Target depth classification method based on polarization characteristics of vector sound field

Abstract

The application discloses a target depth classification method based on a polarization characteristic of a vector sound field, belongs to the field of underwater acoustic signal processing, and comprises the following steps: arranging a vector hydrophone in a shallow sea area to be monitored, collecting horizontal and vertical particle velocity components of a target sound field; selecting a target line spectrum or a target frequency band, performing frequency domain transformation on a target signal and cutting a frequency band, so that two-dimensional vector velocity samples are obtained; constructing a second-order polarization covariance matrix for the two-dimensional vector velocity samples in the selected frequency band, performing eigenvalue decomposition on the second-order polarization covariance matrix, and obtaining two real non-negative eigenvalues arranged in descending order; calculating a polarization degree according to the eigenvalues; comparing the polarization degree with a classification threshold value, and outputting a shallow source / deep source classification result. According to the application, only a single vector hydrophone is needed to realize the method, and a large-aperture array is not needed; moreover, no accurate prior information of an environment and no seabed parameters are needed. In addition, the application has small calculation amount, is convenient for engineering real-time implementation, has low signal-to-noise ratio requirement, and has stable classification performance.

Description

Technical Field

[0001] This invention belongs to the field of underwater acoustic signal processing technology, specifically relating to a target depth classification method based on vector acoustic field polarization characteristics. Background Technology

[0002] In complex marine environments, accurately acquiring the depth attributes of targets is of crucial engineering value for overall situational awareness and subsequent target identification. Due to the combined effects of ocean sound velocity profiles, seabed topographic boundaries, vertical stratification within the water body (such as strata strata), and multipath propagation, the sound field in shallow sea channels exhibits strong modal interference along the vertical dimension. The depth location of the sound source directly determines the excitation efficiency of each normal mode, thus affecting the vector composition, energy proportion, and polarization morphology of the sound field at the receiver. Therefore, stably extracting the depth category features of targets under complex shallow sea conditions has extremely important practical application significance.

[0003] Traditional depth inversion techniques are mostly based on vertical arrays and matched field processing (MFP). These algorithms heavily rely on accurate prior modeling of the marine environment, estimating the sound source location by comparing measured and theoretically calculated sound fields. However, under real sea conditions, the dynamic time-varying nature of sound velocity profiles, the unknown nature of seabed parameters, and deviations in sensor deployment and array attitude can easily lead to severe environmental mismatch. This often results in spatial ambiguity, increased sidelobe levels, and oscillations in the decision results of depth estimation.

[0004] To reduce reliance on precise environmental parameters, research focus has shifted from "absolute depth numerical inversion" to "depth interval classification," which only requires determining whether the target is in near-surface or deep water. While utilizing acoustic field statistical features such as trapped and free mode energy distribution, waveguide invariants, modal scintillation index, and matched subspaces can achieve this interval classification to some extent, such schemes still cannot escape the rigid requirement of large-scale spatial sampling apertures (such as large vertical or horizontal arrays). This makes them difficult to implement in practical scenarios where the platform space is limited, budget is constrained, or covert operations are required.

[0005] Against this backdrop, vector hydrophones demonstrate immense application potential, capable of simultaneously acquiring sound pressure and three-dimensional particle velocity components at a single spatial node, exhibiting excellent spatial direction sensing capabilities, and providing a novel path to break free from the constraints of traditional large-aperture arrays. For shallow sea sound fields formed by the superposition of multiple modes, the coherence relationship between horizontal and vertical particle velocities, the energy proportion of the principal polarization direction, and the uncertainty of polarization states are all related to the modal energy distribution, which in turn is modulated by the sound source depth. If a polarization operator reflecting the second-order statistical structure of the vector sound field can be constructed, robust features for target depth classification can be formed under small aperture or even single-point observation conditions.

[0006] Therefore, it is necessary to propose a target depth classification method based on vector acoustic field polarization features. Summary of the Invention

[0007] To address the problems existing in current technologies, this invention proposes a target depth classification method based on vector acoustic field polarization features. This invention constructs a physically meaningful second-order polarization statistic, utilizes the characteristic that sound source depth alters the spectral structure of the vector acoustic field covariance matrix through modal projection, and finally completes the binary classification decision between surface and underwater targets based on polarization.

[0008] The technical solution adopted by this invention to solve the technical problem is as follows:

[0009] This invention provides a target depth classification method based on vector acoustic field polarization features, comprising the following steps:

[0010] Step 1: Deploy vector hydrophones in the shallow sea area to be monitored to collect the horizontal and vertical particle velocity components of the target sound field.

[0011] Step 2: Select the target line spectrum or target frequency band, perform frequency domain transformation and frequency band truncation on the target signal to obtain two-dimensional vector velocity samples under multiple frequency units or multiple time snapshots;

[0012] Step 3: Construct a second-order polarization covariance matrix for the two-dimensional vector velocity samples within the selected frequency band;

[0013] Step 4: Perform eigenvalue decomposition on the second-order polarization covariance matrix to obtain two real non-negative eigenvalues ​​arranged in descending order;

[0014] Step 5: Calculate the polarizability based on the eigenvalues;

[0015] Step 6: Use polarization as the target depth classification feature, compare it with the classification threshold, and output the shallow / deep source classification result.

[0016] Furthermore, in step two, an appropriate time window is set according to the target line spectrum or target frequency band, and the target signal is segmented; a Fourier transform is performed on each acquired signal segment to obtain the frequency domain complex amplitude of the signal.

[0017] Furthermore, in step two, the receiving depth of the single-vector hydrophone is... Received from depth of The sound field radiated by the sound source has a frequency of The velocity vector at that point is expressed as:

[0018] ;

[0019] in, For horizontal vibration velocity, Vertical vibration velocity;

[0020] Within the framework of normal mode theory, the vector sound field can be expressed as a superposition of modes as follows:

[0021] ;

[0022] in, To match the depth of the sound source The relevant modal excitation coefficients, For depth The modal function at that location, For the first The mathematical expression for the horizontal-vertical polarization direction vector corresponding to the first mode is:

[0023] ;

[0024] in, For horizontal wavenumber, ρ is density, and z is depth.

[0025] Furthermore, in step two, for a point sound source, the distance from the sound source to the depth... The relevant modal excitation coefficients are expressed as projections onto the modal basis functions:

[0026] ;

[0027] in, For source spectrum terms, Let be the modal function at the depth of the sound source. For frequency.

[0028] Furthermore, in step three, a unified second-order statistical computational sub-sub ...

[0029] ;

[0030] in, The receiving depth of a single-vector hydrophone. For the depth of the sound source, This is the conjugate transpose, where v is the vibration velocity vector. For mathematical expectation;

[0031] The first- and second-order variograms characterize the second-order polarization covariance matrix of the vector sound field in the two-dimensional polarization subspace, and their equivalent form is:

[0032] ;

[0033] in, Indicates complex conjugation. For the horizontal particle velocity component, This represents the velocity component of the vertical mass.

[0034] Furthermore, in step three, under the conditions of linear mode superposition, frequency independence, and modal phase independence, the second-order polarization covariance matrix can be decomposed as follows:

[0035] ;

[0036] in:

[0037] ;

[0038] ;

[0039] in, For the first The horizontal-vertical polarization direction vector corresponding to the first mode;

[0040] therefore:

[0041] ;

[0042] The final mapping chain is:

[0043] .

[0044] Furthermore, in step four, the second-order polarization covariance matrix R is decomposed into eigenvalues, and its mathematical expression is:

[0045] ;

[0046] in, The eigenvectors corresponding to the eigenvalues; Eigenvalues ​​represent the energy distribution of the vector sound field along the principal polarization and secondary polarization directions. Corresponding to the dominant coherent component, eigenvalue Corresponding to multimodal interference residual components.

[0047] Furthermore, in step five, the polarizability characterizes the degree of energy dominance in the principal polarization direction, and is defined as:

[0048] ;

[0049] in, Eigenvalues ​​represent the energy distribution of the vector sound field along the principal polarization and secondary polarization directions. Corresponding to the dominant coherent component, eigenvalue Corresponding to multimodal interference residual components.

[0050] Furthermore, in step six, when the polarizability P of the target signal is greater than the classification threshold η, it is a deep source; when the polarizability P of the target signal is less than the classification threshold η, it is a shallow source.

[0051] The target depth classification method based on vector acoustic field polarization features provided by this invention has the following advantages compared with various existing target depth classification schemes:

[0052] (1) Only a single vector hydrophone is needed to complete the entire process of target depth classification, without the need to build a large-aperture hydrophone array that is bulky and difficult to deploy and maintain, which greatly reduces the deployment threshold;

[0053] (2) It does not require prior knowledge of marine environmental information, seabed topography and geological parameters, etc., which reduces the workload of preliminary survey and preparation.

[0054] (3) The overall classification process has a small computational load and low algorithm complexity, which makes it easy to realize real-time calculation output of the classification process in actual marine exploration projects;

[0055] (4) It has low requirements for the signal-to-noise ratio of the received signal, and can still output reliable results in detection scenarios with weak signals. The overall classification performance remains stable and will not fluctuate significantly.

[0056] (5) It can adapt to the complex and varied shallow sea waveguide environment with many interference factors. It can work normally for different hydrological conditions and different seabed topography detection scenarios, and has strong robustness and environmental adaptability. Attached Figure Description

[0057] Figure 1 The flowchart shows a target depth classification method based on vector acoustic field polarization features provided by the present invention.

[0058] Figure 2 This is a schematic diagram of the sound source, vector hydrophone, and two-dimensional particle velocity components in a shallow sea waveguide.

[0059] Figure 3 This is a graph showing the variation of polarization with the depth of the sound source and the depth of the receiver.

[0060] Figure 4 The curve shows the change in polarization with the depth of the sound source when the receiving depth is 70m.

[0061] Figure 5 The ROC curves are shown for different signal-to-noise ratios. Detailed Implementation

[0062] The present invention will be further described in detail below with reference to the accompanying drawings.

[0063] This invention provides a target depth classification method based on vector acoustic field polarization features. It constructs a second-order polarization covariance matrix of the target signal based on the horizontal and vertical vibration velocities obtained by a vector hydrophone, and uses polarization as the depth classification metric to achieve target depth classification.

[0064] See Figure 1 The present invention provides a target depth classification method based on vector acoustic field polarization features, the specific implementation process of which is as follows:

[0065] Step 1: Deploy vector hydrophones in the shallow sea area to be monitored to collect the horizontal and vertical particle velocity components of the target sound field.

[0066] like Figure 2 As shown, in a shallow-sea waveguide, the sound source is placed at a distance from the sea surface. The location, i.e., the depth of the sound source, is Vector hydrophones were deployed at a distance from the sea surface. The position, i.e., the receiving depth, is The horizontal distance between the sound source and the vector hydrophone is r, and the horizontal particle velocity component acquired by the vector hydrophone is... The vertical particle velocity component is .

[0067] Step 2: Select the target line spectrum or target frequency band, perform frequency domain transformation and frequency band truncation on the target signal to obtain two-dimensional vector velocity samples under multiple frequency units or multiple time snapshots;

[0068] Specifically, based on the target line spectrum or target frequency band, an appropriate time window is set to segment the target signal, and a Fourier transform is performed on each acquired signal segment to obtain the frequency domain complex amplitude of the signal. Considering the receiving depth of a single-vector hydrophone is... Received from depth of The sound field radiated by the sound source has a frequency of The velocity vector at that point is expressed as:

[0069] ;

[0070] in, For horizontal vibration velocity, This represents the vertical vibration velocity.

[0071] Within the framework of normal mode theory, the vector sound field can be expressed as a superposition of modes as follows:

[0072] ;

[0073] in, To match the depth of the sound source The relevant modal excitation coefficients, For depth The modal function at that location, For the first The specific mathematical expression for the horizontal-vertical polarization direction vector corresponding to the first mode is as follows:

[0074] ;

[0075] in, For horizontal wavenumber, ρ is density, and z is depth.

[0076] For a point sound source, the depth of the sound source... The relevant modal excitation coefficients can be expressed as projections onto the modal basis functions, as follows:

[0077] ;

[0078] in, For source spectrum terms, Let be the modal function at the depth of the sound source. From formula (4), we can see that the depth of the sound source... The modal excitation coefficients enter the propagation system solely through their values ​​at the source location.

[0079] Step 3: Construct a second-order polarization covariance matrix for the two-dimensional vector velocity samples within the selected frequency band;

[0080] To extract the deep correlation structure, a unified second-order statistical operator is introduced, as follows:

[0081] ;

[0082] in, The receiving depth of a single-vector hydrophone. For the depth of the sound source, This is the conjugate transpose, where v is the vibration velocity vector. Let be the mathematical expectation.

[0083] The first- and second-order variograms characterize the second-order polarization covariance matrix of the vector sound field in the two-dimensional polarization subspace, and their equivalent form is:

[0084] ;

[0085] in, Indicates complex conjugation. For the horizontal particle velocity component, This represents the velocity component of the vertical mass.

[0086] The second-order polarization covariance matrix characterizes the second-order statistical structure of the vector sound field, which can be interpreted as the statistical projection of the sound field in the two-dimensional polarization subspace, and also contains information on the coherence of energy distribution and particle motion direction.

[0087] Under the conditions of linear mode superposition, frequency independence, and modal phase independence, the second-order polarization covariance matrix can be decomposed as follows:

[0088] ;

[0089] in:

[0090] ;

[0091] ;

[0092] in, For the first The horizontal-vertical polarization direction vector corresponding to the first mode.

[0093] therefore:

[0094] ;

[0095] The final mapping chain is:

[0096] ;

[0097] Step 4: Perform eigenvalue decomposition on the second-order polarization covariance matrix R to obtain two real non-negative eigenvalues ​​arranged in descending order;

[0098] The eigenvalue decomposition of the second-order polarization covariance matrix R is expressed as follows:

[0099] ;

[0100] in, Indicates complex conjugation; Eigenvalues ​​represent the energy distribution of the vector sound field along the principal polarization and secondary polarization directions. Corresponding to the dominant coherent component, eigenvalue Corresponding to multimodal interference residual components.

[0101] Step 5: Calculate the polarizability based on the eigenvalues;

[0102] Polarizability is defined as:

[0103] ;

[0104] Polarizability P characterizes the degree of energy dominance in the main polarization direction. In a sound field excited by a deep source, lower-order modes have a higher degree of dominance; while shallow sources cannot excite lower-order modes and are mainly composed of higher-order modes.

[0105] Specifically, the sound source frequency is set to 200Hz, the transmission distance is 5km, and the negative gradient depth is 20-40m. The calculated polarization varies with the sound source depth. With reception depth The change graph is as follows Figure 3 As shown, when the receiving depth When located below the negative gradient layer, the polarization of shallow sources is significantly less than that of deep sources. (Set the receiving depth...) At a depth of 70m, the polarization varies with the sound source depth. The change curve is as follows Figure 4 As shown, the polarization of deep sources is significantly greater than that of shallow sources.

[0106] Step 6: Use polarization as the target depth classification feature and compare it with the classification threshold η to output the shallow / deep source classification result.

[0107] When the polarizability P of the target signal is greater than the classification threshold η, it is a deep source; when the polarizability P of the target signal is less than the classification threshold η, it is a shallow source.

[0108] Specifically, the shallow source depth was set to 15m, the deep source depth to 50m, and the signal-to-noise ratios were set to -9dB, -6dB, and -3dB, respectively. 5000 Monte Carlo experiments were conducted to obtain the ROC curves as shown below. Figure 5 ( For detection probability, As shown in the figure (where the false alarm probability is 0), when the signal-to-noise ratio is -9dB, the classification threshold is determined based on the ROC curve. At this point, the false alarm probability is The detection probability is 0.988.

[0109] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. However, these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A target depth classification method based on vector acoustic field polarization features, characterized by the following steps: Step 1: Deploy vector hydrophones in the shallow sea area to be monitored to collect the horizontal and vertical particle velocity components of the target sound field. Step 2: Select the target line spectrum or target frequency band, perform frequency domain transformation and frequency band truncation on the target signal to obtain two-dimensional vector velocity samples under multiple frequency units or multiple time snapshots; Step 3: Construct a second-order polarization covariance matrix for the two-dimensional vector velocity samples within the selected frequency band; Step 4: Perform eigenvalue decomposition on the second-order polarization covariance matrix to obtain two real non-negative eigenvalues ​​arranged in descending order; Step 5: Calculate the polarizability based on the eigenvalues; Step 6: Use polarization as the target depth classification feature, compare it with the classification threshold, and output the shallow / deep source classification result.

2. The target depth classification method based on vector acoustic field polarization features according to claim 1, characterized in that, In step two, a suitable time window is set according to the target line spectrum or target frequency band, and the target signal is segmented; Fourier transform is performed on each acquired signal segment to obtain the frequency domain complex amplitude of the signal.

3. The target depth classification method based on vector acoustic field polarization features according to claim 1, characterized in that, In step two, the receiving depth of the single-vector hydrophone is Received from depth of The sound field radiated by the sound source has a frequency of The velocity vector at that point is expressed as: ； in, For horizontal vibration velocity, Vertical vibration velocity; Within the framework of normal mode theory, the vector sound field can be expressed as a modal superposition as follows: ； in, To match the depth of the sound source The relevant modal excitation coefficients, For depth The modal function at that location, For the first The mathematical expression for the horizontal-vertical polarization direction vector corresponding to the first mode is: ； in, For horizontal wavenumber, ρ is density, and z is depth.

4. The target depth classification method based on vector acoustic field polarization features according to claim 3, characterized in that, In step two, for a point sound source, the distance from the sound source is... The relevant modal excitation coefficients are expressed as projections onto the modal basis functions: ； in, For source spectrum terms, Let be the modal function at the depth of the sound source. For frequency.

5. The target depth classification method based on vector acoustic field polarization features according to claim 1, characterized in that, In step three, a unified second-order statistical computational operator is introduced: ； in, The receiving depth of a single-vector hydrophone. For the depth of the sound source, This is the conjugate transpose, where v is the vibration velocity vector. For mathematical expectation; The first- and second-order variograms characterize the second-order polarization covariance matrix of the vector sound field in the two-dimensional polarization subspace, and their equivalent form is: ； in, Indicates complex conjugation. For the horizontal particle velocity component, This represents the velocity component of the vertical mass.

6. The target depth classification method based on vector acoustic field polarization features according to claim 5, characterized in that, In step three, under the conditions of linear mode superposition, frequency independence, and modal phase independence, the second-order polarization covariance matrix can be decomposed as follows: ； in: ； ； in, For the first The horizontal-vertical polarization direction vector corresponding to the first mode; therefore: ； The final mapping chain is: 。 7. The target depth classification method based on vector acoustic field polarization features according to claim 1, characterized in that, In step four, the eigenvalue decomposition of the second-order polarization covariance matrix R is performed, and its mathematical expression is: ； in, The eigenvectors corresponding to the eigenvalues; Eigenvalues ​​represent the energy distribution of the vector sound field along the principal polarization and secondary polarization directions. Corresponding to the dominant coherent component, eigenvalue Corresponding to multimodal interference residual components.

8. The target depth classification method based on vector acoustic field polarization features according to claim 1, characterized in that, In step five, the polarizability characterizes the degree of energy dominance in the principal polarization direction, and is defined as follows: ； in, Eigenvalues ​​represent the energy distribution of the vector sound field along the principal polarization and secondary polarization directions. Corresponding to the dominant coherent component, eigenvalue Corresponding to multimodal interference residual components.

9. The target depth classification method based on vector acoustic field polarization features according to claim 1, characterized in that, In step six, when the polarizability P of the target signal is greater than the classification threshold η, it is a deep source; when the polarizability P of the target signal is less than the classification threshold η, it is a shallow source.

Images