
is
ea
s
es,
th
e
ViT
m
o
d
el
p
r
o
d
u
ce
d
l
o
wer
f
als
e
-
n
eg
ativ
e
r
ates
th
an
th
e
C
NN
b
aselin
es,
as
s
h
o
wn
b
y
th
e
co
n
f
u
s
io
n
m
atr
ix
in
Fig
u
r
e
3
.
T
h
e
m
at
r
ix
ex
h
ib
its
s
tr
o
n
g
d
iag
o
n
al
d
o
m
i
n
an
ce
ac
r
o
s
s
all
ten
class
es,
in
d
icatin
g
a
h
ig
h
co
r
r
ec
t
class
if
icatio
n
r
ate.
Mo
s
t
d
is
ea
s
e
ca
teg
o
r
ies,
in
clu
d
in
g
b
ac
ter
ia
l
leaf
b
lig
h
t,
b
ac
ter
ial
leaf
s
tr
ea
k
,
b
ac
ter
ial
p
an
icle
b
lig
h
t,
d
ea
d
h
ea
r
t,
an
d
n
o
r
m
al
leav
es,
s
h
o
w
v
er
y
lim
i
ted
m
is
class
if
icatio
n
,
d
em
o
n
s
tr
atin
g
th
at
th
e
m
o
d
el
lear
n
s
h
ig
h
ly
d
is
cr
im
in
ativ
e
f
ea
tu
r
es
f
o
r
th
ese
class
es.
Ho
wev
er
,
m
in
o
r
c
o
n
f
u
s
io
n
is
o
b
s
er
v
ed
am
o
n
g
v
is
u
ally
s
im
ilar
d
is
ea
s
e
ca
teg
o
r
ies.
I
n
p
ar
ticu
lar
,
b
last
is
o
cc
asio
n
ally
m
is
class
if
ied
as
tu
n
g
r
o
an
d
h
is
p
a,
wh
ich
ca
n
b
e
attr
i
b
u
ted
to
o
v
er
lap
p
i
n
g
lesi
o
n
p
atter
n
s
,
d
is
co
lo
r
atio
n
,
an
d
tex
tu
r
e
s
im
ilar
ities
in
ad
v
an
ce
d
d
is
ea
s
e
s
tag
es.
Simila
r
ly
,
d
o
wn
y
m
ild
ew
s
h
o
ws
lim
ited
co
n
f
u
s
io
n
with
b
last
an
d
b
r
o
wn
s
p
o
t,
as
th
ese
d
is
ea
s
e
s
s
h
ar
e
d
if
f
u
s
e
les
i
o
n
b
o
u
n
d
ar
ies
an
d
co
m
p
ar
ab
le
co
lo
r
v
ar
iatio
n
s
u
n
d
er
n
atu
r
al
f
ield
co
n
d
itio
n
s
.
T
h
e
h
is
p
a
class
ex
h
ib
its
s
o
m
e
co
n
f
u
s
io
n
with
n
o
r
m
al
leav
es
an
d
b
last
,
wh
ich
is
ex
p
ec
ted
s
in
ce
ea
r
ly
-
s
tag
e
h
is
p
a
in
f
estatio
n
s
p
r
o
d
u
ce
s
u
b
tle
v
is
u
al
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
GA
N
-
a
u
g
men
ted
visi
o
n
tr
a
n
s
f
o
r
mer wi
th
b
a
la
n
ce
d
s
yn
th
etic
d
a
ta
g
en
era
tio
n
…
(
S
a
ifu
l
I
s
la
m
)
1315
s
y
m
p
to
m
s
th
at
clo
s
ely
r
esem
b
le
h
ea
lth
y
leaf
tex
tu
r
es.
T
u
n
g
r
o
is
also
o
cc
asio
n
ally
co
n
f
u
s
ed
with
b
last
an
d
h
is
p
a
d
u
e
to
c
o
m
m
o
n
y
ello
wi
n
g
an
d
m
o
s
aic
-
lik
e
p
atter
n
s
o
n
th
e
leaf
s
u
r
f
ac
e.
I
m
p
o
r
tan
tly
,
th
e
n
u
m
b
e
r
o
f
f
alse
n
eg
ativ
es
f
o
r
m
in
o
r
ity
d
is
ea
s
e
class
es
r
em
ain
s
lo
w,
in
d
icatin
g
th
at
th
e
in
teg
r
atio
n
o
f
GAN
-
b
ased
d
at
a
au
g
m
en
tatio
n
ef
f
ec
tiv
ely
im
p
r
o
v
es c
lass
b
alan
ce
an
d
r
ec
all.
Fig
u
r
e
3
.
C
o
n
f
u
s
io
n
m
atr
i
x
af
t
er
f
in
e
tu
n
in
g
T
h
e
co
n
f
u
s
io
n
m
atr
ix
co
n
f
ir
m
s
th
at
th
e
p
r
o
p
o
s
ed
f
r
am
ew
o
r
k
n
o
t o
n
ly
ac
h
iev
es h
ig
h
o
v
er
all
ac
cu
r
ac
y
b
u
t
also
m
ain
tain
s
r
o
b
u
s
t
clas
s
-
wis
e
d
is
cr
im
in
atio
n
,
wh
ich
i
s
ess
en
tial
f
o
r
r
eliab
le
r
ice
lea
f
d
is
ea
s
e
d
iag
n
o
s
is
in
r
ea
l
-
wo
r
ld
ag
r
icu
ltu
r
al
s
ettin
g
s
.
T
ab
le
4
p
r
esen
ts
th
e
class
-
wi
s
e
p
r
ec
is
io
n
,
r
ec
all,
an
d
F1
-
s
co
r
e.
Hig
h
p
r
ec
is
io
n
an
d
r
ec
all
v
alu
es
ar
e
o
b
s
er
v
ed
f
o
r
m
o
s
t
d
is
ea
s
e
ca
teg
o
r
ies,
in
d
icatin
g
r
eliab
le
class
if
icatio
n
p
er
f
o
r
m
an
ce
.
Sli
g
h
tly
lo
we
r
r
ec
all
v
alu
es
f
o
r
v
is
u
ally
s
im
ilar
class
es
s
u
ch
as
b
las
t,
h
is
p
a,
an
d
tu
n
g
r
o
ca
n
b
e
attr
ib
u
ted
to
o
v
er
lap
p
in
g
lesi
o
n
p
atter
n
s
an
d
co
lo
r
s
im
ilar
ities
.
Fin
ally
,
th
e
r
esu
lts
s
h
o
w
th
at
th
e
ViT
ar
ch
itectu
r
e
co
m
b
in
ed
with
GAN
-
au
g
m
en
ted
d
at
a
o
f
f
e
r
s
a
d
ep
en
d
a
b
le
an
d
s
ca
lab
le
m
eth
o
d
f
o
r
au
to
m
ated
r
ice
d
is
e
ase
d
iag
n
o
s
is
.
T
h
is
e
n
s
u
r
es
n
o
t
o
n
ly
im
p
r
o
v
ed
ac
cu
r
ac
y
b
u
t a
ls
o
p
r
ac
tical
ap
p
licab
ilit
y
f
o
r
d
ep
lo
y
m
en
t in
r
ea
l tim
e
in
p
r
ec
is
io
n
ag
r
icu
ltu
r
e.
T
ab
le
4
.
Per
-
class
p
r
ec
is
io
n
,
r
ec
all,
an
d
F1
-
s
co
r
e
C
l
a
s
s
P
r
e
c
i
s
i
o
n
(
%)
R
e
c
a
l
l
(
%)
F1
-
sc
o
r
e
(
%)
B
a
c
t
e
r
i
a
l
l
e
a
f
b
l
i
g
h
t
9
8
.
5
7
9
8
.
2
2
9
8
.
4
0
B
a
c
t
e
r
i
a
l
l
e
a
f
s
t
r
e
a
k
9
7
.
7
1
9
9
.
6
1
9
8
.
6
5
B
a
c
t
e
r
i
a
l
p
a
n
i
c
l
e
b
l
i
g
h
t
9
7
.
2
5
9
6
.
5
0
9
6
.
8
8
B
l
a
s
t
9
1
.
1
2
9
1
.
8
3
9
1
.
4
7
B
r
o
w
n
s
p
o
t
9
0
.
8
1
9
7
.
2
4
9
3
.
9
2
D
e
a
d
h
e
a
r
t
9
8
.
8
3
9
6
.
9
5
9
7
.
8
8
D
o
w
n
y
m
i
l
d
e
w
9
6
.
8
6
9
2
.
5
1
9
4
.
6
4
H
i
sp
a
9
5
.
4
9
9
0
.
0
7
9
2
.
7
0
N
o
r
mal
9
0
.
8
5
9
5
.
0
4
9
2
.
8
9
Tu
n
g
r
o
9
3
.
0
6
9
2
.
3
1
9
2
.
6
8
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
1
6
,
No
.
3
,
J
u
n
e
20
2
6
:
1
3
0
7
-
1
3
1
8
1316
6.
CO
NCLU
SI
O
N
T
h
is
r
esear
ch
p
r
o
p
o
s
ed
a
GA
N
-
au
g
m
en
ted
ViT
f
r
am
ewo
r
k
f
o
r
th
e
ea
r
ly
id
en
tific
atio
n
of
d
is
ea
s
es
of
r
ice
leav
es,
o
f
f
er
in
g
a
n
o
v
el
s
o
lu
tio
n
f
o
r
p
r
ec
is
io
n
ag
r
ic
u
ltu
r
e.
By
u
tili
zin
g
DC
GAN
-
b
a
s
ed
s
y
n
th
etic
d
ata
g
en
er
atio
n
,
th
e
f
r
am
ewo
r
k
ef
f
ec
tiv
ely
ad
d
r
ess
ed
d
ataset
im
b
alan
ce
an
d
im
p
r
o
v
e
d
m
o
d
el
g
en
er
aliza
tio
n
ac
r
o
s
s
m
u
ltip
le
r
ice
d
is
ea
s
e
ca
teg
o
r
ies.
T
h
e
ViT
ar
ch
itectu
r
e
,
en
h
an
ce
d
th
r
o
u
g
h
s
elf
-
atten
tio
n
m
ec
h
an
is
m
s
,
s
u
cc
ess
f
u
lly
ca
p
tu
r
ed
b
o
th
g
lo
b
al
an
d
lo
ca
l
co
n
te
x
tu
al
r
elati
o
n
s
h
ip
s
,
en
a
b
lin
g
ac
cu
r
ate
r
ec
o
g
n
itio
n
o
f
d
is
ea
s
e
ty
p
e
an
d
s
ev
er
ity
ev
en
u
n
d
er
p
r
ac
tical
a
n
d
v
ar
iab
le
f
ield
co
n
d
itio
n
s
.
Ob
s
er
v
atio
n
s
f
r
o
m
th
e
e
x
p
er
im
e
n
ts
s
u
g
g
ested
th
at
th
e
p
r
o
p
o
s
ed
Vi
T
m
o
d
el
ac
h
iev
e
d
a
9
6
.
5
%
o
v
e
r
all
ac
cu
r
ac
y
,
o
u
tp
er
f
o
r
m
in
g
al
l
b
en
ch
m
ar
k
C
NN
m
o
d
els
s
u
ch
as
R
esNet5
0
,
Mo
b
ileNetV2
,
an
d
Den
s
eNe
t1
2
1
.
T
h
e
in
cl
u
s
io
n
o
f
GAN
-
au
g
m
en
ted
d
ata
led
t
o
s
ig
n
if
ican
t
im
p
r
o
v
em
e
n
ts
in
m
in
o
r
ity
class
d
etec
tio
n
,
r
ed
u
ce
d
f
alse
n
e
g
ativ
es,
a
n
d
e
n
h
an
ce
d
r
eliab
ilit
y
f
o
r
ea
r
ly
d
i
a
g
n
o
s
i
s
.
F
u
r
t
h
e
r
m
o
r
e
,
e
v
a
l
u
at
i
o
n
on
an
e
x
t
e
r
n
a
l
f
i
e
l
d
d
a
t
as
et
c
o
n
f
i
r
m
e
d
t
h
e
f
r
a
m
e
w
o
r
k
’
s
s
tr
o
n
g
g
e
n
e
r
a
l
i
z
a
t
i
o
n
ab
ilit
y
,
wh
ile
its
ef
f
icien
t
in
f
er
en
ce
tim
e
on
ed
g
e
d
ev
ices
in
d
icate
s
r
ea
d
in
ess
f
o
r
r
ea
l
-
tim
e
d
ep
lo
y
m
en
t.
Fu
tu
r
e
r
esear
ch
will
co
n
ce
n
tr
ate
o
n
e
x
ten
d
in
g
th
e
d
ataset
ac
r
o
s
s
b
r
o
a
d
er
E
n
v
ir
o
n
m
en
tal
cir
cu
m
s
tan
ce
s
,
in
co
r
p
o
r
atin
g
m
u
lti
-
c
r
o
p
d
is
e
ase
d
etec
tio
n
,
an
d
in
teg
r
atin
g
th
e
m
o
d
el
with
UAVs,
I
o
T
d
ev
ices,
an
d
m
o
b
ile
p
latf
o
r
m
s
to
s
u
p
p
o
r
t la
r
g
e
-
s
ca
le
au
to
m
ated
m
o
n
ito
r
i
n
g
.
F
UNDING
I
NF
O
R
M
A
T
I
O
N
T
h
is
r
esear
ch
is
f
u
n
d
ed
b
y
th
e
I
n
ter
n
atio
n
al
I
s
lam
ic
Un
iv
er
s
ity
o
f
C
h
ittag
o
n
g
a
n
d
Dh
ak
a
Un
iv
er
s
ity
o
f
E
n
g
i
n
ee
r
in
g
an
d
T
ec
h
n
o
lo
g
y
.
AUTHO
R
CO
NT
RI
B
UT
I
O
NS ST
A
T
E
M
E
N
T
T
h
is
jo
u
r
n
al
u
s
es
th
e
C
o
n
t
r
ib
u
to
r
R
o
les
T
a
x
o
n
o
m
y
(
C
R
ed
iT)
to
r
ec
o
g
n
ize
in
d
iv
i
d
u
al
au
th
o
r
co
n
tr
ib
u
tio
n
s
,
r
ed
u
ce
au
th
o
r
s
h
ip
d
is
p
u
tes,
an
d
f
ac
ilit
ate
co
llab
o
r
atio
n
.
Na
m
e
o
f
Aut
ho
r
C
M
So
Va
Fo
I
R
D
O
E
Vi
Su
P
Fu
Saif
u
l I
s
lam
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
Md
.
Nasim
Ak
h
tar
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
Ma
h
ad
i H
asan
✓
✓
✓
✓
✓
✓
✓
✓
✓
A.
N.
M.
R
ez
au
l K
ar
im
✓
✓
✓
✓
✓
✓
✓
I
s
r
at
B
in
t
e
h
Hab
ib
✓
C
:
C
o
n
c
e
p
t
u
a
l
i
z
a
t
i
o
n
M
:
M
e
t
h
o
d
o
l
o
g
y
So
:
So
f
t
w
a
r
e
Va
:
Va
l
i
d
a
t
i
o
n
Fo
:
Fo
r
mal
a
n
a
l
y
s
i
s
I
:
I
n
v
e
s
t
i
g
a
t
i
o
n
R
:
R
e
so
u
r
c
e
s
D
:
D
a
t
a
C
u
r
a
t
i
o
n
O
:
W
r
i
t
i
n
g
-
O
r
i
g
i
n
a
l
D
r
a
f
t
E
:
W
r
i
t
i
n
g
-
R
e
v
i
e
w
&
E
d
i
t
i
n
g
Vi
:
Vi
su
a
l
i
z
a
t
i
o
n
Su
:
Su
p
e
r
v
i
s
i
o
n
P
:
P
r
o
j
e
c
t
a
d
mi
n
i
st
r
a
t
i
o
n
Fu
:
Fu
n
d
i
n
g
a
c
q
u
i
si
t
i
o
n
CO
NF
L
I
C
T
O
F
I
N
T
E
R
E
S
T
ST
A
T
E
M
E
NT
Au
th
o
r
s
s
tate
n
o
co
n
f
lict o
f
in
t
er
est.
RE
F
E
R
E
NC
E
S
[
1
]
E.
F
l
a
h
e
r
t
y
,
“
A
r
o
u
n
d
t
h
e
w
o
r
l
d
i
n
e
i
g
h
t
c
o
m
mo
d
i
t
i
e
s,
e
p
i
so
d
e
t
h
r
e
e
:
R
i
c
e
,
”
G
l
o
b
a
l
L
a
n
d
s
c
a
p
e
s
Fo
r
u
m
,
D
e
c
.
2
0
2
4
.
h
t
t
p
s
:
/
/
t
h
i
n
k
l
a
n
d
sc
a
p
e
.
g
l
o
b
a
l
l
a
n
d
sc
a
p
e
sf
o
r
u
m.
o
r
g
/
7
1
3
3
8
/
a
r
o
u
n
d
-
t
h
e
-
w
o
r
l
d
-
in
-
e
i
g
h
t
-
c
o
mm
o
d
i
t
i
e
s
-
e
p
i
s
o
d
e
-
t
h
r
e
e
-
r
i
c
e
/
(
a
c
c
e
sse
d
F
e
b
.
1
0
,
2
0
2
6
)
.
[
2
]
I
F
T
N
e
x
t
,
“
A
b
o
u
t
3
.
5
b
i
l
l
i
o
n
p
e
o
p
l
e
e
a
t
r
i
c
e
,
w
h
i
c
h
s
u
p
p
l
i
e
s
n
e
a
r
l
y
2
0
%
o
f
g
l
o
b
a
l
h
u
ma
n
p
e
r
c
a
p
i
t
a
e
n
e
r
g
y
,
”
I
FT
Fo
o
d
T
e
c
h
n
o
l
o
g
y
Ma
g
a
zi
n
e
,
N
o
v
.
2
0
1
8
,
[
O
n
l
i
n
e
]
.
A
v
a
i
l
a
b
l
e
:
h
t
t
p
s
:
/
/
w
w
w
.
i
f
t
.
o
r
g
/
n
e
w
s
-
a
n
d
-
p
u
b
l
i
c
a
t
i
o
n
s
/
f
o
o
d
-
t
e
c
h
n
o
l
o
g
y
-
mag
a
z
i
n
e
/
i
ssu
e
s/
2
0
1
8
/
n
o
v
e
m
b
e
r
/
c
o
l
u
mn
s/
i
f
t
n
e
x
t
-
r
i
c
e
-
p
r
o
d
u
c
t
i
o
n
[
3
]
S
.
M
i
r
y
a
l
a
a
n
d
K
.
R
a
s
a
n
e
,
“
En
h
a
n
c
i
n
g
s
u
g
a
r
c
a
n
e
l
e
a
f
d
i
sea
s
e
c
l
a
ss
i
f
i
c
a
t
i
o
n
u
si
n
g
v
i
si
o
n
t
r
a
n
sf
o
r
m
e
r
s
o
v
e
r
C
N
N
s,”
D
i
sc
o
v
e
r
Art
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
,
v
o
l
.
5
,
n
o
.
1
,
p
.
8
9
,
2
0
2
5
,
d
o
i
:
1
0
.
1
0
0
7
/
s
4
4
1
6
3
-
0
2
5
-
0
0
3
4
0
-
7.
[
4
]
C
.
G
.
S
i
m
h
a
d
r
i
,
H
.
K
.
K
o
n
d
a
v
e
e
t
i
,
V
.
K
.
V
a
t
sa
v
a
y
i
,
A
.
M
i
t
r
a
,
a
n
d
P
.
A
n
a
n
t
h
a
c
h
a
r
i
,
“
D
e
e
p
l
e
a
r
n
i
n
g
f
o
r
r
i
c
e
l
e
a
f
d
i
sea
s
e
d
e
t
e
c
t
i
o
n
:
A
sy
st
e
ma
t
i
c
l
i
t
e
r
a
t
u
r
e
r
e
v
i
e
w
o
n
e
merg
i
n
g
t
r
e
n
d
s
,
m
e
t
h
o
d
o
l
o
g
i
e
s
a
n
d
t
e
c
h
n
i
q
u
e
s,
”
I
n
f
o
rm
a
t
i
o
n
Pr
o
c
e
ssi
n
g
i
n
A
g
ri
c
u
l
t
u
r
e
,
v
o
l
.
1
2
,
n
o
.
2
,
p
p
.
1
5
1
–
1
6
8
,
2
0
2
5
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
i
n
p
a
.
2
0
2
4
.
0
4
.
0
0
6
.
[
5
]
S
.
K
.
M
,
A
.
R
,
S
.
K
u
r
i
a
n
.
P
,
a
n
d
S
.
O
.
K
,
“
I
n
t
e
r
p
r
e
t
a
b
l
e
m
u
l
t
i
t
a
s
k
d
e
e
p
l
e
a
r
n
i
n
g
m
o
d
e
l
f
o
r
d
e
t
e
c
t
i
n
g
a
n
d
a
n
a
l
y
z
i
n
g
s
e
v
e
r
i
t
y
o
f
r
i
c
e
b
a
c
t
e
r
i
a
l
l
e
a
f
b
l
i
g
h
t
,
”
S
c
i
e
n
t
i
f
i
c
R
e
p
o
r
t
s
,
v
o
l
.
1
5
,
n
o
.
1
,
p
.
2
7
3
1
3
,
2
0
2
5
,
d
o
i
:
1
0
.
1
0
3
8
/
s
4
1
5
9
8
-
0
2
5
-
1
2
2
7
6
-
0.
[
6
]
M
.
Z.
U
d
d
i
n
,
M
.
N
.
M
a
h
a
m
o
o
d
,
A
.
R
a
y
,
M
.
I
.
P
r
a
ma
n
i
k
,
F
.
A
l
n
a
j
j
a
r
,
a
n
d
M
.
A
.
R
.
A
h
a
d
,
“
E
2
ET
C
A
:
E
n
d
-
to
-
e
n
d
t
r
a
i
n
i
n
g
o
f
C
N
N
a
n
d
a
t
t
e
n
t
i
o
n
e
n
se
mb
l
e
s
f
o
r
r
i
c
e
d
i
s
e
a
se
d
i
a
g
n
o
si
s,
”
J
o
u
r
n
a
l
o
f
I
n
t
e
g
r
a
t
i
v
e
A
g
r
i
c
u
l
t
u
r
e
,
v
o
l
.
2
5
,
n
o
.
2
,
p
p
.
7
5
6
–
7
6
8
,
2
0
2
6
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
j
i
a
.
2
0
2
4
.
0
3
.
0
7
5
.
[
7
]
Z.
J
i
a
n
g
,
Z.
D
o
n
g
,
W
.
J
i
a
n
g
,
a
n
d
Y
.
Y
a
n
g
,
“
R
e
c
o
g
n
i
t
i
o
n
o
f
r
i
c
e
l
e
a
f
d
i
s
e
a
ses
a
n
d
w
h
e
a
t
l
e
a
f
d
i
se
a
ses
b
a
se
d
o
n
mu
l
t
i
-
t
a
sk
d
e
e
p
Evaluation Warning : The document was created with Spire.PDF for Python.