Underwater incoherent sound propagation loss prediction method, device, equipment and medium

Abstract

The invention relates to an underwater incoherent sound propagation loss prediction method and device, equipment and a medium. The method comprises the following steps: generating an underwater incoherent sound propagation loss data set based on different sound source depths, and taking the data set as a training set; constructing a lightweight neural network model based on Ghost convolution and a U-Net architecture; preprocessing sound source depth in the training set by adopting a sound source space coding algorithm; inputting the training set subjected to sound source depth preprocessing into a lightweight neural network model for training until the model converges or reaches a preset training standard to obtain a trained lightweight neural network model, and inputting new unknown sound source depth data into the trained lightweight neural network model after the same preprocessing step to obtain a lightweight neural network model; and predicting distribution of underwater incoherent sound propagation loss corresponding to output unknown sound source depth data. By adopting the method, efficient and accurate prediction of the underwater incoherent sound propagation loss can be realized.

Description

Technical Field

[0001] The present application relates to the field of signal processing technology, and in particular to a method, device, equipment and medium for predicting underwater incoherent sound propagation loss. Background Art

[0002] Sound waves are the only effective carrier for long-distance information transmission in the ocean and are widely used for underwater early warning and detection. The calculation of underwater incoherent sound propagation loss provides a key theoretical basis for sonar performance evaluation. However, traditional methods for calculating underwater incoherent sound propagation loss are highly dependent on complex mathematical equations and numerical models, resulting in a lengthy and time-consuming calculation process.

[0003] In contrast, data-driven methods do not need to solve complex equations, and can simulate the sound propagation process more efficiently. As a model with powerful nonlinear mapping capabilities, neural networks have shown significant advantages in simulating physical processes and reducing computational costs. However, existing neural networks cannot accurately predict underwater incoherent sound propagation losses, and their network parameters are also large in volume, and the prediction takes a long time. Summary of the invention

[0004] Based on this, it is necessary to provide a method, device, equipment and medium for predicting underwater incoherent sound propagation loss to address the technical problems of low accuracy and long prediction time in the above-mentioned underwater incoherent sound propagation loss prediction, so as to achieve efficient and accurate prediction of underwater incoherent sound propagation loss.

[0005] A method for predicting underwater incoherent sound propagation loss, the method comprising: Generate an underwater incoherent sound propagation loss dataset based on different sound source depths, and use the dataset as a training set; Build a lightweight neural network model based on Ghost convolution and U-Net architecture. This model replaces the standard convolution in the U-Net architecture with Ghost convolution. The sound source depth in the training set is preprocessed using a sound source space coding algorithm. The preprocessed sound source depth is encoded into a matrix with the same size as the predicted sound field. The training set after sound source depth preprocessing is input into the lightweight neural network model for training until the model converges or reaches the predetermined training standard, and a trained lightweight neural network model is obtained. The new unknown sound source depth data is subjected to the same preprocessing steps and then input into the trained lightweight neural network model to predict and output the distribution of underwater incoherent sound propagation loss corresponding to the unknown sound source depth data.

[0006] In one embodiment, generating an underwater incoherent sound propagation loss dataset based on different sound source depths includes: Different sound source depths, terrain depths, geoacoustic parameters and sound field settings are taken as input parameters, and the traditional numerical calculation program Bellhop based on ray theory is used to run and generate an underwater incoherent sound propagation loss data set.

[0007] In one embodiment, Bellhop, a traditional numerical calculation program based on ray theory, is used to generate an underwater incoherent sound propagation loss data set, including: Bellhop uses ray theory to simulate the propagation of sound waves in the marine environment and generates a dataset of underwater incoherent sound propagation losses, including intrinsic sound rays and underwater incoherent sound propagation losses. Incoherent sound pressure A reference signal at a receiver location relative to the source To calculate, specifically expressed as: ; ; in, Indicates the number of eigenvalues ​​that contribute to the sound field at a specific receiving position, By The incoherent sound pressure caused by the intrinsic sound rays, and ; and Respectively represent the horizontal and vertical (ie, depth) coordinates of the receiving position.

[0008] In one embodiment, a sound source space coding algorithm is used to pre-process the sound source depth in the training set, including: The study area Discretize into a zero matrix , the matrix elements Represents the zero matrix Line Column location in the study area The relative position within the study area The horizontal and vertical distances are and , and Respectively represent the number of rows and columns of the matrix; Known sound source , for the set underwater sound source depth , in the depth range Search and The closest vertical index , expressed as: ; in, ; Depth to sound source Perform normalization to obtain the normalized sound source depth ; According to the characteristics of cylindrical coordinates, the sound source is set to appear only in the first column of the zero matrix. Therefore, the normalized sound source depth Encoded in zero matrix The first column of the vertical index , and get the matrix after the sound source space encoding ,and ; Among them, the matrix after the sound source space encoding is Reserved at a specific location , so that the matrix contains not only the depth information of the sound source, but also the spatial position information of the sound source.

[0009] In one embodiment, the depth of the sound source Perform normalization to obtain the normalized sound source depth , expressed as: ; in, , and Respectively represent the minimum and maximum value of the sound source depth.

[0010] In one embodiment, the training set after the sound source depth preprocessing is input into the lightweight neural network model for training until the model converges or reaches a predetermined training standard, and a trained lightweight neural network model is obtained, including: Initialize the weights and biases of the lightweight neural network model; The training set after sound source depth preprocessing is input into the lightweight neural network model. Through the forward propagation process, the relevant features of underwater incoherent sound propagation loss are extracted through the multi-layer Ghost convolution and activation function in the model, and the preliminary distribution of underwater incoherent sound propagation loss is predicted and generated. Based on the error between the prediction result and the label value in the training set, a loss function is constructed, and the back propagation algorithm is used to adjust the gradient information, weights and biases in the model layer by layer according to the loss function, and the Adam optimizer is used to optimize the model parameters. The optimization goal is to minimize the prediction error. The lightweight neural network model is iteratively trained and optimized in multiple iteration cycles until the model converges or reaches a predetermined training standard, and the model parameters after training and optimization are saved to obtain a trained lightweight neural network model.

[0011] A device for predicting underwater incoherent sound propagation loss, the device comprising: A data set generation module is used to generate an underwater incoherent sound propagation loss data set based on different sound source depths, and use the data set as a training set; The model building module is used to build a lightweight neural network model based on Ghost convolution and U-Net architecture. This model replaces the standard convolution in the U-Net architecture with Ghost convolution; A preprocessing module, used for preprocessing the sound source depth in the training set by using a sound source space coding algorithm, wherein the preprocessed sound source depth is encoded into a matrix having the same size as the predicted sound field; The model training and prediction module is used to input the training set after sound source depth preprocessing into the lightweight neural network model for training until the model converges or reaches a predetermined training standard, thereby obtaining a trained lightweight neural network model, and input new unknown sound source depth data into the trained lightweight neural network model after the same preprocessing steps, and predict and output the distribution of underwater incoherent sound propagation loss corresponding to the unknown sound source depth data.

[0012] A computer device comprises a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the following steps are implemented: Generate an underwater incoherent sound propagation loss dataset based on different sound source depths, and use the dataset as a training set; Build a lightweight neural network model based on Ghost convolution and U-Net architecture. This model replaces the standard convolution in the U-Net architecture with Ghost convolution. The sound source depth in the training set is preprocessed using a sound source space coding algorithm. The preprocessed sound source depth is encoded into a matrix with the same size as the predicted sound field. The training set after sound source depth preprocessing is input into the lightweight neural network model for training until the model converges or reaches the predetermined training standard, and a trained lightweight neural network model is obtained. The new unknown sound source depth data is subjected to the same preprocessing steps and then input into the trained lightweight neural network model to predict and output the distribution of underwater incoherent sound propagation loss corresponding to the unknown sound source depth data.

[0013] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the following steps: Generate an underwater incoherent sound propagation loss dataset based on different sound source depths, and use the dataset as a training set; Build a lightweight neural network model based on Ghost convolution and U-Net architecture. This model replaces the standard convolution in the U-Net architecture with Ghost convolution. The sound source depth in the training set is preprocessed using a sound source space coding algorithm. The preprocessed sound source depth is encoded into a matrix with the same size as the predicted sound field. The training set after sound source depth preprocessing is input into the lightweight neural network model for training until the model converges or reaches the predetermined training standard, and a trained lightweight neural network model is obtained. The new unknown sound source depth data is subjected to the same preprocessing steps and then input into the trained lightweight neural network model to predict and output the distribution of underwater incoherent sound propagation loss corresponding to the unknown sound source depth data.

[0014] The above-mentioned underwater incoherent sound propagation loss prediction method, device, equipment and medium have the following beneficial effects: 1. Construct a lightweight neural network model based on Ghost convolution and U-Net architecture to predict underwater incoherent sound propagation loss. On the one hand, this model can use the skip connection operation in the U-Net architecture to splice the features of the corresponding layers in the encoder and decoder, retaining the detail information and thus improving the prediction accuracy; on the other hand, the lightweight Ghost convolution is used to replace the standard convolution operation in the U-Net architecture, which effectively reduces the model calculation complexity and the number of model parameters, maintains the expressive power of the convolution operation, and can also improve the model's prediction speed and reasoning efficiency without sacrificing performance.

[0015] 2. By adopting the sound source spatial coding algorithm, the sound source depth is encoded into a matrix with the same size as the predicted sound field. The encoded matrix contains the sound source depth and spatial position information, so that the model can simultaneously obtain the sound source depth and the sound source spatial position information for learning, which greatly gives play to the advantages of the model convolution kernel in processing spatial data and improves the accuracy of the distribution prediction of underwater incoherent sound propagation loss.

[0016] 3. Through the systematic training of lightweight neural network models, end-to-end prediction of underwater incoherent sound propagation loss under deep sea and long distance conditions can be achieved, which reduces hardware requirements and deployment costs, and improves the feasibility of underwater incoherent sound propagation loss prediction on mobile devices or low-power platforms. Compared with traditional numerical calculation models, this lightweight neural network model can significantly improve the efficiency of sound field calculation and resource utilization while maintaining high accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 A schematic diagram of a flow chart of a method for predicting underwater incoherent sound propagation loss in an embodiment; Figure 2A schematic diagram of generating an underwater incoherent sound propagation loss data set in one embodiment; Figure 3 A schematic diagram of a U-Net architecture in one embodiment; Figure 4 A schematic diagram of a Ghost convolution operation in one embodiment; Figure 5 A schematic diagram of lightweight neural network model training and prediction in one embodiment; Figure 6 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0018] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0019] In one embodiment, Figure 1 As shown, a method for predicting underwater incoherent sound propagation loss is provided, comprising the following steps: Step S1, generating an underwater incoherent sound propagation loss data set based on different sound source depths, and using the data set as a training set.

[0020] In step S1, considering that the underwater sound field observation data is scarce and it is difficult to support the effective training of the neural network, this application uses the traditional numerical calculation results as the training label value. In the simulation of the underwater sound field and acoustic channel, the Bellhop model is used to generate the underwater incoherent sound propagation loss data set, such as Figure 2 As shown, it includes: taking different sound source depths, terrain depths, geoacoustic parameters and sound field settings as input parameters, and using the traditional numerical calculation program Bellhop based on ray theory to run and generate an underwater incoherent sound propagation loss data set. Specifically, Bellhop uses ray theory to simulate the propagation process of sound waves in the marine environment, and generates an underwater incoherent sound propagation loss data set including intrinsic sound rays and underwater incoherent sound propagation losses; among them, the underwater incoherent sound propagation loss reflects the degree of energy attenuation during the sound wave propagation process, and the underwater incoherent sound propagation loss Incoherent sound pressure A reference signal at a receiver location relative to the source To calculate, specifically expressed as: ; ; in, Indicates the number of eigenvalues ​​that contribute to the sound field at a specific receiving position, By The incoherent sound pressure caused by the intrinsic sound rays, and ; and Respectively represent the horizontal and vertical (i.e. depth) coordinates of the receiving position. According to the above formula, the incoherent sound pressure is calculated based on the addition of the contribution of each intrinsic sound ray, and the incoherent sound pressure ignores the pressure phase associated with each intrinsic sound ray. Therefore, the incoherent sound pressure is independent of the sound source frequency and sound speed.

[0021] Step S2, constructing a lightweight neural network model based on Ghost convolution and U-Net architecture, which replaces the standard convolution in the U-Net architecture with Ghost convolution.

[0022] Among them, the U-Net structure is a deep learning network architecture widely used in image segmentation tasks. This architecture is particularly suitable for medical image segmentation. It adopts a symmetrical encoder-decoder structure to extract and reconstruct image features by gradually compressing and restoring the spatial resolution of the image. Figure 3 The key feature of the U-Net architecture is the skip connection, which concatenates the features of the corresponding layers of the encoder and decoder, retaining the detailed information and thus improving the accuracy of the model prediction.

[0023] In order to further improve the efficiency and performance of the neural network model in the prediction of underwater incoherent sound propagation loss, this application further uses Ghost convolution to replace the standard convolution operation in the U-Net architecture. Ghost convolution is a lightweight convolution operation that reduces the model calculation complexity and model parameter quantity by introducing additional low-computational cost "Ghost" feature maps while maintaining the expressive power of the convolution operation, such as Figure 4 As shown in the figure. In the U-Net architecture, using Ghost convolution instead of standard convolution can not only effectively reduce computing resource consumption, but also improve the prediction speed and reasoning efficiency of the model without sacrificing performance. This improvement enables U-Net to train and reason more efficiently when processing large-scale data sets, and adapt to more complex application scenarios.

[0024] Step S3, using a sound source space coding algorithm to pre-process the sound source depth in the training set, and the pre-processed sound source depth is encoded into a matrix with the same size as the predicted sound field.

[0025] In step S3, the present application considers that there are limitations in using the sound source depth as a direct input to the neural network model. This input method fails to give full play to the advantages of the convolution kernel in spatial feature extraction. In order to overcome this limitation, the present application further introduces a sound source spatial coding algorithm, which encodes the sound source depth into a matrix with the same size as the predicted sound field, so that the encoded matrix can simultaneously contain the depth and spatial position information of the sound source, thereby fully giving full play to the advantages of the convolution kernel in the lightweight neural network model in spatial feature extraction, improving the accuracy of underwater incoherent sound propagation loss related feature extraction, and thus improving the accuracy of model prediction. The sound source spatial coding algorithm specifically includes the following steps: (1) Regional discretization: Discretize the study area Discretize into a zero matrix , the matrix elements Represents the zero matrix Line Column location in the study area The relative position within the study area The horizontal and vertical distances are and , and Represent the number of rows and columns of the matrix respectively.

[0026] (2) Sound source depth positioning: known sound source , for the set underwater sound source depth , in the depth range Search and The closest vertical index , expressed as: ; in, .

[0027] (3) Depth normalization: Depth of sound source Perform normalization to obtain the normalized sound source depth , expressed as: ; in, , and Respectively represent the minimum and maximum value of the sound source depth.

[0028] (4) Data embedding: According to the characteristics of cylindrical coordinates, the sound source is set to appear only in the first column of the zero matrix. Therefore, the normalized sound source depth Encoded in zero matrix The first column of the vertical index , and get the matrix after the sound source space encoding ,and ; Among them, the matrix after the sound source space encoding is Reserved at a specific location , so that the matrix contains not only the depth information of the sound source, but also the spatial position information of the sound source.

[0029] Step S4, input the training set after sound source depth preprocessing into the lightweight neural network model for training until the model converges or reaches the predetermined training standard, and obtain the trained lightweight neural network model, and input the new unknown sound source depth data into the trained lightweight neural network model after the same preprocessing steps, and predict and output the distribution of underwater incoherent sound propagation loss corresponding to the unknown sound source depth data.

[0030] The training and prediction process of the lightweight neural network model in step S4 is as follows: Figure 5 As shown, the following steps are included: First, the weights and biases of the lightweight neural network model are initialized. Then, the training set after sound source depth preprocessing is input into the lightweight neural network model. Through the forward propagation process, the relevant features of underwater incoherent sound propagation loss are extracted through the multi-layer Ghost convolution and activation function in the model, and the preliminary distribution of underwater incoherent sound propagation loss is predicted and generated. Then, based on the error between the prediction result and the label value in the training set, a loss function is constructed, and the back propagation algorithm is used to adjust the gradient information, weights and biases in the model layer by layer according to the loss function, and the Adam optimizer is used to optimize the model parameters, and the optimization goal is to minimize the prediction error. The lightweight neural network model is iteratively trained and optimized in multiple iteration cycles until the model converges or reaches the predetermined training standard, and the model parameters after training optimization are saved to obtain a trained lightweight neural network model.

[0031] Finally, the new unknown sound source depth data is used as a test set. After the same preprocessing steps, it is input into the trained lightweight neural network model to predict and output the distribution of underwater incoherent sound propagation loss corresponding to the unknown sound source depth data. This process not only ensures the efficient learning of the model on the training data, but also ensures its reliability and accuracy in practical applications, can effectively locate the sound source position, and improve the overall performance and practicality of the model.

[0032] In summary, the method for predicting underwater incoherent sound propagation loss proposed in this application takes the sound source depth as input, and realizes end-to-end prediction of underwater incoherent sound propagation loss under deep-sea and long-distance conditions through systematic training of a lightweight neural network model, which significantly reduces the model parameter scale and training time. In addition, the present application also reduces hardware requirements and deployment costs, and improves the feasibility of underwater incoherent sound propagation loss prediction on mobile devices or low-power platforms. Compared with traditional numerical calculation models, this lightweight neural network model can significantly improve the sound field calculation efficiency and resource utilization while maintaining high accuracy.

[0033] In one embodiment, a device for predicting underwater incoherent sound propagation loss is provided, comprising: A data set generation module is used to generate an underwater incoherent sound propagation loss data set based on different sound source depths, and use the data set as a training set; The model building module is used to build a lightweight neural network model based on Ghost convolution and U-Net architecture. This model replaces the standard convolution in the U-Net architecture with Ghost convolution; A preprocessing module, used for preprocessing the sound source depth in the training set by using a sound source space coding algorithm, wherein the preprocessed sound source depth is encoded into a matrix having the same size as the predicted sound field; The model training and prediction module is used to input the training set after sound source depth preprocessing into the lightweight neural network model for training until the model converges or reaches a predetermined training standard, thereby obtaining a trained lightweight neural network model, and input new unknown sound source depth data into the trained lightweight neural network model after the same preprocessing steps, and predict and output the distribution of underwater incoherent sound propagation loss corresponding to the unknown sound source depth data.

[0034] For the specific definition of the underwater incoherent sound propagation loss prediction device, please refer to the definition of the underwater incoherent sound propagation loss prediction method mentioned above, which will not be repeated here. Each module in the above-mentioned underwater incoherent sound propagation loss prediction device can be implemented in whole or in part through software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above-mentioned modules.

[0035] In one embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as follows: Figure 6As shown. The computer device includes a processor, a memory, a network interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a method for predicting underwater incoherent sound propagation loss is implemented. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covering the display screen, or a key, trackball or touchpad set on the computer device housing, or an external keyboard, touchpad or mouse, etc.

[0036] Those skilled in the art will understand that Figure 6 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0037] In one embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the following steps are implemented: Generate an underwater incoherent sound propagation loss dataset based on different sound source depths, and use the dataset as a training set; Build a lightweight neural network model based on Ghost convolution and U-Net architecture. This model replaces the standard convolution in the U-Net architecture with Ghost convolution. The sound source depth in the training set is preprocessed using a sound source space coding algorithm. The preprocessed sound source depth is encoded into a matrix with the same size as the predicted sound field. The training set after sound source depth preprocessing is input into the lightweight neural network model for training until the model converges or reaches the predetermined training standard, and a trained lightweight neural network model is obtained. The new unknown sound source depth data is subjected to the same preprocessing steps and then input into the trained lightweight neural network model to predict and output the distribution of underwater incoherent sound propagation loss corresponding to the unknown sound source depth data.

[0038] In one embodiment, a computer readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the following steps are implemented: Generate an underwater incoherent sound propagation loss dataset based on different sound source depths, and use the dataset as a training set; Build a lightweight neural network model based on Ghost convolution and U-Net architecture. This model replaces the standard convolution in the U-Net architecture with Ghost convolution. The sound source depth in the training set is preprocessed using a sound source space coding algorithm. The preprocessed sound source depth is encoded into a matrix with the same size as the predicted sound field. The training set after sound source depth preprocessing is input into the lightweight neural network model for training until the model converges or reaches the predetermined training standard, and a trained lightweight neural network model is obtained. The new unknown sound source depth data is subjected to the same preprocessing steps and then input into the trained lightweight neural network model to predict and output the distribution of underwater incoherent sound propagation loss corresponding to the unknown sound source depth data.

[0039] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0040] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0041] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A method for predicting underwater incoherent sound propagation loss, characterized in that: The method comprises: Generate an underwater incoherent sound propagation loss dataset based on different sound source depths, and use the dataset as a training set; Build a lightweight neural network model based on Ghost convolution and U-Net architecture. This model replaces the standard convolution in the U-Net architecture with Ghost convolution. The sound source depth in the training set is preprocessed by using a sound source space coding algorithm, and the preprocessed sound source depth is encoded into a matrix with the same size as the predicted sound field; The training set after sound source depth preprocessing is input into the lightweight neural network model for training until the model converges or reaches a predetermined training standard, thereby obtaining a trained lightweight neural network model. The new unknown sound source depth data is subjected to the same preprocessing steps and then input into the trained lightweight neural network model to predict and output the distribution of underwater incoherent sound propagation loss corresponding to the unknown sound source depth data.

2. The method according to claim 1, characterized in that Generate underwater incoherent sound propagation loss dataset based on different sound source depths, including: Different sound source depths, terrain depths, geoacoustic parameters and sound field settings are taken as input parameters, and the traditional numerical calculation program Bellhop based on ray theory is used to run and generate an underwater incoherent sound propagation loss data set.

3. The method according to claim 2, characterized in that The traditional numerical calculation program Bellhop based on ray theory is used to generate underwater incoherent sound propagation loss data sets, including: Bellhop uses ray theory to simulate the propagation of sound waves in the marine environment and generates a dataset of underwater incoherent sound propagation losses, including intrinsic sound rays and underwater incoherent sound propagation losses. Incoherent sound pressure A reference signal at a receiver location relative to the source To calculate, specifically expressed as: ； ； in, Indicates the number of eigenvalues ​​that contribute to the sound field at a specific receiving position, By The incoherent sound pressure caused by the intrinsic sound rays, and ; and Respectively represent the horizontal and vertical coordinates of the receiving position.

4. The method according to claim 1, characterized in that: The sound source depth in the training set is preprocessed using a sound source space coding algorithm, including: The study area Discretize into a zero matrix , the matrix elements represents the zero matrix Line Column location in the study area The relative position within the study area The horizontal and vertical distances are and , and Respectively represent the number of rows and columns of the matrix; Known sound source , for the set underwater sound source depth , in the depth range Search and The closest vertical index , expressed as: ； in, ; Depth to sound source Perform normalization to obtain the normalized sound source depth ; According to the characteristics of cylindrical coordinates, the sound source is set to appear only in the first column of the zero matrix. Therefore, the normalized sound source depth Encoded in zero matrix The first column of the vertical index , and get the matrix after the sound source space encoding ,and ; Among them, the matrix after the sound source space encoding is Reserved at a specific location , so that the matrix contains not only the depth information of the sound source, but also the spatial position information of the sound source.

5. The method according to claim 4, characterized in that Depth to sound source Perform normalization to obtain the normalized sound source depth , expressed as: ； in, , and Respectively represent the minimum and maximum value of the sound source depth.

6. The method according to claim 1, characterized in that Inputting the training set after the sound source depth preprocessing into the lightweight neural network model for training until the model converges or reaches a predetermined training standard, to obtain a trained lightweight neural network model, including: Initializing the weights and biases of the lightweight neural network model; The training set after sound source depth preprocessing is input into the lightweight neural network model, and through the forward propagation process, the relevant features of underwater incoherent sound propagation loss are extracted through the multi-layer Ghost convolution and activation function in the model, and the preliminary distribution of underwater incoherent sound propagation loss is predicted and generated; Based on the error between the prediction result and the label value in the training set, a loss function is constructed, and the back propagation algorithm is used to adjust the gradient information, weights and biases in the model layer by layer according to the loss function, and the Adam optimizer is used to optimize the model parameters. The optimization goal is to minimize the prediction error. The lightweight neural network model is iteratively trained and optimized in multiple iteration cycles until the model converges or reaches a predetermined training standard, and the model parameters after training optimization are saved to obtain a trained lightweight neural network model.

7. An underwater incoherent sound propagation loss prediction device, characterized in that: The device comprises: A data set generation module is used to generate an underwater incoherent sound propagation loss data set based on different sound source depths, and use the data set as a training set; The model building module is used to build a lightweight neural network model based on Ghost convolution and U-Net architecture. This model replaces the standard convolution in the U-Net architecture with Ghost convolution; A preprocessing module, used for preprocessing the sound source depth in the training set by using a sound source space coding algorithm, wherein the preprocessed sound source depth is encoded into a matrix having the same size as the predicted sound field; The model training and prediction module is used to input the training set after sound source depth preprocessing into the lightweight neural network model for training until the model converges or reaches a predetermined training standard, thereby obtaining a trained lightweight neural network model, and input new unknown sound source depth data into the trained lightweight neural network model after the same preprocessing steps, and predict and output the distribution of underwater incoherent sound propagation loss corresponding to the unknown sound source depth data.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images