Regional sound propagation loss matrix prediction method based on conditional diffusion model

Abstract

The invention discloses a regional sound propagation loss matrix prediction method and device based on a conditional diffusion model, a medium and equipment. The method comprises the following steps: determining topographic data of a whole region of submarine topography according to key nodes of a dramatic change topography; performing omnibearing two-dimensional terrain modeling on the terrain data of the whole area to obtain two-dimensional terrain models of different areas, and calculating two-dimensional terrain feature data of different areas according to the two-dimensional terrain models; inputting the two-dimensional topographic feature data into a sound field calculation model, and outputting a sound propagation loss matrix under a corresponding region condition and labels of different regions; a matrix prediction model is constructed based on a U-Net network, sound field propagation matrixes of different areas of submarine topography are rapidly predicted by using the trained matrix prediction model, and a sonar detection result is obtained according to the predicted sound field propagation matrixes, so that end-to-end sound field modeling and prediction are realized. Compared with a traditional sound field calculation model, professional and complex underwater sound knowledge and tedious use rules are not needed.

Description

Technical Field

[0001] This application relates to the fields of marine acoustics and deep learning technology, and in particular to a method, apparatus, medium and device for predicting regional acoustic propagation loss matrix based on a conditional diffusion model. Background Technology

[0002] In recent years, various methods and models have emerged in the field of marine acoustic field computation. Traditional acoustic field computation methods mainly include wave equation theory and ray theory, but these methods still have some limitations, such as huge computational resource requirements and insufficient adaptability to different marine conditions. A detailed analysis follows:

[0003] Currently, underwater sound wave propagation is mainly studied using two methods: wave theory, which studies the changes in amplitude and phase of the sound signal in the sound field; and ray theory, which treats sound waves as sound ray beams under high-frequency conditions and typically studies the changes in sound intensity in the sound field as a function of the ray beam. Based on these two theories, several classical sound field calculation models have been developed to characterize the underwater sound propagation process, including normal mode models, ray models, and parabolic approximation models. These models generally contain a large number of parameters, involving multiple factors such as sound velocity profiles, seabed topography, and ocean current conditions. The complexity of parameter settings increases the difficulty of using these models. Most physical field-based models are derived from the sound wave equation, but closed-form solutions to the wave equation are often difficult to obtain in marine environments. Therefore, the wave equation usually employs multiple approximate solutions, most of which can be considered as variations of four basic methods: the ray method, normal mode models, parabolic equations, and wavenumber integrals.

[0004] While calculation methods based on acoustic field physics models are accurate, they typically require a long computation time and demand significant computing resources, which limits their efficiency in simulated combat exercises to some extent. Summary of the Invention

[0005] The main objective of this application is to provide a method, apparatus, medium, and device for predicting the regional acoustic propagation loss matrix based on a conditional diffusion model. The aim is to rapidly generate signal strength and channel response characteristics between multiple nodes within an end-to-end framework, thereby providing support for the rapid assessment of sonar detection and communication capabilities.

[0006] To achieve the above objectives, this application provides a method for predicting the regional acoustic propagation loss matrix based on a conditional diffusion model, comprising: adaptively dividing the seabed topography into triangular meshes for different regions using the OceanMesh2D toolbox according to the seabed topography and acoustic model, obtaining key nodes of drastic topography, and determining the topographic data of the entire seabed topography region based on the key nodes of drastic topography; performing omnidirectional two-dimensional topographic modeling on the topographic data of the entire region to obtain two-dimensional topographic models for different regions, and calculating two-dimensional topographic feature data for different regions based on the two-dimensional topographic models; inputting the two-dimensional topographic feature data into the acoustic field calculation model, outputting the acoustic propagation loss matrix under the corresponding regional conditions, and labels for different regions; constructing a matrix prediction model based on a U-Net network, inputting the acoustic propagation loss matrix and corresponding labels into the U-Net network, and using a multilayer perceptron to fuse the sound source depth, sound source frequency, and topographic features as conditional information with the output features of the intermediate layer of the U-Net network to jointly train the matrix prediction model, obtaining the trained matrix prediction model; using the trained matrix prediction model to quickly predict the acoustic field propagation matrix of different regions of the seabed topography, and obtaining sonar detection results based on the predicted acoustic field propagation matrix.

[0007] Optionally, the adaptive triangular meshing of the seabed topography in different regions using the OceanMesh2D toolbox includes: in flat areas, the OceanMesh2D toolbox automatically lowers the mesh resolution; while in ridge and canyon areas, it automatically raises the mesh resolution.

[0008] Optionally, the joint training of the matrix prediction model to obtain the trained matrix prediction model includes: gradually adding noise to the input data through a first control function during the forward diffusion process of the matrix prediction model, transforming it into a pure noise distribution; and reconstructing the pure noise distribution back into the original data through a second control function during the backward diffusion process of the matrix prediction model.

[0009] Optionally, the first control function is:

[0010]

[0011] Where is the intermediate state at step , is the scheduling parameter that controls the noise intensity, and is the random noise sampled from the standard normal distribution;

[0012] The second control function is:

[0013]

[0014] Where is the noise component predicted by the network, represents the cumulative variance, and is the noise standard deviation during denoising.

[0015] Optionally, determining the topographic data of the entire seabed topography region based on the key nodes of the dramatically changed topography includes: slicing the dramatically changed topography based on the key nodes of the dramatically changed topography, and constructing the topographic data of the entire seabed topography region based on the sliced ​​dramatically changed topography.

[0016] Optionally, the terrain features include: maximum depth, minimum depth, maximum slope, average slope, flatness, and number of oscillations.

[0017] Optionally, the sound field calculation model is Bellhop.

[0018] Furthermore, to achieve the above objectives, this application also provides a device for predicting the regional acoustic propagation loss matrix based on a conditional diffusion model, comprising: a terrain data determination module, used to adaptively triangularly mesh the seabed terrain in different regions using the OceanMesh2D toolbox according to the seabed terrain conditions and acoustic model, obtain key nodes of drastically changing terrain, and determine the terrain data of the entire seabed terrain region based on the key nodes of drastically changing terrain; a terrain feature calculation module, used to perform omnidirectional two-dimensional terrain modeling on the terrain data of the entire region, obtain two-dimensional terrain models for different regions, and calculate two-dimensional terrain feature data for different regions based on the two-dimensional terrain models; and a training data acquisition module, used to... The system inputs 3D topographic feature data into the sound field calculation model, outputting the sound propagation loss matrix under the corresponding regional conditions, as well as labels for different regions. The model training module is used to build a matrix prediction model based on the U-Net network. The sound propagation loss matrix and corresponding labels are input into the U-Net network, and the sound source depth, sound source frequency, and topographic features are fused with the output features of the intermediate layers of the U-Net network as conditional information through a multilayer perceptron to jointly train the matrix prediction model, resulting in the trained matrix prediction model. The prediction module is used to quickly predict the sound field propagation matrix of different regions of the seabed topography using the trained matrix prediction model, and obtain the sonar detection results based on the predicted sound field propagation matrix.

[0019] To achieve the above objectives, this application also provides a computer-readable storage medium including instructions that, when executed on a computer, cause the computer to perform the method for predicting the regional acoustic propagation loss matrix based on the conditional diffusion model provided in the above embodiments.

[0020] To achieve the above objectives, this application also provides an electronic device, which includes: at least one processor, a memory, and an input / output unit; wherein the memory is used to store a computer program, and the processor is used to call the computer program stored in the memory to execute the method for predicting the regional acoustic propagation loss matrix based on the conditional diffusion model provided in any of the foregoing embodiments.

[0021] This application proposes a method, apparatus, medium, and device for predicting regional acoustic propagation loss matrices based on a conditional diffusion model. The method involves adaptively dividing the seabed topography into triangular meshes using the OceanMesh2D toolbox based on seabed topography conditions and an acoustic model. This yields key nodes for dramatically changing topography, and the topographic data for the entire seabed region is determined based on these key nodes. A comprehensive two-dimensional topographic model is then performed on this data to obtain two-dimensional topographic models for different regions. Two-dimensional topographic feature data for each region is calculated based on these models. The two-dimensional topographic feature data is input into a sound field calculation model, which outputs the acoustic propagation loss matrix under the corresponding regional conditions, along with labels for different regions. A matrix prediction model is constructed based on a U-Net network, and the acoustic propagation loss matrix is ​​input into the U-Net network. The matrix and corresponding labels are used to fuse the sound source depth, sound source frequency, and terrain features as conditional information with the output features of the intermediate layer of the U-Net network through a multilayer perceptron, thereby jointly training a matrix prediction model. The trained matrix prediction model is then used to quickly predict the sound field propagation matrix of different regions of the seabed topography, and the sonar detection results are obtained based on the predicted sound field propagation matrix. This application uses sound source depth, frequency, and terrain features as guiding conditions and trains the model using a sound field matrix. After training, the corresponding sound field matrix can be directly generated based on the required sound source depth, frequency, and terrain features, achieving end-to-end sound field modeling and prediction. Its advantages lie in the simplicity of the trained model, which does not require complex underwater acoustic knowledge or cumbersome usage rules compared to traditional sound field calculation models. Furthermore, it is faster and requires fewer computational resources than traditional sound field calculation models. Attached Figure Description

[0022] Figure 1 This is a flowchart illustrating an embodiment of the method for predicting the regional acoustic propagation loss matrix based on the conditional diffusion model in this application.

[0023] Figure 2 This is a schematic diagram of the regional seabed topography provided in an embodiment of the method for predicting the regional acoustic propagation loss matrix based on the conditional diffusion model of this application.

[0024] Figure 3 This is a schematic diagram illustrating the adaptive triangular meshing of a selected region using the OceanMesh2D toolbox, as part of an embodiment of the method for predicting the regional acoustic propagation loss matrix based on the conditional diffusion model of this application.

[0025] Figure 4 This is a schematic diagram of adaptive triangular mesh partitioning provided in an embodiment of the method for predicting the regional acoustic propagation loss matrix based on the conditional diffusion model of this application.

[0026] Figure 5 This is a schematic diagram of a training sample provided for an embodiment of the method for predicting the regional acoustic propagation loss matrix based on the conditional diffusion model of this application;

[0027] Figure 6 This is a schematic diagram of the error curve of the model training process provided in an embodiment of the method for predicting the regional acoustic propagation loss matrix based on the conditional diffusion model of this application.

[0028] Figure 7 This is a schematic diagram of a test set provided for an embodiment of the method for predicting the regional acoustic propagation loss matrix based on the conditional diffusion model of this application;

[0029] Figure 8 This is a schematic diagram illustrating the analysis of the regional sound propagation loss matrix prediction method based on the conditional diffusion model provided in this application. After the model training is completed, 100 terrain features are randomly selected in the region, and a test set is constructed using the same method as the training set.

[0030] Figure 9 (a) Comparison of propagation loss (4.06dB) at a receiving depth of 360m for sample 198 provided in an embodiment of the regional acoustic propagation loss matrix prediction method based on the conditional diffusion model of this application. Figure 9 (b) Comparison of propagation loss of sample 84 at a receiving depth of 960m (3.56dB);

[0031] Figure 10 The results are shown in one of the test samples provided for an embodiment of the method for predicting the regional acoustic propagation loss matrix based on the conditional diffusion model of this application.

[0032] Figure 11 This diagram illustrates the model stability test results provided in an embodiment of the method for predicting the regional acoustic propagation loss matrix based on the conditional diffusion model of this application.

[0033] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0034] It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit this application.

[0035] To overcome the shortcomings of existing technologies in rapidly predicting sound fields in regional sea areas, this invention proposes a rapid prediction method for ocean sound propagation matrices based on terrain feature extraction and diffusion model generation. This method combines the OceanMesh2D toolbox to adaptively divide the regional seabed topography into triangular meshes, model the two-dimensional terrain around nodes and extract features, preferably using traditional acoustic models to calculate the sound field and construct a training set, and finally proposes a sound field prediction model for different terrain conditions in the region by combining the diffusion model.

[0036] Specifically, the method for predicting the regional sound propagation loss matrix based on the conditional diffusion model includes: adaptively dividing the dramatically changing seabed topography into triangular meshes using the OceanMesh2D toolbox; then modeling the two-dimensional topography and extracting features, and using a traditional acoustic model to calculate the sound field to form a training set; finally, combining the diffusion model to learn the mapping relationship between the sound field propagation characteristics and the topographic structure under different seabed topographic conditions, learning the sound field distribution law under different topographic structures, and then using the trained model to quickly predict the sound field propagation matrix under different topographic conditions in the region. The entire method and processing can be divided into the following steps:

[0037] Reference Figure 1 The method for predicting the regional sound propagation loss matrix based on the conditional diffusion model provided in the first embodiment of this application can be executed by a processor in a terminal or server. The method for predicting the regional sound propagation loss matrix based on the conditional diffusion model may include the following execution process:

[0038] S10. Based on the seabed topography and acoustic model, use the OceanMesh2D toolbox to adaptively divide the seabed topography of different regions into triangular meshes, obtain the key nodes of the dramatically changing topography, and determine the topographic data of the entire seabed topography region based on the key nodes of the dramatically changing topography.

[0039] In one embodiment of this application, the adaptive triangular mesh generation of seabed topography in different regions using the OceanMesh2D toolbox may include the following execution process:

[0040] In flat areas, the OceanMesh2D toolbox automatically lowers the mesh resolution; while in ridge and canyon areas, it automatically increases the mesh resolution.

[0041] In one embodiment of this application, the process of determining the topographic data of the entire seabed topography region based on the key nodes of the dramatically changing topography may include the following execution process: slicing the dramatically changing topography according to the key nodes of the dramatically changing topography, and constructing the topographic data of the entire seabed topography region based on the sliced ​​dramatically changing topography.

[0042] This application focuses on the seabed topography of the region. It uses the OceanMesh2d toolbox to adaptively divide the regional topography into triangular meshes, selects important nodes with drastic changes in the region, and performs two-dimensional topographic slicing on each node in turn, so as to cover the topography of the entire region with a limited two-dimensional topographic model.

[0043] In one embodiment of this application, the terrain features include: maximum depth, minimum depth, maximum slope, average slope, flatness, and number of oscillations.

[0044] In this application, a diffusion model structure is established that integrates multi-dimensional conditional information, including sound source depth, sound source frequency, and terrain features (including maximum depth, minimum depth, maximum slope, average slope, flatness, and number of oscillations). This structure guides the model to learn the sound field structure corresponding to different conditional information. After training, the model can directly predict the corresponding sound propagation loss matrix based on the required multi-dimensional conditional information.

[0045] Specifically, using the OceanMesh2D toolbox, the mesh resolution is automatically adjusted based on the complexity of the seabed topography and the requirements for sound wave propagation, thus refining the discretization of the seabed topography. In flat areas, the mesh resolution is lower; while in areas with complex topography (such as ridges and canyons), the mesh resolution is finer. This adaptive meshing method allows for detailed analysis of dramatically changing terrain areas, and the data is saved to ensure accurate representation of topographic features.

[0046] S20. Perform comprehensive two-dimensional terrain modeling on the terrain data of the entire area to obtain two-dimensional terrain models for different areas, and calculate two-dimensional terrain feature data for different areas based on the two-dimensional terrain models.

[0047] Specifically, based on the dramatically changed terrain area processed in the previous step, key nodes are selected for 360° omnidirectional two-dimensional terrain modeling. The core idea of ​​this process is to construct multiple two-dimensional terrain slice models of the area by extending a certain distance at fixed angular intervals. Subsequently, feature calculations are performed on all saved two-dimensional terrain models to provide basic data for further sound field calculations and analysis.

[0048] S30. Input the two-dimensional terrain feature data into the sound field calculation model, and output the sound propagation loss matrix under the corresponding regional conditions, as well as the labels of different regions;

[0049] Specifically, the processor converts the two-dimensional terrain data generated in the previous step into a .bty file format and inputs it into a traditional sound field calculation model (such as Bellhop) to calculate the sound propagation loss matrix under the corresponding terrain conditions. These calculation results serve as the model's training set and are assigned corresponding labels based on different types of terrain data. Furthermore, the terrain vector corresponding to each sound field calculation is also recorded to provide additional conditional information for model training. This application utilizes large-scale model training, eliminating the need to understand and analyze complex ocean sound propagation patterns. It only leverages existing sound field calculation data. By analyzing and learning the distribution patterns of sound fields among different sound source depths, frequencies, and terrain types, the model achieves rapid prediction of sound fields under different terrain structures within the coverage area. The computational load is low, the process is simple to implement, and the results are relatively accurate.

[0050] In one embodiment of this application, the sound field calculation model is Bellhop.

[0051] S40. Construct a matrix prediction model based on the U-Net network. Input the sound propagation loss matrix and the corresponding label into the U-Net network. Use a multilayer perceptron to fuse the sound source depth, sound source frequency, and terrain features as conditional information with the output features of the intermediate layer of the U-Net network to jointly train the matrix prediction model and obtain the trained matrix prediction model.

[0052] In one embodiment of this application, the joint training of the matrix prediction model to obtain the trained matrix prediction model may include the following execution process:

[0053] In the forward diffusion process of the matrix prediction model, noise is gradually added to the input data through the first control function, transforming it into a pure noise distribution;

[0054] In the backward diffusion process of the matrix prediction model, the pure noise distribution is reconstructed into the original data through the second control function.

[0055] The first control function is:

[0056]

[0057] Where, x t α represents the intermediate state at step t. t Scheduling parameters for controlling noise intensity, Random noise sampled from a standard normal distribution;

[0058] The second control function is:

[0059]

[0060] Where, ∈ θ The noise component predicted by the network. σ represents the cumulative variance. t This represents the noise standard deviation during denoising.

[0061] Specifically, based on the principle of the diffusion model, a U-Net network is designed to learn complex sound field distribution patterns. The implementation process is divided into forward denoising and backward denoising. The forward diffusion process gradually transforms the original data into a pure noise distribution through a series of denoising steps. Each step is controlled by a well-defined function, the mathematical expression of which is shown in the first control function. This process follows the Markov property, ensuring that the data gradually converges to an isotropic noise distribution. The backward denoising process is the core objective of the diffusion model, reconstructing the original data from a pure noise state. This is parameterized by a deep neural network, and its mathematical formula is shown in the second control function. This model iteratively approximates the true TL data distribution. The processor integrates sound source depth, sound source frequency, and terrain features as conditional information through a multilayer perceptron and the features of the U-Net's intermediate layers. This allows multi-dimensional conditional information to be directly injected into the deep representation of the network, enabling the network to simultaneously consider factors such as sound source depth, sound source frequency, and terrain during sound field prediction.

[0062] S50. The trained matrix prediction model is used to quickly predict the sound field propagation matrix of different areas of the seabed topography, and the sonar detection results are obtained based on the predicted sound field propagation matrix.

[0063] The beneficial effects of this application include:

[0064] (1) This invention utilizes the OceanMesh2D toolbox, which can automatically adjust the mesh resolution according to the complexity of the seabed topography and the requirements of sound wave propagation, and discretize the seabed topography in a refined manner. This method can quickly and easily select prominent nodes of the seabed topography in a region, and perform two-dimensional topography modeling accordingly. By modeling a limited two-dimensional topography, the topography of the entire region can be covered.

[0065] (2) This invention uses sound source depth, frequency, and terrain features as guiding conditions and trains the sound field matrix. After training, the corresponding sound field matrix can be directly generated based on the required sound source depth, frequency, and terrain features, achieving end-to-end sound field modeling and prediction. Its advantages lie in the simplicity of the trained model, which does not require complex underwater acoustic knowledge or cumbersome usage rules compared to traditional sound field calculation models. Furthermore, it is faster and requires fewer computational resources than traditional sound field calculation models.

[0066] (3) Based on the big data training of the model, this invention does not require understanding and analyzing the complex laws of ocean sound propagation. It only uses the existing sound field calculation data. The model analyzes and learns the distribution law of sound field between different sound source depths, sound source frequencies and terrains, and realizes the rapid prediction of sound field under different terrain structures within the coverage area of ​​the region. The amount of calculation is small, the process is simple to implement, and the results are relatively accurate.

[0067] Next, this application will describe some of the accompanying drawings: Figure 7 After the model is trained, 100 terrain features are randomly selected in the region to build a test set using the same method as the training set. The figure shows the terrain selection for the test set. Figure 8 After the model training was completed, 100 terrain features were randomly selected in the area to construct a test set using the same method as the training set. The following is an error analysis of 400 test samples. (a) Comparison of the overall RMSE between the bellhop calculation results and the model prediction results of the test set: the average RMSE is 5.68 dB, the maximum RMSE is 8.87 dB, and the minimum RMSE is 3.56 dB; (b) Comparison of the bellhop calculation results and the model prediction results of the test set at a receiving depth of 360 m: the average RMSE is 5.28 dB, the maximum RMSE is 8.45 dB, and the minimum RMSE is 4.25 dB; (c) Comparison of the bellhop calculation results and the model prediction results of the test set at a receiving depth of 960 m: the average RMSE is 3.84 dB, the maximum RMSE is 5.74 dB, and the minimum RMSE is 2.98 dB. Figure 9 (a) shows the propagation loss of sample 198 at a receiving depth of 360m (4.06dB), and (b) shows the propagation loss of sample 84 at a receiving depth of 960m (3.56dB). Figure 10 This is one of the test samples. (a) is the sample 301, and (b) is the sample 396. Figure 11 To test the model's stability, two test samples were selected and the model was used to make ten consecutive predictions. It can be seen that the model's results for the same sample conditions were almost identical in the ten consecutive predictions, indicating that it is relatively stable. (a) The receiving depth is 360m and (b) The receiving depth is 960m.

[0068] Based on the above embodiments, this application also provides a device for predicting the regional acoustic propagation loss matrix based on a conditional diffusion model. The device includes a terrain data determination module, a terrain feature calculation module, a training data acquisition module, a model training module, and a prediction module. The terrain data determination module is used to adaptively triangularly mesh the seabed terrain in different regions using the OceanMesh2D toolbox, based on the seabed terrain conditions and an acoustic model, to obtain key nodes of dramatically changing terrain, and to determine the terrain data of the entire seabed terrain region based on these key nodes. The terrain feature calculation module is used to perform comprehensive two-dimensional terrain modeling on the terrain data of the entire region, obtaining two-dimensional terrain models for different regions, and calculating the two-dimensional terrain features of different regions based on the two-dimensional terrain models. The system consists of three modules: a 2D terrain feature data acquisition module and a training data acquisition module. The training module inputs the 2D terrain feature data into the sound field calculation model and outputs the sound propagation loss matrix under the corresponding regional conditions, as well as labels for different regions. The model training module is used to build a matrix prediction model based on the U-Net network. The sound propagation loss matrix and corresponding labels are input into the U-Net network. The sound source depth, sound source frequency, and terrain features are used as conditional information and fused with the output features of the intermediate layers of the U-Net network through a multilayer perceptron to jointly train the matrix prediction model and obtain the trained matrix prediction model. The prediction module is used to quickly predict the sound field propagation matrix of different regions of the seabed topography using the trained matrix prediction model and obtain the sonar detection results based on the predicted sound field propagation matrix.

[0069] It is not difficult to see that this embodiment is a system embodiment corresponding to the above method embodiments, and this embodiment can be implemented in conjunction with the above method embodiments. The relevant technical details and technical effects mentioned in the above embodiments are still valid in this embodiment, and will not be repeated here to reduce repetition. Accordingly, the relevant technical details mentioned in this embodiment can also be applied to the above embodiments.

[0070] It is worth mentioning that all modules involved in this embodiment are logical modules. In practical applications, a logical unit can be a physical unit, a part of a physical unit, or a combination of multiple physical units. Furthermore, to highlight the innovative aspects of this application, this embodiment does not introduce units that are not closely related to solving the technical problems proposed in this application; however, this does not mean that other units are absent from this embodiment.

[0071] Another embodiment of this application proposes an electronic device, including: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the method for predicting the regional acoustic propagation loss matrix based on the conditional diffusion model in the above-described method embodiments.

[0072] The memory and processor are connected via a bus, which can include any number of interconnecting buses and bridges, connecting various circuits of one or more processors and the memory. The bus can also connect various other circuits such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art and will not be described further herein. The bus interface provides an interface between the bus and the transceiver. The transceiver can be a single element or multiple elements, such as multiple receivers and transmitters, providing a unit for communicating with various other devices over a transmission medium. Data processed by the processor is transmitted over the wireless medium via an antenna, which further receives data and transmits it to the processor.

[0073] The processor manages the bus and general processing, and also provides various functions, including timing, peripheral interfaces, voltage regulation, power management, and other control functions. Magnetic memory can be used to store data used by the processor during operation.

[0074] Another embodiment of this application relates to a computer-readable storage medium storing a computer program. When executed by a processor, the computer program implements the method embodiments described above.

[0075] That is, those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware. This program is stored in a storage medium and includes several instructions to cause a device (which may be a microcontroller, chip, etc.) or processor to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, ROM (Read-Only Memory), RAM (Random Access Memory), magnetic disks, or optical disks.

[0076] The above are merely preferred embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. A method for predicting the regional sound propagation loss matrix based on a conditional diffusion model, characterized in that, include: Based on the seabed topography and acoustic model, the OceanMesh2D toolbox is used to adaptively divide the seabed topography in different regions into triangular meshes to obtain key nodes of the dramatically changing topography, and the topographic data of the entire seabed topography region is determined based on the key nodes of the dramatically changing topography. A comprehensive two-dimensional terrain model is performed on the terrain data of the entire area to obtain two-dimensional terrain models for different areas, and two-dimensional terrain feature data for different areas are calculated based on the two-dimensional terrain models. Two-dimensional terrain feature data is input into the sound field calculation model, and the sound propagation loss matrix under the corresponding regional conditions and the labels of different regions are output. A matrix prediction model is constructed based on the U-Net network. The sound propagation loss matrix and the corresponding label are input into the U-Net network. The sound source depth, sound source frequency, and terrain features are used as conditional information and fused with the output features of the intermediate layer of the U-Net network through a multilayer perceptron to jointly train the matrix prediction model and obtain the trained matrix prediction model. The trained matrix prediction model is used to quickly predict the sound field propagation matrix of different regions of the seabed topography, and the sonar detection results are obtained based on the predicted sound field propagation matrix.

2. The method for predicting the regional sound propagation loss matrix based on the conditional diffusion model as described in claim 1, characterized in that, The method of adaptively generating triangular meshes for seabed topography in different regions using the OceanMesh2D toolbox includes: In flat areas, the OceanMesh2D toolbox automatically lowers the mesh resolution; while in ridge and canyon areas, it automatically increases the mesh resolution.

3. The method for predicting the regional sound propagation loss matrix based on the conditional diffusion model as described in claim 1, characterized in that, The jointly trained matrix prediction model yields a trained matrix prediction model, including: In the forward diffusion process of the matrix prediction model, noise is gradually added to the input data through the first control function, transforming it into a pure noise distribution; In the backward diffusion process of the matrix prediction model, the pure noise distribution is reconstructed into the original data through the second control function.

4. The method for predicting the regional sound propagation loss matrix based on the conditional diffusion model as described in claim 3, characterized in that, The first control function is: Where, x t α represents the intermediate state at step t. t Scheduling parameters for controlling noise intensity, Random noise sampled from a standard normal distribution; The second control function is: Where, ∈ θ The noise component predicted by the network. σ represents the cumulative variance. t This represents the noise standard deviation during denoising.

5. The method for predicting the regional sound propagation loss matrix based on the conditional diffusion model as described in claim 1, characterized in that, The topographic data for determining the entire area of ​​seabed topography based on key nodes of dramatically changing terrain includes: The dramatic terrain is sliced ​​based on key nodes, and the topographic data of the entire seabed region is constructed based on the sliced ​​dramatic terrain.

6. The method for predicting the regional sound propagation loss matrix based on the conditional diffusion model as described in claim 1, characterized in that, The terrain features include: Maximum depth, minimum depth, maximum slope, average slope, flatness, and number of oscillations.

7. The method for predicting the regional acoustic propagation loss matrix based on the conditional diffusion model as described in claim 1, characterized in that, The sound field calculation model is Bellhop.

8. A device for predicting the regional sound propagation loss matrix based on a conditional diffusion model, characterized in that, include: The terrain data determination module is used to adaptively divide the seabed terrain of different regions into triangular meshes based on the seabed terrain conditions and acoustic model using the OceanMesh2D toolbox, obtain the key nodes of the dramatically changing terrain, and determine the terrain data of the entire seabed terrain region based on the key nodes of the dramatically changing terrain. The terrain feature calculation module is used to perform comprehensive two-dimensional terrain modeling on the terrain data of the entire area, obtain two-dimensional terrain models of different areas, and calculate two-dimensional terrain feature data of different areas based on the two-dimensional terrain models. The training data acquisition module is used to input two-dimensional terrain feature data into the sound field calculation model and output the sound propagation loss matrix under the corresponding regional conditions, as well as the labels of different regions. The model training module is used to build a matrix prediction model based on the U-Net network. The sound propagation loss matrix and the corresponding label are input into the U-Net network. The sound source depth, sound source frequency, and terrain features are used as conditional information and fused with the output features of the intermediate layer of the U-Net network through a multilayer perceptron to jointly train the matrix prediction model and obtain the trained matrix prediction model. The prediction module is used to quickly predict the sound field propagation matrix of different areas of the seabed topography using the trained matrix prediction model, and obtain the sonar detection results based on the predicted sound field propagation matrix.

9. A computer-readable storage medium, characterized in that, It includes instructions that, when executed on a computer, cause the computer to perform the method for predicting the regional acoustic propagation loss matrix based on the conditional diffusion model as described in any one of claims 1 to 7.

10. An electronic device, characterized in that, The electronic device includes: At least one processor, memory, and input / output unit; The memory is used to store computer programs, and the processor is used to call the computer programs stored in the memory to execute the method for predicting the regional acoustic propagation loss matrix based on the conditional diffusion model according to any one of claims 1 to 7.

Images