Method for estimating sound speed of high-resolution three-dimensional sound field in marine search and rescue area and application thereof

Abstract

The present application belongs to the technical field of marine information analysis, and particularly relates to a method for estimating the sound velocity of a high-resolution three-dimensional sound field in a marine search and rescue area and application thereof. The given search and rescue area is gridded, and surface temperature and salinity data are filled into the corresponding grid. According to the temperature, salinity and depth profile values of the search and rescue area, the surface sound velocity is calculated and resampled into the constructed grid. Similarly, the existing sound velocity data on each depth layer in the vertical direction are resampled into the nearest grid point. A vertical-space joint feature dataset is constructed. A neural network model is designed and trained, and the complete vertical structure information from the surface to the target layer is used to reconstruct the sound velocity data of the lower layer. When the data of the next layer is reconstructed, the newly constructed layer is used as the new upper layer data. The present application performs high-resolution three-dimensional numerical layer-by-layer fast estimation on the sound field of the given marine search and rescue area, provides high-precision data support for search and rescue decision-making, and is suitable for large-scale search and rescue sea areas.

Description

Technical Field

[0001] This invention belongs to the field of marine information analysis technology, specifically relating to a method and application for estimating the sound velocity of a high-resolution three-dimensional sound field in marine search and rescue areas. Background Technology

[0002] Maritime search and rescue operations are a race against time, resources, and lives. During the search and rescue process, it is not only necessary to determine whether towed sonar, unmanned underwater vehicle (AUV), or surface ship sonar should be used, but also to determine the towing depth and plan the deployment of the array. These search and rescue decisions all need to be based on real-time or historical sound field data to make optimal choices.

[0003] Sound waves are the most effective carrier of underwater information transmission. Sound field parameters are related to water temperature, pressure, and salinity. To understand the three-dimensional changes of the sound field in a given area, it is necessary to obtain the vertical distribution data of these parameters. Traditionally, obtaining this vertical data relies mainly on shipborne instruments such as the XCTD (X-ray CTD) for on-site measurements. Additionally, Argo buoys can also acquire vertical temperature and salinity profile data of water bodies. Currently, there are over 4,000 Argo buoys operating normally at sea. However, despite the increasing number of Argo buoys, high-resolution coverage of all global ocean areas is still not possible, especially in high-latitude regions, marginal waters, and some sensitive areas where there are few or no buoys. Therefore, current technology cannot yet provide high-resolution estimations of the three-dimensional sound field distribution in specific sea areas such as marine search and rescue zones.

[0004] The development of remote sensing technology has provided new data acquisition methods for obtaining high-resolution, wide-coverage observations of ocean water temperature and salinity. The principle behind the ability of remote sensing data to retrieve sea surface temperature (SST) and salinity (SSS) is primarily based on the physical mechanism of the interaction between seawater and electromagnetic waves (mainly infrared and microwaves). Sensors on satellites or aircraft receive radiation or reflected signals from the sea surface, thereby retrieving water temperature or salinity values. However, it is important to note that remote sensing data only provides temperature and salinity data for the surface layer of the water. In contrast, field measuring instruments or Argo buoys can acquire vertical temperature and salinity profiles. Therefore, can effectively fusing these two types of data enable high-resolution three-dimensional sound field estimation over a large area of ​​the ocean? The answer is yes. The algorithm model of this invention is based on this idea.

[0005] Currently, both domestic and international efforts to acquire high-precision three-dimensional ocean acoustic field data for designated sea areas (e.g., search and rescue zones) face numerous challenges. Firstly, in the data acquisition phase, sensor arrays struggle to simultaneously acquire comprehensive, high-precision three-dimensional acoustic field information. Secondly, the acquired acoustic data often exhibits sparsity, low signal-to-noise ratio, and low resolution. These challenges restrict the accurate construction of the three-dimensional acoustic field. Summary of the Invention

[0006] To address the problems existing in the prior art, this invention provides a method and application for estimating the sound velocity of a high-resolution three-dimensional sound field in marine search and rescue areas. It effectively integrates remote sensing and on-site detection data to achieve high-resolution three-dimensional sound field estimation over a large area of ​​the sea.

[0007] The technical solution adopted by this invention to solve its technical problem is as follows: a method for estimating the sound velocity of a high-resolution three-dimensional sound field in a marine search and rescue area, which involves constructing a three-dimensional sound field model for estimation, and the steps are as follows:

[0008] S1. Gridding: The designated search and rescue area is gridded, and surface temperature and salinity data are filled into the corresponding grids.

[0009] S2. Data filling: Calculate the surface sound velocity based on the profile values ​​of temperature, salinity and depth in the search and rescue area, and resample the surface sound velocity data into the constructed grid; similarly, resample the existing sound velocity data of each depth layer vertically to the nearest grid point to ensure that the data of each layer has the same spatial resolution and coordinate grid system.

[0010] S3. Construct a vertical-spatial joint feature dataset. Construct formatted samples from all valid grid data points to form a pre-training dataset. The formatted samples take a multi-dimensional data input as input and output a scalar value at the corresponding position. The multi-dimensional data includes the number of channels and the depth index.

[0011] S4. Based on the vertical-spatial joint feature dataset, design and train a neural network model to reconstruct missing values ​​in the lower-layer sound velocity data using complete vertical structural information from the surface layer to the upper layer of the target layer.

[0012] S5. When the data is reconstructed at the next lower level, the newly constructed layer is used as the new upper-level data.

[0013] Preferably, in step S2, the surface sound velocity is calculated using the following formula:

[0014] ;

[0015] In the formula, C is the speed of sound, T is the temperature, S is the salinity, and D is the water depth; the formula is applicable within the range of 0℃ ≤ T < 30℃. <S<40,0m≤D<8000m。

[0016] Preferably, step S2 further includes filling in missing data. The co-kriging method is applied, using the upper layer sound velocity data as an auxiliary variable, to perform preliminary spatial interpolation estimation on the missing data of the lower layer sound velocity data, filling in the missing data to the nearest grid point, so that the effective data volume on each depth layer reaches m% of the total number of grid data points, where m is a set parameter.

[0017] Preferably, m% is 30%.

[0018] Preferably, in step S3, the construction of the vertical-spatial joint feature dataset is as follows:

[0019] S31. Suppose the data is a three-dimensional grid with size represented as (depth D, height H, width W), where the depth direction D represents different layers from the surface to the deepest layer;

[0020] S32. For the depth index of the nth layer, n-1, the position of the known valid point is denoted as {(i,j)|i is the row index, j is the column index};

[0021] S33. For each valid point (i,j), construct a three-dimensional data block from layer 1 to layer (n-1) as input. Considering the spatial continuity of the data, take a k×k window centered at (i,j) in the horizontal direction as the neighborhood. The shape of the input data block is (n-1,k,k). Utilizing the local continuity in the vertical direction, the vertical (n-1) is regarded as the depth dimension. Construct a 4-dimensional input data block (1,n-1,k,k), which represents treating the input as a 3D image with 1 channel, a depth of n-1, and height and width of k. The output is a scalar.

[0022] S34, the corresponding output value is the value of the nth layer at position (i,j).

[0023] Preferably, in step S4, the neural network model is a multi-scale convolutional neural network, which simultaneously captures vertical correlation and spatial continuity through three-dimensional convolutional kernels, uses mean squared error as the loss function, uses 70% of the data in the vertical-spatial joint feature dataset for training and 30% for validation, and saves the model that meets the index requirements for the rapid estimation of missing data in the target layer.

[0024] Preferably, in step S5, for the missing data position (i,j) in the target layer, the data from the 1st layer to the (n-1)th layer form a vector of length n-1. A k×k window neighborhood is taken in the horizontal direction with (i,j) as the center. The input data is (1,n-1,k,k). The input is to the model, and the scalar value estimated by the model is the estimated value of the missing data in the target layer.

[0025] Preferably, the latitude and longitude of the search and rescue area are gridded according to the required accuracy of the acoustic field data; the surface temperature and salinity values ​​of the sea area are obtained through satellite remote sensing monitoring data or data from on-site detection by sensors carried by UAVs; and the temperature and salinity profile values ​​are monitored through Argo buoys or drop-out XCTD meters.

[0026] The aforementioned method for estimating the sound velocity of a high-resolution three-dimensional sound field in the marine search and rescue area can be applied to obtain real-time three-dimensional sound field data for a given sea area, or to establish three-dimensional sound field data based on historical observation data.

[0027] The above-mentioned method for estimating the sound velocity of a high-resolution three-dimensional sound field in the marine search and rescue area is applied to marine search and rescue.

[0028] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0029] 1. This application proposes a data estimation method based on a three-dimensional sound field estimation model, which performs high-resolution three-dimensional numerical estimation of the sound field of a given marine search and rescue area in a progressive manner, providing high-precision and high-resolution three-dimensional stereo data support for search and rescue decision-making. It is applicable to a wide range of search and rescue sea areas, and gets rid of the previous dilemma that high-precision gridded three-dimensional sound field data could not be obtained quickly during search and rescue, thus failing to quickly guide the preferred search and rescue area and the use of search and rescue tools.

[0030] 2. In this application, the three-dimensional sound field estimation model is constructed based on remote sensing data and field detection data. It integrates field detection data based on XCTD, Argo buoys, etc., and temperature, salinity and depth data obtained by remote sensing detection methods. The data is comprehensive and the estimation results are reliable.

[0031] 3. The application of this estimation method can not only obtain real-time three-dimensional sound field data of the search and rescue area, but also be applied to historical archived Argo data and satellite remote sensing data to obtain a large amount of historical three-dimensional sound field datasets for a given sea area; these sound field datasets can be directly integrated into search and rescue decision-making during search and rescue. Attached Figure Description

[0032] Figure 1 This is a schematic diagram of the algorithm flow for the estimation model in this invention.

[0033] Figure 2 A surface sound velocity field map obtained from remote sensing data for a given search and rescue area.

[0034] Figure 3 A schematic diagram showing the location of stations in a given search and rescue area, based on vertical depth data obtained from Argo buoys.

[0035] Figure 4 For based on Figure 2 and Figure 3 The data plots show the average sound velocity field estimated at depths of 200m, 400m, and 700m using the algorithm of this invention.

[0036] Figure 5 This is a comparison chart of the reconstructed data at a water depth of 700m obtained by the algorithm of this invention and the field measured data; where r represents the correlation coefficient, RMSE represents the root mean square error, and MAPE represents the mean percentage error. Detailed Implementation

[0037] To facilitate understanding of the present invention, it will be described in more detail below with reference to the accompanying drawings and specific embodiments. However, the present invention can be implemented in many different forms and is not limited to the embodiments described in this specification. Rather, these embodiments are provided to provide a more thorough and complete understanding of the disclosure of the present invention.

[0038] This invention addresses the current challenges by proposing a model for rapidly estimating the three-dimensional sound field of a given area using vertical temperature, salinity, and depth data measured by field instruments (e.g., XCTD, Argo buoys) and surface temperature, salinity, and depth data measured by airborne sensors. The model is cost-effective and easy to operate. Furthermore, the algorithm can not only obtain real-time three-dimensional sound field data for a given sea area (e.g., a search and rescue zone), but also utilize archived Argo data and satellite remote sensing data to obtain a large historical three-dimensional sound field dataset for that area. This dataset can be directly integrated into search and rescue decision-making.

[0039] Combination Figure 1 Understood, the specific plan is as follows:

[0040] 1. Determine the approximate search and rescue area (e.g., [lat1:lat2, log1:log2]), and adjust the latitude and longitude of the area according to the required precision of the sound field data (e.g., ...). A gridded structure is constructed. Surface temperature and salinity values ​​obtained from remote sensing data (which can be satellite remote sensing data or data from on-site detection by sensors mounted on UAVs) are resampled into the constructed grid. Simultaneously, searches are conducted during the search and rescue operation. If Argo buoys are present in the search area, temperature and salinity profile values ​​monitored by the Argo buoys can be used. If no Argo buoys are present, multiple XCTDs can be deployed at randomly selected points in the search area to obtain temperature and salinity profile values. The more deployment points, the higher the accuracy of the estimated 3D sound field data. The corresponding sound speed values ​​can be calculated from the temperature and salinity values ​​at the same depth layer using the following formula.

[0041] ;

[0042] In the formula, C, T, S, and D represent the speed of sound, temperature, salinity, and water depth, respectively; the formula is applicable within the range of 0℃ ≤ T < 30℃. <S<40,0m≤D<8000m。

[0043] 2. Resample the obtained surface sound velocity data into the constructed grid. Similarly, resample the existing sound velocity data at each depth layer (the depth layer can be set according to actual needs, such as 10m, 30m, 50m, 70m, etc.) to the nearest grid point to ensure that the data of each layer has the same spatial resolution and coordinate grid system. At this point, we have obtained the surface (upper layer) sound velocity data SV_L1 and the lower layer sound velocity data SV_L2 (most data is missing). Here, the surface sound velocity data serves as the initial layer, with valid data covering the entire grid. The number of valid data points in the lower layer is determined by the number of Argo buoys or XCTDs deployed, requiring at least two. If the amount of valid data in SV_L2 reaches m% (generally 30%) of the total number of grid data points, then proceed to the next step of constructing the dataset required for neural network training. Otherwise, apply the co-Kriging method, using the upper layer sound velocity data SV_L1 as an auxiliary variable to perform preliminary spatial interpolation estimation for the missing data in the lower layer sound velocity data SV_L2. The co-kriging method considers the spatial correlation and cross-correlation between the main variable (SV_L2) and the auxiliary variable (SV_L1), which can improve the accuracy and reliability of the lower-level data estimation until the effective data in the lower-level sound velocity data SV_L2 reaches m% of the total number of grid data points.

[0044] First, after resampling the data to the nearest grid point at each depth layer, it is determined whether the number of existing sound velocity data points in the depth layer to be reconstructed reaches m% of the total number of grid points. If yes, the dataset construction phase begins; otherwise, known and unknown points in the upper and lower layer data are marked, spatial data structure analysis is performed, and the missing values ​​are reconstructed using the co-kriging algorithm. High-confidence reconstructed data are added to the known point set. When the number of known points reaches m% of the total number of grid points, the dataset construction phase begins; otherwise, spatial data structure analysis is performed again until the number of known points reaches m% of the total number of grid points, at which point the reconstruction of missing values ​​at that depth layer stops.

[0045] 3. Constructing the Dataset. This step requires constructing a dataset to train a convolutional neural network (CNN) to estimate the missing data in the target layer (layer n). It's important to note that this algorithm considers not only the data relationship between layers n and (n-1) but also the entire vertical data structure from the surface to the top. Therefore, the dataset construction doesn't just use data from layer (n-1), but rather the entire vertical data from the surface to layer (n-1). That is, for a valid data point in layer n (position (i,j)), its input x_data is a three-dimensional data block (the vertical data vector at position (i,j) from layer 1 to layer (n-1)). In fact, since the data from layer 1 to layer (n-1) at each position (i,j) constitutes a vector (of length n-1), for the entire grid, we obtain one such vector for each valid point in layer n, for a total of t vectors, where t is the amount of valid data already present in layer n. Therefore, we can construct the dataset as follows:

[0046] (1) Assume our data is a three-dimensional grid with size (depth D, height H, width W). Wherein, the depth direction D represents different layers (from the surface to the deep layer).

[0047] (2) For the nth layer (the depth index is n-1, because the index of the first layer is 0), we know a portion of the valid data points. The positions of these valid points are denoted as {(i,j)|where i is the row index and j is the column index}.

[0048] (3) For each valid point (i,j), we construct an input data block. This input data block is a three-dimensional data block from layer 1 to layer (n-1) (i.e., depth 0 to depth n-2). Considering the spatial continuity of the data, we take a neighborhood (k×k window) centered at (i,j) in the horizontal direction. Therefore, the shape of the input data block is (n-1,k,k). Note: If (i,j) is close to the boundary, the boundary value needs to be copied for filling. However, it should be noted that as sample data, the vertical (n-1) can be regarded as the number of channels or as the depth dimension. Here we need to utilize the local continuity of the vertical direction, so we regard it as a depth dimension. That is, we regard the input as a 3D image with 1 channel, a depth of n-1, and a height and width of k.

[0049] (4) The corresponding output value is the value of the nth layer at position (i,j) (i.e., a scalar).

[0050] In this way, we construct a sample: the input is a 4-dimensional (1, n-1, k, k) data set, and the output is a scalar. Then, we construct such a sample using all valid points, forming the pre-training dataset.

[0051] 4. Convolutional Neural Network Model Training. Based on the vertical-spatial joint feature dataset constructed in the above steps, a multi-scale convolutional neural network is designed to estimate the missing data in the target layer using the complete vertical structural information from the surface layer to the upper layer of the target layer. The model captures both vertical correlation and spatial continuity simultaneously through three-dimensional convolutional kernels, using mean squared error (MSE) as the loss function. 70% of the data in the dataset is used for training, and 30% is used for validation. The model that meets the performance requirements is saved for rapid estimation of missing data in the target layer.

[0052] The algorithm of this invention provides a specific example of network structure design as follows (which can be replaced by other neural network architectures in practical applications):

[0053] The input to a 3D convolutional neural network is (batch_size, channels, depth, height, width), where batch_size can be set manually, for example, batch_size=32. If we want to estimate the data of the 11th layer, treating the vertical dimension as the depth dimension, then the number of channels is 1. Taking k=5 (i.e., a 5x5 neighborhood), the input shape is (32, 1, 10, 5, 5). That is, we treat the input as a single-channel 3D image with a depth of 10 and a height and width of 5. The neural network architecture is designed as follows:

[0054] - Input: (batch_size, 1, 10, 5, 5);

[0055] -3D convolutional layer: 16 3x3x3 convolutional kernels, padding=1, activation function ReLU;

[0056] -3D convolutional layer: 32 3x3x3 convolutional kernels, padding=1, activation function ReLU;

[0057] - 3D convolutional layer: 64 3x3x3 convolutional kernels, padding=1, activation function ReLU;

[0058] -Global average pooling;

[0059] - Fully connected layer: 64->32, activation function ReLU;

[0060] - Fully connected layer: 32->1.

[0061] 5. Lower Layer Missing Data Estimation. The model trained in the above steps is used for fast estimation of missing data in the target layer (layer n). For the missing data position (i,j) in the target layer, the data from layers 1 to (n-1) form a vector (length n-1). A neighborhood (k×k window) is taken horizontally centered at (i,j). Then the model input data (1,n-1,k,k) is obtained. This data is then input into the model, and the scalar value estimated by the model is the estimated value of the missing data in the target layer.

[0062] 6. The estimated data of the nth layer is returned in the Dataset construction of the model algorithm, providing input for the estimation of the missing data of the (n+1)th layer.

[0063] Example 1: Applying the algorithm of this invention, in a designated search and rescue area (15-40°N, 145-170°E), based on the monthly average surface sound velocity field data obtained from remote sensing data in January 2022, and the monthly average vertical sound velocity data from 10 stations measured by Argo buoys, we estimated the high-resolution, full-depth planar layer monthly average sound velocity field data in January 2022 at depths of 200m, 400m, and 700m, consistent with the resolution of the surface sound velocity field. Suitable surface and vertical data can be collected for applications in other time periods and sea areas. The results of this example are as follows... Figure 2-5 As shown. Existing methods (such as...) Figure 3 The measured vertical sound velocity data are only some discrete point profiles. The algorithm of this invention ( Figure 4 This means that satellite data, combined with discrete point profile data, can be used to reconstruct a large-scale, continuous, high-resolution three-dimensional sound velocity field. Among these, Figure 5 This is a scatter plot comparing the reconstructed data at a water depth of 700m obtained by the algorithm of this invention with the field-measured data. At a water depth of 700m, the reconstructed data and the field-measured data are compared. The results show that the correlation coefficient between the two is r=0.978, the root mean square error RMSE=1.141m / s, and the mean percentage error MAPE is 0.07%. This indicates that the sound field data reconstructed by the algorithm of this invention not only has high resolution but also ensures data accuracy.

[0064] The technical features of the above embodiments can be combined in any way. For the sake of brevity, 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.

Claims

1. A method for estimating the sound velocity in a high-resolution three-dimensional sound field in a marine search and rescue area, characterized in that, The steps for constructing a three-dimensional sound field model for estimation are as follows: S1. Gridding: The designated search and rescue area is gridded, and surface temperature and salinity data are filled into the corresponding grids. S2. Data filling: Calculate the surface sound velocity based on the profile values ​​of temperature, salinity and depth in the search and rescue area, and resample the surface sound velocity data into the constructed grid. Similarly, existing sound velocity data at each depth layer are resampled to the nearest grid point to ensure that the data at each layer has the same spatial resolution and coordinate grid system; S3. Construct a vertical-spatial joint feature dataset. Construct formatted samples from all valid grid data points to form a pre-training dataset. The formatted samples take a multi-dimensional data input as input and output a scalar value at the corresponding position. The multi-dimensional data includes the number of channels and the depth index. The steps for constructing the vertical-spatial joint feature dataset are as follows: S31. Suppose the data is a three-dimensional grid with size represented as (depth D, height H, width W), where the depth direction D represents different layers from the surface to the deepest layer; S32. For the depth index of the nth layer, n-1, the position of the known valid point is denoted as {(i,j)|i is the row index, j is the column index}; S33. For each valid point (i,j), construct a three-dimensional data block from layer 1 to layer (n-1) as input. Considering the spatial continuity of the data, take a k×k window centered at (i,j) in the horizontal direction as the neighborhood. The shape of the input data block is (n-1,k,k). Utilizing the local continuity in the vertical direction, the vertical (n-1) is regarded as the depth dimension. Construct a 4-dimensional input data block (1,n-1,k,k), which represents treating the input as a 3D image with 1 channel, a depth of n-1, and height and width of k. The output is a scalar. S34, the corresponding output value is the value of the nth layer at position (i,j); S4. Based on the vertical-spatial joint feature dataset, design and train a neural network model to reconstruct missing values ​​in the lower-layer sound velocity data using complete vertical structural information from the surface layer to the upper layer of the target layer. The neural network model is a multi-scale convolutional neural network that simultaneously captures vertical correlation and spatial continuity through three-dimensional convolutional kernels. It uses mean squared error as the loss function and uses 70% of the data in the vertical-spatial joint feature dataset for training and 30% for validation. The model that meets the target requirements is saved for the rapid estimation of missing data in the target layer. S5. When reconstructing the data at the next lower layer, the newly constructed layer is used as the new upper layer data. For the missing data position (i,j) in the target layer, the data from the 1st layer to the (n-1)th layer form a vector of length n-1. Take a k×k window neighborhood with (i,j) as the center in the horizontal direction. The input data is (1,n-1,k,k). Input the model, and the scalar value estimated by the model is the estimated value of the missing data in the target layer.

2. The method for estimating the sound velocity of a high-resolution three-dimensional sound field in a marine search and rescue area according to claim 1, characterized in that, In step S2, the surface sound velocity is calculated using the following formula: C=1448.96+4.591*T-5.304*10 -2 *T 2 +2.374*10 -4 *T 3 +1.340*(S-35)+1.630*10 -2 *D+1.675*10 -7 *D 2 -1.025*10 -2 *T*(S-35)-7.139*10 -13 *T*D 3 ; In the formula, C is the speed of sound, T is the temperature, S is the salinity, and D is the water depth; the formula is applicable within the range of 0℃ ≤ T < 30℃. <S<40,0m≤D<8000m。 3. The method for estimating the sound velocity of a high-resolution three-dimensional sound field in a marine search and rescue area according to claim 2, characterized in that, Step S2 also includes filling in missing data. The co-kriging method is applied, using the upper layer sound velocity data as an auxiliary variable, to perform preliminary spatial interpolation estimation on the missing data of the lower layer sound velocity data, filling in the missing data to the nearest grid point, so that the effective data volume on each depth layer reaches m% of the total number of grid data points, where m is a set parameter.

4. The method for estimating the sound velocity of a high-resolution three-dimensional sound field in a marine search and rescue area according to claim 3, characterized in that, m% is taken as 30%.

5. The method for estimating the sound velocity of a high-resolution three-dimensional sound field in a marine search and rescue area according to claim 1, characterized in that, The latitude and longitude of the search and rescue area are gridded according to the required accuracy of the acoustic field data; the surface temperature and salinity values ​​of the sea area are obtained through satellite remote sensing monitoring data or data detected on-site by sensors carried by UAVs; and the temperature and salinity profile values ​​are monitored through Argo buoys or drop-out XCTD meters.

6. Application of a high-resolution three-dimensional sound field sound velocity estimation method in marine search and rescue areas, characterized in that... The sound velocity estimation method for high-resolution three-dimensional sound field in marine search and rescue areas as described in any one of claims 1-5 can be applied to obtain three-dimensional sound field data of a given sea area in real time, or to establish three-dimensional sound field data based on historical observation data.

7. The application of the sound velocity estimation method for high-resolution three-dimensional sound field in marine search and rescue areas as described in claim 6, characterized in that, The sound speed estimation method described above is applied to marine search and rescue.

Images