A method for seabed topography inversion based on fully connected deep neural networks

Abstract

This invention discloses a seabed topography inversion method based on a fully connected deep neural network, comprising: acquiring shipborne depth sounding data and performing data preprocessing; acquiring a gravity anomaly dataset, decomposing the gravity anomaly field into long-wave and short-wave components, fitting the difference to obtain residual long-wave and residual short-wave gravity anomalies; constructing a sea-depth inversion model, using the gravity anomaly data and latitude and longitude data as input; training the sea-depth inversion model with the neural network through shipborne depth sounding data, outputting sea-depth inversion results, and evaluating and optimizing the accuracy of the sea-depth inversion model. This invention improves the accuracy of seabed topography inversion by combining shipborne depth sounding data, long-wave gravity anomaly data, short-wave gravity anomaly data, residual short-wave gravity anomaly data, and residual long-wave gravity anomaly data. Using a fully connected deep neural network model further enhances the model's inversion capability, enabling the model to more accurately capture topographic changes and achieve superior depth inversion accuracy.

Description

Technical Field

[0001] This invention relates to the field of marine surveying and mapping technology, and in particular to a method for seabed topography inversion based on a fully connected deep neural network. Background Technology

[0002] Seabed topographic surveying is a fundamental marine mapping task. Traditional methods primarily rely on ships as surveying tools, typically combining ship positioning and sonar technologies to detect ocean depth and seabed obstacles. However, the influence of ship speed, navigation routes, and natural conditions leads to unevenness in depth measurement results, leaving many data gaps in the vast ocean and limiting our understanding of seabed topography. These gaps restrict progress in scientific research, resource development, and ecological protection.

[0003] Satellite altimetry technology, with its all-weather, uninterrupted operation capabilities, extensive global coverage, and rapid information acquisition, has brought innovative technical approaches to seafloor topography mapping. Gravity anomalies and seafloor topography are significantly correlated within specific wavebands. Therefore, ocean gravity anomaly data obtained through altimetry technology can be used to invert seafloor topography, providing data support for Earth system science research.

[0004] In studies using gravity anomaly data for seafloor topography inversion, commonly used methods include gravity geological methods and their improved GGM methods, admittance function methods, the Smith & Sandwell method (S&S), and least squares methods. These methods provide reliable models to a certain extent, but they still have several limitations. For example, while gravity geological methods and their improved methods offer high accuracy, they require dense control point data and neglect the nonlinear relationship between gravity anomalies and water depth, resulting in poor inversion accuracy in shallow sea areas. The admittance function method utilizes the correlation between water depth and topography to perform model transformation in the frequency domain, making it suitable for sea areas lacking ship survey data, but its spatial resolution is low. Therefore, some researchers have begun to combine gradient anomalies with gravity anomalies to obtain better inversion results. However, the nonlinear iterative least squares method still has shortcomings in computational efficiency and processing time compared to other modeling methods. Therefore, further exploration and development of more efficient algorithms are urgently needed to improve modeling efficiency and inversion accuracy. Summary of the Invention

[0005] The purpose of this section is to outline some aspects of embodiments of the present invention and to briefly describe some preferred embodiments. Simplifications or omissions may be made in this section, as well as in the abstract and title of this application, to avoid obscuring the purpose of these documents; however, such simplifications or omissions should not be construed as limiting the scope of the invention.

[0006] In view of the aforementioned existing problems, this invention is proposed. Therefore, this invention provides a seabed topography inversion method based on a fully connected deep neural network to solve the above problems.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0008] In a first aspect, the present invention provides a method for seabed topography inversion based on a fully connected deep neural network, comprising: acquiring ship-measured water depth data and performing data preprocessing;

[0009] Based on the ship-measured water depth data, the gravity anomaly data is decomposed into short-wave gravity anomaly data, long-wave gravity anomaly data, residual short-wave gravity anomaly data, and residual long-wave gravity anomaly data. Specifically, after decomposing the gravity anomaly field into long-wave and short-wave components, the residual long-wave gravity anomaly and residual short-wave gravity anomaly are obtained by fitting and subtracting them from the ship-measured water depth data, respectively.

[0010] A sea depth inversion model is constructed. The gravity anomaly dataset and latitude and longitude data are used as inputs. The sea depth prediction model is inverted by a neural network. The neural network of the sea depth inversion model is trained by ship-measured water depth data. The sea depth inversion results are output. The accuracy of the sea depth inversion model is evaluated and optimized based on the ship-measured water depth data.

[0011] As a preferred embodiment of the seabed topography inversion method based on a fully connected deep neural network described in this invention, the step of acquiring ship-measured water depth data and performing data preprocessing includes:

[0012] Error filtering is performed on the ship-measured depth data, and the depth of the world ocean seabed topography map is interpolated to the ship-measured depth points to calculate the difference between the ship-measured depth values ​​and the actual depth values.

[0013] Based on the differences, the standard deviations of the two are calculated, and depth sounding points with differences exceeding three times the standard deviation are removed.

[0014] Based on a control point to checkpoint ratio of 4:1, control points and checkpoints are randomly obtained.

[0015] As a preferred embodiment of the seabed topography inversion method based on fully connected deep neural networks described in this invention, the obtained gravity anomaly dataset includes:

[0016] The gravity anomaly field is decomposed into long-wavelength and short-wavelength components. The short-wavelength gravity anomaly of the control point is calculated. The long-wavelength gravity anomaly of the control point is obtained by subtracting the short-wavelength gravity anomaly of the control point from the gravity anomaly of the control point.

[0017] The long-wave gravity anomaly of the control point is gridded to construct a long-wave gravity anomaly field. The long-wave gravity anomaly of the unknown point is obtained by interpolation, and the short-wave gravity anomaly of the unknown point is solved.

[0018] The linear relationship between the long and short wave components and the ship-measured water depth is determined by a linear regression model. The ship-measured water depth is multiplied by the corresponding fitting slope to obtain the reference values ​​of the long wave and short wave gravity anomalies. The long wave and short wave gravity anomalies are subtracted from the reference values ​​to obtain the residual long wave gravity anomaly and the residual short wave gravity anomaly.

[0019] As a preferred embodiment of the seabed topography inversion method based on fully connected deep neural networks described in this invention, the construction of the sea depth inversion model includes:

[0020] The input layer contains longitude, latitude, and gravity anomaly data, including shortwave gravity anomaly, longwave gravity anomaly, residual shortwave gravity anomaly, and residual longwave gravity anomaly.

[0021] There are four hidden layers. The first hidden layer has 16 neurons, the second hidden layer has 32 neurons, the third hidden layer has 256 neurons, and the fourth hidden layer has 512 neurons. The neurons in each layer are interconnected through weights and biases.

[0022] The output layer, via the hidden layer, outputs the ocean depth measurement results.

[0023] As a preferred embodiment of the seabed topography inversion method based on a fully connected deep neural network described in this invention, the sea depth inversion model is expressed as follows:

[0024] ;

[0025] in, This represents the input gravitational field signal. y Represents the output of a neuron. , These represent the input weights and the neuron biases, respectively. Indicates the activation function;

[0026] In the model initialization phase, weights and biases are assigned initial values. During training, the weights of each neuron are adjusted based on the inversion differences. The weight iteration formula is expressed as:

[0027] ;

[0028] in, This indicates the adjusted weights. Indicates the learning rate. Indicates the initial weights.loss This represents a loss function that depends on the mean squared error.

[0029] As a preferred embodiment of the seabed topography inversion method based on a fully connected deep neural network described in this invention, the neural network inversion includes:

[0030] The location information of the ship's depth measurement control points and gravity anomaly data are divided into training set and test set;

[0031] The training dataset and the test dataset are normalized according to standard, and the ocean depth inversion model is trained accordingly.

[0032] Initialize the ocean depth inversion model by randomly generating initial weights and biases;

[0033] The network was trained for 40 epochs using a loss function based on the root mean square error of inversion depth and shipborne depth, with an initial learning rate of 0.001. The model parameters were then adjusted using the Adam optimization algorithm.

[0034] The gravity signal obtained from the height measurement is fed into the trained ocean depth inversion model to obtain the inversion water depth of the measurement area.

[0035] As a preferred embodiment of the seabed topography inversion method based on fully connected deep neural networks described in this invention, the accuracy evaluation optimization includes:

[0036] Multiple comparison models were interpolated to the check point using bicubic interpolation and compared with the ship-measured water depth data at the check point.

[0037] The values ​​of standard deviation, root mean square error, and mean absolute percentage error are used as accuracy evaluation indicators for the ocean depth inversion model and other comparative models.

[0038] Secondly, the present invention provides a seabed topography inversion system based on a fully connected deep neural network, comprising:

[0039] The data acquisition module is used to acquire ship-measured water depth data and perform data preprocessing.

[0040] The data processing module is used to decompose gravity anomaly data into short-wave gravity anomaly data, long-wave gravity anomaly data, residual short-wave gravity anomaly data, and residual long-wave gravity anomaly data based on the ship-measured water depth data. Specifically, after decomposing the gravity anomaly field into long-wave and short-wave components, the residual long-wave gravity anomaly and residual short-wave gravity anomaly are obtained by fitting and subtracting them from the ship-measured water depth data, respectively.

[0041] The model building module is used to construct a sea depth inversion model. It takes the gravity anomaly dataset and latitude and longitude data as input, performs neural network inversion on the sea depth prediction model, trains the neural network on the sea depth inversion model using ship-measured water depth data, outputs the sea depth inversion results, and evaluates and optimizes the accuracy of the sea depth inversion model based on the ship-measured water depth data.

[0042] Thirdly, the present invention provides an electronic device, comprising:

[0043] Memory and processor;

[0044] The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, they implement the steps of the seabed topography inversion method based on a fully connected deep neural network.

[0045] Fourthly, the present invention provides a computer-readable storage medium storing computer-executable instructions that, when executed by a processor, implement the steps of the seabed topography inversion method based on a fully connected deep neural network.

[0046] Compared with existing technologies, the beneficial effects of this invention are as follows: By combining shipborne bathymetry data, long-wave gravity anomaly data, and short-wave gravity anomaly data, this invention can fully utilize multi-source information to improve the accuracy of seabed topography inversion. Using a fully connected deep neural network model, it can capture complex nonlinear relationships, improve modeling efficiency, and further enhance the model's inversion capability. The model can more accurately capture topographic changes and achieve better inversion depth accuracy. By decomposing gravity anomaly data into long-wave and short-wave components, it can better analyze seabed topographic features at different scales. Residual gravity anomalies further refine topographic features and improve the resolution of local topography. The trained model can complete large-scale seabed topography inversion in a short time and is suitable for large-scale marine mapping tasks. Attached Figure Description

[0047] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0048] Figure 1 This is a schematic diagram of the overall process of the seabed topography inversion method based on a fully connected deep neural network according to an embodiment of the present invention;

[0049] Figure 2This is a schematic diagram of the seabed topography inversion process of the seabed topography inversion method based on a fully connected deep neural network according to an embodiment of the present invention;

[0050] Figure 3 This is a schematic diagram showing the distribution of the research area and the shipborne bathymetry trajectory of the seabed topography inversion method based on a fully connected deep neural network according to an embodiment of the present invention.

[0051] Figure 4 This is a schematic diagram of the sea depth inversion model of the seabed topography inversion method based on a fully connected deep neural network according to an embodiment of the present invention;

[0052] Figure 5 The above is a histogram and scatter plot showing the difference between the FCD_Depth_SCS, GGM_Depth, GEBCO_2024, SIOv25.1 and DTU18 models and ship-measured water depth at the check point of the seabed topography inversion method based on fully connected deep neural networks described in one embodiment of the present invention.

[0053] Figure 6 This is a schematic diagram showing the distribution of different regions A, B, C, and D in the seabed topography inversion method based on a fully connected deep neural network according to an embodiment of the present invention. Detailed Implementation

[0054] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0055] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0056] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0057] This invention is described in detail with reference to the schematic diagrams. When detailing the embodiments of this invention, for ease of explanation, the cross-sectional views illustrating the device structure may be partially enlarged, not adhering to the usual scale. Furthermore, the schematic diagrams are merely examples and should not be construed as limiting the scope of protection of this invention. In actual fabrication, the three-dimensional spatial dimensions of length, width, and depth should be included.

[0058] Furthermore, in the description of this invention, it should be noted that the terms "upper," "lower," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. These terms are used solely for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. In addition, the terms "first," "second," or "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0059] Unless otherwise explicitly specified and limited, the terms "installation," "connection," and "joining" in this invention should be interpreted broadly. For example, they can refer to fixed connections, detachable connections, or integral connections; similarly, they can refer to mechanical connections, electrical connections, or direct connections, or indirect connections through an intermediate medium, or internal connections between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0060] Example:

[0061] Reference Figures 1-5 Tables 1-3, representing an embodiment of the present invention, provide a method for seabed topography inversion based on a fully connected deep neural network, such as... Figure 1 , 2 As shown, it includes:

[0062] S101, acquire ship-measured water depth data and perform data preprocessing;

[0063] S102, based on ship-measured water depth data, the gravity anomaly data is decomposed into short-wave gravity anomaly data, long-wave gravity anomaly data, residual short-wave gravity anomaly data, and residual long-wave gravity anomaly data. Among them, after the gravity anomaly field is decomposed into long-wave and short-wave components, the residual long-wave gravity anomaly and residual short-wave gravity anomaly are obtained by fitting and subtracting them from the ship-measured water depth data respectively.

[0064] S103, Construct a sea depth inversion model. Using gravity anomaly dataset and latitude and longitude data as input, perform neural network inversion on the sea depth prediction model. Train the neural network on the sea depth inversion model using ship-measured water depth data, output the sea depth inversion results, and evaluate and optimize the accuracy of the sea depth inversion model based on the ship-measured water depth data.

[0065] It should be noted that due to the influence of ship speed, navigation route and natural conditions, the water depth measurement results are not uniform. There are still many data gaps in the vast ocean, which makes our understanding of seabed topography relatively limited. Gravity anomalies and seabed topography are significantly correlated in specific wavebands. Ocean gravity anomaly data obtained through altimetry technology can be used to invert seabed topography.

[0066] Preferred, such as Figure 3 As shown, acquiring ship-measured water depth data and performing data preprocessing includes:

[0067] Error filtering is performed on the ship-measured water depth data, the water depth of GEBCO is interpolated to the ship-measured water depth point, and the difference between the ship-measured water depth value and the water depth value is calculated.

[0068] Based on the differences, the standard deviations of the two are calculated, and depth sounding points with differences exceeding three times the standard deviation are removed;

[0069] Based on a control point to checkpoint ratio of 4:1, control points and checkpoints are randomly obtained.

[0070] Specifically, a dataset of shipborne depth soundings from 108°E to 121°E and 6°N to 23°N was obtained from the South China Sea. This dataset consisted of shipborne depth soundings and could be obtained from the U.S. National Environmental Information Center. The dataset was preprocessed. Based on the GEBCO_2024 model, the "3σ" principle was used to filter the shipborne depth soundings for errors. Specifically, the depths from the GEBCO_2024 model were interpolated to the shipborne depth points, and the differences between the two values ​​were calculated. Based on these differences, the standard deviations of both values ​​were calculated, and depth soundings with differences exceeding three standard deviations were removed. Ultimately, 8653 data points were removed, leaving 802379 shipborne depth soundings, a removal rate of approximately 1.07%.

[0071] Based on the principle of a 4:1 ratio of control points to check points, 641,903 ship depth measurement points were randomly selected as control points, and the remaining 160,476 ship depth measurement points were selected as check points. The control points were used only for training the neural network model, while the check points were used specifically to evaluate the accuracy of the ocean depth inversion model (FCD_Depth_SCS).

[0072] Preferably, in step S102, the gravity anomaly dataset includes:

[0073] The gravity anomaly field is decomposed into long-wave and short-wave components. The short-wave gravity anomaly of the control point is calculated. The long-wave gravity anomaly of the control point is obtained by subtracting the short-wave gravity anomaly of the control point from the gravity anomaly of the control point.

[0074] The long-wave gravity anomalies of the control points are gridded to construct a long-wave gravity anomaly field. The long-wave gravity anomalies of the unknown points are obtained by interpolation, and the short-wave gravity anomalies of the unknown points are solved.

[0075] The linear relationship between the long and short wave components and the ship-measured water depth is determined by a linear regression model. The ship-measured water depth is multiplied by the corresponding fitting slope to obtain the reference values ​​of the long wave and short wave gravity anomalies. The long wave and short wave gravity anomalies are subtracted from the reference values ​​to obtain the residual long wave gravity anomaly and the residual short wave gravity anomaly.

[0076] Specifically, after obtaining the long-wave and short-wave gravity anomalies, a linear regression model is used to determine the linear relationship between the long-wave and short-wave gravity anomalies and the ship-measured water depth values. The ship-measured water depth values ​​are used as input variables, and the long-wave and short-wave gravity anomaly data are used as target variables. The parameters of the model, namely the slope and intercept of the linear equation, are calculated to minimize the difference between the model's inversion on the training data and the actual values. The reference long-wave and short-wave gravity anomaly data are obtained by multiplying the ship-measured water depth data with the fitted slope. The residual long-wave and short-wave gravity anomalies are obtained by subtracting the reference long-wave and short-wave gravity anomalies from the long-wave and short-wave gravity anomalies. Therefore, the gravity anomaly data feature dataset consists of short-wave gravity anomaly, long-wave gravity anomaly, residual short-wave gravity anomaly, and residual long-wave gravity anomaly data.

[0077] Since the relationship between gravity (GA) and sounding data is not linear, the GA field is decomposed into long-wave and short-wave components. The short-wave gravity anomaly mainly originates from subtle changes in local bedrock topography, while the long-wave gravity anomaly originates from changes in mass distribution at deeper levels within the Earth's crust.

[0078] In one alternative implementation, the shortwave gravity anomaly at the control point is represented as:

[0079] ;

[0080] in, Indicates the first The shortwave gravity anomaly at point G, where G is the gravitational constant. It is the density difference constant. Control points The ocean depth is denoted as D, which is the reference ocean depth and is set as the maximum water depth covered by the study area, -5174 m.

[0081] Preferably, in step S103, as Figure 4 As shown, constructing the ocean depth inversion model includes:

[0082] The input layer takes longitude, latitude, and gravity anomaly data as input data. The gravity anomaly data includes shortwave gravity anomaly, longwave gravity anomaly, residual shortwave gravity anomaly, and residual longwave gravity anomaly.

[0083] There are four hidden layers. The first hidden layer has 16 neurons, the second hidden layer has 32 neurons, the third hidden layer has 256 neurons, and the fourth hidden layer has 512 neurons. The neurons in each layer are interconnected through weights and biases.

[0084] The output layer, via the hidden layer, outputs the ocean depth measurement results.

[0085] It should be noted that ocean topography information is complex, and the inversion model needs to combine multiple gravity field elements. Due to the complex characteristics of the input data, tasks such as classification and regression need to be performed. Therefore, the ocean depth inversion model is constructed as a fully connected neural network (FC-DNN). The ocean depth inversion model contains an input layer, multiple hidden layers and an output layer. The hidden layers can help extract relevant data features, but they may also lead to increased training time and overfitting. In order to achieve the best balance, four hidden layers are set. Each neuron in the ocean depth inversion model uses a specific operation to extract data features. The neurons in each layer are interconnected through weights and biases.

[0086] The preferred deep-sea depth inversion model is expressed as follows:

[0087] ;

[0088] in, This represents the input gravitational field signal. y Represents the output of a neuron. , These represent the input weights and the neuron biases, respectively. Indicates the activation function;

[0089] In the model initialization phase, weights and biases are assigned initial values. During training, the weights of each neuron are adjusted based on the inversion differences. The weight iteration formula is expressed as:

[0090] ;

[0091] in, This indicates the adjusted weights. Indicates the learning rate. Indicates the initial weights. loss This represents a loss function that depends on the mean squared error.

[0092] During model training, a reverse calculation is performed for each forward calculation, and the weights are adjusted multiple times to obtain the optimal value, so that the difference between the inverted water depth and the ship's loading depth is minimized.

[0093] It should be noted that the activation function used in this model is the Rectified Linear Unit (ReLU) activation function. Before the activation function, the dataset needs to be normalized to ensure that the dataset falls within the range of 0 to 1. The ocean depth inversion model also has the function of backpropagation to optimize the network. This function is mainly used to adjust the weights of the ocean depth inversion model. During the model initialization phase, the weights and biases of the neural network are given initial values. During training, the weights of each neuron are adjusted according to the differences in inversion.

[0094] Preferably, in step S103, performing neural network inversion includes:

[0095] The location information of ship depth control points and gravity anomaly data are divided into training and testing sets to train the ocean depth inversion model. High-quality ship depth data is used as training data, and the remaining inspection data is used for testing.

[0096] The training data and gravity field data are normalized by maximum and minimum to ensure that the data falls between 0 and 1, which greatly reduces the difference in input data.

[0097] Initialize the ocean depth inversion model by randomly generating initial weights and biases;

[0098] The network was trained for 40 epochs using a loss function based on the root mean square error of inversion depth and shipborne depth with an initial learning rate of 0.001. In order to update the weights according to the learning rate and loss function, the model parameters were adjusted by the Adam optimization algorithm to optimize performance.

[0099] The gravity signal obtained from the height measurement is fed into the trained ocean depth inversion model to obtain the inversion water depth of the measurement area.

[0100] Preferably, the accuracy assessment optimization includes:

[0101] Multiple comparison models were interpolated to the check point using bicubic interpolation and compared with the ship-measured water depth data at the check point.

[0102] The values ​​of standard deviation, root mean square error, and mean absolute percentage error are used as accuracy evaluation indicators for the ocean depth inversion model and other comparative models.

[0103] Specifically, to evaluate the accuracy of the established FCD_Depth_SCS model, this study used the gravity geological method (GGM) to invert the seabed topography model of the South China Sea, named the GGM_Depth model. In addition, the GEBCO_2024, SIOv25.1, and DTU18 models were introduced. Using bicubic interpolation, the FCD_Depth_SCS, GGM_Depth, GEBCO_2024, SIOv25.1, and DTU18 models were interpolated to checkpoints and compared with ship-gauge depth data at the checkpoints. Standard deviation (STD), root mean square error (RMS), and mean absolute percentage error (MPAE / %) were used as evaluation indicators.

[0104] Mean Absolute Percentage Error (MAPE / %) is a widely used metric for evaluating inversion accuracy. It primarily measures the difference between the inverted value and the actual value. The calculation formula is as follows:

[0105] ;

[0106] in, Indicates the first One ocean depth inversion value, For the first A measured water depth value, This represents the number of checkpoints.

[0107] like Figure 5 The table shows the histogram and scatter plot of the difference between the FCD_Depth_SCS, GGM_Depth, GEBCO_2024, SIOv25.1 and DTU18 models and the ship-measured water depth at the checkpoint: the first column is the FCD_Depth_SCS model, the second column is the GGM_Depth model, the third column is the GEBCO_2024 model, the fourth column is the SIOv25.1 model, and the fifth column is the DTU18 model.

[0108] Table 1 shows the statistical results of comparing the FCD_Depth_SCS, GGM_Depth, GEBCO_2024, SIOv25.1, and DTU18 models with ship-measured water depth values ​​at the checkpoints.

[0109] Table 1: Comparison of water depth values ​​measured by different models and ships (unit: m)

[0110] Model MAX MIN MEAN STD RMS MAPE / % FCD_Depth_SCS 1245.725 -857.709 0.187 44.755 44.756 2.903 GGM_Depth 1133.773 -1524.209 0.321 58.874 58.875 3.663 GEBCO_2024 842.874 -842.579 2.181 82.234 82.263 3.733 SIOv25.1 2137.575 -1964.993 6.786 108.241 108.454 5.149 DTU18 2130.733 -2762.660 31.904 186.967 189.670 9.095

[0111] By comparing the difference between the water depth inverted by the FCD_Depth_SCS model and the actual depth measurement at the checkpoint, the standard deviation (STD) was 44.755 m, and the mean absolute percentage error (MAPE) of the difference was 2.903%. Both are better than the STD (58.874 m, 82.234 m, 108.241 m, 189.670 m) and MAPE (3.663%, 3.733%, 5.149%, 9.095%) of the difference between the GGM_Depth model, GEBCO_2024 model, SIOv25.1 model, and DTU18 model and the measured water depth at the checkpoint, respectively. This shows smaller inversion dispersion and higher accuracy, indicating that the FCD_Depth_SCS model obtained by fusing gravity anomaly data through the FCDNN neural network is effective.

[0112] Table 2 shows the accuracy performance of five models—FCD_Depth_SCS, GGM_Depth, GEBCO_2024, SIOv25.1, and DTU18—at different distances from the coastline.

[0113] Table 2: Accuracy results of different models at different distances from the coastline (unit: m)

[0114]

[0115] Table 3 presents the statistical results of comparing the FCD_Depth_SCS, GGM_Depth, GEBCO_2024, SIOv25.1, and DTU18 models with ship-measured water depth data in water depths greater than 500 m and less than 500 m, respectively.

[0116] Table 3: Comparison of different models with ship-measured water depth data at different water depth ranges (unit: m)

[0117]

[0118] Comparison with measured sea depth values ​​at checkpoints at different depth ranges and distances from the shore showed that the FCD_Depth_SCS model performed stably in deep-water inversion, while other models such as SIOv25.1 and DTU18 showed significant error fluctuations at greater water depths. Furthermore, the FCD_Depth_SCS model had a greater advantage in offshore areas. As the distance from the coastline decreased, the accuracy of the model gradually decreased, indicating that the FCD_Depth_SCS model was superior to other models in terms of inversion accuracy and stability.

[0119] It should be noted that this invention, by combining shipborne bathymetry data, long-wave gravity anomaly data, and short-wave gravity anomaly data, can fully utilize multi-source information to improve the accuracy of seabed topography inversion. Using a fully connected deep neural network model, it can capture complex nonlinear relationships, further enhancing the model's inversion capability. The model can more accurately capture topographic changes, resulting in superior water depth inversion accuracy. By decomposing gravity anomaly data into long-wave and short-wave components, it can better analyze seabed topographic features at different scales. Residual gravity anomalies further refine topographic features, improving the resolution of local topography. The trained model can complete large-scale seabed topography inversion in a short time, making it suitable for large-scale marine mapping tasks.

[0120] The above is an illustrative scheme of a seabed topography inversion method based on a fully connected deep neural network according to this embodiment. It should be noted that the technical solution of this seabed topography inversion system based on a fully connected deep neural network belongs to the same concept as the technical solution of the seabed topography inversion method based on a fully connected deep neural network described above. Details not described in detail in the technical solution of the seabed topography inversion system based on a fully connected deep neural network in this embodiment can be found in the description of the technical solution of the seabed topography inversion method based on a fully connected deep neural network described above.

[0121] The seabed topography inversion system based on a fully connected deep neural network in this embodiment includes:

[0122] The data acquisition module is used to acquire ship-measured water depth data and perform data preprocessing.

[0123] The data processing module is used to decompose gravity anomaly data into short-wave gravity anomaly data, long-wave gravity anomaly data, residual short-wave gravity anomaly data, and residual long-wave gravity anomaly data based on ship-measured water depth data. Specifically, after decomposing the gravity anomaly field into long-wave and short-wave components, the residual long-wave gravity anomaly and residual short-wave gravity anomaly are obtained by fitting and subtracting them from the ship-measured water depth data, respectively.

[0124] The model building module is used to construct an ocean depth inversion model. It takes gravity anomaly dataset and latitude and longitude data as input, performs neural network inversion on the ocean depth prediction model, trains the ocean depth inversion model with ship-measured water depth data, outputs ocean depth inversion results, and evaluates and optimizes the accuracy of the ocean depth inversion model based on ship-measured water depth data.

[0125] This embodiment also provides an electronic device suitable for seabed topography inversion based on fully connected deep neural networks, including:

[0126] The system includes a memory and a processor. The memory stores computer-executable instructions, and the processor executes these instructions to implement the seabed topography inversion method based on a fully connected deep neural network as proposed in the above embodiments.

[0127] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the seabed topography inversion method based on a fully connected deep neural network as proposed in the above embodiments.

[0128] The storage medium proposed in this embodiment and the seabed topography inversion method based on fully connected deep neural networks proposed in the above embodiments belong to the same inventive concept. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0129] Based on the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.

[0130] Example:

[0131] Reference Figure 6 Table 4, along with an embodiment of the present invention, provides a method for seabed topography inversion based on a fully connected deep neural network. To verify its beneficial effects, scientific demonstration is conducted through economic benefit calculations and simulation experiments.

[0132] To further verify the model's effectiveness, compare the inversion results of different sea areas, and further analyze the correlation between the accuracy of the seabed topography model based on neural network inversion and the size of the sea area, and to provide an implementation case for the subsequent inversion of the global seabed topography model, four sea areas of different sizes were selected in the South China Sea for experiments.

[0133] Table 4 presents the comparative statistical results of the inversion results for four sea areas A, B, C, and D using the FCDNN neural network and ship-measured water depth data. It also provides the comparison results of the FCD_Depth_SCS model with ship-measured water depth in each sea area.

[0134] Table 4: Comparison of depth inversion model with ship-measured water depth in various sea areas

[0135]

[0136] This invention selected four sea areas of different sizes (A, B, C, and D) in the South China Sea to analyze the correlation between model accuracy and sea area size and topography. (See Table 4 and...) Figure 6 The results show that the non-regional model has a significant advantage in flat or less varied regions A and C, with both STD and MAPE lower than the regional model. In region B, although the topography is more complex, the seabed is flat, and the non-regional model still shows a slightly better performance than the regional model. In complex topographic regions (such as region D), the STD of the difference between the regional model FCD_Depth_SCS_D and the ship-measured depth at the checkpoint is 98.529 m, while the STD of the non-regional model is 163.879 m, and the MAPE (2.008%) is also better. This indicates that the regional model can more accurately capture topographic changes and has better accuracy in retrieving water depth.

[0137] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for seabed topography inversion based on a fully connected deep neural network, characterized in that, include: Acquire ship-measured water depth data and gravity anomaly data, and perform data preprocessing; Based on the ship-measured water depth data, the gravity anomaly data is decomposed into short-wave gravity anomaly data, long-wave gravity anomaly data, residual short-wave gravity anomaly data, and residual long-wave gravity anomaly data. Specifically, after decomposing the gravity anomaly field into long-wave and short-wave components, the residual long-wave gravity anomaly and residual short-wave gravity anomaly are obtained by fitting and subtracting them from the ship-measured water depth data, respectively. A sea depth inversion model is constructed. Gravity anomaly data and latitude and longitude data are used as inputs. The sea depth inversion model is inverted by a neural network. The neural network is trained on the sea depth inversion model by ship-measured water depth data. The sea depth inversion results are output. The accuracy of the sea depth inversion model is evaluated and optimized based on the ship-measured water depth data. Gravity anomaly data includes: The gravity anomaly field is decomposed into long-wave and short-wave components. The short-wave gravity anomaly of the control point is calculated. The long-wave gravity anomaly of the control point is obtained by subtracting the short-wave gravity anomaly of the control point from the gravity anomaly of the control point. The long-wave gravity anomaly of the control point is gridded to construct a long-wave gravity anomaly field. The long-wave gravity anomaly of the unknown point is obtained by interpolation, and the short-wave gravity anomaly of the unknown point is solved. The linear relationship between the long and short wave components and the ship-measured water depth is determined by a linear regression model. The ship-measured water depth is multiplied by the corresponding fitting slope to obtain the reference values ​​of the long wave and short wave gravity anomalies. The long wave and short wave gravity anomalies are subtracted from the reference values ​​to obtain the residual long wave gravity anomaly and the residual short wave gravity anomaly. The construction of the ocean depth inversion model includes: The input layer contains longitude, latitude, and gravity anomaly data, including shortwave gravity anomaly, longwave gravity anomaly, residual shortwave gravity anomaly, and residual longwave gravity anomaly. There are four hidden layers. The first hidden layer has 16 neurons, the second hidden layer has 32 neurons, the third hidden layer has 256 neurons, and the fourth hidden layer has 512 neurons. The neurons in each layer are interconnected through weights and biases. The output layer, via the hidden layer, outputs the ocean depth inversion result.

2. The seabed topography inversion method based on a fully connected deep neural network as described in claim 1, characterized in that, The acquisition of ship-measured water depth data and the subsequent data preprocessing include: Error screening is performed on the ship-measured depth data, and the depth of the world ocean seabed topography model is interpolated to the ship-measured depth points to calculate the difference between the ship-measured depth values ​​and the actual ship-measured depth values. Based on the differences, the standard deviations of the two are calculated, and depth sounding points with differences exceeding three times the standard deviation are removed. Based on a control point to checkpoint ratio of 4:1, control points and checkpoints are randomly obtained.

3. The seabed topography inversion method based on a fully connected deep neural network as described in claim 1, characterized in that, The ocean depth inversion model is expressed as follows: ； in, This represents the input gravitational field signal. y Represents the output of a neuron. , These represent the input weights and the neuron biases, respectively. Indicates the activation function; In the model initialization phase, weights and biases are assigned initial values. During training, the weights of each neuron are adjusted based on the inversion differences. The weight iteration formula is expressed as: ； in, This indicates the adjusted weights. Indicates the learning rate. Indicates the initial weights. loss This represents a loss function that depends on the mean squared error.

4. The seabed topography inversion method based on a fully connected deep neural network as described in claim 1, characterized in that, The neural network inversion includes: The location information of the ship's depth measurement control points and gravity anomaly data are divided into training set and test set; The training dataset and the test dataset are normalized according to standard, and the ocean depth inversion model is trained accordingly. Initialize the ocean depth inversion model by randomly generating initial weights and biases; The network was trained for 40 epochs using a loss function based on the root mean square error of inversion depth and shipborne depth, with an initial learning rate of 0.

001. The model parameters were then adjusted using the Adam optimization algorithm. The gravity signal obtained from the height measurement is fed into the trained ocean depth inversion model to obtain the inversion water depth of the measurement area.

5. The seabed topography inversion method based on a fully connected deep neural network as described in claim 4, characterized in that, The accuracy assessment optimization includes: Multiple comparison models were interpolated to the check point using bicubic interpolation and compared with the ship-measured water depth data at the check point. The values ​​of standard deviation, root mean square error, and mean absolute percentage error are used as indicators of model accuracy.

6. A seabed topography inversion system based on a fully connected deep neural network, used in the seabed topography inversion method based on a fully connected deep neural network as described in any one of claims 1-5, characterized in that, include, The data acquisition module is used to acquire ship-measured water depth data and perform data preprocessing. The data processing module is used to decompose gravity anomaly data into short-wave gravity anomaly data, long-wave gravity anomaly data, residual short-wave gravity anomaly data, and residual long-wave gravity anomaly data based on the ship-measured water depth data. Specifically, after decomposing the gravity anomaly field into long-wave and short-wave components, the residual long-wave gravity anomaly and residual short-wave gravity anomaly are obtained by fitting and subtracting them from the ship-measured water depth data, respectively. The model building module is used to build an ocean depth inversion model. It takes gravity anomaly data and latitude and longitude data as input, performs neural network inversion on the ocean depth inversion model, trains the neural network on the ocean depth inversion model with ship-measured water depth data, outputs ocean depth inversion results, and evaluates and optimizes the accuracy of the ocean depth inversion model based on ship-measured water depth data. Gravity anomaly data includes: The gravity anomaly field is decomposed into long-wave and short-wave components. The short-wave gravity anomaly of the control point is calculated. The long-wave gravity anomaly of the control point is obtained by subtracting the short-wave gravity anomaly of the control point from the gravity anomaly of the control point. The long-wave gravity anomaly of the control point is gridded to construct a long-wave gravity anomaly field. The long-wave gravity anomaly of the unknown point is obtained by interpolation, and the short-wave gravity anomaly of the unknown point is solved. The linear relationship between the long and short wave components and the ship-measured water depth is determined by a linear regression model. The ship-measured water depth is multiplied by the corresponding fitting slope to obtain the reference values ​​of the long wave and short wave gravity anomalies. The long wave and short wave gravity anomalies are subtracted from the reference values ​​to obtain the residual long wave gravity anomaly and the residual short wave gravity anomaly. The construction of the ocean depth inversion model includes: The input layer contains longitude, latitude, and gravity anomaly data, including shortwave gravity anomaly, longwave gravity anomaly, residual shortwave gravity anomaly, and residual longwave gravity anomaly. There are four hidden layers. The first hidden layer has 16 neurons, the second hidden layer has 32 neurons, the third hidden layer has 256 neurons, and the fourth hidden layer has 512 neurons. The neurons in each layer are interconnected through weights and biases. The output layer, via the hidden layer, outputs the ocean depth inversion result.

7. An electronic device, characterized in that, include: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions, which, when executed by the processor, implement the steps of the seabed topography inversion method based on a fully connected deep neural network as described in any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, It stores computer-executable instructions that, when executed by a processor, implement the steps of the seabed topography inversion method based on a fully connected deep neural network as described in any one of claims 1 to 5.

Images