CN117518142A - Semi-physical simulation data set construction method for underwater sound source localization - Google Patents
Semi-physical simulation data set construction method for underwater sound source localization Download PDFInfo
- Publication number
- CN117518142A CN117518142A CN202311058053.4A CN202311058053A CN117518142A CN 117518142 A CN117518142 A CN 117518142A CN 202311058053 A CN202311058053 A CN 202311058053A CN 117518142 A CN117518142 A CN 117518142A
- Authority
- CN
- China
- Prior art keywords
- sound source
- semi
- physical simulation
- simulation data
- hydrophone
- 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.)
- Pending
Links
- 238000004088 simulation Methods 0.000 title claims abstract description 31
- 230000004807 localization Effects 0.000 title claims abstract description 11
- 238000010276 construction Methods 0.000 title claims abstract description 9
- 239000011159 matrix material Substances 0.000 claims abstract description 30
- 238000004364 calculation method Methods 0.000 claims abstract description 15
- 230000007613 environmental effect Effects 0.000 claims abstract description 13
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 claims abstract description 13
- 238000000034 method Methods 0.000 claims abstract description 11
- 230000005236 sound signal Effects 0.000 claims abstract description 4
- 238000012360 testing method Methods 0.000 claims description 15
- 238000012549 training Methods 0.000 claims description 13
- 238000013527 convolutional neural network Methods 0.000 description 11
- 238000013528 artificial neural network Methods 0.000 description 8
- 238000004422 calculation algorithm Methods 0.000 description 6
- 238000005259 measurement Methods 0.000 description 5
- 238000013135 deep learning Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 230000006870 function Effects 0.000 description 3
- 238000013480 data collection Methods 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 206010027336 Menstruation delayed Diseases 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000007635 classification algorithm Methods 0.000 description 1
- 238000009795 derivation Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 230000004069 differentiation Effects 0.000 description 1
- 239000006185 dispersion Substances 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000001228 spectrum Methods 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
- 239000002699 waste material Substances 0.000 description 1
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
- G01S7/00—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00
- G01S7/52—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S15/00
- G01S7/539—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S15/00 using analysis of echo signal for target characterisation; Target signature; Target cross-section
-
- 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
- G01S15/00—Systems using the reflection or reradiation of acoustic waves, e.g. sonar systems
- G01S15/88—Sonar systems specially adapted for specific applications
Landscapes
- Engineering & Computer Science (AREA)
- Radar, Positioning & Navigation (AREA)
- Remote Sensing (AREA)
- Physics & Mathematics (AREA)
- Computer Networks & Wireless Communication (AREA)
- General Physics & Mathematics (AREA)
- Acoustics & Sound (AREA)
- Measurement Of Velocity Or Position Using Acoustic Or Ultrasonic Waves (AREA)
Abstract
The invention discloses a semi-physical simulation data set construction method for underwater sound source localization, which comprises the following steps of 1, measuring the position information of N hydrophones arranged in an experimental water area, and measuring and analyzing the self-noise and environmental noise information of each hydrophone; step 2, setting N hydrophones in the experimental water area at a space plane Z=0, and setting the hydrophones in the experimental water area at Z=Z 0 Setting M sound sources at the position; step 3, let z=z 0 Is pixelized; step 4, generating sound signals and position information of the sound source through a random number function, and obtaining a tag value according to the position information; step 5, acoustic signal [ S 'by acoustic source'] M×1 And formula (1) to obtain [ R ]] N×1 The method comprises the steps of carrying out a first treatment on the surface of the Step 6, R is as follows] N×1 Adding self-noise and environmental noise information to obtain new matrix R 0 ]The beam forming image matrix S is obtained through calculation of (2) 0 ]To obtain ([ S) 0 ]Label value), step 7, repeating the steps4 to 6, constituting a dataset.
Description
Technical Field
The invention belongs to the field of metering test and the field of acoustics, and particularly relates to a semi-physical simulation data set construction method for underwater sound source positioning.
Background
The problem of underwater sound source positioning is always an important branch in the research of the underwater sound field, and various underwater positioning methods are provided for different application scenes. Almost all positioning methods exist, and the positioning accuracy and the calculation force consumption are inversely proportional. The low-precision measurement can achieve immediate calculation, but the positioning error is larger; the calculation power consumption of high-precision measurement is extremely high, and the calculation power consumption of a high-precision positioning algorithm cannot be maintained in the calculation power configuration of a general computer, so that a great amount of time is required for offline processing of data after a test. Therefore, the calculation force consumption can be concentrated in a preparation part before the test in a mode of training the neural network, so that the real-time calculation force requirement of a high-precision positioning algorithm in the test process can be reduced, and the high-precision real-time positioning measurement can be realized as much as possible.
The quality of the data set determines the training speed and the final effect of training under the condition that the design of the neural network is unchanged. For this purpose, the following requirements should be fulfilled: the data volume of the data set is large enough to ensure that enough data is mined during model training; the single data set should have the necessary difference to avoid over-fitting of model training; the data form and the label should be set reasonably, so that unnecessary waste of calculation force is avoided.
However, the underwater positioning test has high cost, long period and multiple devices, and the data collection by the experiment cannot be operated according to the single data collection requirement, so that the difference and the label setting cannot be accurately designed through the experiment.
Disclosure of Invention
The invention aims to provide a semi-physical simulation data set construction method for underwater sound source positioning, which aims to solve the problems in the background technology.
In order to achieve the above purpose, the present invention provides the following technical solutions:
a semi-physical simulation data set construction method for underwater sound source localization comprises the following steps:
step 1, measuring the position information of N hydrophones arranged in an experimental water area, and measuring and analyzing the self-noise and environmental noise information of each hydrophone;
step 2, setting N hydrophones { R1, R2, & gt, RN } in the experimental water area to be at a space plane z=0, and at the space plane z=z 0 Setting M sound sources { S1, S2, };
step 3, let z=z 0 Any pixel point is set as a sound source label or a non-sound source label;
step 4, generating the sound signal S 'of the sound source through the random number function'] M×1 ＝[S S1 (t)，S S2 (t)，...，S SM (t)]The position information is used for obtaining a label value according to the position information;
step 5, acoustic signal [ S 'by acoustic source'] M×1 And formula (1) to obtain [ R ]] N×1 ，
[R] N×1 ＝[T] N×M [S] M×1 (1)
Wherein T is nm For transmitting matrix [ T ]]Element T of (a) nm ＝exp(-j2πf 0 d mn )/d mn ，d nm For the spacing, f, between the nth hydrophone Rn and the mth sound source Sm 0 Is a characteristic frequency for locating a target;
S m is a matrix S]Element S of (3) m ＝fft(S Sm (t)，f 0 ) M=1, 2,; step 6, R is as follows] N×1 Adding self-noise and environmental noise information to obtain new matrix R 0 ]The beam forming image matrix S is obtained through calculation of (2) 0 ]Obtain hydrophone semi-physical simulation data pair ([ S) 0 ]A tag value),
[S 0 ] M′×1 ＝[T] N′×M′ -1 [R 0 ] N′×1 (2)
wherein, [ S ] 0 ]For a matrix of beamformed images, M 'represents the number of pixels of the beamformed image, and N' represents the new matrix R 0 ]N' =n;
and 7, repeating the steps 4 to 6 to obtain batch hydrophone semi-physical simulation data pairs to form a data set.
Preferably, the data sets are sorted and grouped according to the element positions of the sound sources, a plurality of generated test sets are randomly selected from each group, and the unselected parts are used as training sets.
Preferably, the sorting packet includes the steps of:
step 7.1, performing primary grouping on hydrophone semi-physical simulation data pairs according to the sound source number M;
step 7.2, for z=z 0 Numbering all the pixels of the image;
and 7.3, calculating the number sum of the hydrophone semi-physical simulation data pair associated M sound sources, and carrying out secondary grouping according to the number sum.
Preferably, if the numbers are the same, calculating the number products of the hydrophone semi-physical simulation data on the associated M sound sources, and carrying out three-level grouping according to the number products.
Compared with the prior art, the invention has the beneficial effects that:
aiming at the problem of underwater sound source localization, the invention constructs a data set suitable for a convolutional neural network by combining real experimental data with analog calculation through semi-physical simulation; in the construction of a data set, experimental measurement data are used for analyzing the influence of the noise of the water area where the sensor is located on the sensor, and a semi-physical simulation mode is used for simulating the actual situation as far as possible; meanwhile, aiming at the characteristics of the convolutional neural network, a regression algorithm in the positioning problem is converted into a classification algorithm in a label setting mode, so that the requirement of deep learning on a data set in the application process of underwater sound source positioning is met;
according to the invention, after the data sets are sorted and grouped, the test sets are randomly selected from each group, so that the situation that the training effect of the neural network cannot be effectively verified due to the fact that the positions of sound sources in the test sets are concentrated in a certain area of a plane is prevented.
Drawings
FIG. 1 is a beam forming image matrix S 0 ]And visualizing the result.
Detailed Description
The invention discloses a method for constructing a semi-physical simulation data set for underwater sound source localization, which is used for training a convolutional neural network (convolution neural network, CNN for short).
The convolutional neural network (Convolutional Neural Network, CNN) is a type of feedforward neural network that includes convolutional calculations and has a deep structure, and is one of representative algorithms of deep learning (deep learning). Convolutional neural networks have the capability of token learning (representation learning) and are capable of performing a translation-invariant classification of input information in their hierarchical structure, and are therefore also referred to as "translation-invariant artificial neural networks".
The construction method of the convolutional neural network semi-physical simulation data set for underwater sound source localization specifically comprises 6 steps.
And step 1, measuring the position information of N hydrophones arranged in the experimental water area, and measuring and analyzing the self-noise and environmental noise information of each hydrophone.
In the invention, the environmental data is an important component of experimental data, and comprises environmental noise information of an experimental water area, hydrophone self-noise information and hydrophone position information. The hydrophone position information can be manually adjusted, but the composition of the environmental noise and the self-noise information is complex, and the actual measurement mode is needed to be used to be as close to the actual situation as possible.
Specifically, the self-noise and environmental noise information should be measured in the early, middle and late periods of the experimental water area, respectively, and the specific measuring method is shown in JJF 1651-2017, specification for calibration of 20 Hz-100 kHz underwater noise sources. In the measured sound pressure spectrum density level of the underwater noise, f is selected 0 The values at are measured multiple times as data set noise. This is a conventional technical means in the art, and is not described in detail herein.
Step 2, setting N hydrophones { R1, R2, & gt, RN } in the experimental water area to be at a space plane z=0, and at the space plane z=z 0 Set M sound sources { S1, S2, }, SM }.
Here, N hydrophones present in a spatial mid-plane z=0 are given the numbers { R1, R2, }, RN } (R denotes a recter), where each hydrophone is atThe position in the plane is P R (x n ，y n 0), n= {1,2,... At the same height Z in the space plane 0 The numbers of the M sound sources are { S1, S2, & SM }, m= {1,2, & M }, respectively, (S represents Source), and the corresponding positions of the respective sound sources are P S (x m ，y m ，Z 0 )。
Step 3, let z=z 0 Any pixel is set as a sound source label or a non-sound source label.
The labels are target values of self iteration of the neural network algorithm, unnecessary calculation power consumption can be reduced by effectively setting the labels of the data set, iteration efficiency is improved, and understanding is facilitated. CNN architecture is mainly used for data classification and identification, while underwater positioning generally uses regression algorithm architecture, so labels need to be designed for CNN architecture. Because of the positioning problem, the position of the sound source is more focused, and then whether the sound source exists in each pixel in the space can be used as a label. Thus, the regression problem can be treated as a classification problem, and the task of setting the data set label is completed.
Step 4, generating the sound signal S 'of the sound source through the random number function'] M×1 And the position information, and obtain the label value according to the position information.
In the invention, the signals sent by the M sound sources which work independently are S Sm (t), the acoustic signals of the M acoustic sources can be represented by a matrix S'] M×1 ＝[S S1 (t)，S S2 (t)，...，S Sm (t)，...，S SM (t)]And (3) representing. The position information of the M sound sources is determined by a random number function at the same time, the position information is a specific pixel position, the position without the sound source is regarded as the condition that the sound source is present but not working, the position without the sound source is set as a non-sound source label when the label is set, and the position with the sound source is set as a sound source label.
Step 5, acoustic signal [ S 'by acoustic source'] M×1 And formula (1) to obtain [ R ]] N×1 ，
[R] N×1 ＝[T] N×M [S] M×1 (2)
Wherein T is nm For transmitting matrix [ T ]]Element T of (a) nm ＝exp(-j2πf 0 d mn )/d mn ，f 0 For artificially selected characteristic frequencies, d, for locating objects nm The distance between the nth hydrophone Rn and the mth sound source Sm;
S m is a matrix S]Element S of (3) m ＝fft(S Sm (t)，f 0 )；
R n Is a matrix [ R]Element R of (a) n ＝fft(S Rn (t)，f 0 )。
In step 5 of the present invention, the derivation step of formula (2) is as follows: due to the time-domain signal S received by the hydrophone Rn (t) an acoustic signal [ S 'which can pass through the acoustic source'] M×1 Acquisition, expressed as
Wherein S is Rn Representing the signal received by the nth hydrophone Rn, S Sm Represents the signal sent by the mth sound source Sm, d nm C is the sound speed of water, and t represents time;
the equation is fourier transformed and expressed in matrix form to obtain equation (2).
Step 6, R is as follows] N×1 Adding the environmental noise information and the hydrophone self-noise information to obtain a new matrix [ R ] 0 ]The beam forming image matrix S is obtained through the calculation of 3) 0 ]Obtain hydrophone semi-physical simulation data pair ([ S) 0 ]A label),
[S 0 ] M′×1 ＝[T] N′×M′ -1 [R 0 ] N′×1 (3)
wherein, [ S ] 0 ]For a matrix of beamformed images, M 'represents the number of pixels of the beamformed image, and N' represents the new matrix R 0 ]Here, N' =n is preset.
In the invention, through T nm ＝exp(-j2πf 0 d mn )/d mn Calculating to obtain T nm Then obtains the transfer matrix T]Through S m ＝fft(S Sm (t)，f 0 ) Obtain S m Then obtain the matrix S]And combining the values of (2) to obtain a matrix [ R ]]Is a value of (2); matrix [ R ]]After adding the environmental noise information and hydrophone self-noise information to the values of (2) to obtain a beam forming image matrix S through a method (3) 0 ]For example, a beam forming image matrix S 0 ]The visual result of (2) is shown in fig. 1, and the sound source position is (0, 17.8284, -34.2941).
And 7, repeating the steps 4 to 6 to obtain a batch of hydrophone semi-physical simulation data pairs, sorting and grouping according to the element positions of the sound sources, randomly selecting a plurality of generated test sets in each group, and taking the unselected parts as training sets.
After the semi-physical simulation data pair is generated, all the data pair needs to be divided into a training set and a testing set. In order to avoid the problem that the test set is insufficient in differentiation, such as the sound source position is concentrated in the same area, and the like, the model is over-simulated, so that the obtained batch hydrophone semi-physical simulation data pairs need to be selected in groups. Specifically, the ordering packet includes the steps of:
step 7.1, performing primary grouping on hydrophone semi-physical simulation data pairs according to the sound source number M;
step 7.2, for z=z 0 Numbering all the pixels of the image;
step 7.3, calculating the number sum of the M sound sources which are related by hydrophone semi-physical simulation data, and carrying out secondary grouping according to the number sum, which is equivalent to classifying according to the distribution position of the sound sources;
and 7.4, if the numbers are the same, calculating the number products of the hydrophone semi-physical simulation data on the M associated sound sources, and carrying out three-level grouping according to the number products, wherein the three-level grouping is equivalent to the classification according to the dispersion of the sound source distribution.
In step 7.2 of the present invention, after numbering the pixel points, the position of each sound source may be represented by a pixel point number, there are M sound sources having M pixel point numbers, and the M pixel point numbers of the M sound sources are added to obtain a sum of the numbers.
In the invention, the training set and the testing set in the data set are conveniently selected in a sorting and grouping mode, so that the sound source positions corresponding to the data of the testing set are uniformly distributed on the positioning plane. The positions of sound sources in the test set are prevented from being concentrated in a certain area of the plane, so that the training effect of the neural network is effectively verified.
Claims (4)
1. The semi-physical simulation data set construction method for underwater sound source localization is characterized by comprising the following steps of:
step 1, measuring the position information of N hydrophones arranged in an experimental water area, and measuring and analyzing the self-noise and environmental noise information of each hydrophone;
step 2, setting N hydrophones { R1, R2, & gt, RN } in the experimental water area to be at a space plane z=0, and at the space plane z=z 0 Setting M sound sources { S1, S2, };
step 3, let z=z 0 Any pixel point is set as a sound source label or a non-sound source label;
step 4, generating the sound signal S 'of the sound source through the random number function'] M×1 ＝[S S1 (t)，S S2 (t)，...，S SM (t)]The position information is used for obtaining a label value according to the position information;
step 5, acoustic signal [ S 'by acoustic source'] M×1 And formula (1) to obtain [ R ]] N×1 ，
[R] N×1 ＝[T] N×M [S] M×1 (1)
Wherein T is nm For transmitting matrix [ T ]]Element T of (a) nm ＝exp(-j2πf 0 d mn )/d mn ，d nm For the spacing, f, between the nth hydrophone Rn and the mth sound source Sm 0 Is a characteristic frequency for locating a target;
S m is a matrix S]Element S of (3) m ＝fft(S Sm (t)，f 0 ) M=1, 2,; step 6, R is as follows] N×1 Adding self-noise and environmental noise information to obtain new matrix R 0 ]The beam forming image matrix S is obtained through calculation of (2) 0 ]Obtain hydrophone semi-physical simulation data pair ([ S) 0 ]A tag value),
[S 0 ] M′×1 ＝[T] N′×M′ -1 [R 0 ] N′×1 (2)
wherein, [ S ] 0 ]For a matrix of beamformed images, M 'represents the number of pixels of the beamformed image, and N' represents the new matrix R 0 ]N' =n;
and 7, repeating the steps 4 to 6 to obtain batch hydrophone semi-physical simulation data pairs to form a data set.
2. The method for constructing semi-physical simulation data sets for underwater sound source localization according to claim 1, wherein the data sets are sorted and grouped according to the element positions of the sound sources, a plurality of generated test sets are randomly selected from each group, and unselected parts are used as training sets.
3. A method of constructing a semi-physical simulation data set for underwater sound source localization as claimed in claim 2, wherein said sorting group comprises the steps of:
step 7.1, performing primary grouping on hydrophone semi-physical simulation data pairs according to the sound source number M;
step 7.2, for z=z 0 Numbering all the pixels of the image;
and 7.3, calculating the number sum of the hydrophone semi-physical simulation data pair associated M sound sources, and carrying out secondary grouping according to the number sum.
4. A method of constructing a semi-physical simulation data set for underwater sound source localization as claimed in claim 3, wherein if the numbers are the same, the number product of hydrophone semi-physical simulation data pair associated M sound sources is calculated and three-level grouping is performed according to the number product.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311058053.4A CN117518142A (en) | 2023-08-21 | 2023-08-21 | Semi-physical simulation data set construction method for underwater sound source localization |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311058053.4A CN117518142A (en) | 2023-08-21 | 2023-08-21 | Semi-physical simulation data set construction method for underwater sound source localization |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117518142A true CN117518142A (en) | 2024-02-06 |
Family
ID=89759148
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311058053.4A Pending CN117518142A (en) | 2023-08-21 | 2023-08-21 | Semi-physical simulation data set construction method for underwater sound source localization |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117518142A (en) |
-
2023
- 2023-08-21 CN CN202311058053.4A patent/CN117518142A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110074813B (en) | Ultrasonic image reconstruction method and system | |
| CN112084974B (en) | Multi-label rolling bearing fault diagnosis method based on meta-learning | |
| CN109597043A (en) | Radar Signal Recognition method based on quantum particle swarm convolutional neural networks | |
| CN109190665A (en) | A kind of general image classification method and device based on semi-supervised generation confrontation network | |
| CN109685848A (en) | A kind of neural network coordinate transformation method of three-dimensional point cloud and three-dimension sensor | |
| CN112884059B (en) | Small sample radar working mode classification method fusing priori knowledge | |
| US11948347B2 (en) | Fusion model training using distance metrics | |
| CN112990082B (en) | Detection and identification method of underwater sound pulse signal | |
| Guo et al. | Estimating cement compressive strength using three-dimensional microstructure images and deep belief network | |
| CN110363218A (en) | A kind of embryo's noninvasively estimating method and device | |
| Tamosiunas et al. | Investigating cosmological GAN emulators using latent space interpolation | |
| CN108630228A (en) | Sound quality recognition methods, device, system and vehicle | |
| Khavalko et al. | Image classification and recognition on the base of autoassociative neural network usage | |
| CN113109794B (en) | Deep sea sound source depth setting method based on deep neural network in strong noise environment | |
| CN117518142A (en) | Semi-physical simulation data set construction method for underwater sound source localization | |
| CN109506763B (en) | Vector hydrophone sparse array arrangement method based on teaching and learning optimization | |
| CN116183718A (en) | Microcrack positioning and evaluating method based on Lamb wave nonlinear mixing and LT network | |
| CN112215351B (en) | Enhanced multi-scale convolution neural network soft measurement method | |
| CN111965601A (en) | Underwater sound source passive positioning method based on nuclear extreme learning machine | |
| CN111931412A (en) | Underwater target noise LOFAR spectrogram simulation method based on generative countermeasure network | |
| CN112711873A (en) | Method, system and medium for carrying out damage positioning based on lamb wave and neural network | |
| Roy et al. | Interpretability of an Interaction Network for identifying $ H\rightarrow b\bar {b} $ jets | |
| CN113109795A (en) | Deep sea direct sound zone target depth estimation method based on deep neural network | |
| CN111027597A (en) | Strawberry disease and pest rapid classification method and device | |
| Rege et al. | Evaluating the distribution learning capabilities of GANs |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |