Signal feature extraction apparatus, signal feature extraction method, and program
The purpose of the present invention is to accurately identify an appearance position of a desired waveform pattern. A signal feature extraction apparatus extracts a feature peak value of a waveform pattern from an output signal of a feature extraction filter that extracts a waveform pattern of an input signal. A local maximum value detection part 1 stores a time index and a value of the output signal when an absolute value of the output signal is equal to or more than a predetermined threshold value and takes a local maximum value. A feature peak value selection part 2 selects a feature peak value on the basis of a difference between a time interval between the stored time index and a time index of a feature peak value selected immediately before and an average time interval at which the feature peak value can occur, and a difference between a logarithmic amplitude and phase of the output signal corresponding to the stored time index and an average logarithmic amplitude and average phase that the feature peak value can take.
1. A signal feature extraction apparatus for extracting a feature peak value of a waveform pattern from an output signal of a feature extraction filter that extracts a waveform pattern of an input signal,
wherein the feature extraction filter is a linear or nonlinear filter that outputs as the output signal a complex signal having a real part and an imaginary part mutually convertible or approximately convertible by Hilbert transform,
the signal feature extraction apparatus comprising:
a local maximum value detection part that stores a time index and a value of the output signal when an absolute value of the output signal is equal to or more than a predetermined threshold value and takes a local maximum value;
a feature peak value selection part that selects the feature peak value on the basis of a difference between a time interval between the stored time index and a time index of a feature peak value selected immediately before and an average time interval at which the feature peak value can occur, and a difference between a logarithmic amplitude and phase of the output signal corresponding to the stored time index and an average logarithmic amplitude and average phase that the feature peak value can take; and
a local maximum value rejection part that rejects the stored value of the output signal except the value of the output signal selected as the feature peak value.
2. The signal feature extraction apparatus according to claim 1 , wherein
the feature peak value selection part selects an absolute value of the output signal corresponding to a time index calculated by a following formula as the feature peak value:
argmin
m
λ
[
w
1
m
λ
-
n
0
-
Δ
T
2
Δ
T
2
+
w
2
log
(
y
(
m
λ
)
)
-
log_A
2
+
w
3
angle
(
y
(
m
λ
)
/
e
j
Φ
)
2
w
4
log_A
2
+
w
5
Φ
2
]
where, N denotes a number of the stored time indexes, m 1 , . . . , m N denote the stored time indexes, y(m 1 ), . . . , y(m N ) denote the output signals corresponding to the time indexes m 1 , . . . , m N , λ denotes each integer of 1 to N, ΔT denotes the average time interval, log_A denotes the average logarithmic amplitude, Φ denotes the average phase, w 1 , w 2 , w 3 , w 4 , and w 5 denote predetermined weighting factors, angle (⋅) denotes a phase angle of a complex number ⋅, and n 0 denotes an initial value of the time index.
3. The signal feature extraction apparatus according to claim 1 , wherein
the input signal is an electrocardiogram, and
the feature peak value corresponds to the waveform pattern indicating a heartbeat.
4. A signal feature extraction apparatus for extracting a feature peak value of a waveform pattern from an output signal of a feature extraction filter that extracts a waveform pattern of an input signal,
wherein the feature extraction filter is a linear or nonlinear filter that outputs as the output signal a complex signal having a real part and an imaginary part mutually convertible or approximately convertible by Hilbert transform,
the signal feature extraction apparatus comprising:
a local maximum value detection part that stores a time index and a value of the output signal when an absolute value of the output signal is equal to or more than a predetermined threshold value and takes a local maximum value;
a feature peak value selection part that selects the feature peak value on the basis of a difference between a time interval between the stored time index and a time index of a feature peak value selected immediately before and an average time interval at which the feature peak value can occur, a difference between an amplitude of the output signal corresponding to the stored time index and an average amplitude the feature peak value can take, and a difference between a phase of the output signal corresponding to the stored time index and an average phase the feature peak value can take; and
a local maximum value rejection part that rejects the stored value of the output signal except the value of the output signal selected as the feature peak value.
5. The signal feature extraction apparatus according to claim 4 , wherein
the input signal is an electrocardiogram, and
the feature peak value corresponds to the waveform pattern indicating a heartbeat.
6. A non-transitory computer-readable recording medium on which a program recorded thereon for causing a computer to function as the signal feature extraction apparatus according to any of claim 1 , 2 , or 4 .
7. A signal feature extraction method for extracting a feature peak value of a waveform pattern from an output signal of a feature extraction filter that extracts a waveform pattern of an input signal,
wherein the feature extraction filter is a linear or nonlinear filter that outputs as the output signal a complex signal having a real part and an imaginary part mutually convertible or approximately convertible by Hilbert transform,
the signal feature extraction method comprising:
a local maximum value detection step in which a local maximum value detection part stores a time index and a value of the output signal when an absolute value of the output signal is equal to or more than a predetermined threshold value and takes a local maximum value;
a feature peak value selection step in which a feature peak value selection part selects the feature peak value on the basis of a difference between a time interval between the stored time index and a time index of a feature peak value selected immediately before and an average time interval at which the feature peak value can occur, and a difference between a logarithmic amplitude and phase of the output signal corresponding to the stored time index and an average logarithmic amplitude and average phase that the feature peak value can take; and
a local maximum value rejection step in which a local maximum value rejection part rejects the stored value of the output signal except the value of the output signal selected as the feature peak value.
8. The signal feature extraction method according to claim 7 , wherein
the feature peak value selection step selects an absolute value of the output signal corresponding to a time index calculated by a following formula as the feature peak value:
argmin
m
λ
[
w
1
m
λ
-
n
0
-
Δ
T
2
Δ
T
2
+
w
2
log
(
y
(
m
λ
)
)
-
log_A
2
+
w
3
angle
(
y
(
m
λ
)
/
e
j
Φ
)
2
w
4
log_A
2
+
w
5
Φ
2
]
where, N denotes a number of the stored time indexes, m 1 , . . . , m N denote the stored time indexes, y(m 1 ), . . . , y(m N ) denote the output signals corresponding to the time indexes m 1 , . . . , m N , λ denotes each integer of 1 to N, ΔT denotes the average time interval, log_A denotes the average logarithmic amplitude, Φ denotes the average phase, w 1 , w 2 , w 3 , w 4 , and w 5 denote predetermined weighting factors, angle (⋅) denotes a phase angle of a complex number ⋅, and n 0 denotes an initial value of the time index.
9. The signal feature extraction method according to claim 7 , wherein
the input signal is an electrocardiogram, and
the feature peak value corresponds to the waveform pattern indicating a heartbeat.
10. A signal feature extraction method for extracting a feature peak value of a waveform pattern from an output signal of a feature extraction filter that extracts a waveform pattern of an input signal,
wherein the feature extraction filter is a linear or nonlinear filter that outputs as the output signal a complex signal having a real part and an imaginary part mutually convertible or approximately convertible by Hilbert transform,
the signal feature extraction method comprising:
a local maximum value detection step in which a local maximum value detection part stores a time index and a value of the output signal when an absolute value of the output signal is equal to or more than a predetermined threshold value and takes a local maximum value;
a feature peak value selection step in which a feature peak value selection part selects the feature peak value on the basis of a difference between a time interval between the stored time index and a time index of a feature peak value selected immediately before and an average time interval at which the feature peak value can occur, a difference between an amplitude of the output signal corresponding to the stored time index and an average amplitude the feature peak value can take, and a difference between a phase of the output signal corresponding to the stored time index and an average phase the feature peak value can take; and
a local maximum value rejection step in which a local maximum value rejection part rejects the stored value of the output signal except the value of the output signal selected as the feature peak value.
11. The signal feature extraction method according to claim 10 , wherein
the input signal is an electrocardiogram, and
the feature peak value corresponds to the waveform pattern indicating a heartbeat.