Method for detecting diaper wetness based on radio signal technology
A method for detecting diaper wetness based on a ratio signal technology comprises: sending a frequency-modulated continuous wave signal to a diaper through a radar device to obtain a transmission signal S T (t) and a return signal S R (t); detecting a static environment factor and a dynamic environment factor, and obtaining continuous radar snapshots; eliminating interference of environmental factors by using a filtering template based on the continuous radar snapshots and object respiration; obtaining an outline of the diaper by using wavelet synchronous compression transformation, and dividing into a training sample set and a testing sample set; and inputting the training sample set and the test sample set into a heuristic mobility network for training until a loss function is converged, obtaining the trained heuristic mobility network, and obtaining diaper wetness of the test sample set.
1 . A method for detecting diaper wetness based on a radio signal technology, comprising the following steps:
sending a frequency-modulated continuous wave signal to a diaper through a radar device to obtain a transmission signal S T (t) and a return signal S R (t) of the radar device;
detecting a static environment factor and a dynamic environment factor based on the transmission signal S T (t) and the return signal S R (t), and obtaining continuous radar snapshots, wherein obtaining the continuous radar snapshots comprises: multiplying the transmission signal S T (t) and the return signal S R (t), and filtering a part with the highest frequency to obtain a signal S M1 (t); detecting the static environment factor through range-fast Fourier transform to obtain a signal S M2 (t), wherein S M2 (t)=rangeFFT(S M1 (t)); and detecting the static environment factor and the dynamic environment factor through Doppler-fast Fourier transform to obtain the continuous radar snapshots CRS(t), wherein CRS(t)=dopplerFFT(S M2 (t 1 , t 2 , . . . , t 64 ));
eliminating interference of the static environment factor and the dynamic environment factor by using a filtering template based on the continuous radar snapshots and an individual's breath to obtain interference-eliminated continuous radar snapshots CRS″, wherein eliminating interference comprises: filtering multipath interference by collecting dot products of a CRS 0 filtering template on an empty bed and signals of the continuous radar snapshots: CRS′=CRS*CRS 0 ; and filtering noise and interference of the static environment factor and the dynamic environment factor by collecting dot products of a CRS filtering template in which a user lies on a bed without wearing the diaper and signals of the continuous radar snapshots: CRS″=CRS′*CRS b ; wherein CRS 0 is CRS of the empty bed, and CRS b is CRS in which the user lies on the bed without wearing the diaper;
obtaining based on the continuous radar snapshots after eliminating the interference, an outline of the diaper by using wavelet synchronous compression transformation, and dividing the outline of the diaper into a training sample set and a testing sample set; and
inputting the training sample set and the testing sample set into a heuristic mobility network for training until a loss function is converged, obtaining a trained heuristic mobility network, and obtaining diaper wetness of the testing sample set in feature presentation, wherein inputting comprises: extracting feature representation by using a Resnet network; representing a migration difference between the training sample set and the testing sample set through the heuristic mobility network; and eliminating the migration difference by using a generative adversarial network mechanism until the loss function of the heuristic mobility network is converged to obtain the trained heuristic mobility network;
wherein the Resnet network is a ResNet34 network, and wherein the generative adversarial network mechanism is implemented by an adversarial network (AdNet) coupled to the extracted feature representation.
2 . The method according to claim 1 , wherein expressions of the transmission signal S T (t) and the return signal S R (t) of the radar device are as follows:
S
T
(
t
)
=
cos
(
∫
0
t
f
(
t
′
)
dt
)
=
cos
(
2
π
(
f
c
t
+
B
s
t
2
2
T
s
)
)
;
S
R
(
t
)
=
S
T
(
t
-
τ
)
=
cos
(
2
π
(
f
c
(
t
-
τ
)
+
B
s
(
t
-
τ
)
2
2
T
s
)
)
;
wherein f c is a start frequency, B s represents a scanning bandwidth, T s represents scanning time, τ represents reflection delay of the transmission signal, and attenuation of an amplitude is ignored.
3 . The method according to claim 2 ,
S T (t) S R (t) S M1 (t)
S
M
1
(
t
)
=
cos
(
2
π
(
f
c
τ
-
B
s
(
τ
2
-
2
t
τ
)
2
T
s
)
)
S M2 (t)
S
M
2
(
t
)
=
rangeFFT
(
S
M
1
(
t
)
)
CRS(t)
CRS
(
t
)
=
dopplerFFT
(
S
M
2
(
t
1
,
t
2
,
…
,
t
6
4
)
)
wherein S M2 (t 1 , t 2 , . . . , t 64 ) represents a time sequence of S M2 .
4 . The method according to claim 1 , wherein the step of obtaining the outline of the diaper comprises the following steps:
eliminating smearing on a time-frequency diagram of the diaper by using the wavelet synchronous compression transformation to obtain a first representation of the outline of the diaper;
Wsst
f
(
w
b
,
τ
)
=
(
Δ
w
)
-
1
∑
a
k
:
A
W
f
(
a
k
,
τ
)
a
k
-
3
2
(
Δ
a
)
k
;
wherein w b is a central value of
[
w
b
-
1
2
Δ
w
,
w
b
+
1
2
Δ
w
]
,
τ represents a resolution time domain, Δw=w b −w b-1 , a k represents a discrete value of the scale, and A represents
❘
"\[LeftBracketingBar]"
w
(
a
k
,
τ
)
-
w
b
❘
"\[RightBracketingBar]"
≤
Δ
w
2
;
W
f
=
a
k
-
1
2
∫
R
f
(
r
)
ψ
(
t
-
τ
a
k
)
dt
,
f (t) represents a signal corresponding to a wet diaper, ψ( ) represents a wavelet basis function, (Δa) k =a k −a k-1 ;
refining a time-frequency representation through a plurality of wavelet synchronous compression transformations to obtain a second representation of the outline of the diaper;
Wsst
[
M
]
(
w
l
,
τ
)
=
∫
-
∞
+
∞
Wsst
f
(
w
b
,
τ
)
δ
(
w
l
-
w
~
b
[
M
]
(
w
b
,
τ
)
)
dw
b
;
wherein w 1 represents an instantaneous frequency, δ( ) represents a Kronecker Dirac function, and
w
~
b
[
M
]
(
w
b
,
τ
)
represents instantaneous frequency estimation of Wsst [M] section;
limiting a value of M by the following constraint conditions;
∫
-
∞
+
∞
∫
-
∞
+
∞
❘
"\[LeftBracketingBar]"
w
~
b
[
M
]
(
w
b
,
τ
)
-
w
~
b
[
M
-
1
]
(
w
b
,
τ
)
❘
"\[RightBracketingBar]"
d
τ
dw
b
<
ε
;
wherein ε is set to 1e-8.
5 . The method according to claim 1 ,
wherein the loss function L(Ttm)—of the heuristic mobility network is:
L
(
Ttm
)
=
L
Rem
+
L
Pro
;
where
L
Rem
=
E
x
i
n
∼
D
N
log
D
(
F
i
)
+
E
x
j
a
∼
D
A
log
[
1
-
D
(
F
i
)
]
;
L
Pro
=
L
CrossEntropy
(
F
i
-
H
(
x
i
)
,
y
i
)
;
wherein L Rem represents a loss function to eliminate the migration difference, L Pro represents a loss function to predict diaper wetness; F i represents extracted features, H(x i ) represents the migration difference,
E
x
i
n
∼
D
N
represents expectation of a source domain, D(F i ) represents distribution of the extracted features,
E
x
j
a
∼
D
A
represents expectation of a target domain, L CrossEntropy represents calculating cross entropy of two variables, y i represents sample labels,
x
i
n
represents a signal from the source domain, D N represents a source domain set,
x
j
a
represents a signal from the target domain, and D A represents a target domain set.