Int
ern
at
i
onal
Journ
al of Ele
ctrical
an
d
Co
mput
er
En
gin
eeri
ng
(IJ
E
C
E)
Vo
l.
8
, No
.
6
,
Decem
ber
201
8
, p
p.
4239
~
4252
IS
S
N: 20
88
-
8708
,
DOI: 10
.11
591/
ijece
.
v8
i
6
.
pp
4239
-
42
52
4239
Journ
al h
om
e
page
:
http:
//
ia
es
core
.c
om/
journa
ls
/i
ndex.
ph
p/IJECE
Sm
ar
t
Mic
rowave O
ven
with Im
age Clas
sificatio
n and
Tempe
ratur
e Recomm
endation A
lgorithm
Tareq
Kh
an
School
of Engin
ee
ring
T
ec
hnolo
g
y
,
E
aste
rn
Mic
higa
n
Univ
ersity
,
US
A
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
Ma
y
25
, 201
8
Re
vised
Ju
l
5
,
201
8
Accepte
d
Aug
8
, 2
01
8
W
hen
food
is
w
armed
in
a
m
ic
r
owave
oven
,
the
user
guesses
th
e
est
imated
ti
m
e
for
the
he
a
ti
ng.
Thi
s
cognitive
proc
ess
of
guessing
ca
n
be
inc
orr
ect
-
result
ing
the
fi
nal
food
te
m
pe
rat
ure
to
be
to
o
hot
or
stil
l
c
old.
In
thi
s
rese
arc
h
,
a
no
vel
cl
osed
-
loop
m
ic
r
owave
o
ven
is
d
esign
ed
whic
h
aut
om
at
i
c
al
ly
suggests
the
t
arg
e
t
te
m
per
at
ure
o
f
a
food
b
y
le
ar
ning
from
pre
vious
exp
erienc
es
and
th
e
hea
t
ing
stops
a
utomati
c
al
l
y
wh
en
th
e
foo
d
te
m
per
at
ur
e
re
a
che
s
the
ta
rg
et
te
m
per
at
ur
e.
The
proposed
m
ic
rowave
ca
ptur
es
and
c
la
ss
ifi
es
the
fo
od
image,
and
rec
om
m
ends
the
ta
rg
e
t
te
m
per
at
ur
e,
th
us
the
user
do
es
not
n
ee
d
to
remem
ber
th
e
ta
rg
et
food
te
m
per
at
ur
e
ea
c
h
ti
m
e
the
sam
e
food
is
warm
ed
.
The
al
gor
it
h
m
gra
dua
l
l
y
le
arn
s
th
e
t
y
pe
o
f
foods
tha
t
are
used
in
tha
t
hou
sehold
and
be
co
m
es
s
m
art
er
in
the
recom
me
ndation.
The
proposed
al
gor
i
thm
ca
n
r
ec
om
m
end
ta
rge
t
te
m
per
at
ur
e
wit
h
an
a
cc
ur
acy
of
86.
31%
for
sol
i
d
food
and
100
%
for
li
qu
id
food.
A
prototy
pe
of
th
e
pro
posed
m
ic
rowave
is
dev
el
ope
d
using
th
e
embedde
d
s
y
s
tem
and
t
este
d.
Ke
yw
or
d:
C
losed
-
lo
op
E
m
bed
ded
s
yst
e
m
H
ist
ogram
I
m
age
c
la
ssific
at
ion
T
em
per
at
ur
e
s
ensin
g
Copyright
©
201
8
Instit
ut
e
o
f Ad
vanc
ed
Engi
n
ee
r
ing
and
S
cienc
e
.
Al
l
rights re
serv
ed
.
Corres
pond
in
g
Aut
h
or
:
Tareq K
ha
n,
School
of E
ng
i
neer
i
ng Tec
hnology,
East
ern Mi
chig
an Un
i
ver
sit
y,
118 Si
ll
H
al
l,
Yp
sil
a
nti, MI
4819
7,
USA.
Em
a
il
: t
areq
.khan@em
ic
h.
ed
u
1.
INTROD
U
CTION
Digital
te
chnol
og
y,
sen
sors,
a
nd
le
arn
i
ng
al
gorithm
s
are
ch
ang
i
ng
the
wa
y
we
interact
with
dev
ic
e
s
and
a
ppli
ances
.
Re
cent
ad
vances
in
m
iniat
ure
sens
or
s
,
em
b
edd
e
d
process
or
s
,
an
d
wi
reless
te
chnolo
gies
have
cause
d
a
ra
pid
gro
wth
in
sm
art
ap
plica
ti
ons
su
c
h
as
c
ontrolli
ng
a
nd
m
on
it
ori
ng
hom
e
ap
pliances
[
1]
-
[
4]
,
m
on
it
or
ing
of
sm
art
par
king
lots
[
5]
,
[
6],
a
uto
m
at
ic
m
e
ter
rea
ding
(
AMR)
usi
ng
sm
art
m
et
ers
[7
]
et
c.
A
m
ic
ro
wa
ve
ov
en
(c
omm
on
ly
ref
e
rr
e
d
to
as
a
m
ic
ro
wa
ve)
is
a
kitchen
a
ppli
ance
that
heat
s
and
c
ooks
f
ood
by
exposi
ng
it
to
el
ect
ro
m
agn
et
ic
rad
ia
ti
on
i
n
the
m
ic
ro
w
ave
f
reque
ncy
range.
It
has
beco
m
e
a
com
m
on
app
li
anc
e
in
th
e
m
od
ern
kitc
he
n
si
nce
it
s
release
to
t
he
publ
ic
for
re
side
ntial
and
com
m
e
rcial
us
e
in
19
67
[
8
].
Since the
n, s
om
e so
rt
of
ti
m
i
ng m
echan
ism
,
whethe
r
it
b
e
thro
ugh dial
s or
thro
ugh p
us
h
-
bu
tt
on
pro
gr
a
m
m
ing
,
is use
d
to
contr
ol ho
w
lo
n
g
th
e m
ic
ro
wa
ve h
eat
s t
he
f
ood.
In
this
process
,
the
us
e
r
nee
ds
to
est
im
at
e
t
he
e
xact
ti
m
e
m
ental
ly
by
co
ns
ide
rin
g
se
ve
ral
va
riables
su
c
h
as
the
cu
rr
e
nt
te
m
per
at
ur
e
of
the
foo
d
(
f
or
i
ns
ta
nce
,
the
foo
d
co
ul
d
be
ta
ken
out
f
ro
m
ref
ri
ge
rator,
fr
eeze
r,
or
c
ou
ld
be
at
r
oom
tem
per
at
ur
e
),
qu
a
ntit
y
and
t
her
m
al
pr
opert
ie
s
of
t
he
f
ood,
powe
r
le
vel
of
t
he
m
ic
ro
wa
ve
ov
en,
a
nd
finall
y
the
desire
d
ta
r
get
tem
per
at
ur
e
of
th
e
f
ood.
This
co
gnit
ive
process
of
cal
c
ulati
ng
the
exact
requi
red
ti
m
e
is
co
m
plex
and
m
a
y
be
est
i
m
a
te
d
incorrect
ly
.
If
a
lon
g
er
ti
m
e
is
est
i
m
at
ed
than
the
require
d
tim
e,
then
the
f
ood
beco
m
es
too
hot
to
eat
.
It
m
a
y
cause
burn
in
the
m
ou
th
if
the
food
is
co
nsum
ed
with
a
s
poon
without
c
heck
i
ng
the
te
m
per
a
ture
by
to
uch
i
ng
it
first.
Wh
en
the
f
ood
is
t
oo
hot,
t
he
us
e
r
nee
ds
to
wait
u
ntil
the
food
c
oo
ls
dow
n.
I
f
a
short
er
tim
e
is
est
i
m
at
ed
than
the
required
ti
m
e,
then
the
us
e
r
needs
to
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
8
, N
o.
6
,
Dece
m
ber
2
01
8
:
4239
-
4252
424
0
rep
eat
t
he
proc
ess
of
c
hec
king
the
c
urren
t
te
m
per
at
ur
e o
f
t
he
foo
d
by
to
uc
h,
e
stim
at
ing
tim
e,
and
t
hen h
eat
ing
again u
ntil
the
desire
d
ta
r
get t
e
m
per
at
ur
e is
r
eached
. T
his
process is
ti
resom
e and
e
rro
r
-
pro
ne.
In
this
pa
per
,
a
so
l
ution
of
t
he
a
bove
pro
bl
e
m
s
is
pro
po
se
d.
The
ove
rall
op
e
rati
on
of
th
e
syst
e
m
is
sh
ow
n
in
Fi
g
ure
1.
T
he
m
ai
n
con
t
rib
ution
s
of the
p
a
per are
m
entioned
below.
1.
In
this
pr
oj
ect
,
a
cl
os
e
d
-
l
oo
p
m
ic
ro
wa
ve
ov
e
n
is
desig
ned
wh
ic
h
co
ntinuo
us
ly
m
e
asur
e
s
the
foo
d
tem
per
at
ur
e
w
it
ho
ut
physi
ca
l
con
ta
ct
wh
i
le
the
fo
od
is
bein
g
heate
d,
an
d
sto
ps
the
m
ic
ro
wa
ve
autom
at
ic
ally
wh
e
n
t
he
f
ood
tem
per
at
ur
e
r
eaches
t
he
ta
r
get
te
m
per
at
ure.
I
n
this
proc
ess,
the
us
e
r
doe
s
no
t
nee
d
to
gu
ess
the
e
xtract
require
d
ti
m
e
m
entally
.
This
is
m
or
e
conven
ie
nt
to
us
e
an
d
en
sures
t
he
exact t
ar
get tem
per
at
ur
e of t
he fo
od.
2.
In
this
resea
rc
h,
an
a
uto
m
at
i
c
ta
rg
et
tem
per
at
ur
e
reco
m
m
end
at
io
n
al
gori
thm
is
al
so
pr
opos
e
d.
Wh
e
n
a
foo
d
is
inse
rte
d
in
the prop
ose
d
m
ic
ro
wa
ve
ov
e
n
a
nd
the
door
is
cl
os
e
d,
it
ta
ke
s
a
n
im
age
of
the
f
ood
a
nd
store
s
the
c
olor
patte
rn
(such
as
histogram
)
of
the
f
ood
i
n
a
database.
Once
a
tem
per
at
ur
e
is
set
fo
r
th
at
foo
d
us
i
ng
pus
h
butt
on
s
witc
hes,
t
hat
te
m
p
eratur
e
is
assigne
d
t
o
the
st
or
e
d
histo
gr
a
m
in
the
data
ba
se.
Lat
er,
w
he
n
th
e
us
er
heats
th
e
sam
e
kin
d
of
foo
d
agai
n
(e
ven
i
f
the
foo
d
is
on
a
dif
fer
e
nt
plate
or
c
up)
,
the
pr
opos
e
d
al
gorithm
automa
ti
cally
reco
m
m
end
s
the
t
e
m
per
at
ure
w
hi
ch
wa
s
assig
ne
d
for
that
f
oo
d
pr
e
viously
.
T
he
us
e
r
just
pr
e
sses
the
‘sta
rt’
butt
on
a
nd
do
not
need
to
r
e
m
e
m
ber
or
re
-
enter
the
ta
r
ge
t
tem
per
at
ur
e
.
The
al
gorithm
cal
culat
es
the
colo
r
patte
rn
s
i
m
i
la
rity
between
the
curre
nt
fo
od
an
d
th
e
histo
gr
am
s
stored
in
the
database.
T
he
cl
ose
st
m
at
ch
histog
ram
’s
assigne
d
tem
per
at
ure
is
reco
m
m
end
e
d
as
the
ta
r
get
te
m
per
at
ur
e.
If
t
her
e
is
no
si
gnific
ant
m
at
ch
between
t
he
c
urren
t
foo
d
c
olo
r
patte
r
n
a
nd
t
he
store
d
histo
gr
a
m
s,
then
the
new
histo
gr
am
fo
r
that
f
ood
and
it
s
desire
d
tem
per
at
ure
is
add
e
d
to
t
h
e
database
.
I
n
this
way,
t
he
m
ic
ro
wa
ve
ov
en
gra
du
al
ly
le
arn
s
t
he
ty
pe
of
foo
ds
that
are
us
e
d
in
t
hat
hous
e
hold a
nd
beco
m
es sm
art
er in rec
omm
e
nd
i
ng or
pr
e
dic
ti
ng
the
tar
get te
m
per
at
ure
.
3.
Fo
r
20
18,
tota
l
un
it
s
hip
m
ents
of
m
ic
ro
wav
e
ov
e
ns
a
re
pro
j
ect
ed
to
rea
ch
12.69
m
il
l
i
on
unit
s
[
9
]
a
nd
96%
of
U
.S.
hom
es
us
e
a
m
i
crowa
ve
ove
n
[
10
]
.
T
he
s
urv
ey
in
[
11
]
sho
ws
that
Am
eri
cans
ar
e
us
i
ng
m
ic
ro
wa
ves
to
war
m
and
he
at
m
or
e,
rathe
r
pre
par
e
dis
he
s
from
scratch.
T
he
m
ark
et
repo
rt
in
[
12
]
publishe
d
i
n
Decem
ber
20
17
sta
te
s
t
hat
m
ic
ro
wa
ves
la
g
be
hind
oth
e
r
m
ajo
r
ap
pliances
i
n
te
rm
s
of
“sm
art”
featur
es.
The
pro
po
sed
novel
m
a
chine
le
ar
ning
base
d
aut
onom
ou
s
m
ic
ro
w
ave
can
fill
this
m
ark
et
and
res
earch
g
a
p.
(
a
)
C
a
m
er
a
a
n
d
i
n
f
r
a
r
ed
tem
p
er
a
tu
r
e
s
en
s
o
r
Au
to
m
a
ti
c
a
l
l
y
s
u
g
g
es
ted
ta
r
g
et
te
m
p
e
r
a
tu
r
e
f
r
o
m
e
x
p
e
r
i
e
n
c
e
C
u
r
r
en
t
f
o
o
d
te
m
p
e
r
a
tu
r
e
(
b)
Au
t
o
m
a
t
i
c
a
l
l
y
s
t
o
p
s
w
h
en
t
a
r
g
et
t
em
p
er
a
t
u
r
e
i
s
r
ea
c
h
ed
Fig
ure
1.
The
ov
e
rall
oper
at
ion
of t
he pr
opose
d
syst
em
. (
a)
when a
foo
d
is
inser
te
d
a
nd th
e door is cl
os
e
d,
t
he
pro
po
se
d
m
ic
ro
wa
ve su
ggest
s the tar
get
foo
d
te
m
per
at
ur
e
us
in
g
im
age classi
ficat
ion
al
gorithm
; (b
)
Afte
r
the
“Sta
rt” butt
on i
s presse
d,
t
he m
ic
ro
wa
ve hea
ts t
he
f
ood u
ntil
f
oo
d
te
m
per
at
ur
e
r
eac
hes
t
he
targ
et
tem
per
a
ture
Seve
ral
work
s
are
found
in
the
li
te
ratur
e
about
m
easur
ing
the
tem
per
at
ur
e
of
the
sam
ple
or
f
oo
d
wh
il
e
they
a
re
heated
by
m
ic
ro
wa
ve
i
n
r
eal
tim
e.
T
he
us
e
of
co
nve
ntion
al
el
ect
ric
tem
per
at
ur
e
se
nsor
s
insid
e
a
m
ic
ro
wa
ve
ov
e
n
is
pr
ob
l
e
m
at
ic
becau
se
of
the
st
ron
g
el
ect
ro
m
agnet
ic
fiel
d
env
i
ronm
ent.
Fibe
r
-
op
ti
c
sens
or
s
[
13
]
a
r
e
inse
ns
it
ive
to
el
ect
ro
m
agn
et
ic
fiel
ds
.
H
owever,
they
re
quire
direct
co
nta
ct
with
t
he
f
oo
d
a
nd
requires
cl
eani
n
g
eac
h
ti
m
e
t
he
foo
d
is
warm
ed.
T
his
is
i
ncon
ven
ie
nt.
I
n
[14],
t
he
us
e
of
a
n
in
frare
d
(I
R
)
fibero
ptic ra
dio
m
et
er is p
rop
os
e
d
f
or
non
-
c
on
ta
ct
te
m
per
at
ur
e m
easur
em
ents insi
de
a
m
ic
rowav
e
ove
n. In thi
s
m
et
ho
d,
t
he
ti
p
of
the
IR
fiber
is
sit
uated
directl
y
abov
e
the
sam
ple
and
tra
ns
m
it
s
the
therm
al
rad
ia
ti
on
e
m
itted
by
t
he
heate
d
sam
ple
to
a
ra
dio
m
et
er.
In
[
15]
,
t
her
m
al
i
m
aging
te
c
hn
i
qu
e
w
it
h
f
orward
-
lo
ok
i
ng
infr
a
re
d
(F
L
IR
)
cam
era
is
us
ed,
w
he
re
te
m
per
at
ur
e
a
nd
oth
er
data
is
tran
sferred
t
o
a
PC
fo
r
m
on
it
or
i
ng
a
n
d
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N:
20
88
-
8708
Smart Micr
ow
ave
Oven w
it
h Image
Cl
as
sif
ic
ation a
n
d.
.
..
(
Ta
re
q
K
han)
4241
con
t
ro
ll
in
g
th
e
powe
r
le
vel
of
the
m
ic
ro
wa
ve.
I
n
[16],
FL
IR
cam
era
is
use
d
t
o
ca
pture
the
the
rm
al
i
mage
of
the
foo
d
su
r
fac
e.
Howe
ver
,
F
LIR
cam
era
s
a
re
ex
pen
si
ve
[
17
]
an
d
can
si
gn
i
ficantl
y
increase
the
pri
ce
of
the
m
ic
ro
wa
ve
ov
en.
I
n
[18],
tem
per
at
ur
e
m
on
it
or
ing
in
side
a
m
ic
ro
wav
e
ov
e
n
is
pr
op
os
e
d
by
us
in
g
conve
ntion
al
c
olor
cha
r
ge
-
c
ouple
d
dev
ic
e
(
CC
D)
cam
era.
This
m
et
ho
d
cou
l
d
be
lo
w
cost,
bu
t
m
ay
no
t
be
accurate.
An
i
nexpe
ns
ive
te
m
per
at
ur
e
co
ntro
ll
er
[
19]
and
op
ti
m
u
m
po
w
er
co
ntr
ol
strat
egies
f
or
m
ic
r
ow
a
v
e
ov
e
n
a
re
disc
usse
d
in
[
20
]
,
[21].
H
oweve
r,
t
hey
do
not
discuss
the
a
ut
oma
ti
c
sh
utdo
wn
of
m
ic
ro
wa
ve
powe
r
wh
e
n
t
he fo
od
tem
per
at
ur
e
r
e
aches t
he desir
ed
te
m
per
at
ur
e
.
A
rece
nt
heati
ng
el
em
ent
based
ove
n
(
no
t
m
ic
ro
wa
ve
-
bas
ed
heati
ng
)
in
[22],
us
e
s
a
hi
gh
de
fi
niti
on
(HD)
cam
era
to
ide
ntify
from
a
c
omm
on
pr
e
def
i
ned
set
of
foo
ds
that
are
pu
t
i
ns
ide
a
nd
reco
m
m
end
co
ok
i
ng
tim
es
and
temperat
ur
e.
A
r
e
sist
ance
t
e
m
per
at
ur
e
detect
ors
(RTD
)
te
m
per
at
ur
e
se
nsor
pr
obe
is
inse
rted
m
anu
al
ly
into
the
f
ood
a
nd
use
r
ca
n
get
a
noti
ficat
ion
on
s
m
art
dev
ic
es
wh
e
n
t
he
foo
d
is
co
oked
.
H
oweve
r
,
this
m
et
ho
d
of
tem
per
at
ure
sensing
re
quir
es
ph
ysi
cal
co
ntact
of
th
e
tem
per
at
ure
se
ns
or
with
the
food.
More
ov
e
r,
the
foo
d
recog
niti
on
al
go
rithm
is
fixed
to
a
set
of
pre
def
i
ned
f
oods
,
th
us
it
do
es
not
le
arn
a
ny
new
foo
d
it
e
m
and
un
a
ble
to
reco
m
m
end
their
tem
per
at
ur
es.
In
this
propo
sed
resea
rch,
foo
d
tem
per
at
ur
e
is
m
easur
ed
us
i
ng
IR
te
m
per
at
ure
sens
or.
It
is
insensiti
ve
to
m
ic
ro
wa
ves,
m
uch
c
hea
per
t
ha
n
FL
IR
cam
er
a,
an
d
do
e
s
not
nee
d
ph
ysi
cal
c
on
ta
ct
with
the
f
oo
d.
In
this
pro
pose
d
w
ork
,
the
autom
at
ic
sh
ut
down
of
m
ic
ro
wa
ve
powe
r
w
hen
t
he
food
te
m
per
a
ture
reac
hes
t
he
desire
d
te
m
p
eratur
e
is
al
s
o
discusse
d
us
in
g
a
m
ic
ro
con
tr
oller
-
base
d
em
bed
de
d
syst
em
.
The
pro
po
se
d
food
cl
assi
ficat
ion
al
gorithm
gr
a
dual
ly
le
arn
s
the
f
ood
it
e
m
s
by
i
m
age p
r
ocessi
ng and
rec
omm
end
s t
heir
ta
r
get tem
per
at
ures;
r
at
he
r
tha
n a p
red
e
fine
d
se
t of f
oods.
2.
RESEA
R
CH MET
HO
D
2.1.
Fo
od
Clas
si
ficat
i
on
u
sing
I
ma
ge
Pr
ocessi
ng
and Tem
pe
rature
Reco
m
mendatio
n
Algo
ri
th
m
T
he
pro
po
se
d
m
ic
ro
wa
ve
oven
ca
ptures
a
n
i
m
age
of
t
he
f
ood
w
hen
it
’s
door
is
cl
os
e
d.
The
im
age
is
then
processe
d
and
the
al
go
rithm
trie
s
to
classify
the
im
a
ge
by
c
om
par
ing
it
s
c
olor
pa
tt
ern
with
the
store
d
colo
r
patte
r
ns
i
n
the
datab
ase.
If
t
he
f
ood
is
cl
assifi
ed,
th
en
it
reco
m
m
end
s
the
te
m
per
at
ur
e
t
hat
is
assi
gn
e
d
t
o
that
f
ood.
I
f
th
e
foo
d
is
uncl
assifi
ed,
the
n
it
’
s
col
or
patte
r
n
and
de
sired
te
m
per
at
ur
e
is
a
dd
e
d
t
o
the
dat
abase
.
A
bl
oc
k
dia
gra
m
of
the
pro
po
s
ed
cl
assi
ficat
ion
al
gorit
hm
is
sh
own
i
n
Fig
ure
2
.
A
bri
ef
descr
i
ptio
n
of
th
e
al
gorithm
is d
isc
us
se
d belo
w.
I
s
E
q
u
a
l
I
s
F
o
o
d
S
o
l
i
d
R
e
m
o
v
e
B
a
c
k
g
r
o
u
n
d
f
r
o
m
S
o
l
i
d
F
o
o
d
R
e
m
o
v
e
B
a
c
k
g
r
o
u
n
d
f
r
o
m
L
i
q
u
i
d
F
o
o
d
G
e
n
e
r
a
t
e
H
i
s
t
o
g
r
a
m
C
l
a
s
s
i
f
y
F
o
o
d
I
s
C
l
a
s
s
i
f
e
d
?
R
e
c
o
m
m
e
n
d
T
e
m
p
e
r
a
t
u
r
e
A
d
d
i
n
D
a
t
a
b
a
s
e
C
a
p
t
u
r
e
F
o
o
d
I
m
a
g
e
Y
e
s
N
o
E
m
p
t
y
I
m
a
g
e
N
o
N
o
Y
e
s
E
n
d
Y
e
s
D
a
t
a
b
a
s
e
S
t
a
r
t
Fig
ure
2.
Bl
oc
k diag
ram
o
f
th
e f
ood
cl
assifi
c
at
ion
a
nd tem
per
at
ur
e
r
ec
omm
end
at
ion al
gorithm
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
8
, N
o.
6
,
Dece
m
ber
2
01
8
:
4239
-
4252
4242
2.1.1.
Existence
of F
ood
Det
ec
tion
As
s
oon
as
t
he
door
of
the
m
ic
rowav
e
oven
is
cl
os
ed
,
an
i
m
age,
I
captured
,
is
captu
red
us
i
ng
a
n
HD
ca
m
era
m
ou
nt
ed
on
the
r
oof
of
the
oven
.
T
he
im
age
is
an
RGB
col
or
im
age
of
siz
e
320
×
240.
T
he
fir
st
ste
p
of
the
al
go
rith
m
is
to
determ
ine
wh
et
her
t
he
re
is
food
i
n
the
oven
or
it
is
e
m
pty.
To
do
that,
a
preex
ist
ing
i
m
age
of
the
m
i
cro
wa
ve
ov
en
with
out
an
y
fo
od,
I
em
pt
y
,
is
com
par
ed
with
I
captured
.
If
there
is
a
sign
ific
ant
m
at
ch,
then
th
e
m
ic
ro
wa
ve
i
s
em
pty
and
the
al
gorit
hm
t
erm
inate
s
witho
ut
te
m
per
at
ure
reco
m
m
e
nd
a
ti
on
.
If
there
is
a
sign
ific
ant
m
is
m
atch
,
the
n
there
is
fo
od
in
the
m
ic
ro
wa
ve
an
d
it
go
es
to
the
nex
t
blo
c
k
of
the
al
gorithm
.
As
the
m
ic
ro
wav
e
tur
ntable
tray
can
r
otate
,
it
cou
l
d
be
i
n
a
di
ff
e
ren
t
posit
io
n
tha
n
the
pree
xisti
ng
I
em
pt
y
i
m
age.
S
o,
a
h
a
r
d pixel
by p
i
xel co
m
pa
rison
betwee
n
I
captured
an
d
I
em
pt
y
will
n
ot work.
In
Fig
ure
3,
an
exam
ple
of
a
pr
ee
xisti
ng
em
pty
i
m
age
(a),
a
captu
red
im
a
ge
without
f
oo
d
(c
),
a
nd
a
captu
red
im
age
with
food
(e)
al
on
g
with
th
ei
r
gr
ay
scal
e
hi
stog
ram
s
(b
)
,
(d)
an
d
(f)
are
sh
ow
n
nam
el
y.
Her
e
we
see
that,
e
ven
t
hough
(c
)
is
ro
ta
te
d
co
m
par
ed
wit
h
(
a),
their
histo
gram
s
(d
)
an
d
(
b)
a
re
quit
e
sim
il
ar
because
histo
gram
s
do
no
t
con
ta
in
the
po
s
it
ion
inform
at
i
on
of
the
pixe
ls
[1
2].
O
n
the
oth
e
r
ha
nd,
the
histo
gr
am
of
the
im
age
with
foo
d,
(
f),
is
qu
it
e
diff
ere
nt
f
r
om
(b
)
an
d
(d)
.
To
c
om
par
e
the
im
ages,
h
ist
ogram
diff
e
re
nce
bet
ween
t
he
tw
o
i
m
ages
are
cal
c
ulate
d.
If
t
he
di
ff
ere
nce
of
t
he
histo
gr
am
s
is
le
ss
than
a
t
hresh
old
,
HIST_DIFF_THRESHOLD
,
then
the
i
m
ages
are
consi
der
e
d
e
qual
,
el
se
they
are
con
si
der
e
d
diff
e
ren
t.
E
xperi
m
ents
hav
e
been
co
nducte
d
with
10
em
pty
i
m
ages
,
I
em
pt
y
,
hav
i
ng
di
ff
e
ren
t
r
ot
at
ion
s
of
the
tur
ntable
tray
and
a
thres
ho
l
d
val
ue
is
assigne
d
as
100.
T
he
ste
ps
of
determ
ining
w
hethe
r
th
e
i
m
ages
are
equ
al
are
s
how
n
in
the
ps
e
udoc
od
e
in Fi
g
ure
4.
(a)
(c)
(e)
(b)
(d)
(f)
Fig
ure
3. (
a
) P
reex
ist
in
g
im
a
ge wh
e
n n
o food i
n o
ven
;
(
b) H
ist
ogram
o
f
(a
); (
c
)
Ca
pt
ur
e
d
im
age w
he
n no
foo
d
in
ove
n,
t
he
tray
c
ould
be
rotat
ed
c
om
par
ed
w
it
h (a
); (d)
H
ist
og
ram
o
f
(c
); (
e
)
Ca
ptured
im
age w
he
n
foo
d
is
pr
ese
nt
; (f)
Histogram
of (
e
)
I
captured_
gray
:= grayscale of I
captured
I
empty_gray
:= grayscale of I
empty
h1 := histogram of I
captured_gray
h
2 := histogram of I
empty_gray
Calculate the average of histogram differences as d
12
:
256
hh
d
IsEqual := (
d <
HIST_DIFF_THRESHOLD
)
Fig
ure
4. Pse
udoc
ode
for
c
om
par
ing
im
ages
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N:
20
88
-
8708
Smart Micr
ow
ave
Oven w
it
h Image
Cl
as
sif
ic
ation a
n
d.
.
..
(
Ta
re
q
K
han)
4243
2.1.2.
So
li
d
an
d
Li
quid F
ood C
l
assi
ficat
i
on
Af
te
r
t
he
exist
ence of th
e f
oo
d
is determ
ined,
the
nex
t
bloc
k
of the alg
or
i
thm
d
et
er
m
ine
s w
het
her
the
foo
d
is
so
li
d
or
li
quid
.
Li
quids
(co
m
m
on
ly
ref
err
e
d
as
dr
inks
)
are
ge
ner
al
ly
serve
d
in
cu
ps
or
glasses,
wh
e
reas
so
li
ds
are
serv
e
d
on
plate
s
or
bow
ls.
Cl
assify
ing
wh
et
he
r
the
f
ood
is
so
li
d
or
li
qu
id
is
required
to
detect
and
rem
ov
e
the
foo
d
ba
ckgr
ound
in
the
ne
xt
blo
c
k
of
the
al
g
or
it
hm
.
The
pr
em
is
e
fo
r
cl
assify
in
g
so
li
d
and
li
quid
foo
d
is
that,
in
m
os
t
cases,
the
te
xture
of
so
li
d
f
ood
is
coa
rse,
wh
e
reas
the
te
xture
of
li
qu
i
d
foo
d
is
sm
oo
th
an
d
ho
m
og
enous.
S
o,
the
var
ia
nce
(
σ
2
)
of
so
li
d
f
ood
im
age
will
be
m
uch
highe
r
tha
n
t
he
var
i
a
nce
of
li
qu
id
f
ood.
In
Fig
ure
5,
t
he
colo
r
im
age
of
a
s
olid
foo
d
is
sh
ow
n
in
(a
),
it
s
gray
scal
e
i
m
age
is
sh
ow
n
in
(b)
wit
h
a
32
×
32
pix
el
blo
c
k
draw
n
in
m
agen
ta
color
at
m
idd
le
,
and
the
3D
plo
t
of
the
bl
ock
where
heigh
t
is
pro
portion
al
to
the
pix
el
va
lu
e
i
s
sh
own
i
n
(
c).
Sim
il
arl
y,
(d
)
,
(e),
an
d
(
f)
sh
ows
t
he
i
m
ages
f
or
a
li
qui
d
f
ood
nam
ely.
Her
e
,
we
see
that
th
e
pix
el
val
ues
i
n
the
bl
oc
k
c
ha
ng
e
m
or
e
in
s
olid
foo
d
as
show
n
i
n
(c)
,
whereas
the
pi
xel
val
ue
s
are
al
m
os
t
the
sam
e
in
li
qu
id
f
ood
as
s
ho
wn
in
(f).
For
this
exam
ple,
the
var
ia
nce
of
(c)
is
746.
9
an
d
the
var
ia
nce
of
(f)
is
1.
7
.
The
var
i
ance
is
cal
culat
ed
us
in
g
(
2)
a
nd
(3)
w
her
e
,
N
is
the
total
pix
el
in
the
blo
c
k,
x
i
is
the
i
th
pix
el
value,
an
d
m
is
the
m
ean
of
al
l
pix
el
values
i
n
the
bl
ock.
Sim
ula
ti
on
s
ha
ve
been
cond
ucted
with
263
so
li
d
foo
d
an
d
5
3
li
quid
food
im
ages
fo
r
a
blo
c
k
siz
e
of
32
×
32
in
th
e
m
idd
le
.
It
ha
s
bee
n
fou
nd
t
hat
the
m
axi
m
u
m
var
ia
nce
f
or
li
qui
d
foo
d
is
4.6
.
A
thres
ho
l
d
,
MI
N
_S
OL
ID
_F
OO
D_
VA
R
,
is
assig
ne
d
as
10
in
the
al
gorith
m
.
If
the
blo
c
k
var
ia
nce
is
l
ess
tha
n
or
eq
ual
to
the
t
hr
e
sh
ol
d,
the
n
the
f
ood
is
cl
as
sifie
d
a
s
li
qu
id,
else i
t i
s
classi
fied
as
s
olid.
(a)
(b)
(c)
(d)
(e)
(f)
Fig
ure
5. (
a
)
C
olor im
age o
f
s
olid
foo
d; (b)
Gr
ay
scal
e im
a
ge of
(
a
) wit
h
a
b
loc
k d
rawn i
n
m
agen
ta
c
olor at
m
idd
le
; (c)
3D
plo
t
of the
blo
c
k
in
(b) w
he
re
heig
ht is
propo
rtion
al
t
o
the
p
i
xel v
al
ue; (d
) C
olo
r
im
age o
f
li
qu
id
f
ood; (e)
G
r
ay
scal
e i
m
age
of (d)
w
it
h
a
blo
c
k d
rawn
in m
agen
ta
co
l
or at m
idd
le
; (f)
3D
pl
ot of the
blo
c
k
in
(
e
) whe
re
heig
ht is
pro
po
rtion
al
t
o
t
he pixel
value
2.1.3.
Back
ground Rem
oval
In
this
co
ntext
,
on
ly
the
foo
d
portio
n
of
the
captu
re
d
im
age
is
def
in
ed
as
foregr
ound,
as
sh
ow
n
inside
the
gr
ee
n
r
e
gion
in
Fi
g
ur
e
6(
a
),
a
nd
the
rest
of
the
i
m
age
is
def
ine
d
as
bac
kgr
ound
as
s
how
n
in
Fig
ure
6(
a
)
a
nd
Fig
ure
6(b
).
The
c
olor
in
form
at
i
on
of
t
he
f
oo
d
is
c
onta
ined
in
the
f
or
e
gr
ound,
a
nd
no
t
in
the
backg
rou
nd.
T
he
backg
rou
nd
m
a
y
var
y
if
t
he
sam
e
food
is
ser
ve
d
in
a
di
ff
ere
nt
ty
pe
of
plate
,
bowl,
or
c
up
.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
8
, N
o.
6
,
Dece
m
ber
2
01
8
:
4239
-
4252
4244
So
,
the
al
gorithm
re
m
ov
es
the
backgro
und
f
ro
m
the
i
m
age
so
that
the
ge
ner
at
e
d
histo
gra
m
(d
isc
us
se
d
in
the
nex
t
sect
io
n)
do
no
t
co
ntain
backgro
und
c
olor
inf
or
m
at
i
on,
an
d
thu
s
t
he
foo
d
can
be
cl
assifi
ed
accuratel
y
even if it
is
put o
n
a
dif
fer
e
nt
con
ta
ine
r.
(a)
(b)
Fig
ure
6. (a
)
F
or
e
gro
und (insi
de
the
gree
n
r
e
gion)
and
bac
kgr
ound
(outsid
e the
gr
ee
n re
gi
on) of
so
li
d f
ood;
(b) fo
regr
ound
(the ora
nge
part
)
an
d bac
kground (
ou
tsi
de
th
e oran
ge part
)
of li
qu
i
d
f
ood
So
li
d
f
ood
is
gen
e
rall
y
pu
t
on
a
plate
or
po
t.
When
food
is
pla
ced
on
a
plate
,
it
is
gen
e
rall
y
su
r
rou
nded
by
s
m
oo
th
te
xtu
r
e
hav
i
ng
a
m
on
ot
onous
c
ol
or
of
t
he
plate
a
s
show
n
in
Fig
ure
6
(a
)
.
F
or
so
li
d
foo
d,
t
he
f
oreg
rou
nd
has
hi
gh
va
riance,
as
s
how
n
in
Fig
ure
5
(c)
,
a
nd
it
s
backg
rou
nd
ha
s
lo
w
var
ia
nce
s
in
t
he
pix
el
val
ues
.
This
he
ur
ist
ic
is
us
ed
to
det
ect
the
backgroun
d
of
the
s
olid
f
ood.
Liq
uid
f
ood
or
dr
ink
is
gen
e
rall
y
serv
e
d
in
a
m
ug
or
glass.
F
or
li
qu
i
d
f
ood,
the
for
egro
und
has
lo
w
va
riance
,
as
sh
ow
n
in
Fig
ure
5
(f),
and
it
is
s
urrounde
d
by
t
he
edg
e
o
f
the
c
onta
iner
a
s
sho
wn
i
n
Fi
g
ure
6
(
b).
T
he
bl
oc
k
co
ve
rin
g
an
edg
e
,
as
sh
ow
n
in
Fig
ure
6
(b)
in
blac
k
col
or,
will
ha
ve
hi
gher
vari
ance
in
pix
el
values
.
T
hese
high
va
riance
edg
e
blo
c
ks
a
re
us
e
d
to
d
et
ect
t
he back
gro
und o
f t
he
li
qu
i
d foo
d.
To
aut
om
at
e
t
he
bac
kgr
ound
re
m
ov
al
p
r
oc
ess,
a
cop
y
of
the
i
m
age
is
first
conver
te
d
to
gr
ay
scal
e
and
the
n
di
vide
d
into
f
our
qu
adr
a
nts
–
top
-
r
igh
t,
to
p
-
le
ft,
bott
om
-
le
ft,
and
bo
tt
om
-
righ
t
nam
ely
as
sh
ow
n
in
Fig
ure
7.
T
hen
sta
rting
f
r
om
the
center
of
t
he
gray
scal
e
im
age,
the
va
ri
ance
is
cal
cula
te
d
f
or
eac
h
10
×
10
blo
c
k
s
acc
ordi
ng
to
the
horiz
on
ta
l
se
qu
e
nce
(i.e.
row
a
fter
row
in
raster
s
can
fa
sh
i
on)
as
show
n
Fi
g
ure
7(
a
)
at
the
to
p
-
rig
ht
quad
ra
nt.
F
or
s
ol
id
f
ood,
if
t
he
var
ia
nce
of
CO
NS
EC
_B
G_
BL
K
num
ber
of
c
ons
ecuti
ve
blo
c
ks
is
le
ss
than
a
th
resh
ol
d
val
ue,
BG_V
AR_THRESHOLD
,
then
the
ba
ckgr
ound
is
reached
a
nd
the
rem
ai
nin
g
bloc
ks
of
that
row
are
m
ade
blac
k
on
th
e
act
ual
color
i
m
age.
For
li
quid
foo
d,
if
the
var
ia
nce
of
a
bl
ock
is
great
er
than
a
thres
ho
l
d
val
ue
,
EDGE_VAR_THRESHOLD
,
t
hen
t
he
bac
kgr
oun
d
is
reac
he
d
an
d
the
rem
ai
nin
g
blo
c
ks
of
that
row
are
m
ade
black
on
the
act
ual
color
im
age.
The
n
this
pr
oces
s
is
rep
eat
ed
f
or
th
e
re
m
a
ining
top
-
le
ft
,
bo
tt
om
-
le
ft,
a
nd
bo
tt
om
-
righ
t
qua
dr
a
nts
acc
ordi
ng
t
o
t
he
seq
uen
ce
s
hown
in
Fig
ure
7
(a).
Af
te
r
th
at
,
the
entire
proce
ss
is
rep
eat
ed
a
ccordin
g
to
t
he
ve
rtic
al
seq
uen
ce
a
s
sho
wn
i
n
Fig
ure
7
(
b).
A
nother
si
m
il
ar
horizo
ntal
an
d
ver
ti
cal
seq
uence
are
re
peated
wh
e
re
it
searc
hes
f
or
fu
ll
y
bl
ack
bl
ock
a
nd
on
ce
fou
nd,
it
fill
s
the
rest
of
t
he
r
ow
or
the
c
ol
um
n
with
black
nam
ely.
The
ba
ckgr
ound
of
t
he
im
ages
sh
own
in
Fi
g
ure
5(a)
a
nd
Fi
gure
5(d
)
is
rem
ov
ed
us
in
g
the
pro
po
se
d
al
gorithm
and
the
res
ults
are
sh
ow
n
in
Fig
ure
8
(a
)
a
nd
Fi
g
ure
8
(b) nam
ely.
(a)
(b)
Fig
ure
7. Bl
oc
k varia
nce c
he
ckin
g
se
quence
s in
t
he
f
our
qu
adr
a
nts
of the i
m
age: (a)
hor
iz
on
ta
l
pass;
(b)
ver
ti
cal
p
a
ss
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N:
20
88
-
8708
Smart Micr
ow
ave
Oven w
it
h Image
Cl
as
sif
ic
ation a
n
d.
.
..
(
Ta
re
q
K
han)
4245
(a)
(b)
Fig
ure
8. (
a
)
B
ackgr
ound r
em
ov
e
d
from
Fig
ure
4
(a
); (b)
bac
kgr
ound r
em
ov
e
d
from
Fig
ure
4 (
d)
2.1.4.
Histogr
am Ge
nera
tion
Ther
e
a
re
se
ve
ral
featu
re
extr
act
ion
te
ch
niques
av
ai
la
ble
in
the
li
te
ratur
e
ta
rg
et
in
g
dif
fe
ren
t
ty
pes
of
i
m
ages
[23]
,
[
24]
.
For
instan
ce,
a
face
rec
ogniti
on
al
gor
it
h
m
will
extract
featur
es
s
uch
as
betwee
n
-
ey
e
distance,
widt
h
-
le
ngth
rati
o
et
c.
[25].
H
owev
er,
the
cl
assifi
c
at
ion
of
f
ood
i
m
age
is
qu
it
e
diff
e
re
nt
as
the
food
do
e
s
not
ha
ve
any
fixe
d
s
ha
pe
or
te
xture.
A
foo
d
will
ha
ve
dif
fer
e
nt
sh
a
pe
eac
h
tim
e
wh
e
n
it
is
plac
ed
on
a
plate
or
wh
e
n
i
t
is
sti
rr
ed.
T
he
ov
e
rall
colo
r
patte
rn
(i.e.
histogram
)
of
the
foo
d
is
the
only
sign
ific
ant
fe
at
ur
e
for
cl
assify
in
g
foo
d.
I
n
the
propose
d
al
gorit
hm
,
the
histogram
of
the
i
m
a
ge,
h_
te
st
,
is
ge
ner
at
e
d
f
or
re
d
(R)
,
gr
ee
n
(
G)
,
a
nd
blu
e
(B
)
cha
nnel
s
a
fter
rem
ov
ing
the
bac
kgr
ound
.
Th
e
blac
k
pix
el
co
unt
in
the
hi
sto
gr
a
m
(i.e.
R=
0
,
G=
0,
B=
0)
is
set
t
o
ze
r
o
as
blac
k
is
use
d
t
o
c
ov
e
r
t
he
bac
kgr
ound
and
it
has
no
e
ff
ect
on
the
f
ood
col
or
patte
rn.
The
ge
ner
at
e
d
histo
gr
am
s
of
Fig
ure
8(
a)
a
nd
Fig
ure
8(
b)
are
s
how
n
in
Fig
ure
9(
a
)
an
d
Fig
ure
9(
b)
nam
ely.
(a)
(b)
Fig
ure
9. (a
) H
ist
og
ram
o
f
Fi
g
ure
7(
a
); (b
) Hist
ogram
o
f F
ig
ure
7(b)
2.1.5.
Matchin
g Sco
re C
alcula
tion
In
t
his
ste
p,
t
he
ge
ne
rated
te
st
histogram
,
h_te
st
,
is
co
m
par
ed
with
the
previ
ou
sly
gen
e
rated
histo
gr
am
s
stored
in
the
data
base
an
d
m
a
tch
in
g
sco
res
are
cal
culat
ed.
Let
’s
con
si
der
th
at
,
a
fo
od
was
heated
pr
e
viously
an
d
it
s
histo
gr
am
,
h_
c
omp
,
is
al
r
eady
sto
red
in
the
data
base.
Now,
w
he
n
th
e
sam
e
fo
od
is
heate
d
again,
the
f
oo
d
can
be
in
a
diff
ere
nt
plac
e
on
the
plate
,
in
a
ro
ta
te
d
po
sit
io
n,
scal
e
d
(i.e.
dif
fere
nt
fo
od
qu
a
ntit
y),
or
sti
rr
ed
c
om
par
ed
with
its
pr
e
vi
ou
sly
store
d
hi
stog
ram
.
The
al
gorithm
sh
ould
consi
der
al
l
these
trans
form
ations
an
d
s
houl
d
i
nd
ic
at
e
t
hey
a
re
the
sam
e
food.
T
he
tra
ns
l
at
ion
(i.e.
dif
f
eren
t
posit
ion
in
the
plate
)
a
nd
t
he
r
otati
on
of
the food w
il
l
not
ha
ve
m
uch
ef
fect
on
the h
ist
ogr
a
m
becau
se po
sit
ion
in
form
ation
o
f
pix
el
s
is
no
t
st
or
e
d
in
histo
gr
a
m
s.
The
diff
e
ren
t
sti
r
rin
g
co
nd
it
io
n
ca
n
ca
us
e
s
om
e
m
inor
di
ff
e
ren
ces
,
but
the
ov
e
rall
colo
r
pa
tt
ern
will
be
s
i
m
i
la
r.
Fig
ure
10(a)
s
how
s
th
e
sa
m
e
fo
od
a
s
sh
ow
n
in
Fi
g
ure
5(a),
howe
ve
r
,
th
e
foo
d
is
r
otate
d
an
d
in
a
diff
e
r
ent
sti
rr
i
ng
co
nd
it
io
n.
Fig
ure
10(
b)
sho
ws
t
he
im
age
a
fter
backg
rou
nd
re
m
ov
al
and
Fig
ure
10(
c)
sho
ws
the
hi
stog
ram
of
Fi
g
ure
10
(b).
I
f
we
com
par
e
th
e
sh
ape
s
of
the
histogram
s
in
Fig
ure
9(a
)
with
t
he
histo
gr
am
s
sh
own
in
Fig
ure
10(c),
we
see
that
they
are
qu
it
e
sim
i
la
r.
This
sho
ws
t
hat
the
histo
gr
am
s
the
sa
m
e
fo
od
ha
ving
a
dif
fer
e
nt
translat
io
n,
ro
ta
ti
on
an
d
st
irrin
g
co
ndit
ion
will
be
sim
i
la
r.
I
f
there
is
scal
in
g
of
the
f
ood,
then
one
of
the
histo
gr
am
s
is
scal
ed
befor
e
c
om
par
ing.
T
he
sm
al
le
r
area
histo
gr
am
is
sc
al
ed
to
the
la
r
ge
r
area
hist
ogr
a
m
us
ing
the
pse
udoc
ode
as
s
how
n
in
Fig
ure
11.
T
his
is
do
ne
f
or
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
8
, N
o.
6
,
Dece
m
ber
2
01
8
:
4239
-
4252
4246
each c
olor c
ha
nn
el
.
(a)
(b)
(c)
Fig
ure
10
.
(a)
The
sam
e foo
d as s
how
n
in
Fi
g
ure
5(
a
), h
ow
ever, in
a
diff
e
ren
t
ro
ta
ti
on a
nd sti
r
rin
g
c
ondi
ti
on
;
(b)
Ba
c
kgrou
nd
rem
ov
ed fr
om
Fig
ure
10(a);
(
c)
Histo
gr
am
of F
i
g
ure
10(
b)
, whic
h
is
quit
e sim
il
ar to
Fig
ure
9
(a
)
1.
Area_h_test := sum (h_test)
2.
Area_h_comp := sum (h_comp)
3.
If Area_h_test >=
Area_h_comp Then
4.
h_comp := h_comp * (Area_h_test / Area_h_comp)
5.
Else
6.
h_test := h_test * (Area_h_comp / Area_h_test)
7.
End
Fig
ure
11
.
Pse
udoc
od
e
for sc
al
ing
histo
gr
a
m
s
Af
te
r
scal
in
g,
Eucli
dian
dista
nce
is
cal
c
ulate
d
betwee
n
t
he
histo
gr
am
s
f
or
eac
h
c
olor
c
ha
nn
el
us
i
ng
(1).
The
n
t
he
total
m
at
ching
sc
or
e
is
cal
culat
ed
by
a
ddin
g
the
dista
nces,
d,
f
or
R
,
G
,
a
nd
B
c
hannel
histo
gr
am
s.
Th
e sm
aller th
e
distance (
or m
at
c
hing sc
or
e
), t
he
b
et
te
r
t
he
m
a
tc
h.
2
te
st
c
om
p
d
h
h
(1)
2.1.6.
Te
mpera
tu
re
Recomm
end
ati
on
Ma
tc
hin
g
sco
r
es
are
cal
culat
ed
betwee
n
the
ge
ner
at
ed
te
st
histo
gr
am
,
h_
t
est
,
an
d
with
the
f
ood
it
e
m
histo
gr
am
s
stored
in
the
database.
T
he
assigne
d
te
m
per
at
ur
e
is
looke
d
up
for
the
f
ood
it
e
m
wh
ic
h
has
th
e
m
ini
m
u
m
m
at
c
hing
sco
re.
T
his
tem
per
at
ur
e
i
s
reco
m
m
end
ed
by
the
al
gori
thm
to
the
us
er
.
The
u
s
er
can
then
increase
or
de
crease
t
he
rec
omm
end
ed
te
m
per
at
ur
e
if
wan
ts
by
pus
h
bu
tt
on
switc
hes
or
can
ac
cept
the
reco
m
m
end
ed
tem
per
at
ur
e
for
the
f
ood.
I
f
th
ere
is
no
sig
nif
ic
ant
m
a
tc
h
between
t
he
gen
e
r
at
ed
te
st
histo
gram
,
h_
te
st
,
a
nd
wit
h
the
histogra
m
s
stored
in
t
he
data
base
,
(i.e.
the
m
in
i
m
u
m
m
a
tc
hin
g
sc
or
e
is
gr
eat
er
t
han
a
thres
ho
l
d
val
ue
,
MAX_MS
),
then
the
f
ood
is
consi
der
e
d
as
a
new
foo
d
it
e
m
.
In
t
his
case,
a
def
a
ult
tem
per
at
ur
e
is
reco
m
m
end
e
d
by
the
al
gorithm
to
the
us
er
.
The
u
se
r
m
ay
change
t
his
te
m
per
at
ur
e
by
push
bu
tt
on
s
witc
hes.
Af
te
r
the
us
e
r
pr
ess
the
sta
rt
bu
tt
on
o
n
t
he
m
ic
ro
wa
ve,
t
he
h_te
st
an
d
it
s
assig
ned
te
m
per
at
ur
e
is
a
dd
e
d
to
the
database
as
a
new
f
ood
it
em
.
In
this
wa
y,
the
al
go
rith
m
gr
ad
ually
l
earn
s
wh
at
te
m
per
at
ur
e
it
s
houl
d
reco
m
m
end
f
or
diff
e
re
nt
f
oods
us
e
d
i
n
that
h
ouse
hold.
Th
e
al
gorithm
al
so
st
or
es
the
dat
e
w
hen
a
ne
w
i
tem
is
create
d, ho
w
m
any tim
es an
it
e
m
is u
sed
sinc
e it
s
creati
on
a
nd the a
ve
rag
e
tim
e o
f
the
day
when i
t i
s u
se
d.
2.2.
Te
mpera
tu
re
Feedb
ack Clo
sed
-
Lo
op S
ys
t
em
Af
te
r
the
ta
rg
e
t
foo
d
te
m
per
at
ur
e
is
set
,
the
pro
posed
m
icr
owa
ve
ov
e
n
c
on
ti
nues
to
he
at
the
f
oo
d
it
e
m
un
ti
l
it
re
aches
the
ta
rg
e
t
tem
per
at
ur
e.
On
ce
t
he
ta
r
ge
t
tem
per
at
ur
e
i
s
reac
hed,
the
heati
ng
is
tur
ne
d
of
f
autom
at
ic
ally.
Ever
y
physi
cal
obje
ct
ra
diate
s
inf
rare
d
(
IR)
wa
ves
pro
por
ti
on
al
to
it
s
te
m
per
at
ur
e.
T
he
f
oo
d
tem
per
at
ur
e
is
m
easur
ed
us
in
g
a
non
-
c
onta
ct
IR
tem
per
at
ur
e
sens
or
[
26]
.
The
sens
or
is
m
ou
nted
on
th
e
ou
te
r
side
of
the
m
ic
rowa
ve
ov
e
n
cavit
y
roo
f
th
r
ough
a
4
-
m
m
ho
le
.
T
he
m
ic
ro
wa
ve
has
a
wav
el
e
ng
t
h
of
arou
nd
120
m
m
[2
7]
and
i
nfrar
e
d
has
a
wa
velen
gth
i
n
the
range
of
1
m
m
-
75
0
nm
[2
8].
As
t
he
m
ic
ro
wa
ves
ar
e
long
wav
e
s,
t
hey
do
no
t
pas
s
th
e
4
-
m
m
ho
le
a
nd
do
no
t
dam
age
t
he
se
nsor
.
The
IR
wa
ves
can
pass
th
rou
gh
the
ho
le
a
nd th
us t
he fo
od tem
per
at
ur
e ca
n be s
e
ns
e
d.
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N:
20
88
-
8708
Smart Micr
ow
ave
Oven w
it
h Image
Cl
as
sif
ic
ation a
n
d.
.
..
(
Ta
re
q
K
han)
4247
3.
PROT
OT
YP
E IMPLE
MENTATI
ON
An
em
bed
de
d
syst
e
m
is
i
m
pl
e
m
ented
in
a
c
omm
ercial
m
icr
owa
ve
ov
e
n
[
29
]
a
fter
disc
onnecti
ng
it
s
m
echan
ic
al
ti
m
ing
m
echan
ism
.
The
hard
war
e
a
nd
the
f
irm
war
e
par
t
of
the
prototype
are
bri
efly
de
scribe
d
belo
w.
3.1.
Ha
rdw
are
De
sign
The
blo
c
k
diag
ram
of
the
ha
r
dw
a
re
i
s
s
how
n
in
Fig
ure
12.
The
si
ng
le
board
c
om
pu
te
r,
Ra
sp
be
rr
y
P
i
(RPi)
v3
[
30
]
,
is
us
ed
as
the
process
or.
A
s
im
age
processi
ng
nee
ds
high
m
e
m
or
y
and
proces
sin
g
spe
e
d
[
31]
,
the
RPi
is
ch
os
e
n
over
othe
r
m
ic
ro
con
tr
ollers
s
uch
as
Adva
nced
Vi
rtual
RISC
(
AV
R
)
an
d
Pe
rip
her
al
In
te
r
face
C
on
t
ro
ll
er
(PIC)
w
hich
has
le
sse
r
m
e
m
or
y
capa
ci
ty
and
s
peed.
An
RPi
H
D
c
a
m
era
m
od
ule
[32]
is
m
ou
nted
at
the
center
of
the
roof
of
the
m
i
crowa
ve
oven
prototype
th
rough
a
4
-
m
m
ho
le
.
To
get
a
sh
ar
per
i
m
age,
the
fo
c
us
le
ng
t
h
is
a
dju
ste
d
to
the
heig
ht
of
the
oven
ca
vity
wh
ic
h
is
15
c
m
.
A
non
-
co
nt
act
IR
tem
per
at
ur
e
se
ns
or
[
33]
is
m
ounted
sli
ghtl
y
away
from
the
center
of
the
roof
of
the
oven
t
hroug
h
a
4
-
mm
ho
le
.
The
se
nsor
has
a
fiel
d
of
view
(FO
V)
of
5
degrees
a
nd
di
sta
nce
f
ro
m
the
sens
or
t
o
the
tur
ntable
tray
is
15
cm
.
Thu
s
,
the
se
ns
or
get
s
the
ave
rag
e
t
e
m
per
at
ure
of
a
ci
rc
ular
a
rea
hav
i
ng
a
ra
dius
of
15
×
tan
(
5)
=
1.3
c
m
.
Wh
e
n
the
tur
ntable
tray
r
otate
s,
the
sen
so
r
gets
the
te
m
per
at
ur
e
of
di
ff
ere
nt
port
io
ns
of
the
f
ood
as
it
is
placed
sli
gh
tl
y
away
from
the
center
of
t
he
oven
cavit
y.
T
he
sens
or
is
inte
rf
ace
d
with
t
he
RPi
us
i
n
g
tw
o
wire
I2
C
protoc
ol.
A
16×
2
cha
rac
te
r
li
qu
id
c
rysta
l
disp
la
y
(LC
D)
[34]
is
interface
d
with
t
he
RPi
us
in
g
un
iversal
asy
nchron
ous
r
ecei
ver
-
t
ransm
it
te
r
(U
ART
)
prot
oco
l.
T
o
i
nteract
with
the
us
er
,
the
desig
n
co
ntains
fou
r
push
-
bu
tt
on
keys
la
beled
Up
,
D
ow
n
,
Sta
rt,
an
d
Stop
.
A
buzzer
is
inclu
ded
in
the
desig
n
to
gen
e
rate
beep
so
un
ds
.
On
e
of
t
he
door
s
witc
hes
of
t
he
oven
is
c
onnected
wit
h
an
inter
rupt
pin
of
RPi
,
an
d
t
he
oth
e
r
door
s
wi
tc
h
i
s
connecte
d
to
t
he
path
of
the
AC
ci
rcu
it
so
that
cur
re
nt
can
only
flow
wh
e
n
the
door
is
cl
os
ed.
A
dayl
igh
t
wh
it
e
LE
D
bulb
[
35
]
is
conne
ct
ed
thr
ough
th
e
con
ta
ct
s
of
a
sing
le
pole
sin
gle
thr
ough
no
rm
ally
op
en
(SPST
-
NO)
relay
[
36]
.
A
so
li
d
-
sta
te
relay
(S
SR)
[37]
is
us
e
d
to
turn
on/o
ff
th
e
m
agn
et
r
on
c
ircuit
for
gen
e
rati
ng
m
ic
ro
wa
ve
a
nd
t
he
m
oto
r
c
onnecte
d
t
o
the
turntable
tray
.
Th
e pow
e
r
s
up
ply
f
or
the
RPi
boar
d
an
d
t
he
ca
m
era
is
su
ppli
ed
us
i
ng
a
11
0V
AC
to
5.
1V
DC
a
dap
te
r
[38].
T
he
RPi
bo
a
rd
gen
e
rates
3.3
V
DC
an
d
it
is
us
ed
to
powe
r
the
LC
D,
IR tem
per
at
ur
e
senso
r,
a
nd
buzzer
.
R
P
i
L
C
D
R
P
i
c
a
m
e
r
a
K
e
y
p
a
d
I
R
t
e
m
p
e
r
a
t
u
r
e
s
e
n
s
o
r
B
u
z
z
e
r
A
C
B
u
l
b
M
o
t
o
r
T
o
m
a
g
n
e
t
r
o
n
c
i
r
c
u
i
t
S
S
R
R
e
l
a
y
D
o
o
r
s
w
i
t
c
h
D
o
o
r
s
w
i
t
c
h
Fig
ure
12. Blo
ck diag
ram
o
f
t
he har
dware
3.2.
Fir
mw
are D
e
s
ign
A
Deb
ia
n
-
base
d
Li
nux
operat
ing
syst
em
,
Rasp
bia
n
[
30
]
,
is
instal
le
d
on
a
16
GB
S
D
ca
r
d
of
the
RPi
bo
a
r
d.
The
fir
m
war
e
fo
r
the
pro
po
se
d
m
ic
r
ow
a
ve
is
de
velop
e
d
in
Pyt
ho
n
la
ngua
ge.
T
he
fi
rm
war
e
is
bu
il
t
on
two
la
ye
rs
-
the
dr
i
ver
la
ye
r
a
nd
t
he
a
pp
li
ca
ti
on
la
ye
r.
T
he
dr
i
ver
la
ye
r
consi
sts
of
l
ow
-
le
vel
firm
war
e
f
or
acce
ssing
diff
e
ren
t
ha
rdwar
e
per
i
ph
e
rals.
Th
e
ap
plica
ti
on
l
ay
er
acce
ss
the
ha
rdwa
re
by
c
al
li
ng
the
f
unc
ti
on
s
of the
dr
i
ver la
ye
r.
A pseud
oc
od
e
of t
he
a
p
pl
ic
at
ion
lay
er is
sh
ow
n
in
Fi
g
ure
13.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
8
, N
o.
6
,
Dece
m
ber
2
01
8
:
4239
-
4252
4248
On DoorClo
se
E
vent
:
BulbOn()
I
:= Ca
pt
ure
Im
a
ge(
)
BulbOff
C
:= Cla
ss
if
y
Im
age
(I)
If
(C
=
NO
T EMP
TY)
Tr
:
=
Ge
t
Requi
red
Te
m
p(C)
BulbOn()
Hea
tOn()
isF
ini
sh :
=
False
while
(NO
T
isF
ini
sh)
FIF
O.a
dd
(Rea
d
IRTe
m
p())
Tc
:= FIF
O.a
vg
()
Displa
y
LCD
()
Butt
onInput
()
wait
(SA
MP
LE
_
RATE_DEL
AY
)
is_food_rea
d
y
:= (T
c
>
= Tr)
isF
ini
sh :
=
is_fo
od_re
ad
y
OR
is_
stop_pressed
OR i
s_door_openne
d
BulbOff()
Hea
tOff()
if
is_food_re
ad
y
Pla
y
Bu
zzer()
Fig
ure
13. Pse
udoc
od
e
for ap
plica
ti
on
lay
er
4.
RESU
LT
S
AND DI
SCUS
S
ION
4.1.
Simul
at
i
on
Re
sults
The
pro
pose
d
im
age
cl
assifi
cat
ion
an
d
te
m
p
eratur
e
reco
m
m
end
at
ion
al
gorithm
,
as
discuss
e
d
in
Sec.
2.1,
has
been
i
m
ple
m
ented
in
MATLAB
f
or
si
m
ulati
on
on
a
per
s
on
al
co
m
pu
te
r
(P
C).
I
n
the
sim
ulatio
n,
26
3
i
m
ages
of
s
olid
f
ood,
53
i
m
a
ges
of
li
qu
i
d
f
ood,
a
nd
10
e
m
pt
y
i
m
ages
(i.e.
w
he
n
no
f
ood
i
n
the
m
ic
r
ow
a
ve
)
wer
e
us
ed
.
Al
l
i
m
ages
wer
e
captur
e
d
by
the
cam
era
mo
unte
d
on
t
he
top
of
the
m
ic
rowav
e
.
Dif
f
eren
t
scenari
os
wer
e
create
d
with
diff
e
re
nt
foo
ds
placed
in
dif
f
eren
t
c
on
ta
ine
r
s.
Thi
rteen
dif
fer
e
nt
ty
pes
of
so
li
d
foo
ds
an
d
3
plate
s
hav
in
g
a
c
olor
of
wh
it
e,
blu
e,
a
nd
red
wer
e
us
ed
.
Eac
h
so
li
d
foo
d
w
as
placed
on
e
ach
ty
pe
of
t
he
plate
s
in
tur
n.
T
he
n
the
or
ie
ntati
on
of
the
p
la
te
wa
s
c
hange
d
a
nd
t
he
foo
d
wa
s
sti
rred
be
f
or
e
putt
ing
it
into
t
he
m
ic
ro
wav
e
s
o
t
hat
the
sam
e
foo
d
i
s
in
a
diff
e
re
nt
sit
uation
eac
h
tim
e.
Nine
di
f
fer
e
nt
ty
pes
of
dri
nks
(i.e.
li
qu
i
d
f
ood)
wer
e
al
s
o
use
d
in
the
te
st.
Fig
ure
14
sho
ws
so
m
e
sa
m
ple
i
m
ages
of
the
f
oods
us
ed
in
the
te
sts.
Durin
g
t
he
si
m
ula
ti
on
,
t
he
ta
rg
et
te
m
per
at
ur
e
for
eac
h
t
ype
of
foo
d
w
as
assig
ne
d
by
ge
ner
at
in
g
norm
al
l
y
distribu
te
d
ra
ndom
nu
m
ber
s
ha
vin
g
a
m
ean
(
μ
)
of
10
0
a
nd
sta
nd
a
rd
de
via
ti
on
(σ
)
of
15
.
The
accuracy
of
th
e
i
m
age
cl
assif
ic
a
ti
on
an
d
te
m
per
at
ur
e
rec
omm
end
at
io
n
al
gorithm
is
s
how
n
in
Table
1.
Th
e
pro
po
se
d
al
gor
it
h
m
can
disti
nguis
h
betwee
n
an
em
pty
and
a
m
ic
ro
wa
ve
oven
with
f
ood
w
it
h
100%
accu
r
acy
.
It
can
al
s
o
cl
a
ssify
w
hethe
r
the
f
ood
is
so
l
id
or
li
quid
ha
ving
a
n
acc
uracy
of
100
%.
The
n
the
al
go
rithm
rem
ov
es
the
ba
ckgr
ound
f
rom
the
i
m
age,
gen
e
rate
hist
ogram
s,
cl
assify
the
i
m
age,
an
d
rec
omm
end
a
ta
rg
et
tem
per
at
ur
e
.
I
t
can
reco
m
m
e
nd
the
e
xact
ta
rg
et
te
m
per
at
ur
e
with
an
acc
ur
acy
of
86.
31%
for
so
li
d
an
d
10
0%
for
li
qu
i
d
f
ood.
As
the
al
go
rithm
is
le
arn
in
g,
it
need
s
to
detect
new
f
ood
it
e
m
s.
W
he
n
the
al
gorithm
cl
as
sifie
s
a
new
foo
d
im
age
as
one
of
the
f
ood
it
e
m
s
that
are
al
read
y
in
the
da
ta
base,
it
is
co
ns
ide
red
a
s
in
correct
detect
ion
of
ne
w
f
ood.
Co
nve
rsely
,
w
he
n
the
al
gorithm
cl
as
sifie
s
a
f
ood
i
m
age
that
is
al
read
y
in
the
da
ta
base
as
a
new
foo
d
it
e
m
,
i
t
is
con
sidere
d
as
inc
orrect
detect
ion
of
new
foo
d.
The
pr
opos
e
d
al
gorithm
can
detect
new f
ood i
tems ha
ving a
n
acc
ur
acy
of 89.
35
% for
so
li
d an
d 9
2.45% f
or li
quid
f
ood.
Evaluation Warning : The document was created with Spire.PDF for Python.