Seabed sea mountain terrain inversion method based on removal and recovery method

Abstract

The invention discloses a submarine sea mountain terrain inversion method based on a removal and recovery method, and the method comprises the steps: removing the influence of a sea mountain on a boundary based on a water depth forward modeling formula through the removal and recovery method by employing water depth forward modeling gravity abnormality and a vertical gravity gradient, and obtaining a high-precision sea mountain boundary terrain; and high-precision inversion is carried out on the sea and mountain main body by combining a GGM method. According to the method, the influence of sea and mountain topographic relief on boundary topography inversion in the linear relation is removed, the nonlinear relation in the linear relation is captured through deep learning, meanwhile, the water depth of a sea and mountain body is obtained through inversion of high-precision ship survey water depth points in combination with a GGM, and finally the high-precision submarine topography of the complete sea and mountain is obtained.

Description

Technical Field

[0001] The invention belongs to the technical field of terrain inversion, and in particular relates to a method for inverting seabed and seamount terrain based on a removal restoration method. Background Art

[0002] The ocean occupies most of the earth's surface, which contains a large number of minerals and resources. Seabed topography data is the basis for studying related disciplines such as marine biology, marine chemistry and marine geophysics. Since the invention of acoustic bathymetry, two high-precision shipborne water depth detection methods based on echo sounding technology have emerged. Research ships equipped with single-beam and multi-beam can carry out high-precision and high-resolution seabed topography surveys. However, shipborne water depth measurement still has defects such as low detection efficiency and uneven distribution of shipborne sounding points. Therefore, it takes a lot of financial and material resources to conduct a comprehensive detection of seabed topography through echo sounding technology. At present, about 25% of the seabed area has been explored by shipborne water depth detection methods, and 75% of the marine topography is still in blank areas. The current way to obtain the water depth of blank waters is generally to obtain it through gravity data inversion. Its advantages are wide coverage and no influence from the marine environment, but its inversion accuracy and efficiency are limited.

[0003] The Parker formula method provides a theoretical basis for directly inverting the seabed topography using gravity anomalies in frequency. Based on the Parker formula, Dixon et al. used altimetry data in 1983 to confirm that there is a high correlation between the seabed topography and the geoid fluctuations in the 50-300km band. Studies on the frequency domain characteristics of gravity anomalies show that there is a linear correlation between the seabed topography and gravity anomalies in the 15-160km band. The GGM method based on Bouguer correction in space has certain defects. It relies on the linear relationship between water depth and gravity parameters and ignores the nonlinear relationship. The linear relationship is easily affected by the terrain fluctuations. The inversion accuracy. In addition, the GGM method is also highly dependent on the quality of the input ship-measured water depth. The ship-measured water depth data determines the inversion accuracy of the GGM method.

[0004] Based on the forward modeling formula of water depth, the present invention uses the forward modeling gravity anomaly and vertical gravity gradient of water depth to remove the influence of seamounts on their boundaries through the removal and recovery method, obtains high-precision seamount boundary terrain, and combines the GGM method to perform high-precision inversion of the main body of the seamount. This method not only removes the influence of seamount terrain fluctuations on boundary terrain inversion in the linear relationship, but also captures the nonlinear relationship through deep learning. At the same time, it also combines GGM to invert the use of high-precision ship-measured water depth points to obtain the water depth of the main body of the seamount, and finally obtains the high-precision seabed terrain of the complete seamount. Summary of the invention

[0005] The purpose of the present invention is to design a method for inverting the seabed seamount topography based on the removal and restoration method. The method uses the forward depth gravity anomaly and the vertical gravity gradient to remove the influence of the seamount on its boundary based on the forward depth modeling formula, obtain high-precision seamount boundary topography, and combine the GGM method to perform high-precision inversion of the main body of the seamount.

[0006] In order to achieve the above object, the technical solution adopted by the present invention is as follows:

[0007] A method for inverting seamount topography based on a removal and restoration method, characterized in that: the method is as follows:

[0008] S1: Obtain satellite gravity field data and shipborne water depth data to ensure that the data can cover the study area;

[0009] S2: pre-process the data and remove outliers;

[0010] S3: The long-wave component of the seamount boundary gravity anomaly VG / vertical gravity gradient VGG is calculated by the forward modeling formula, and the short-wave component is calculated by combining the corresponding observation value. The short-wave component value is used as one of the input data for predicting the seamount boundary water depth by the fully connected deep network model;

[0011] S4: Calculate the short-wave gravity anomaly component of the control point through the forward modeling formula, and use the short-wave gravity anomaly component value as one of the input data for predicting the water depth inside the seamount by the fully connected depth network model;

[0012] S5: Construct a fully connected deep neural network model for gravity field inversion of water depth. The fully connected deep neural network model contains several hidden layers. The model learns the nonlinear mapping relationship between satellite gravity field data and shipborne bathymetric data, predicts the relationship between the seabed depth and the actual shipborne measured depth, and finally realizes the inversion of the seabed topography.

[0013] Furthermore, the obtaining of satellite-observed gravity field data and shipborne water depth data in S1 specifically includes: obtaining ocean gravity field model data and shipborne water depth data from a public database or data source.

[0014] Furthermore, in S2, the data is compared with the GEBCO water depth model of the world ocean floor topography map, and points with a difference of more than 10% are eliminated.

[0015] Furthermore, the specific process of obtaining the long-wave component of the gravity anomaly VG or the vertical gravity gradient VGG in S2 is as follows: the gravity formula of a single grid is calculated using the following formula:

[0016]

[0017] Where: Ω is a rectangular prism with uniform density and mass distribution under water; the coordinate system is established with the ground center of the rectangular prism Ω as the origin, P(x, y, z) is any point on the sea surface, and G is the gravitational constant 6.672×10 -11 Nm 2 kg -2 , ρ is the density of Ω, (ξ, ψ, ζ) are the integral variables, and the coordinates of the point on the prism are any point on the prism Ω;

[0018] VG and VGG generated by Ω at point P can be expressed as:

[0019]

[0020] in

[0021] Where: length, width and height are 2a, 2b and Hh respectively, where H is the maximum depth, i.e. the distance from the bottom to the sea level, h is the distance from the rectangular prism to the sea surface; r is the distance from any point on the prism Ω to any point on the sea surface where P(x, y, z) is;

[0022] The above are based on gravity anomalies or vertical gravity gradients generated from a single prism;

[0023] The study area R = {(x, y); -a≤x≤a, -a≤y≤a} is a square area on the sea surface, Ω = {(x, y, z), (x, y)∈R, Hh(x, y)≤z≤H} is the terrain corresponding to the R area, where h(x, y) is the height from the top of Ω to the sea surface; the area R is grid-divided, [-a, a] is equally divided into 2N sub-areas R ij , using the superposition principle, we can calculate that VG generated by Ω at sea level can be expressed as:

[0024]

[0025] Where: h ij The distance from the rectangular prism representing the current point to the sea surface; Represents the calculation and accumulation of gravity anomalies for each sub-area;

[0026] Where: J(h, x, y) represents the gravity anomaly generated in the sub-grid area after the four points around the grid are superimposed.

[0027]

[0028] Where: J(h, x, y) represents the superposition value of four points around the grid after being divided into sub-grids. This value is regarded as the gravity anomaly generated in the sub-grid area. Due to the lack of parameter calculation, it is only regarded as an intermediate calculation step;

[0029] The influence of seawater can be removed by replacing ρ with Δρ (Δρ = ρs-ρw), where ρs and ρw are the densities of bedrock and seawater, respectively; J R (H, x, y) Since the maximum depth H is a fixed value, that is, the J calculated for each sub-region R (H, x, y) are all the same value, so the final VG generated by Ω at sea level can be expressed as:

[0030]

[0031] VGG can be obtained in the same way:

[0032]

[0033]

[0034] Where: I(h, x, y) represents the superposition value of four points around the grid after being divided into sub-grids. This value is regarded as the vertical gravity gradient generated in the sub-grid area. Due to the lack of parameter calculation, it is only regarded as an intermediate calculation step;

[0035] The above is the calculated gravity anomaly or the long-wave component of the vertical gravity gradient.

[0036] Furthermore, the shortwave component can be obtained by subtracting the longwave component from the observed value:

[0037] ξ short =ξ 观测 -ξ long , where ξ shhort represents the shortwave component, ξ 观测 represents the observed value, ξ long Represents the long wave component.

[0038] The shortwave component value is used as one of the input values ​​of the fully connected deep network model to predict the water depth of the seamount boundary.

[0039] Furthermore, the specific process of calculating the shortwave gravity anomaly component in S4 is as follows:

[0040] The shortwave component of gravity anomaly ΔGs(j) at the control point is calculated by the following formula:

[0041] ΔG s (j) = 2πGΔρ(E(j)-D);

[0042] ΔG s (j) is the short-wave gravity anomaly of the control point, G is the gravitational constant 6.672×10 -8 cm 3 / g·s, Δρ is the density coefficient, E(j) is the water depth of the control point, and D is the reference water depth, which is usually the maximum water depth;

[0043] The long-wave component of the gravity anomaly ΔGl(j) of the control point is calculated by the following formula, where ΔG(j) is the gravity anomaly of the control point: ΔG l (j) = ΔG(j) - ΔGs(j);

[0044] The spatial interpolation method obtains the long-wave gravity data of the regular grid in the inversion area, and further calculates the short-wave gravity grid data of the point where the water depth is measured without a ship. The calculation method is as follows: ΔG s (j) = ΔG(j) - ΔG l (j)

[0045] The short-wave gravity anomaly component value is used as one of the input values ​​of the fully connected deep network model to predict the water depth inside the seamount. Furthermore, the fully connected deep neural network model for gravity field inversion water depth constructed in S5 specifically includes: the fully connected deep neural network FC-DNN structure network contains 3 hidden layers, with 20, 10 and 5 neurons in each layer respectively; the output layer is used to predict the depth data of the target location; in this network, the learning rate is set to 0.001, the training round epoch is set to 30, and the number of samples per batch is 120; the activation function uses ReLU, the optimizer selects Adam, and the loss function is Mean Squared Error, MSE. The input data for the inversion of the interior of the seamount are: longitude, latitude, short-wave gravity anomaly, and the output data is water depth; the input data for the inversion of the seamount boundary are: longitude, latitude, gravity anomaly / vertical gravity gradient short-wave component, and the output data is water depth.

[0046] Through the above technical scheme, the present invention not only removes the influence of the seamount terrain fluctuation on the boundary terrain inversion in the linear relationship, but also captures the nonlinear relationship therein through deep learning, and at the same time combines GGM to invert the use of high-precision ship-measured water depth points to obtain the water depth of the main body of the seamount, and finally obtains the high-precision seabed topography of the complete seamount.

[0047] The specific advantages are:

[0048] 1. Improved the accuracy of water depth inversion of seamounts: The present invention combines the removal and restoration method inversion with the GGM method inversion, and can simultaneously invert the water depth of the main body and the boundary of the seamount with high accuracy. And through FC-DNN, it can effectively learn the complex nonlinear mapping relationship between satellite gravity field data and shipborne single-beam bathymetric data, thereby improving the accuracy of water depth inversion.

[0049] 2. Improved the efficiency of water depth inversion for seamounts: The present invention uses deep learning as the basic inversion method. Its advantage is that the model only needs to be thoroughly trained once before it can be used on any seamount, and no secondary training is required, which greatly saves the time of water depth inversion in conventional methods.

[0050] 3. Innovative introduction of the short-wave component of the gravity anomaly / vertical gravity gradient: We found that by removing the long-wave component of the gravity anomaly / vertical gravity gradient generated by the seamount and then performing water depth inversion on the remaining short-wave component, the accuracy of water depth inversion on the boundary can be significantly improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 GGM inverted water depth map, where the left side is the boundary 10km and the right side is the boundary 20km.

[0052] Figure 2 Cross-sectional water depth map, the left one is the cross-section along the longitude, and the right one is the cross-section along the latitude.

[0053] Figure 3 The final result of the shortwave gravity parameter inversion boundary combined with the GGM inversion of the main body of the seamount, including (a) VG shortwave inversion 10km boundary; (b) VG shortwave inversion 20km boundary (c) VGG shortwave inversion 10km boundary; (d) VGG shortwave inversion 20km boundary. DETAILED DESCRIPTION

[0054] The present invention will be further described below by way of examples and drawings:

[0055] like Figure 1-3 As shown in the figure, a method for inverting the seabed seamount topography based on the removal and restoration method is as follows:

[0056] 1. Obtain satellite-observed gravity field data and ship-borne water depth data:

[0057] In this embodiment, the ocean gravity anomaly data grid model and shipborne single-beam bathymetric data are obtained from a public database or data source to ensure that the acquired data covers the study area and that the quality and accuracy of the data meet the needs of the research.

[0058] 2. Preprocess the data

[0059] In this step, by comparing with the GEBCO water depth model, points with a difference of more than 10% are eliminated. This ensures that the data in the training data set is accurate and reliable, thereby improving the inversion accuracy. In this embodiment, the gebco2022 terrain model is imported to obtain the water depth points of the gebco model at the same position of the ship-borne water depth measurement points. With 10% as the limit, the ship-borne water depth measurement points with a difference of more than 10% are eliminated.

[0060] 3. The long-wave component of the seamount boundary gravity anomaly VG or the vertical gravity gradient VGG is calculated by the forward formula, and the short-wave component is calculated by combining the corresponding observations:

[0061] The gravity formula for a single grid is calculated using the following formula:

[0062]

[0063] Where: Ω is a rectangular prism with uniform density and mass distribution under water; the coordinate system is established with the ground center of the rectangular prism Ω as the origin, P (x, y, z) is an arbitrary point on the sea surface, and G is the gravitational constant 6.672×10 -11 Nm 2 kg -2 , ρ is the density of Ω, (ξ, ψ, ζ) are the integral variables, and the coordinates of the point on the prism are any point on the prism Ω.

[0064] VG and VGG generated by Ω at point P can be expressed as:

[0065]

[0066] in

[0067] In the formula: the length, width and height are 2a, 2b, and Hh respectively, where H is the maximum depth, that is, the distance from the bottom to the sea level, and h is the distance from the rectangular prism to the sea surface; r is the distance between any point on the prism Ω and P(x, y, z) is any point on the sea surface. The above is the gravity anomaly or vertical gravity gradient generated based on a single prism.

[0068] The study area R = {(x, y); -a≤x≤a, -a≤y≤a} is a square area on the sea surface, Ω = {(x, y, z); (x, y)∈R, Hh(x, y)≤z≤H} is the terrain corresponding to the R area, where h(x, y) is the height from the top of Ω to the sea surface. The area R is gridded, [-a, a] is equally divided into 2N sub-areas Rij, and the VG generated by Ω at sea level can be calculated using the superposition principle, which can be expressed as:

[0069]

[0070] Where: h ij The distance from the rectangular prism representing the current point to the sea surface; Represents the calculation and accumulation of gravity anomalies for each sub-area;

[0071]

[0072] Where: J(h, x, y) represents the superposition value of four points around the grid after being divided into sub-grids. This value is regarded as the gravity anomaly generated in the sub-grid area. Due to the lack of parameter calculation, it is only regarded as an intermediate calculation step;

[0073] The influence of seawater can be removed by replacing ρ with Δρ (Δρ = ρs-ρw), where ρs and ρw are the densities of bedrock and seawater, respectively; J R (H, x, y) Since the maximum depth H is a fixed value, that is, the J calculated for each sub-region R (H, x, y) are all the same value, so the final VG generated by Ω at sea level can be expressed as:

[0074]

[0075] VGG can be obtained in the same way:

[0076]

[0077] Among them, J R (H, x, y) and I R (H, x, y) is a part removed from the formula at the beginning. It is for the sake of the beauty of the formula and the convenience of calculation, and has no practical significance. The height of the prism is Hh, So the formula is as follows: the maximum depth of H is always fixed, h ij Each sub-region is different, so J R (H, x, y) and I R (H, x, y) is a fixed value that can be used to add up the symbols. N represents the sub-grids that are divided and does not require special explanation.

[0078]

[0079] Where: I(h, x, y) represents the superposition value of four points around the grid after being divided into sub-grids. This value is regarded as the vertical gravity gradient generated in the sub-grid area. Due to the lack of parameter calculation, it is only regarded as an intermediate calculation step;

[0080] The above is the calculated gravity anomaly or the long-wave component of the vertical gravity gradient.

[0081] The shortwave component can be obtained by subtracting the longwave component from the observed value:

[0082] ξ short =ξ 观测 -ξ long

[0083] The shortwave component value is used as one of the inputs of the fully connected deep network model to predict the water depth of seamount boundaries.

[0084] 4. The shortwave gravity anomaly component of the control point is calculated by the formula, and the value of the shortwave gravity anomaly component is used as one of the inputs of the fully connected deep network model to predict the water depth inside the seamount.

[0085] The short-wave component of gravity anomaly ΔG at the control point s (j) is calculated by the following formula:

[0086] ΔG s (j) = 2πGΔρ(E(j)-D)

[0087] ΔG s (j) is the short-wave gravity anomaly of the control point, G is the gravitational constant 6.672×10 -8 cm 3 / g·s, Δρ is the density coefficient, E(j) is the water depth of the control point, and D is the reference water depth, which is usually the maximum water depth.

[0088] The long-wave component of gravity anomaly ΔG at the control point l (j) is calculated by the following formula, where ΔG(j) is the gravity anomaly of the control point:

[0089] ΔG l (j) = ΔG(j) - ΔG s (j)

[0090] The spatial interpolation method obtains the long-wave gravity data of the regular grid in the inversion area, and further calculates the short-wave gravity grid data of the points where the water depth is measured without a ship. The calculation method is as follows:

[0091] ΔG s (j) = ΔG(j) - ΔG l (j)

[0092] The shortwave gravity anomaly component value is used as one of the inputs of the fully connected deep network model to predict the water depth inside the seamount.

[0093] 5. FC-DNN model design

[0094] The above-mentioned fully connected deep neural network (FC-DNN) structure network contains 3 hidden layers, with 20, 10 and 5 neurons in each layer respectively. The output layer is used to predict the depth data of the target location. In this network, the learning rate is set to 0.001, the training round (epoch) is set to 30, and the number of samples in each batch (batch size) is 120. The activation function uses ReLU, the optimizer selects Adam, and the loss function is Mean Squared Error (MSE). The input data for the inversion of the interior of the seamount are: longitude, latitude, short-wave gravity anomaly, and the output data is water depth; the input data for the inversion of the seamount boundary are: longitude, latitude, gravity anomaly / vertical gravity gradient short-wave component, and the output data is water depth.

[0095] 6. Combining the seamount boundary topography obtained in step 3 above with the main seamount topography obtained in step 4 above, the topography and water depth of the complete seamount are finally obtained.

[0096] like Figure 1 The main seamount terrain and the boundary terrain affected by the undulations of the main seamount terrain were obtained by the GGM method using the shipborne water depth data and gravity anomaly data of the main seamount. The results show that the boundary accuracy of the GGM method inverting the boundary terrain is low due to the influence of the seamount. The rms error of the 10km boundary is 66.8m, and the error of the 20km boundary is 108.85m.

[0097] In addition, the depth error between the boundary and the main body of the seamount can be clearly observed by making a profile, such as Figure 2 We made profiles along the longitude and latitude lines, and drew profile scatter plots with 1km intervals. The results show that the error increases with the distance outside the main area of ​​the seamount.

[0098] The boundary water depth of 10km / 20km is further obtained by inverting the shortwave gravity parameters, and combined with the main body of the seamount obtained by the GGM method to obtain the final complete seamount topography, such as Figure 3 The results show that in the final topography obtained by combining the 10km boundary / 20km boundary obtained by VG shortwave inversion with the main body of the seamount of GGM, the rms error of the terrain obtained with a boundary of 10km is 44.06m, and the rms error of the terrain obtained with a boundary of 20km is 41.80m; in the final topography obtained by combining the 10km boundary / 20km boundary obtained by VGG shortwave inversion with the main body of the seamount of GGM, the rms error of the terrain obtained with a boundary of 10km is 45.54m, and the rms error of the terrain obtained with a boundary of 20km is 49.93m.

[0099] By removing the influence of seamounts by the removal and restoration method, we finally obtained the seamount topography. When the boundary is 10km, the accuracy of the results obtained by VG parameters is improved by 34.04%, and the accuracy of the results obtained by VGG parameters is improved by 31.82%. When the boundary is 20km, the accuracy of the results obtained by VG parameters is improved by 61.60%, and the accuracy of the results obtained by VGG parameters is improved by 53.81%.

[0100] The above are all preferred embodiments of the present invention. For ordinary technicians in this technical field, without departing from the principle of the present invention, various equivalent modifications to the present invention belong to the protection scope of the claims attached to this application.

Claims

1. A method for inverting seamount topography based on the removal and restoration method, characterized in that: Here’s how: S1: Obtain satellite gravity field data and shipborne water depth data to ensure that the data can cover the study area; S2: pre-process the data and remove outliers; S3: The long-wave component of the seamount boundary gravity anomaly VG or the vertical gravity gradient VGG is calculated by the forward modeling formula, and the short-wave component is calculated by combining the corresponding observation value. The short-wave component value is used as one of the input data for the fully connected deep network model to predict the seamount boundary water depth; S4: Calculate the short-wave gravity anomaly component of the control point through the forward modeling formula, and use the short-wave gravity anomaly component value as one of the input data for predicting the water depth inside the seamount by the fully connected depth network model; S5: Construct a fully connected deep neural network model for gravity field inversion of water depth. The fully connected deep neural network model contains several hidden layers. The model learns the nonlinear mapping relationship between satellite gravity field data and shipborne bathymetric data, predicts the relationship between the seabed depth and the actual shipborne measured depth, and finally realizes the inversion of the seabed topography.

2. The method for inverting seamount topography based on the removal and restoration method according to claim 1, characterized in that: The step of acquiring satellite-observed gravity field data and ship-borne water depth data in S1 specifically includes: acquiring ocean gravity field model data and ship-borne water depth data from a public database or data source.

3. The method for inverting seamount topography based on the removal and restoration method according to claim 1, characterized in that: In S2, the data are compared with the GEBCO bathymetric model of the world ocean floor topography map, and points with differences exceeding 10% are eliminated.

4. The method for inverting seamount topography based on the removal and restoration method according to claim 1, characterized in that: The specific process of obtaining the long-wave component of the gravity anomaly VG or the vertical gravity gradient VGG in S2 is as follows: The gravity formula for a single grid is calculated using the following formula: Where: Ω is a rectangular prism with uniform density and mass distribution under water; the coordinate system is established with the ground center of the rectangular prism Ω as the origin, P(x, y, z) is any point on the sea surface, and G is the gravitational constant 6.672×10 -11 Nm 2 kg -2 , ρ is the density of Ω, (ξ, ψ, ζ) are the integral variables, and the coordinates of the point on the prism are any point on the prism Ω; VG and VGG generated by Ω at point P can be expressed as: in Where: length, width and height are 2a, 2b and Hh respectively, where H is the maximum depth, i.e. the distance from the bottom to the sea level, h is the distance from the rectangular prism to the sea surface; r is the distance from any point on the prism Ω to any point on the sea surface where P(x, y, z) is; The above are based on gravity anomalies or vertical gravity gradients generated from a single prism; The study area R = {(x, y); -a≤x≤a, -a≤y≤a} is a square area on the sea surface, Ω = {(x, y, z), (x, y)∈R, Hh(x, y)≤z≤H} is the terrain corresponding to the R area, where h(x, y) is the height from the top of Ω to the sea surface; the area R is grid-divided, [-a, a] is equally divided into 2N sub-areas R ij , using the superposition principle, we can calculate that VG generated by Ω at sea level can be expressed as: Where: h ij The distance from the rectangular prism representing the current sub-area to the sea surface; Represents the calculation of gravity anomaly in the current sub-area; Among them: J(h,x,y) represents the superposition value of four points around the grid after being divided into sub-grids. This value is regarded as the gravity anomaly generated in the sub-grid area. Due to the lack of parameter calculation, it is only regarded as an intermediate calculation step; The influence of seawater can be removed by replacing ρ with Δρ (Δρ = ρs-ρw), where ρs and ρw are the densities of bedrock and seawater, respectively; J R (H, x, y) Since the maximum depth H is a fixed value, that is, the J calculated for each sub-region R (H, x, y) are all the same value, so the final VG generated by Ω at sea level can be expressed as: VGG can be obtained in the same way: Where: I(h, x, y) represents the superposition value of four points around the grid after being divided into sub-grids. This value is regarded as the vertical gravity gradient generated in the sub-grid area. Due to the lack of parameter calculation, it is only regarded as an intermediate calculation step; The above is the calculated gravity anomaly or the long-wave component of the vertical gravity gradient.

5. The method for inverting seamount topography based on the removal and restoration method according to claim 4, characterized in that: The shortwave component can be obtained by subtracting the longwave component from the observed value: ξ short =ξ 观测 -ξ long , where ξ short represents the shortwave component, ζ 观测 represents the observed value, ζ long Represents the long wave component. The shortwave component value is used as one of the input values ​​of the fully connected deep network model to predict the water depth of the seamount boundary.

6. The method for inverting seamount topography based on the removal and restoration method according to claim 1, characterized in that: The specific process of calculating the shortwave gravity anomaly component in S4 is as follows: The shortwave component of gravity anomaly ΔGs(j) at the control point is calculated by the following formula: ΔGs(j)=2πGΔρ(E(j)-D); ΔG s (j) is the short-wave gravity anomaly of the control point, G is the gravitational constant 6.672×10 -8 cm 3 / g·s, Δρ is the density coefficient, E(j) is the water depth at the control point, and D is the reference water depth, which is usually the maximum water depth; The long-wave component of gravity anomaly ΔG at the control point l (j) is calculated by the following formula, where ΔG(j) is the gravity anomaly of the control point: ΔG L (j)=ΔG(j)-ΔGs(j); The spatial interpolation method obtains the long-wave gravity data of the regular grid in the inversion area, and further calculates the short-wave gravity grid data of the points where the water depth is measured without a ship. The calculation method is as follows: ΔG s (j)=ΔG(j)-ΔG l (j) The shortwave gravity anomaly component is used as one of the input values ​​of the fully connected depth network model to predict the water depth inside the seamount.

7. The method for inverting seamount topography based on the removal and restoration method according to claim 1, characterized in that: The fully connected deep neural network model for gravity field inversion water depth constructed in S5 specifically includes: the fully connected deep neural network FC-DNN structure network contains 3 hidden layers, and the number of neurons in each layer is 20, 10 and 5 respectively; the output layer is used to predict the depth data of the target location; in this network, the learning rate is set to 0.001, the training round epoch is set to 30, and the number of samples in each batch is 120; the activation function uses ReLU, the optimizer selects Adam, and the loss function is Mean Squared Error, MSE. The input data for seamount interior inversion are: longitude, latitude, shortwave gravity anomaly, and the output data is water depth; the input data for seamount boundary inversion are: longitude, latitude, gravity anomaly or vertical gravity gradient shortwave component, and the output data is water depth.

Images