A shallow sea low-frequency near-field stratified seabed geoacoustic inversion method and device

Abstract

A method for inverting the shallow sea low-frequency near-field stratified seabed geoacoustic parameters, comprising: respectively obtaining the pulse of the direct wave and the pulse of the seabed reflected wave; converting the pulse time-domain signal into a frequency-domain signal and calculating the actual reflection coefficient; constructing a neural network model based on the cross-self attention mechanism based on the reflection coefficient matrix of the frequency-domain signal, the input of the neural network being the simulated reflection coefficient matrix and the output being the predicted value of the geoacoustic parameters; wherein, the simulated reflection coefficient matrix is obtained by forward modeling; training the model by means of multi-task learning, and using the trained model to predict the actual reflection coefficient matrix, wherein the actual reflection coefficient matrix is the input of the trained model. This method can achieve efficient and accurate multi-parameter inversion, and effectively improve the inversion accuracy and stability of the near-field seabed stratification and its geoacoustic parameters.

Description

Technical Field

[0001] The present invention relates to the technical field of acoustic inversion, and particularly to a method and device for shallow sea low-frequency near-field layered seabed geoacoustic inversion. Background Art

[0002] The acoustic properties of seabed sediments have a significant impact on sound propagation. Especially in the shallow sea environment, after multiple reflections of sound waves by the seabed, the geoacoustic parameters of sediments (such as sound speed, density, attenuation coefficient, sediment layer thickness, etc.) are of great significance for sound field modeling, target detection, and sonar performance evaluation. The methods for obtaining the geoacoustic parameters of sediments mainly include two types: direct measurement and geoacoustic inversion. The direct measurement technology has a high cost, and the sampling depth and range are limited. The geoacoustic inversion technology penetrates the seabed with sound signals and uses the depth information and geoacoustic characteristics contained therein to obtain the acoustic parameters of deeper layers and a wider area. However, in the currently known geoacoustic inversion technologies, the coupling problem between the sound speed and thickness in each seabed layer is usually not solved. Summary of the Invention

[0003] To solve the problems existing in the prior art, the embodiments of the present application provide a method, device, computing device, computer storage medium, and product containing a computer program for shallow sea low-frequency near-field layered seabed geoacoustic inversion, which can achieve efficient and accurate multi-parameter inversion and effectively improve the inversion accuracy and stability of the near-field seabed layer and its geoacoustic parameters.

[0004] In a first aspect, the embodiments of the present application provide a method for shallow sea low-frequency near-field layered seabed geoacoustic inversion, including: respectively obtaining the pulse of the direct wave and the pulse of the seabed reflected wave; converting the pulse time-domain signal into a frequency-domain signal and calculating the actual reflection coefficient; based on the reflection coefficient matrix of the frequency-domain signal, constructing a neural network model based on the cross-self-attention mechanism, where the input of the neural network is the simulated reflection coefficient matrix and the output is the predicted value of the geoacoustic parameter; wherein, the simulated reflection coefficient matrix is obtained by forward modeling; training the model in a multi-task learning manner, and using the trained model to predict the actual reflection coefficient matrix, where the actual reflection coefficient matrix is the input of the trained model.

[0005] In some possible implementation manners, constructing a neural network model based on the cross-self-attention mechanism includes: constructing a forward model, where the input of the forward model is the seabed layer and the geoacoustic parameter, and the output is the reflection coefficient matrix; dividing the reflection coefficient matrix into two branches in two directions of frequency and angle, and embedding the marker vector; respectively adding feature marker vectors to each branch and then performing position encoding; respectively inputting all the marker vectors into the L-layer multi-channel self-attention mechanism module, using residual connections between layers, and calculating the self-attention weight within the l-th layer; fusing the reflection coefficient features in the angle domain and the frequency domain to obtain the global feature, and mapping the global feature to the geoacoustic parameter vector.

[0006] In some possible implementations, the model is trained through multi-task learning, including: adopting a multi-task learning method based on gradient normalization, taking the prediction of each ground sound parameter as a separate sub-task, and assigning a separate task weight to each sub-task; using the automatic differentiation method to calculate the gradient norm of the loss of each sub-task on the neural network weight parameters, and calculating the task gradient loss according to the training rate of each sub-task; updating the gradient norms of all sub-tasks to make the gradient norms close to the standard gradient, updating the neural network weight parameters and normalizing the weights of the sub-tasks.

[0007] In some possible implementations, when assigning a separate task weight to each sub-task, the total loss function of the k-th training cycle in the neural network model is calculated according to the following formula:

[0008]

[0009] where, L sum (k) represents the multi-task total loss, N t represents the number of sub-tasks, ω i (k) represents the task weight of the i-th sub-task, L i (k) represents the loss of the i-th sub-task, and k represents the k-th frequency point.

[0010] In some possible implementations, the gradient norm is calculated according to the following formula:

[0011]

[0012] where, represents the gradient norm of the sub-task loss on the neural network weight parameter W, W represents the neural network weight parameter, represents the partial differential operator, and i represents the i-th sub-task.

[0013] In some possible implementations, the calculation of the task gradient loss is calculated according to the following formula:

[0014]

[0015] where, L grad represents the gradient loss, N t represents the number of sub-tasks, represents the average gradient of all sub-tasks, and there is r i (k) represents the relative training rate of the i-th sub-task, and there is α represents the intensity parameter used to control the gradient normalization of the sub-task.

[0016] In some possible implementation manners, the processing within the angle branch can be performed according to the following formula:

[0017]

[0018] In the formula, Q h represents the angle branch query matrix, K h represents the angle branch key matrix, V h represents the angle branch value matrix, h represents the number of channels of the multi-channel self-attention mechanism, and each channel can learn different feature representations. represents the learnable projection matrix of the angle branch query matrix. represents the learnable projection matrix of the angle branch key matrix. represents the learnable projection matrix of the angle branch value matrix, Atten represents the self-attention weight, d θ represents the feature vector dimension of the angle branch, represents the frequency marking vector of the l-1 layer, and T represents the transpose.

[0019] In some possible implementation manners, the processing within the frequency branch can be performed according to the following formula:

[0020]

[0021] In the formula, Q h ′ represents the frequency branch query matrix, K h ′ represents the frequency branch key matrix, V h ′ represents the frequency branch value matrix, h represents the number of channels of the multi-channel self-attention mechanism, and each channel can learn different feature representations. represents the learnable projection matrix of the frequency branch query matrix. represents the learnable projection matrix of the frequency branch key matrix. represents the learnable projection matrix of the frequency branch value matrix, Atten represents the self-attention weight, d F represents the feature vector dimension of the frequency branch, represents the frequency marking vector of the l-1 layer, and T represents the transpose.

[0022] In some possible implementation manners, the reflection coefficient features that fuse the angle domain and the frequency domain can be performed according to the following formula:

[0023]

[0024] In the formula, q represents the cross query vector, W q represents the linear mapping matrix, represents the CLS calculated through the self-attention mechanism θ , Kh 'The key matrix representing the frequency branch, The frequency marking vector representing the L layer, The projection matrix V representing the key matrix within the frequency branch, h 'The projection matrix V representing the value matrix within the frequency branch, h 'The frequency branch value matrix, and CrossAtten represents the cross-attention weight between the angle branch and the frequency branch.

[0025] In a second aspect, an apparatus for inverting the shallow sea low-frequency near-field stratified seabed geoacoustic parameters provided by an embodiment of the present application includes: an acquisition module configured to acquire the pulse of the direct wave and the pulse of the seabed reflected wave respectively; a processing module configured to convert the time-domain signal of the pulse into a frequency-domain signal and calculate the actual reflection coefficient; the processing module is further configured to construct a neural network model based on the cross-self-attention mechanism based on the reflection coefficient matrix in the frequency domain, where the input of the neural network is the simulated reflection coefficient matrix and the output is the predicted value of the geoacoustic parameter; wherein, the simulated reflection coefficient matrix is obtained by forward modeling; the processing module is further configured to train the model in a multi-task learning manner and use the trained model to predict the actual reflection coefficient matrix, where the actual reflection coefficient matrix is the input of the trained model.

[0026] In a third aspect, an embodiment of the present application provides a computer-readable storage medium, including computer-readable instructions, when the computer reads and executes the computer-readable instructions, causing the computer to execute the method according to any one of the first aspect.

[0027] In a fourth aspect, an embodiment of the present application provides a computing device, including a processor and a memory, where computer program instructions are stored in the memory, and when the computer program instructions are run by the processor, the method according to any one of the first aspect is executed.

[0028] In a fifth aspect, an embodiment of the present application provides a product including a computer program, when the computer program product runs on a processor, causing the processor to execute the method according to any one of the first aspect. Description of the Drawings

[0029] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention, and those of ordinary skill in the art can obtain other drawings without creative efforts based on these drawings.

[0030] Figure 1 It is a schematic diagram of an application scenario of a method for inverting shallow sea low-frequency near-field stratified seabed geoacoustic parameters provided by an embodiment of the present application;

[0031] Figure 2 It is a schematic flow diagram of a shallow - sea low - frequency near - field stratified seabed acoustic inversion method provided by an embodiment of the present application;

[0032] Figure 3 It is a schematic diagram of a near - field time - domain signal provided by an embodiment of the present application;

[0033] Figure 4 It is a schematic diagram of the pulse compression result of the acoustic signals received by all hydrophones on a vertical array provided by an embodiment of the present application;

[0034] Figure 5 It is a schematic diagram of the correction result of the offset position of the hydrophone on a vertical array provided by an embodiment of the present application;

[0035] Figure 6 It is a schematic diagram of the frequency - domain amplitude of the direct wave and the seabed reflection wave provided by an embodiment of the present application;

[0036] Figure 7 It is a schematic diagram of the average energy at the center frequencies of the direct wave and the seabed reflection wave provided by an embodiment of the present application;

[0037] Figure 8 It is a schematic diagram of the actual reflection coefficient of a near - field stratified seabed provided by an embodiment of the present application;

[0038] Figure 9 It is a schematic diagram of a cross - self - attention mechanism neural network model provided by an embodiment of the present application;

[0039] Figure 10 It is a schematic diagram of the seabed reflection coefficient of a near - field stratified seabed environment provided by an embodiment of the present application;

[0040] Figure 11 It is a schematic diagram of the global sensitivity distribution of an objective function near the optimal solution provided by an embodiment of the present application;

[0041] Figure 12 It is a schematic diagram of the normalized global attention weight distribution of a cross - self - attention mechanism model;

[0042] Figure 13 It is a schematic diagram of the structure of a shallow - sea low - frequency near - field stratified seabed acoustic inversion device provided by an embodiment of the present application;

[0043] Figure 14 It is a schematic diagram of the structure of a computing device provided by an embodiment of the present application. Specific embodiments

[0044] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Apparently, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0045] The term "and / or" in this document describes the associated relationship of associated objects and indicates that three relationships may exist. For example, A and / or B may represent: A exists alone, A and B exist simultaneously, and B exists alone. The symbol " / " in this document indicates that the associated objects are in an "or" relationship. For example, A / B means A or B.

[0046] The terms "first", "second", etc. in the description and claims of this document are used to distinguish different objects, rather than to describe a specific order of the objects. For example, the first response message and the second response message are used to distinguish different response messages, rather than to describe the specific order of the response messages.

[0047] In the embodiments of this application, words such as "exemplary" or "for example" are used to represent examples, illustrations, or explanations. Any embodiment or design solution described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or having more advantages than other embodiments or design solutions. Rather, the use of words such as "exemplary" or "for example" is intended to present relevant concepts in a specific manner.

[0048] In the description of the embodiments of this application, unless otherwise specified, the meaning of "a plurality of" refers to two or more. For example, a plurality of processing units refers to two or more processing units, and a plurality of elements refers to two or more elements.

[0049] To facilitate the understanding of the embodiments of this application, the following will further explain with specific examples in conjunction with the accompanying drawings. The examples do not constitute a limitation to the embodiments of the present invention.

[0050] According to the distance of the sound source, ground sound inversion can be divided into near-field and mid-far-field types. Mid-far-field ground sound inversion can obtain the seabed acoustic parameters in a large sea area, but it requires the sound source and receiver to be located in different places. This multi-base inversion method is relatively complex in actual operation. Near-field ground sound inversion can arrange the sound source and receiving equipment on a single base to achieve motorized real-time inversion, which has strong practicability. In addition, the incident angle of the near-field sound signal on the seabed is larger and the penetration depth into the seabed is deeper, so a more refined seabed stratification structure and its ground acoustic parameters can be obtained. Traditional methods for obtaining near-field seabed stratification mainly include the layer stripping method and the matched field method. The layer stripping method requires the time interval between the reflection signals of each seabed layer to be greater than the signal pulse width, so a broadband high-frequency strong pulse sound source needs to be used. The matched field method has the problem of multi-extremum, is easy to fall into local extreme points, and has low computational efficiency. In addition, both the layer stripping method and the matched field method cannot solve the coupling problem of sound velocity and thickness in each seabed layer.

[0051] In view of this, in order to solve the multi-parameter coupling problem in traditional near-field inversion methods, an embodiment of the present application provides a shallow sea low-frequency near-field stratified seabed ground sound inversion method, which is implemented based on deep learning technology. Deep learning technology can train a neural network model to adaptively extract features from acoustic data. According to the sensitivity differences of different ground acoustic parameters, a suitable neural network model is constructed and an effective optimization algorithm is adopted, which is expected to solve the multi-parameter coupling problem in seabed stratification and its ground acoustic parameter inversion. The acoustic reflection coefficient of the stratified seabed shows an oscillatory structure with frequency and grazing angle, and its oscillatory characteristics are related to the seabed stratification and its ground acoustic parameters. By analyzing the near-field pulse signal, the seabed reflection coefficients at different angles and frequencies are obtained. A neural network model based on the cross-self-attention mechanism is trained by using the multi-task learning method to respectively extract the features of the reflection coefficient in the frequency domain and the angle domain, and learn the relationship between these features and the ground acoustic parameters, so as to realize the inversion of the ground acoustic parameters of the near-field stratified seabed, providing a theoretical and technical basis for realizing the motorized real-time inversion of the seabed stratification structure and its ground acoustic parameters.

[0052] Exemplarily, Figure 1 Fig. shows a schematic diagram of the application scenario of a shallow sea low-frequency near-field stratified seabed ground sound inversion method provided by an embodiment of the present application. As Figure 1 shown, a plurality of hydrophones are distributed on the vertical receiving array, and the hydrophones are used to receive acoustic signals. The ship stops near the receiving array and lowers the transmitting transducer. In this embodiment, the UW350 low-frequency sound source transducer can be selected as the transducer to transmit a linear frequency modulated signal of 30 - 1000 Hz. The transducer is used as the signal source to invert the ground acoustic parameters of the near-field shallow surface seabed. Therefore, the influence of the sediment layer attenuation coefficient can be ignored, and the parameters to be inverted are the seabed surface sound velocity, the seabed second layer sound velocity, and the thickness of the seabed surface.

[0053] The inversion method provided by the embodiments of the present application can be applied to this scenario. Exemplarily, Figure 2 FIG. shows a schematic flow chart of a shallow sea low-frequency near-field stratified seabed acoustic inversion method provided by the embodiments of the present application. As Figure 2 shown, the method may include the following steps:

[0054] S21: Obtain the pulses of the direct wave and the seabed reflected wave respectively.

[0055] In this embodiment, the transmitting transducer emits an acoustic wave signal to the seabed, and a plurality of hydrophones are used to receive the returned signal. On the basis of considering the correction of the transmitting spectrum level of the transmitting transducer and the Doppler effect, the received signal and the sound source signal are correlated to obtain a broadband pulse signal, denoted as P(t). According to the position of the sound source transducer, the position of the hydrophone, the local seawater sound velocity profile, and the corresponding relationship between the propagation distance and arrival time of the direct wave acoustic ray at each receiving depth of the hydrophone array, the formation attitude of the hydrophone array is calibrated. Analyze the near-field multi-path arrival structure, and pick up the pulse signals of the direct wave and the seabed reflected wave. Among them, the ray method can be used to analyze the near-field multi-path arrival structure.

[0056] Taking the Figure 1 shown application scenario as an example, there are 17 hydrophones on the vertical array mooring buoy for receiving acoustic signals. The experimental ship stops near the mooring buoy and hoists the UW350 transmitting transducer to emit a linear frequency modulated signal of 30 - 1000 Hz. Figure 3 FIG. shows the near-field time-domain signal after pulse compression. As Figure 3 shown, for the near-field pulse signal after pulse compression of the received signal, the solid line is the pulse compression result considering the correction of the transmitting sound source level, and the dotted line is the direct pulse compression result. It can be seen that the pulse compression considering the sound source level correction can suppress the sidelobes. Use the ray model to analyze the near-field multi-path arrival structure, and pick up the pulse peaks of the direct wave and the seabed reflected wave. Figure 4 FIG. shows the pulse compression results of the acoustic signals received by all the hydrophones on the vertical array. As Figure 4 shown, the circles in the figure mark the arrival times of the direct wave, the seabed reflected wave, and the sea surface reflected wave in the signals of each hydrophone calculated by Bellhop. Among them, the seabed reflected wave, the direct wave, and the sea surface reflected wave are clearly distinguishable from the 1st to 12th hydrophones on the seabed. Figure 5 FIG. shows the result of positioning the offset positions of each hydrophone on the vertical array by using information such as the arrival time difference of the direct wave of each hydrophone, the depth position of the sound source, the seawater depth and the seawater sound velocity at the vertical array position.

[0057] S22: Convert the pulse time-domain signal into a frequency-domain signal, and calculate the actual reflection coefficient.

[0058] In this embodiment, the pulse signals of the direct wave and the seabed reflection wave are respectively intercepted, and a fast Fourier transform is performed to convert the time-domain signal into a frequency-domain signal. The entire frequency range is divided according to a certain center frequency and a one-third octave bandwidth to obtain several frequency bands. The average energy of each frequency point within each frequency band is calculated as the energy at the center frequency, and the amplitudes of the direct wave and the seabed reflection wave are calculated respectively. After compensating for the spreading loss, the actual reflection coefficient is calculated.

[0059] In some possible embodiments, the pulse signal of the direct wave is denoted as P d (t), and the pulse signal of the seabed reflection wave is denoted as P r (t). The conversion of the time-domain signal into a frequency-domain signal is performed according to the following formula:

[0060]

[0061] where P(f k ) represents the time-domain signal at the k-th frequency point, f k represents the k-th frequency point, and there is f k = k·f s / N, f s represents the sampling rate, N represents the number of time-domain sampling points, P(n) represents the time-domain signal at the n-th time point, and j represents the imaginary unit.

[0062] In some possible embodiments, the actual reflection coefficient is calculated according to the following formula:

[0063]

[0064] where R obs (f,θ) represents the actual reflection coefficient, P r (f) represents the amplitude of the reflection wave at frequency f, P d (f) represents the amplitude of the direct wave at frequency f, D r (θ) represents the propagation distance of the reflection wave, D d (θ) represents the propagation distance of the direct wave. The propagation distances D r (θ) and D d (θ) can be calculated from the depths of the sound source transducer and the hydrophone, the horizontal distance between the two, and the sea depth.

[0065] Please continue to refer to Figures 1 - 5 , and respectively intercept the pulse signal P Figure 4 of the direct wave and the pulse signal P d (t) of the seabed reflection wave of the 1st to 13th hydrophones in r (t), and then perform a fast Fourier transform to convert the time-domain signal into a frequency-domain signal. The sampling rate f sis 16000 Hz. Taking the actual signal with a sound source depth of 73.6 m, a hydrophone depth of 45.58 m, and a horizontal distance of 221.98 m as an example, the frequency-domain amplitudes of the direct wave and the seabed reflection wave are as Figure 6 shown. For the frequency-domain signal from 30 Hz to 1 kHz, the center frequencies are selected every 50 Hz from 150 Hz to 1 kHz, and the bandwidth is one-third octave for each. A total of 18 frequency bands are obtained. The average energy of each frequency point within each frequency band is calculated as the energy at the center frequency, as shown in Figure 7 shown. The amplitudes of the direct wave and the seabed reflection wave are calculated respectively, and the actual reflection coefficient is calculated after compensating for the spreading loss. Figure 8 is the actual seabed reflection coefficient with a grazing angle of 31.73 degrees.

[0066] S23: Based on the reflection coefficient matrix of the frequency-domain signal, a neural network model based on the cross-self-attention mechanism is constructed. The input of the neural network is the simulated reflection coefficient matrix, and the output is the predicted values of the geoacoustic parameters.

[0067] In this embodiment, first, the wavenumber integration method is used as the forward model for calculating the seabed reflection coefficient in the near-field stratified seabed environment. All parameter combinations are traversed at a certain step size within the value range of the geoacoustic parameters. For each set of geoacoustic parameters, the simulated reflection coefficient matrix is calculated, and the reflection coefficient matrix is denoted as R cal (f,θ). All the samples together form a data set for training the deep neural network. The neural network model can select a neural network model based on the cross-self-attention mechanism. The input of the model is the reflection coefficient matrix R cal of all seabed models (f,θ), and the output is the predicted values of multiple geoacoustic parameters.

[0068] Specifically, the forward model uses the wavenumber integration method to simulate the seabed reflection coefficient. It calculates the simulated reflection coefficient matrix based on the known geoacoustic parameters (such as the properties of the seabed, the sound speed profile of the seawater, etc.). These simulated reflection coefficient matrices are used as a data set for training the neural network model. Exemplarily, Figure 9 shows a schematic diagram of a cross-self-attention mechanism neural network model provided by an embodiment of the present application. As shown in Figure 9 shown, the neural network model divides R cal (f,θ) into two branches in the frequency and angle directions respectively and embeds the marker vectors. Feature marker vectors CLS θ and CLS fAfter that, position encoding is performed. All the token vectors are respectively input into the multi-channel self-attention mechanism modules of L layers, and residual connections are used between layers to calculate the self-attention weights within the l-th layer. Among them, the angular branch contains the frequency-domain reflection coefficient vectors at various angles, and the frequency branch contains the angular-domain reflection coefficient vectors at various frequencies. The global features are obtained by processing and fusing the reflection coefficient features in the angular domain and the frequency domain, and then the global features are mapped to the geoacoustic parameter vectors. Among them, i represents the number of predicted geoacoustic parameters.

[0069] In some possible embodiments, the processing within the angular branch can be carried out according to the following formula:

[0070]

[0071] In the formula, Q h represents the query matrix of the angular branch, K h represents the key matrix of the angular branch, V h represents the value matrix of the angular branch, h represents the number of channels of the multi-channel self-attention mechanism, and each channel can learn different feature representations. represents the learnable projection matrix of the query matrix of the angular branch. represents the learnable projection matrix of the key matrix of the angular branch. represents the learnable projection matrix of the value matrix of the angular branch, Atten represents the self-attention weight, d θ represents the feature vector dimension of the angular branch, represents the frequency token vector of the (l - 1)-th layer, and T represents the transpose.

[0072] The feature vector dimension of the branch is processed by softmax to convert the matrix inner product into a normalized weight coefficient, and then after being processed by the multi-channel self-attention mechanism modules of L layers, the sequence representation of the angular branch is obtained. Among them, is the CLS calculated through the self-attention mechanism. θ Since CLS θ calculates self-attention with all the frequency token vectors at all angles, it contains the global features of R cal (f, θ) in the angular dimension.

[0073] In some possible embodiments, the processing within the frequency branch can be carried out according to the following formula:

[0074]

[0075] In the formula, Q h ′ represents the query matrix of the frequency branch, K h ′ represents the key matrix of the frequency branch, V h' represents the frequency branch value matrix, h represents the number of channels of the multi-channel self-attention mechanism, and each channel can learn different feature representations. The learnable projection matrix representing the frequency branch query matrix. The learnable projection matrix representing the frequency branch key matrix. The learnable projection matrix representing the frequency branch value matrix, Atten represents the self-attention weight, d F Represents the feature vector dimension of the frequency branch. Represents the frequency token vector of the l-1 layer, T represents the transpose.

[0076] The processing flow of the frequency branch is the same as that of the angle branch. After passing through the L-layer multi-channel self-attention mechanism module, the sequence representation of the frequency branch is output. Then map the angle domain CLS θ to the feature space of the frequency branch and calculate the cross-attention weight.

[0077] In some possible embodiments, the reflection coefficient features that fuse the angle domain and the frequency domain can be performed according to the following formula:

[0078]

[0079] In the formula, q represents the cross-query vector, W q represents the linear mapping matrix. Represents the CLS calculated through the self-attention mechanism. θ , K h ' represents the frequency branch key matrix. Represents the frequency token vector of the L layer. Represents the projection matrix of the frequency branch inner key matrix, V h ' represents the projection matrix of the frequency branch inner value matrix, V h ' represents the frequency branch value matrix, CrossAtten represents the cross-attention weight between the angle branch and the frequency branch.

[0080] Since contains the frequency domain feature information extracted by the frequency branch feature token vector CLS F Therefore, CrossAtten in the above formula realizes the fusion of the angle domain and frequency domain features.

[0081] For reference Figures 1 - 9, in this example, the wavenumber integration method OASR is used as the forward model for calculating the seabed reflection coefficient in a near-field stratified seabed environment. Since the sound source level of the experimental transmitted signal is small and the energy of the transmitted waves in each layer below the seabed is very small, a two-layer model is considered first. All parameter combinations are traversed at a certain step within the range of geoacoustic parameters, and the simulated reflection coefficient matrix R is calculated for each set of geoacoustic parameters. cal (f,θ), and all the samples together form a dataset for training the deep neural network. In this example, the parameter settings of the dataset are shown in the following table:

[0082] Parameters to be inverted Value range Step size Unit Surface sound velocity [1500,1700] 10 m / s Sound velocity of the second layer [1500,1700] 10 m / s Surface thickness [0,10] 0.5 m

[0083] Construct a neural network model based on the cross-self-attention mechanism. The input of the model is R cal (f,θ), and the output is the predicted values of the seabed surface sound speed, the seabed second-layer sound speed, and the seabed surface thickness. The model divides R cal (f,θ) into two branches in the frequency and angle directions respectively and embeds the token vectors. The angle branch contains the frequency-domain reflection coefficient vectors at each angle, and the frequency branch contains the angle-domain reflection coefficient vectors at each frequency. Feature token vectors CLS θ and CLS f are added to each branch respectively and then position encoding is performed. All the token vectors are input into a 5-layer (i.e., L = 5) multi-channel self-attention mechanism module, and residual connections are used between layers. During the calculation, the feature vector dimension d θ = 192 of the angle branch and the vector dimension d f = 96 of the frequency branch are taken for calculation. Through the processing of the angle branch and the frequency branch, the reflection coefficient features in the angle domain and the frequency domain are fused to obtain the global features, and then the global features are mapped to the geoacoustic parameter vector i = 3 represents the number of predicted geoacoustic parameters, thus realizing the prediction of geoacoustic parameters based on the cross-self-attention mechanism.

[0084] S24: Train the model in the way of multi-task learning, and use the trained model to predict the actual reflection coefficient matrix.

[0085] In this embodiment, a multi-task learning method based on gradient normalization is used. The prediction of each geosound parameter is regarded as a separate sub-task, and a separate task weight is assigned to each sub-task. Using the automatic differentiation method of the deep learning framework, the gradient norm of the loss of each sub-task on the neural network weight parameter W can be efficiently calculated. Then, according to the training rate of each sub-task, the task gradient loss is calculated to adjust the gradient of each sub-task to make it close to the standard gradient. The gradient of the gradient on the sub-task weight is calculated through automatic differentiation, and the gradient norms of all sub-tasks are updated to make them close to the standard gradient. Then, the neural network weight parameter is updated and the weights of the sub-tasks are normalized. This process is repeated until the model training ends. After the model training ends, when the actual reflection coefficient matrix R obs (f,θ) is input, the predicted values of i geosound parameters can be output, thus realizing the simultaneous inversion of multiple geosound parameters.

[0086] In some possible embodiments, when assigning a separate task weight to each sub-task, the total loss function in the k-th training cycle of the neural network model is calculated according to the following formula:

[0087]

[0088] In the formula, L sum (k) represents the total multi-task loss, N t represents the number of sub-tasks, ω i (k) represents the task weight of the i-th sub-task, L i (k) represents the loss of the i-th sub-task, and k represents the k-th frequency point.

[0089] In some possible embodiments, the gradient norm can be calculated according to the following formula:

[0090]

[0091] In the formula, represents the gradient norm of the sub-task loss on the neural network weight parameter W, W represents the neural network weight parameter, represents the partial differential operator, and i represents the i-th sub-task.

[0092] In some possible embodiments, to calculate the task gradient loss, it can be calculated according to the following formula:

[0093]

[0094] In the formula, L grad represents the gradient loss, N t represents the number of sub-tasks, represents the average gradient of all sub-tasks, and there is r i(k) represents the relative training rate of the i-th sub-task, and there is α represents the intensity parameter used to control the normalization of the sub-task gradient.

[0095] After the model optimization converges, input the actual reflection coefficient matrix R obs (f, θ), and the predicted values of 3 geoacoustic parameters can be obtained as the output, thus realizing the simultaneous inversion of the sound speed of the seabed surface layer, the sound speed of the second layer, and the surface layer thickness. In Figure 1 In the application scenario shown, the inversion results in the experimental sea area are that the sound speed of the seabed surface layer is 1600 - 1610 m / s, the sound speed of the second layer is 1540 - 1550 m / s, and the surface layer thickness is 2.5 - 3 m. This prediction result is basically consistent with the Figure 1 seabed sampling results shown.

[0096] To further verify the practicability of this method, in addition to conducting experiments in the Figure 1 application scenario shown, the feature extraction ability of the neural network model with the cross - autocorrelation degree mechanism was also verified using a simulation environment. In the simulation environment, the sound speed of the seabed surface layer is 1550 m / s, the sound speed of the second layer of the seabed is 1650 m / s, and the surface layer thickness is 5 m. Calculate the emission coefficient of the layered seabed based on OASR, with the angle range of 20 - 60 degrees and the frequency range of 600 Hz - 1 kHz. The results are as Figure 10 shown. Near the optimal solution (i.e., the true value) of the geoacoustic parameters, calculate the partial derivatives of the reflection coefficient at each angle point and frequency point with respect to the three geoacoustic parameters, and its root mean square is used as the global sensitivity of the objective function near the optimal solution, as Figure 11 shown. Use a dataset with the same angle and frequency range as the simulation environment to train the cross - self - attention mechanism model of the present invention. After the model optimization converges, the model - normalized global attention weights are as Figure 12 shown. By comparing Figure 10 and Figure 11 , it can be found that the region with a higher sensitivity of the objective function is consistent with the region where the seabed reflection coefficient changes drastically, indicating that the seabed reflection coefficient in these regions is more important for the inverted geoacoustic parameters. By comparing Figure 11 and Figure 12 , it can be found that the distribution of the global attention weights is consistent with the global sensitivity of the objective function, indicating that the cross - self - attention mechanism model proposed in the embodiments of this application can effectively focus on the more critical local features in the reflection coefficient matrix, thereby improving the accuracy of the inversion.

[0097] The above is the shallow - sea low - frequency near - field layered seabed geo - acoustic inversion method provided by the embodiments of the present application. By utilizing the oscillation characteristics of the layered seabed reflection coefficient in the angular domain and frequency domain, a neural network model is trained using a multi - task learning method to achieve accurate and efficient acquisition of layered seabed acoustic parameters. A transducer is used to emit low - frequency broadband signals, and a near - field hydrophone array is used to receive acoustic signals. Considering the source spectral level and Doppler effect of the transmitting transducer, the source signal is compensated and then the received signal is pulse - compressed to extract the seabed reflected wave and direct wave. The seabed reflection coefficient at each frequency and angle is calculated, a neural network model based on the cross - self - attention mechanism is constructed to extract the features of the reflection coefficient matrix in the angular domain and frequency domain, and a multi - task learning method based on gradient normalization is used to train the model to achieve efficient and accurate multi - parameter inversion, effectively improving the inversion accuracy and stability of the near - field seabed stratification and its geo - acoustic parameters.

[0098] It can be understood that the magnitudes of the sequence numbers of the steps in the above - mentioned various embodiments do not mean the order of execution. The order of execution of each process should be determined by its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of the present application. In addition, in some possible implementation manners, the steps in the above - mentioned embodiments can be selectively executed according to the actual situation, can be partially executed, or can be fully executed, which is not limited here. Any feature of any embodiment of the present application, in whole or in part, can be freely combined anywhere without contradiction. The combined technical solutions are also within the scope of the present application.

[0099] A shallow - sea low - frequency near - field layered seabed geo - acoustic inversion method based on deep learning proposed by the present invention, and the system is configured with a transmitting transducer and a hydrophone array. The present invention places the transmitting transducer and the hydrophone array closely in water. The hydrophone receives the broadband linear frequency - modulated acoustic signal radiated by the transducer source, removes the Doppler effect and the source spectral level and performs pulse compression to obtain a broadband pulse signal; the attitude of the receiving array is corrected by using the depth and relative position of the transmitting transducer, the depth of the hydrophone, and the arrival time information of the direct wave; the arrival times of the direct wave, the sea - surface reflected wave, and the seabed reflected wave are obtained by using the sea - water depth and sound - speed information, and the channel signals with three completely separated propagation paths are selected to intercept the direct - wave and seabed - reflected - wave signals; the grazing angle corresponding to the seabed reflected wave is calculated according to the relative positions of the transmitting transducer and the hydrophone; finally, the seabed reflection coefficient matrix is calculated, and the angular and frequency oscillation structures of the seabed reflection coefficient are input into the trained neural network model to obtain the seabed stratification structure and its geo - acoustic parameters.

[0100] In the process of pulse compression of the received original signal, this method not only removes the Doppler effect caused by the movement of the sound source but also corrects the source signal considering the source level of the transmitting transducer, making the main lobe of the compressed pulse signal strong and the side lobes weak, which is beneficial to the resolution and interception of the direct wave and the reflected wave. A neural network model based on the cross-self-attention mechanism is constructed. The self-attention weights are calculated separately in the angle branch and the frequency branch, and the cross-attention mechanism is used to fuse the reflection coefficient features in the angle domain and the frequency domain to obtain global features. Then, the global features are mapped to the geoacoustic parameter vector. Through this feature extraction mechanism of separating first and then fusing, it is expected to improve the sensitivity of the parameters to be inverted and reduce the coupling between the parameters. Based on the multi-task learning method of gradient normalization, the prediction of each geoacoustic parameter is regarded as a separate sub-task, and a separate task weight is assigned to each sub-task, further eliminating the multi-parameter coupling problem.

[0101] It should be noted that since the source level is low in the example experiment used in the introduction of the method, the seabed reflection signal only contains the reflected waves of the surface layer and the second layer, so only the geoacoustic parameters of the seabed surface layer and the second layer are obtained. The numerical simulation results show that this method is applicable to the seabed stratification of more layers and the acquisition of their geoacoustic parameters. In practical applications, by increasing the transmission power of the transmitting transducer, the seabed stratification structure and geoacoustic parameters of deeper layers can be obtained. Therefore, the above examples are only used to illustrate the technical solutions of the present invention rather than to limit them. This near-field inversion method is also applicable to the operation mode of a ship towing a transmitting transducer and a hydrophone array at the same time, and this mode can better reflect the practicality of this inversion method in the mobile acquisition of seabed stratification and its geoacoustic parameters.

[0102] Based on the method in the above embodiments, the embodiments of the present application further provide a shallow sea low-frequency near-field stratified seabed geoacoustic inversion device. Exemplarily, Figure 13 A shallow sea low-frequency near-field stratified seabed geoacoustic inversion device is shown. As Figure 13 shown, the device 1300 includes an acquisition module 1301 and a processing module 1302.

[0103] Among them, the acquisition module 1301 is used to acquire the pulse of the direct wave and the pulse of the seabed reflected wave respectively;

[0104] The processing module 1302 is used to convert the time-domain signal of the pulse into a frequency-domain signal and calculate the actual reflection coefficient;

[0105] The processing module 1302 is further used to construct a neural network model based on the cross-self-attention mechanism based on the reflection coefficient matrix of the frequency-domain signal. The input of the neural network is the simulated reflection coefficient matrix, and the output is the predicted value of the geoacoustic parameter; wherein, the simulated reflection coefficient matrix is obtained by forward model simulation;

[0106] The processing module 1302 is further configured to train the model by means of multitask learning, and use the trained model to predict the actual reflection coefficient matrix, where the actual reflection coefficient matrix is the input of the trained model.

[0107] It should be understood that the above device is used to execute the method in the above embodiment. For the corresponding program module in the device, its implementation principle and technical effect are similar to the description in the above method. The working process of the device can refer to the corresponding process in the above method, which will not be elaborated here.

[0108] The present application further provides a computing device 1400. As Figure 14 shown, the computing device 1400 includes: a bus 1402, a processor 1404, a memory 1406, and a communication interface 1408. The processor 1404, the memory 1406, and the communication interface 1408 communicate with each other through the bus 1402. The computing device 1400 can be a server or a terminal device. It should be understood that the present application does not limit the number of processors and memories in the computing device 1400.

[0109] The bus 1402 can be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For the sake of convenience of representation, Figure 14 only one line is shown in the figure, but it does not mean that there is only one bus or one type of bus. The bus 1404 can include a path for transmitting information between various components (such as the memory 1406, the processor 1404, and the communication interface 1408) of the computing device 1400.

[0110] The processor 1404 can include any one or more of a central processing unit (CPU), a graphics processing unit (GPU), a microprocessor (MP), or a digital signal processor (DSP), etc.

[0111] The memory 1406 may include volatile memory, such as random access memory (RAM). The processor 1404 may also include non-volatile memory, such as read-only memory (ROM), flash memory, a hard disk drive (HDD), or a solid state drive (SSD).

[0112] The executable program code is stored in the memory 1406, and the processor 1404 executes the executable program code to respectively implement the functions of the foregoing obtaining module 1301 and processing module 1302, thereby implementing all or part of the steps of the method in the foregoing embodiments. That is, instructions for executing all or part of the steps of the method in the foregoing embodiments are stored on the memory 1406.

[0113] Alternatively, executable code is stored in the memory 1406, and the processor 1404 executes the executable code to respectively implement the functions of the foregoing shallow sea low-frequency near-field stratified seabed geoacoustic inversion device 1300, thereby implementing all or part of the steps of the method in the foregoing embodiments. That is, instructions for executing all or part of the steps of the method in the foregoing embodiments are stored on the memory 1406.

[0114] The communication interface 1408 uses a transceiver module such as, but not limited to, a network interface card or a transceiver to implement communication between the computing device 1400 and other devices or a communication network.

[0115] Based on the method in the foregoing embodiments, an embodiment of the present application provides a computer-readable storage medium storing a computer program, which, when running on a processor, causes the processor to execute the methods in the foregoing embodiments.

[0116] Based on the method in the foregoing embodiments, an embodiment of the present application provides a computer program product, which, when running on a processor, causes the processor to execute the methods in the foregoing embodiments.

[0117] It can be understood that the processor in the embodiments of the present application may be a central processing unit (CPU), or may also be other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), field programmable gate arrays (FPGAs), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. The general-purpose processor may be a microprocessor or any conventional processor.

[0118] The method steps in the embodiments of the present application may be implemented in a hardware manner or by a processor executing software instructions. The software instructions may be composed of corresponding software modules, and the software modules may be stored in a random access memory (RAM), flash memory, read-only memory (ROM), programmable ROM (PROM), erasable PROM (EPROM), electrically erasable PROM (EEPROM), registers, hard disks, removable hard disks, CD-ROMs, or any other form of storage medium well-known in the art. An exemplary storage medium is coupled to the processor so that the processor can read information from the storage medium and write information to the storage medium. Of course, the storage medium may also be a component of the processor. The processor and the storage medium may be located in an ASIC.

[0119] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the processes or functions described in the embodiments of the present application are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transmitted through the computer-readable storage medium. The computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center in a wired manner (such as coaxial cable, optical fiber, digital subscriber line (DSL)) or a wireless manner (such as infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that includes one or more integrated available media. The available medium can be a magnetic medium (for example, a floppy disk, a hard disk, a magnetic tape), an optical medium (for example, a DVD), or a semiconductor medium (for example, a solid state disk (SSD)), etc.

[0120] It can be understood that the various digital numbers involved in the embodiments of the present application are only for the convenience of description and are not used to limit the scope of the embodiments of the present application.

Claims

1. A shallow water low-frequency near-field layered seabed geoacoustic inversion method, characterized in that: The method comprises: Obtain direct wave pulses and seabed reflected wave pulses respectively; Convert the pulse time domain signal into a frequency domain signal and calculate the actual reflection coefficient; Based on the reflection coefficient matrix of the frequency domain signal, a neural network model based on the cross-self-attention mechanism is constructed, the input of the neural network is the simulated reflection coefficient matrix, and the output is the predicted value of the geoacoustic parameter; wherein the simulated reflection coefficient matrix is ​​obtained by simulation of the forward model; the construction of the neural network model based on the cross-self-attention mechanism includes: constructing a forward model, the forward model uses a wavenumber integration method to simulate the seabed reflection coefficient, and calculates the simulated reflection coefficient matrix according to the known geoacoustic parameters. The forward model inputs the seabed layer and the geoacoustic parameters, and the output is the reflection coefficient matrix; the reflection coefficient matrix is ​​divided into two branches in terms of frequency and angle, and a label vector is embedded; after adding a feature label vector to each branch, position encoding is performed; all label vectors are respectively input into an L-layer multi-channel self-attention mechanism module, and residual connections are used between the layers, and the self-attention weight is calculated in the lth layer; the reflection coefficient features in the angle domain and the frequency domain are integrated to obtain global features, and the global features are mapped to the geoacoustic parameter vector; wherein the processing in the angle branch is performed according to the following formula: In the formula, Q h Characterize the angle branch query matrix, K h Characterizing the angular branch bond matrix, V h represents the angle branch value matrix, h represents the number of channels of the multi-channel self-attention mechanism, and each channel can learn different feature representations. A learnable projection matrix that represents the query matrix of the angle branch, A learnable projection matrix that represents the angular branch bond matrix, The learnable projection matrix representing the angle branch value matrix, Atten represents the self-attention weight, d θ The dimension of the feature vector representing the angle branch, represents the frequency label vector of layer l-1, and T represents the transpose; The processing within the frequency branch is performed according to the following formula: In the formula, Q h ′ represents the frequency branch query matrix, K h ′ represents the frequency branch bond matrix, V h ′ represents the frequency branch value matrix, h represents the number of channels of the multi-channel self-attention mechanism, and each channel can learn different feature representations. A learnable projection matrix representing the frequency branch query matrix, A learnable projection matrix that represents the frequency branch bond matrix, The learnable projection matrix representing the frequency branch value matrix, Atten represents the self-attention weight, d F The dimension of the feature vector representing the frequency branch, represents the frequency label vector of layer l-1, and T represents the transpose; The model is trained by multi-task learning, and the trained model is used to predict the actual reflection coefficient matrix, wherein the actual reflection coefficient matrix is ​​the input of the trained model.

2. The method according to claim 1, characterized in that The model is trained by multi-task learning, including: A multi-task learning method based on gradient normalization is adopted to treat the prediction of each geoacoustic parameter as a separate subtask, and a separate task weight is assigned to each subtask; Using an automatic differentiation method, the gradient norm of each subtask loss on the neural network weight parameter is calculated, and according to the training rate of each subtask, the task gradient loss is calculated; Update the gradient norms of all subtasks to make them close to the standard gradient, update the neural network weight parameters and normalize the weights of the subtasks.

3. The method according to claim 2, characterized in that When assigning a separate task weight to each subtask, the total loss function for the kth training cycle in the neural network model is calculated according to the following formula: Where, L sum (k) represents the total multi-task loss, Nt represents the number of subtasks, ω i (k) Characterization i The task weight of each subtask is L i (k) represents the loss of the i-th subtask, and k represents the k-th frequency point.

4. The method according to claim 2, characterized in that: The gradient norm is calculated according to the following formula: In the formula, Represents the gradient norm of the subtask loss on the neural network weight parameter W, where W represents the neural network weight parameter, represents the partial differential operator, and i represents the i-th subtask.

5. The method according to claim 1, characterized in that The calculation task gradient loss is calculated according to the following formula: Where, L grad Characterize the gradient loss, N t Represents the number of subtasks, represents the average gradient of all subtasks, and has r i (k) represents the relative training rate of the i-th subtask, and has α represents the strength parameter used to control the normalization of subtask gradients.

6. The method according to claim 1, characterized in that The reflection coefficient characteristics of the angular domain and the frequency domain can be integrated according to the following formula: Where q represents the cross query vector, W q Represents the linear mapping matrix, Representation of CLS calculated by self-attention mechanism θ , K h ′ represents the frequency branch characteristic bond matrix, The frequency label vector representing the L layer, The projection matrix representing the bond matrix within the frequency branch, V h ′ represents the projection matrix of the frequency branch internal value matrix, V' h Represents the frequency branch value matrix, and CrossAtten represents the cross-attention weights of the angle branch and the frequency branch.

7. A shallow sea low-frequency near-field layered seabed geoacoustic inversion device, characterized in that: The device comprises: An acquisition module, used for respectively acquiring pulses of direct waves and pulses of seabed reflected waves; A processing module, used for converting the time domain signal of the pulse into a frequency domain signal and calculating an actual reflection coefficient; The processing module is also used to construct a neural network model based on the cross-self-attention mechanism based on the reflection coefficient matrix of the frequency domain signal, the input of the neural network is the simulated reflection coefficient matrix, and the output is the predicted value of the geoacoustic parameter; wherein the simulated reflection coefficient matrix is ​​obtained by forward model simulation; the construction of the neural network model based on the cross-self-attention mechanism includes: constructing a forward model, the forward model uses a wave number integration method to simulate the seabed reflection coefficient, and calculates the simulated reflection coefficient matrix according to the known geoacoustic parameters. The forward model inputs the seabed layer and the geoacoustic parameters, and the output is the reflection coefficient matrix; the reflection coefficient matrix is ​​divided into two branches in terms of frequency and angle, and a label vector is embedded; after adding a feature label vector to each branch, position encoding is performed; all label vectors are respectively input into an L-layer multi-channel self-attention mechanism module, residual connections are used between the layers, and self-attention weights are calculated in the lth layer; the reflection coefficient features in the angle domain and the frequency domain are integrated to obtain global features, and the global features are mapped to the geoacoustic parameter vector; wherein the processing in the angle branch is performed according to the following formula: In the formula, Q h Characterize the angle branch query matrix, K h Characterizing the angular branch bond matrix, V h represents the angle branch value matrix, h represents the number of channels of the multi-channel self-attention mechanism, and each channel can learn different feature representations. A learnable projection matrix that represents the query matrix of the angle branch, A learnable projection matrix that represents the angular branch bond matrix, The learnable projection matrix representing the angle branch value matrix, Atten represents the self-attention weight, d θ The dimension of the feature vector representing the angle branch, represents the frequency label vector of layer l-1, and T represents the transpose; The processing within the frequency branch is performed according to the following formula: In the formula, Q h ′ represents the frequency branch query matrix, K h ′ represents the frequency branch bond matrix, V h ′ represents the frequency branch value matrix, h represents the number of channels of the multi-channel self-attention mechanism, and each channel can learn different feature representations. A learnable projection matrix representing the frequency branch query matrix, A learnable projection matrix that represents the frequency branch bond matrix, The learnable projection matrix representing the frequency branch value matrix, Atten represents the self-attention weight, d F The dimension of the feature vector representing the frequency branch, represents the frequency label vector of layer l-1, and T represents the transpose; The processing module is also used to train the model by multi-task learning, and use the trained model to predict the actual reflection coefficient matrix, wherein the actual reflection coefficient matrix is ​​the input of the trained model.

Images