AU2020287677A1 - Sound source position estimation method, readable storage medium and computer device - Google Patents
Sound source position estimation method, readable storage medium and computer device Download PDFInfo
- Publication number
- AU2020287677A1 AU2020287677A1 AU2020287677A AU2020287677A AU2020287677A1 AU 2020287677 A1 AU2020287677 A1 AU 2020287677A1 AU 2020287677 A AU2020287677 A AU 2020287677A AU 2020287677 A AU2020287677 A AU 2020287677A AU 2020287677 A1 AU2020287677 A1 AU 2020287677A1
- Authority
- AU
- Australia
- Prior art keywords
- signal
- window
- signals
- sound source
- length
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G01S3/00—Direction-finders for determining the direction from which infrasonic, sonic, ultrasonic, or electromagnetic waves, or particle emission, not having a directional significance, are being received
- G01S3/80—Direction-finders for determining the direction from which infrasonic, sonic, ultrasonic, or electromagnetic waves, or particle emission, not having a directional significance, are being received using ultrasonic, sonic or infrasonic waves
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G01S5/00—Position-fixing by co-ordinating two or more direction or position line determinations; Position-fixing by co-ordinating two or more distance determinations
- G01S5/18—Position-fixing by co-ordinating two or more direction or position line determinations; Position-fixing by co-ordinating two or more distance determinations using ultrasonic, sonic, or infrasonic waves
- G01S5/20—Position of source determined by a plurality of spaced direction-finders
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Remote Sensing (AREA)
- Radar, Positioning & Navigation (AREA)
- Theoretical Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Measurement Of Velocity Or Position Using Acoustic Or Ultrasonic Waves (AREA)
Abstract
Disclosed is a sound source position estimation method. The method comprises: a single-vector hydrophone receiving multichannel signals sent by a sound source in the sea; fusing the received multichannel signals into an instant single-channel sound intensity signal by means of combined sliding of a fixed window and a dynamic window, dividing the instant single-channel sound intensity signal into signal segments containing a sufficient amount of information, and with the premise that the information amount is guaranteed, reducing data quantity, and improving the operating speed; performing self-supplementation on the signal by means of an expectation maximization algorithm, and expanding distances between the various signal segments, such that the resolution between the signal segments is improved, and information lost during interception in the previous step is supplemented to a certain extent; and estimating the position of the sound source by means of a recurrent neural network and using self-supplemented equal-length signals. According to the method, only the single-vector hydrophone is needed to collect signals, such that not only are arrangement difficulty and usage costs simplified, but the application range is also expanded. Further disclosed are a readable storage medium and a computer device.
Description
Technical Field
The present disclosure relates to the field of sound source position estimation technologies, and in particular, to a sound source position estimation method, a readable storage medium, and a computer device.
Related Art
The description in this section merely provides background information related to the present disclosure, and does not necessarily constitute the prior art.
With rapid and continuous growth in economy in China, population also continuously increases, and people's demands for resources and requirements for consumption levels also increase greatly accordingly. People are developing and utilizing the limited resources on the land to the greatest extent, and are also facing resource shortage at the same time. Therefore, new fields and new resources need to be further researched and developed on the basis of maximizing the application of existing resources. The ocean area accounts for more than 70% of the surface area of the earth, and therefore, it is crucial to develop and utilize marine resources. In recent years, countries in the world are quite fiercely competing for marine resources. How to develop and utilize marine resources to make full use of the resources has become a major issue that attracts much attention in recent years. China is in a unique position, with a total national territorial area of approximately 9.6 million square kilometers, of which the ocean area accounts for approximately 3% of the total area, and most of the ocean waters are shallow sea waters. Therefore, technical research on shallow sea waters will be of greater significance for human survival and development.
Compared with the deep sea environment, the temporal and spatial variability and uncertainty of the shallow sea environment have a severer impact on signal propagation. In addition, both reflected signals from the shallow sea floor and human activities in the shallow sea cause aliasing of a target signal, further affecting sound source position estimation. Therefore, sound source target estimation in the shallow sea environment has been a research difficulty in this field currently. Compared with a conventional sound pressure hydrophone, a vector hydrophone can acquire a sound pressure signal and triaxial acoustic particle velocity signals in orthogonal directions at a common point, and is of great practical value.
The inventor of the present disclosure finds during research that: currently, a single vector hydrophone is mostly used to estimate an azimuth direction of arrival (DOA) and a pitch DOA of a target, and to determine a position distance of the target, cross-estimation needs to be performed by using a vector hydrophone array due to an inherent defect of the single vector hydrophone, namely, insufficient distance resolution. Moreover, in actual engineering application of the single vector hydrophone, on the one hand, it is difficult for an actual parameter to meet a requirement of an ideal electroacoustic parameter characteristic due to limitations of process conditions, restricting the accuracy of position estimation of the single vector hydrophone; on the other hand, the single vector hydrophone is easily influenced by the environment to cause uncertain changes in attitude, so that acquisition of a real position of the target is further influenced. Such reasons result in insufficient application of the single vector hydrophone in target position estimation.
To overcome the shortcomings in the related art, the present disclosure provides a sound source position estimation method, a readable storage medium, and a computer device. Compared with a conventional sound source estimation model in which a complex vector hydrophone array needs to be deployed to receive a signal, in the sound source estimation method, only a single vector hydrophone is required to acquire a signal, thereby reducing deployment difficulty and use costs, and expanding the application scope.
To achieve the foregoing objective, the present disclosure uses the following technical solutions:
According to a first aspect, the present disclosure provides a sound source position estimation method.
The sound source position estimation method includes the following steps:
receiving, by a single vector hydrophone, multichannel signals from a sound source in the ocean;
fusing the received multichannel signals into an instantaneous single-channel sound intensity signal through fixed-dynamic window combined sliding, and dividing the received multichannel signals into signal segments including a sufficient information amount, to reduce a data amount and improve an operation speed while ensuring the information amount;
performing signal self-supplement by using the expectation maximization algorithm, and increasing a distance between the signal segments, to improve a resolution between the signal segments and supplement information lost in the previous step of capture to some extent;
estimating, through a recurrent neural network, a position of the sound source by using equal-length signals obtained after the self-supplement.
In some possible implementations, the multichannel signals are four-channel signals, including three acoustic particle velocity signals in orthogonal directions: an acoustic
particle velocity v- in an x -axis direction, an acoustic particle velocity vY in a Y -axis
direction, and an acoustic particle velocity vz in a z -axis direction, and a scalar sound
pressure signal P .
As a further limitation, the multichannel signals are fused into the instantaneous single-channel sound intensity signal by using a fixed window, dynamic windows of all lengths are traversed, a steepest ascent section of a Shannon entropy is found, an optimum dynamic window is determined, the instantaneous single-channel sound intensity signal in the fixed window are dynamically captured as unequal-length signals based on the Shannon entropy by using the optimum dynamic window, and signal self-supplement is performed on the captured unequal-length signals by using the expectation maximization algorithm.
As a still further limitation, the dividing the received multichannel signals into signal segments including a sufficient information amount through fixed-dynamic window combined sliding specifically includes the following steps:
401: specifying a fixed window length if and an initial window start point t/A for the
acquired four-channel signals P, v ", and v
402: fusing, by using a fixed window W *f whose window length and start point are
respectively I and tf, , four-channel information in the window, to obtain an
instantaneous single-channel sound intensity signal whose length is if;
403: capturing a dynamic window Wd, whose window length and start point are
respectively 1 and / in the instantaneous single-channel sound intensity signal '
, where the length of the signal in the dynamic window d, is as short as possible provided that the signal includes a sufficient information amount; and
404: returning to 402, updating a start point tf of a fixed window f'+ based on a
signal overlap rate q, and performing a cyclic operation.
As a still further limitation, in step 402, synchronous sliding is performed in channel signals through a time window of a fixed size, signals are extracted, and the information is fused into the instantaneous single-channel sound intensity signal by using a cross-spectrum method; and the step specifically includes the following steps:
501: specifying a fixed window length / and a window start point t/A according to a signal fusion degree;
502: capturing signal segments of a window sizeif by using the same start point tf
in the signal channels of the sound pressure P and the vibration velocities Vv and
VZ W W W in the axial directions, where corresponding window signals are A, xr, e , and
pvr and
503: calculating, based on the cross-spectrum method, the instantaneous single-channel
sound intensity signal ', obtained after the window signals are fused, to implement multi-sensor information fusion, where a calculation formula of the instantaneous single-channel sound intensity signal obtained after the fusion is:
Re[S ] Re[S, ] Re[S ] I=p(t).(t)pRe[S(f)] V(Re[S (f)] 'W) (WRe[S (f]
where S P,, SP , and -= are respectively cross-spectrum functions of three
components x, y, and z, is a spectrum function of 2 is a frequency,
Re[ ] indicates Laplace transform, 0 and 9 are respectively a pitch direction of arrival (DOA) and an azimuth DOA of a sound source relative to a vector hydrophone, where the pitch DOA and the azimuth DOA respectively take the XOY plane and the x
-axis as 0°, and p(t) vx(t) vy,(t), and v(t) are respectively a sound pressure signal and acoustic particle velocity signals in different directions that are received by the vector hydrophone at a moment t.
As a still further limitation, in step 403, the dynamic windows of all lengths are
traversed for the instantaneous single-channel sound intensity signal ', in the fixed window, and the steepest ascent section of the Shannon entropy is found as the optimum
dynamic window d; and the step specifically includes the following steps:
601: traversing the entire fixed window 't of the captured instantaneous
single-channel sound intensity signal I starting from the start point tf of the fixed window, and calculating a Shannon entropy of signals of all lengths by using the following
formula, to construct a Shannon entropy signal S',
Shannon(X )=Ip(x,).-og =- p(x,). log p(x,) i=1 ~ p(x,) i
where X iis a possible value of a random event X, Shannon(X) is a Shannon entropy included in the random event X, m is a total quantity of random events, and
P(x1 is a probability of occurrence of xi
602: finding a steepest ascent section of S'i according to a derivation result S, of
S1 , marking the length as Si, and jumping to step 604;
603: marking the length as if no steepest ascent section is found in SI, W considering that the signal in the fixed window is an invalid signal or a noise signal, where the following two conditions are satisfied:
when S'i is relatively small, it is considered that the signal is a null signal or Shannon
entropy content of the signal in the fixed window is insufficient, and 10, where 0 is a preset minimum captured length; and
when S'i is relatively large, it is considered that the signal is a noise signal or a valid
signal including a relatively high Shannon entropy, and s71, where 11 is a preset maximum captured length;
604: capturing a signal segment whose length is starting from the start point tfi in
the fixed window '' as the dynamic window a', and marking an end time of the
window ast
As a still further limitation, signal self-supplement is performed by using the expectation maximization algorithm, segmented unequal-length signals are equivalent to
observation data X, supplemented equal-length signals are equivalent to complete data Y,
a supplementary signal is equivalent to unobserved data Z , and a result is iterated by using the expectation maximization algorithm to obtain a maximum value 6* of a parameter 0, that is, when a maximum value of a maximum likelihood function L(0) based on Y is obtained, optimal solutions of a mean u and a variance ' of a complete dataset are obtained, and an unknown dataset Z is obtained based on an observed dataset X, to supplement the complete dataset Y, which specifically includes the following steps:
701: assuming that an iteration quantity t = 0, initializing a parameter vector 0(0), 0
being a parameter vector formed by the mean and the variance of the dataset Y, and
calculating an initial maximum likelihood function
n
L(O)j = L (xj,---x,; 0) = p(x,; 0)-.... p(x,,;O = p (xi;0
) 702: obtaining Q (z) from 0t, and ensuring that when 0' is specified, the
equal sign in In(E(X))E[In(X)]is satisfied, to establish a lower bound of ;
Q z) 0(t L(O0) 703: fixing Qt)()and taking as a variable, taking a derivative of in
step 702, and obtaining (t+0 according to the formula
L (O(t1 )) I Qz) in =L (OW) ) ;and
704: if , ending the iterative calculation; otherwise, assuming
that= t +1, and returning to step 702, where a threshold 6 is a specified very small value;
where i represents a distribution of unknown data Z, P( 0(t)) is a
probability of occurrence of xz) under a condition 0 (t), the superscript i indicates an th value of a corresponding parameter, - is the threshold and is an initially specified small value used as a standard for ending the iteration, and E[ ] is a mathematical expectation.
As a still further limitation, the estimating, through a recurrent neural network, a position of the sound source by using equal-length signals obtained after the self-supplement specifically includes: using the supplemented signal segments as an input by using the expectation maximization algorithm, outputting azimuth DOAs and distances of the sound source in different signal segments, and performing cross-validation on estimation results of the different signal segments, to implement accurate positioning of the position of the sound source.
According to a second aspect, the present disclosure provides a computer-readable storage medium, storing a computer program, where when the program is executed by a processor, steps in the sound source position estimation method of the present disclosure are implemented.
According to a third aspect, the present disclosure provides a computer device, including a memory, a processor, and a computer program stored in the memory and capable of being run on the processor, where when the processor executes the program, steps in the sound source position estimation method of the present disclosure are implemented.
Compared with the prior art, the present disclosure has the following beneficial effects:
1. Compared with the conventional sound source estimation model, the sound source position estimation method of the present disclosure avoids a problem of deploying a complex vector hydrophone array to receive a signal. In the sound source estimation method of this application, only a single vector hydrophone is required to acquire a signal, thereby reducing deployment difficulty and use costs, and expanding the application scope.
2. In the sound source position estimation method of the present disclosure, a short-duration signal sample is divided into a large quantity of signal segments through dynamic window-fixed window combined sliding. Through mutual validation of the signal segments, the accuracy and stability of position estimation are improved. The data amount is reduced and the operation speed is improved while ensuring the information amount.
3. In the sound source position estimation method of the present disclosure, large samples are required to train a network only in an early stage, and complex operations are not required during use, thereby implementing real-time tracking of high-speed and highly maneuverable target trajectories.
FIG. 1 shows a relationship between a single vector hydrophone and a sound source position according to Embodiment 1 of the present disclosure.
FIG. 2 is a flowchart of a sound source position estimation method according to Embodiment 1 of the present disclosure.
FIG. 3 is a flowchart of fixed-dynamic window combined sliding according to Embodiment 1 of the present disclosure.
FIG. 4 is a curve graph of an estimation result of a position of an ultra-low frequency sound source according to Embodiment 1 of the present disclosure.
It should be noted that the following detailed descriptions are all exemplary and are intended to provide a further description of the present disclosure. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by a person of ordinary skill in the art to which the present disclosure belongs.
It should be noted that the terms used herein are merely used for describing specific implementations, and are not intended to limit exemplary implementations of the present disclosure. As used herein, the singular form is intended to include the plural form, unless the context clearly indicates otherwise. In addition, it should further be understood that terms "comprise" and/or "include" used in this specification indicate that there are features, steps, operations, devices, components, and/or combinations thereof.
Embodiment 1:
As shown in FIG. 1 and FIG. 2, Embodiment 1 of the present disclosure provides a sound source position estimation method, including the following steps:
receiving, by a single vector hydrophone, multichannel signals from a sound source in the ocean, where the multichannel signals are four-channel signals, including three acoustic
particle velocity signals in orthogonal directions: an acoustic particle velocity v- in an X
-axis direction, an acoustic particle velocity vy in a Y -axis direction, and an acoustic
particle velocity v in a z -axis direction, and a scalar sound pressure signal P ;
fusing the received multichannel signals into an instantaneous single-channel sound intensity signal through fixed-dynamic window combined sliding, and dividing the received multichannel signals into signal segments including a sufficient information amount, to reduce a data amount and improve an operation speed while ensuring the information amount;
performing signal self-supplement by using the expectation maximization algorithm, and increasing a distance between the signal segments, to improve a resolution between the signal segments and supplement information lost in the previous step of capture to some extent;
estimating, through a recurrent neural network, a position of the sound source by using equal-length signals obtained after the self-supplement.
The multichannel signals are fused into the instantaneous single-channel sound intensity signal by using the fixed window, dynamic windows of all lengths are traversed, a steepest ascent section of a Shannon entropy is found, an optimum dynamic window is determined, the instantaneous single-channel sound intensity signal in the fixed window are dynamically captured as unequal-length signals based on the Shannon entropy by using the optimum dynamic window, and signal self-supplement is performed on the captured unequal-length signals by using the expectation maximization algorithm.
As shown in FIG. 3, the dividing the received multichannel signals into signal segments including a sufficient information amount through fixed-dynamic window combined sliding specifically includes the following steps:
301: specifying a fixed window length If and an initial window start point tf for
the acquired four-channel signalsP, v, v , and
W 302: fusing, by using a fixed window f whose window length and start point are
respectively If and tf , four-channel information in the window, to obtain an
instantaneous single-channel sound intensity signal whose length is ;
303: capturing a dynamic window di whose window length and start point are 1i t W respectively s and f in the instantaneous single-channel sound intensity signal ',
where the length of the signal in the dynamic window di is as short as possible provided that the signal includes a sufficient information amount; and
304: returning to 302, updating a start point f, of a fixed window based on a
signal overlap rate q, and performing a cyclic operation.
In step 302, synchronous sliding is performed in channel signals through a time window of a fixed size, signals are extracted, and the information is fused into the instantaneous single-channel sound intensity signal by using a cross-spectrum method; and the step specifically includes the following steps:
401: specifying a fixed window length If and a window start point t ,according to a signal fusion degree;
402: capturing signal segments of a window size f by using the same start point tf
in the signal channels of the sound pressure P and the vibration velocities v- , and
v2 in the axial directions, where corresponding window signals are , and
Wvzi and
403: calculating, based on the cross-spectrum method, the instantaneous single-channel
sound intensity signal ', obtained after the window signals are fused, to implement multi-sensor information fusion, where a calculation formula of the instantaneous single-channel sound intensity signal obtained after the fusion is:
Re[SPV ] Re[Sp, ] Re[S ]_ Re[S (f)] Re[S (f)] Re[S (f)]
A specific derivation process of the instantaneous single-channel sound intensity signal is:
Assuming that a sound signal P propagates in an isotropic noise field, a vector
hydrophone Q receives the signal, and outputs of the vector hydrophone have the following relationships:
Soundpressure p(t)=p(t)+p(t) (2)
Acoustic particle velocity component x v(t)= v(t)+v,1(t) (3)
Acoustic particle velocity component Y v,(t)= v, (t)+ v,, (t) (4)
Acoustic particle velocity component z v,(t)=v.(t)+v'(t) (5)
In the foregoing formulas, subscripts "s " and "n " respectively represent signal and noise. If noise sources are independent of each other and a mean is 0, a sound intensity in the x direction is:
I= (t)-v,(t) =([p'(t) + p" (01 W -(, ++v,,(t)]
= p,(t)-v.(t)±+ p,(t)-v,.(t)+p,(t)-v,(t)+p,(t)-v,.(t) =pS(t)-v, (t) (6)
Similarly:
=p,(t) vy(t) '2 P, t).v2 }t) (7)
It can be learned from the foregoing formula that the sound intensity obtained from the
outputs P, v , and vz of the vector hydrophone does not include noise energy, that is, the vector hydrophone can resist an isotropic noise.
At a moment t, a sound pressure signal and acoustic particle velocity signals in
different directions that are received by the vector hydrophone are respectively p(t)
v,(t) v,(t),and v,(t)
An approximate spatial position of the target is estimated by using the cross-spectrum method. First, a cross-correlation operation is performed on the sound pressure P and each of the acoustic particle velocity components, to obtain the following cross-correlation functions:
R =0p(t)vt-r)dt=0p(t)p(t-r)cosOcospdt
R =(t)v,t-r)dt=JT p(t)p(t-r)cos0sinqdt
T R = p(t)v{t-r)dt=JT p(t)p(t-r)sinOdt (8)
Then, Fourier transform is performed on the foregoing cross-correlation functions, to obtain cross-spectrum functions of the cross-correlation functions:
S =S 2(f)cosOcosqP SP =SP f cos0sinq
sp = S (f)sin0 (9)
Herein, is a spectrum function ofP 2 is a frequency, Re[ ] indicates Laplace transform, and 0, and 9 are respectively a pitch direction of arrival (DOA) and an azimuth DOA of a sound source relative to a vector hydrophone, where the pitch DOA and the azimuth DOA respectively take the XY plane and the x -axis as0°.
In this case, the azimuth DOA and pitch DOA of the target are:
9= arctan S Sil, (f)
0 = arctan S)V (f S ~ v(f+ Mf (10)
The sound intensity of the vector hydrophone may be obtained with reference to formulas (6), (7), and (10):
I=(IxcosOcosp+I cos0sinp+isinO)
=p (t) -v1,, (T)cos 0cos p +p, (t) -v, (t)cos 0sinp +p, (t) - v ,(t)sin 0 Re[S ] Re ] Re[S p(t)-v,(t) [S2 +p(t)-v (t) 2 +p(t)-v,(t) eS2 Re[S (f )] Re[S (f )] Re[S (f)] ($
In step 303, the dynamic windows of all lengths are traversed for the instantaneous
single-channel sound intensity signal w'i in the fixed window, and the steepest ascent
section of the Shannon entropy is found as the optimum dynamic window di; and the step specifically includes the following steps:
601: traversing the entire fixed window 't of the captured instantaneous
single-channel sound intensity signal I starting from the start point t/f of the fixed window, and calculating a Shannon entropy of signals of all lengths by using the following
formula, to construct a Shannon entropy signal s,,
1 Shannon(X)= p(x). log = -$p(x,). log p(x,) P(x1 ) i (12)
where X' is a possible value of a random event X, Shannon(X) is a Shannon entropy included in the random event X, m is a total quantity of random events, and
P(x1 is a probability of occurrence of x;
602: finding a steepest ascent section of S'i according to a derivation result S of
S 1 , marking the length as s, and jumping to step 604;
603: marking the length as si if no steepest ascent section is found in S'i W considering that the signal in the fixed window is an invalid signal or a noise signal, where the following two conditions are satisfied:
when S'i is relatively small, it is considered that the signal is a null signal or Shannon
entropy content of the signal in the fixed window is insufficient, and 0 0, where I is a preset minimum captured length; and
when S'i is relatively large, it is considered that the signal is a noise signal or a valid
signal including a relatively high Shannon entropy, and where 11 is a preset maximum captured length;
604: capturing a signal segment whose length is si starting from the start point tf in
the fixed window h as the dynamic window di, and marking an end time of the
window ast
For the shortcoming of different lengths of the signals obtained after fixed-dynamic window segmentation, signal self-supplement is performed by using the expectation maximization algorithm (EM algorithm). The segmented unequal-length signals are
equivalent to observation data X, the supplemented equal-length signals are equivalent to
complete data Y, and the supplementary signal is equivalent to unobserved data Z. The equal lengths of the signals facilitate subsequent calculation and comparison.
A result is iterated by using the expectation maximization algorithm to obtain a maximum value 0 of a parameter 0, that is, when a maximum value of a maximum likelihood function L(0) based on Y is obtained, optimal solutions of a mean ui and a variance(7 of a complete dataset are obtained, and an unknown dataset Z is obtained based on an observed dataset X , to supplement the complete dataset Y , which specifically includes the following steps:
701: assuming that an iteration quantity t = 0, initializing a parameter vector 0(0), 0
being a parameter vector formed by the mean and the variance of the dataset Y, and
calculating an initial maximum likelihood function
n
702: obtaining Q ( )(from 0), and ensuring that when 0 is specified, the
equal sign in n(E(X))E[n(X)]is satisfied, to establish a lower bound of ;
703: fixing Q() and taking 0 as a variable, taking a derivative of ))in
step 702, and obtaining O(t*1) according to the formula
L (O(t1 )) I Qz() n =L (OW) I' QQz() ZW; n)d and
704: if , ending the iterative calculation; otherwise, assuming
that= t +1, and returning to step 702, where a threshold 6 is a specified very small value;
A specific iteration process is as follows:
Assuming that Z represents missing data, that is, unobserved data, X is observed
data, referred to as incomplete data, a sum of the missing data Z and the incomplete data
X is defined as complete data Y, and X is a function of Y, there are the following relations:
[L(O)= Inp(XIO) -+>max] > [Inp(Yl0) -+max] > 0 -0>o*] (13)
L(O)=L(x),---,x");)=p( x ;0). ... p(x'";O)=ip(x ;0) =1 (14)
{ (15)
p(XIO)is a probability density function of an observed dataset, is a
probability density function of a complete dataset, and ui and 0 are respectively a mean and a variance of a probability density function.
To find a maximum value of the likelihood function L(O) is to find 0 in a parameter space & to maximize the likelihood function when sample points
{ are fixed, specifically: 0* = argmax., 0 L(0) (16)
Since L(0) and InL(0) have extreme values at the same 0 , a logarithmic operation is performed on the likelihood function:
ln(L (0))= lnp(x(i);o) =1(17)
A maximum likelihood estimate 0* of 0 may be obtained by solving the following equation:
d InLO)=0 dO (18)
Therefore, formula (13) can be transformed into:
L(O)= Q(z('))lIn f' 0) Q1(z(') (19)
Qi zf')):= p(zf' x');O) P ( (0 (20)
Qi represents a distribution of unknown data Z , and satisfies the following conditions:
(z)> 0 (21)
from related definitions of a mathematical expectation and the Jensen's inequality:
E[f(x)]= f(x)p(x) (22 i (22)
ln(E(X))> E[ln(X)] (23) It is learned with reference to formula (19) that:
0 L(6)= In E PX'
(-1 (24)
It is leaked with reference to formulas (19) and (20) that, in a tt h iteration:
Q z p( xZ ;0 = lnInQ, (z('))~()z';() Ek~l =Y~ Q z
) > Q) z l =L(6
) I Q, (()InP(O Z(;01).= W IW Q,(?)) (25)
Formula (25) may be seen as a process of finding a lower bound of L(O) Through
continuous iteration, the lower bound is increased until the lower bound
) converges to the vicinity the likelihood function L(0) when the parameter 0 reaches a
maximum value 0*. In this case, the iteration ends.
The estimating, through a recurrent neural network, a position of the sound source by using equal-length signals obtained after the self-supplement specifically includes: using the supplemented signal segments as an input by using the expectation maximization algorithm, outputting azimuth DOAs and distances of the sound source in different signal segments, and performing cross-validation on estimation results of the different signal segments, to implement accurate positioning of the position of the sound source.
To further describe an implementation process of the method, the method is tested by using a signal acquired by a single vector hydrophone deployed at a position when a ship is sailing. Upon testing, it is found that a sound source position can be located in a very short time by using the method, with accuracy of 1.5 m. Compared with a conventional method, only a single vector hydrophone is required, not only improving the positioning accuracy, but also improving the stability. An estimation result is shown in FIG. 4.
Embodiment 2:
Embodiment 2 of the present disclosure provides a computer-readable storage medium, storing a computer program, where when the program is executed by a processor, steps in the sound source position estimation method in Embodiment 1 of the present disclosure are implemented.
Embodiment 3:
Embodiment 3 of the present disclosure provides a computer device, including a memory, a processor, and a computer program stored in the memory and capable of being run on the processor, where when the processor executes the program, steps in the sound source position estimation method in Embodiment 1 of the present disclosure are implemented.
The foregoing descriptions are merely exemplary embodiments of the present disclosure, but are not intended to limit the present disclosure. The present disclosure may include various modifications and changes for a person skilled in the art. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present disclosure shall fall within the protection scope of the present disclosure.
Claims (10)
1. A sound source position estimation method, comprising the following steps:
receiving, by a single vector hydrophone, multichannel signals from a sound source in the ocean;
fusing the received multichannel signals into an instantaneous single-channel sound intensity signal through fixed-dynamic window combined sliding, and dividing the received multichannel signals into signal segments comprising a sufficient information amount;
performing signal self-supplement by using the expectation maximization algorithm, and increasing a distance between the signal segments; and
estimating, through a recurrent neural network, a position of the sound source by using equal-length signals obtained after the self-supplement.
2. The sound source position estimation method according to claim 1, wherein the multichannel signals are four-channel signals, comprising three acoustic particle velocity
signals in orthogonal directions: an acoustic particle velocity vx in an X -axis direction, an
acoustic particle velocity v' in a Y -axis direction, and an acoustic particle velocity v
in a z -axis direction, and a scalar sound pressure signal P .
3. The sound source position estimation method according to claim 2, wherein the multichannel signals are fused into the instantaneous single-channel sound intensity signal by using a fixed window, dynamic windows of all lengths are traversed, a steepest ascent section of a Shannon entropy is found, an optimum dynamic window is determined, the instantaneous single-channel sound intensity signal in the fixed window are dynamically captured as unequal-length signals based on the Shannon entropy by using the optimum dynamic window, and signal self-supplement is performed on the captured unequal-length signals by using the expectation maximization algorithm.
4. The sound source position estimation method according to claim 3, wherein the dividing the received multichannel signals into signal segments comprising a sufficient information amount through fixed-dynamic window combined sliding specifically comprises the following steps:
401: specifying a fixed window length If and an initial window start point tf for
the acquired four-channel signalsP, ,v ,and v
W 402: fusing, by using a fixed window f whose window length and start point are
respectively If and tf , four-channel information in the window, to obtain an
instantaneous single-channel sound intensity signal whose length is ;
403: capturing a dynamic window di whose window length and start point are Ii t W respectively S and f in the instantaneous single-channel sound intensity signal i,
wherein the length of the signal in the dynamic window di is as short as possible provided that the signal comprises a sufficient information amount; and
404: returning to 402, updating a start point tf of a fixed window based on a
signal overlap rate q, and performing a cyclic operation.
5. The sound source position estimation method according to claim 4, wherein in step 402, synchronous sliding is performed in channel signals through a time window of a fixed size, signals are extracted, and the information is fused into the instantaneous single-channel sound intensity signal by using a cross-spectrum method; and the step specifically comprises the following steps:
501: specifying a fixed window length If and a window start point tf according to a signal fusion degree;
502: capturing signal segments of a window size f by using the same start point tf
in the signal channels of the sound pressure P and the vibration velocities v , and V, in the axial directions, wherein corresponding window signals are A, , and "zi; and
503: calculating, based on the cross-spectrum method, the instantaneous w single-channel sound intensity signal 'i obtained after the window signals are fused, to implement multi-sensor information fusion, wherein a calculation formula of the instantaneous single-channel sound intensity signal obtained after the fusion is:
Re[S ] Re[S, ] Re[S V] I=p(t)-v(t) 'Ji)+pft)-v,(t) *( +pft)-v,(t1 Re[S(f)] Re[S(f)] (t) Re[S'(f)]
" wherein S , , and SP= are respectively cross-spectrum functions of three
components x, y, and z, is a spectrum function of 2 is a frequency, Re[ ] indicates Laplace transform, 0 and 9 are respectively a pitch direction of arrival (DOA) and an azimuth DOA of a sound source relative to a vector hydrophone, wherein the pitch DOA and the azimuth DOA respectively take the XOY plane and the x
-axis as 0°, and p(t) vx(t) vy,(t), and v(t) are respectively a sound pressure signal and acoustic particle velocity signals in different directions that are received by the vector hydrophone at a moment t.
6. The sound source position estimation method according to claim 4, wherein in step 403, the dynamic windows of all lengths are traversed for the instantaneous single-channel
sound intensity signal 'i in the fixed window, and the steepest ascent section of the
Shannon entropy is found as the optimum dynamic window di; and the step specifically comprises the following steps:
601: traversing the entire fixed window 't of the captured instantaneous
single-channel sound intensity signal I starting from the start point tf of the fixed window, and calculating a Shannon entropy of signals of all lengths by using the following
formula, to construct a Shannon entropy signal S,
Shannon(X)= p(x). log - =- p(x, ). logp(x) i=1 p( x,) j
wherein i is a possible value of a random event X, Shannon(X) is a Shannon entropy comprised in the random event X, m is a total quantity of random events, and
P(x1 is a probability of occurrence of xi;
602: finding a steepest ascent section of '' according to a derivation result S' of
S1 ,, marking the length as , and jumping to step 604;
603: marking the length as if no steepest ascent section is found in Si, W considering that the signal in the fixed window is an invalid signal or a noise signal, wherein the following two conditions are satisfied:
when S'i is relatively small, it is considered that the signal is a null signal or
Shannon entropy content of the signal in the fixed window is insufficient, and 0,
wherein 10 is a preset minimum captured length; and
when S'i is relatively large, it is considered that the signal is a noise signal or a valid
signal comprising a relatively high Shannon entropy, and i 1, wherein 11 is a preset maximum captured length;
604: capturing a signal segment whose length is starting from the start point tf W W in the fixed window I as the dynamic window ,and marking an end time of the
window as tde
7. The sound source position estimation method according to claim 1, wherein
signal self-supplement is performed by using the expectation maximization algorithm, segmented unequal-length signals are equivalent to observation data X, supplemented equal-length signals are equivalent to complete data Y , a supplementary signal is equivalent to unobserved data Z , and a result is iterated by using the expectation maximization algorithm to obtain a maximum value 0* of a parameter 0, that is, when a maximum value of a maximum likelihood function L(0) based on Y is obtained, optimal solutions of a mean u and a variance'a of a complete dataset are obtained, and an unknown dataset Z is obtained based on an observed dataset X, to supplement the complete dataset Y, which specifically comprises the following steps:
701: assuming that an iteration quantity t = 0, initializing a parameter vector 0), 0
being a parameter vector formed by the mean and the variance of the dataset Y, and
calculating an initial maximum likelihood function
n
L())= L (xj,---x,,; 0) = p( xl; 0)-.... p( x,;6)= p (xi; 0)
702: obtaining Qt) z()) from 0(t), and ensuring that when 0 is specified, the
equal sign in n(E(X))E[n(X)]is satisfied, to establish a lower bound of ;
703: fixing ( and taking0( as a variable, taking a derivative of ))in
step 702, and obtaining O(t1) according to the formula
L (O(t1 )) Q(z)ln =L (6 ;and
704: if () () , ending the iterative calculation; otherwise, assuming
that tt +1, and returning to step 702, wherein a threshold 6 is a specified very small value; wherein ' represents a distribution of unknown data Z , P(i)'Z )is a probability of occurrence of x '), z' under a condition 0 ), the superscript i indicates an Ith value of a corresponding parameter, - is the threshold and is an initially specified small value used as a standard for ending the iteration, and E[ ] is a mathematical expectation.
8. The sound source position estimation method according to claim 1, wherein the estimating, through a recurrent neural network, a position of the sound source by using equal-length signals obtained after the self-supplement specifically comprises: using the supplemented signal segments as an input by using the expectation maximization algorithm, outputting azimuth DOAs and distances of the sound source in different signal segments, and performing cross-validation on estimation results of the different signal segments, to implement accurate positioning of the position of the sound source.
9. A computer-readable storage medium, storing a computer program, wherein when the program is executed by a processor, steps in the sound source position estimation method according to any one of claims 1 to 8 are implemented.
10. A computer device, comprising a memory, a processor, and a computer program stored in the memory and capable of being run on the processor, wherein when the processor executes the program, steps in the sound source position estimation method according to any one of claims 1 to 8 are implemented.
Applications Claiming Priority (3)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910493374.4 | 2019-06-06 | ||
| CN201910493374.4A CN110146846B (en) | 2019-06-06 | 2019-06-06 | Sound source position estimation method, readable storage medium and computer equipment |
| PCT/CN2020/089894 WO2020244359A1 (en) | 2019-06-06 | 2020-05-13 | Sound source position estimation method, readable storage medium and computer device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| AU2020287677A1 true AU2020287677A1 (en) | 2021-03-04 |
| AU2020287677B2 AU2020287677B2 (en) | 2022-07-07 |
Family
ID=67590874
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| AU2020287677A Active AU2020287677B2 (en) | 2019-06-06 | 2020-05-13 | Sound source position estimation method, readable storage medium and computer device |
Country Status (3)
| Country | Link |
|---|---|
| CN (1) | CN110146846B (en) |
| AU (1) | AU2020287677B2 (en) |
| WO (1) | WO2020244359A1 (en) |
Families Citing this family (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110146846B (en) * | 2019-06-06 | 2021-04-13 | 青岛理工大学 | Sound source position estimation method, readable storage medium and computer equipment |
| CN111427008B (en) * | 2020-04-19 | 2022-09-16 | 中国电子科技集团公司电子科学研究院 | Position determination method for underwater autonomous vehicle cluster and related device |
| CN111580048B (en) * | 2020-05-09 | 2020-12-29 | 中国科学院声学研究所 | Broadband sound source depth estimation method using single-vector hydrophone |
| CN112415467B (en) * | 2020-11-06 | 2022-10-25 | 中国海洋大学 | Single-vector subsurface buoy target positioning implementation method based on neural network |
| CN113009417B (en) * | 2021-02-05 | 2022-09-20 | 中国人民解放军国防科技大学 | Submarine acoustic array formation estimation method utilizing sound field interference characteristics |
| CN113009419B (en) * | 2021-02-25 | 2021-11-09 | 中国科学院声学研究所 | Target depth estimation method based on frequency domain cross-correlation matching |
| CN114910901B (en) * | 2022-05-30 | 2024-07-12 | 深圳市大族机器人有限公司 | High-precision multi-sensor fusion ranging system of cooperative robot |
| CN116068494B (en) * | 2023-03-31 | 2023-07-18 | 中国人民解放军96901部队 | Wide-area secondary sound source positioning method based on secondary sound propagation model |
Family Cites Families (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6804167B2 (en) * | 2003-02-25 | 2004-10-12 | Lockheed Martin Corporation | Bi-directional temporal correlation SONAR |
| CN101539597B (en) * | 2009-04-29 | 2011-01-26 | 哈尔滨工程大学 | Multi-target method for distinguishing radiation noises having same frequency band |
| CN202383285U (en) * | 2010-12-30 | 2012-08-15 | 中国科学院声学研究所 | Networked underwater sound positioning node system |
| CN102708374B (en) * | 2012-01-06 | 2014-06-25 | 香港理工大学 | Method and device for classifying remote images by integrating edge information and support vector machine |
| CN102981146B (en) * | 2012-11-19 | 2014-10-22 | 哈尔滨工程大学 | Passive positioning method of single-vector hydrophone |
| CN103267966B (en) * | 2013-05-20 | 2014-12-31 | 西北工业大学 | Orientation estimation method for single two-dimensional differential pressure type vector hydrophone |
| CN103927401B (en) * | 2014-04-16 | 2018-02-16 | 公安部第一研究所 | The high-speed signal acquisition and processing system of a kind of multi-channel sampling |
| CN103954346B (en) * | 2014-05-16 | 2016-06-22 | 中国电子科技集团公司第四十九研究所 | There is the magnetic complex three-dimensional vector hydrophone of target location and discrimination function and the target location of this hydrophone and method of discrimination |
| CN104848900B (en) * | 2015-05-26 | 2017-03-22 | 国家海洋技术中心 | Array-type ocean acoustical signal measurement system |
| US10063965B2 (en) * | 2016-06-01 | 2018-08-28 | Google Llc | Sound source estimation using neural networks |
| CN106249244B (en) * | 2016-07-07 | 2018-10-26 | 哈尔滨工程大学 | A kind of acoustic vector circle battle array wide-band coherent signal source direction estimation method |
| CN108337002A (en) * | 2018-02-09 | 2018-07-27 | 黄淮学院 | A kind of multiple channel acousto signal reconfiguring method |
| CN109492580B (en) * | 2018-11-08 | 2020-08-07 | 北方工业大学 | Multi-size aerial image positioning method based on neighborhood significance reference of full convolution network |
| CN110146846B (en) * | 2019-06-06 | 2021-04-13 | 青岛理工大学 | Sound source position estimation method, readable storage medium and computer equipment |
-
2019
- 2019-06-06 CN CN201910493374.4A patent/CN110146846B/en active Active
-
2020
- 2020-05-13 AU AU2020287677A patent/AU2020287677B2/en active Active
- 2020-05-13 WO PCT/CN2020/089894 patent/WO2020244359A1/en active Application Filing
Also Published As
| Publication number | Publication date |
|---|---|
| WO2020244359A1 (en) | 2020-12-10 |
| AU2020287677B2 (en) | 2022-07-07 |
| CN110146846A (en) | 2019-08-20 |
| CN110146846B (en) | 2021-04-13 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| AU2020287677A1 (en) | Sound source position estimation method, readable storage medium and computer device | |
| Wu et al. | A dynamic time sequence recognition and knowledge mining method based on the hidden Markov models (HMMs) for pipeline safety monitoring with Φ-OTDR | |
| CN103439688B (en) | Sound source positioning system and method used for distributed microphone arrays | |
| CN104076355B (en) | Tracking before Dim targets detection in strong clutter environment based on dynamic programming | |
| Zhong et al. | Particle filtering approaches for multiple acoustic source detection and 2-D direction of arrival estimation using a single acoustic vector sensor | |
| Zhong et al. | Particle filtering and posterior Cramér-Rao bound for 2-D direction of arrival tracking using an acoustic vector sensor | |
| Li et al. | An automated operational modal analysis algorithm and its application to concrete dams | |
| Ravela | Quantifying uncertainty for coherent structures | |
| CN110501683B (en) | Sea-land clutter classification method based on four-dimensional data characteristics | |
| CN113640768B (en) | Low-resolution radar target identification method based on wavelet transformation | |
| Chen et al. | Coal mine microseismic identification and first-arrival picking based on Conv-LSTM-Unet | |
| Park et al. | Track detection of low observable targets using a motion model | |
| CN108269581B (en) | Double-microphone time delay difference estimation method based on frequency domain coherent function | |
| Zhou et al. | Method for predicting depth-averaged current velocities of underwater gliders based on data feature analysis | |
| CN105303009A (en) | Super-resolution spectrum estimation method based on compressed sensing and regular MFOCUSS | |
| Zhong et al. | Particle filtering for 2-D direction of arrival tracking using an acoustic vector sensor | |
| Lv et al. | Location method based on support vector machine for distributed Sagnac fiber sensing system | |
| Mušicki | Doppler-aided target tracking in heavy clutter | |
| Zhang et al. | Underdetermined mixing matrix estimation based on single source detection | |
| Liu et al. | Work mode identification of phased array radar with denoising auto-encoder | |
| Yao et al. | High-precision time delay estimation of narrowband radio signal by PHAT-LSTM | |
| Jun et al. | Research on non-stationary blind separation method for unfixed signal based on extended joint diagonalization | |
| Zhao et al. | A study of individual identification of radiation source based on feature extraction and deep learning | |
| Ghassemi et al. | A method for constructing the observer trajectory in bearings-only tracking of targets with a Markovian model | |
| Li et al. | Underwater source localization in the presence of strong interference |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| FGA | Letters patent sealed or granted (standard patent) |