Novel GEO SAR deep sea underwater geography imaging and inversion method

Abstract

The invention discloses a novel GEO SAR deep sea underwater geography imaging and inversion method, and belongs to the technical field of radar remote sensing and ocean exploration. The problem that in the prior art, a traditional deep sea underwater geography imaging and inversion method is low in precision is solved. According to the method, a deep sea underwater dynamic model is constructed through a layered full-nonlinear hydrodynamic equation set; gEO SAR sea surface echo signals are obtained and preprocessed, and re-processing and focusing imaging are carried out on the GEO SAR sea surface echo signals to obtain a focused sea surface wave complex image; extracting sea surface wave multi-dimensional parameters according to the deep sea underwater dynamic model to obtain sea surface elevation and radial flow velocity contribution of the current mode; performing two-dimensional surface flow field inversion to obtain a corrected two-dimensional flow field; and training the network structure, and outputting deep sea underwater topography inversion through the obtained trained deep sea underwater topography inversion network model. According to the method, the underwater geography imaging and inversion accuracy is improved, and the method can be applied to sea surface wave imaging quality evaluation.

Description

Technical Field

[0001] This invention relates to a novel GEO SAR deep-sea underwater topography imaging and inversion method, belonging to the field of radar remote sensing and marine exploration technology. Background Technology

[0002] Accurate underwater topographic mapping is crucial for submarine engineering, marine resource exploration, shipping safety, and monitoring of natural disasters such as tsunamis. Traditional depth measurement mainly relies on shipborne sonar, which, while highly accurate, is slow, costly, and has limited spatial coverage, making it difficult to meet the needs of large-scale, long-term continuous observation. Optical remote sensing can quickly estimate water depth in shallow waters by utilizing the attenuation and refraction properties of light, but its applicability is significantly limited in deep or turbid waters because light has difficulty penetrating the water.

[0003] In the deep-sea environment, complex seabed topography can generate topographic waves and internal waves. These waves propagate upward in the stratified seawater and modulate the sea surface height and surface current field. The texture and wave characteristics formed on the sea surface can be observed by synthetic aperture radar (SAR), thus providing a new approach for indirect inversion of underwater topography. Geostationary orbit synthetic aperture radar (GEO SAR) has a nearly fixed sub-satellite trajectories, which can perform long-term, multi-frame continuous imaging of the same deep-sea area, providing conditions for capturing the temporal evolution of topography-induced sea surface waves.

[0004] However, existing technologies for deep-sea underwater topography imaging and inversion under GEO SAR conditions still face two main problems: First, sea surface waves induced by deep-sea topography undergo nonlinear and spatially variable propagation in stratified water bodies. Their complex motion characteristics, such as wave crest curvature, enhanced anisotropy, and local vortices, cause severe temporal decorrelation during long synthetic aperture integration, leading to image blurring or even loss. Existing imaging methods, which are mostly based on first-order or approximately linear motion models, cannot adequately handle such high-order, anisotropic motions. Second, many water depth inversion methods based on SAR or other remote sensing data generally treat "underwater topography-sea surface texture" as an approximately direct mapping, ignoring the filtering and distortion effects of deep-sea stratified propagation on topographic signals. Most methods are only suitable for converting static images to water depth using single-frame observations in shallow water backgrounds, making it difficult to fully utilize the temporal response information of sea surface height and two-dimensional flow field provided by GEO SAR, and difficult to characterize the complex propagation relationship between deep-sea topography and sea surface observation results.

[0005] In summary, a novel GEOSAR deep-sea underwater topography imaging and inversion method is needed to achieve clear imaging with a long synthetic aperture while fully considering layered propagation and complex sea surface movements, and to achieve high-precision inversion of deep-sea underwater topography by combining multi-frame sea surface multidimensional information. Summary of the Invention

[0006] A brief overview of the invention is given below to provide a basic understanding of certain aspects of it. It should be understood that this overview is not an exhaustive summary of the invention. It is not intended to identify key or essential parts of the invention, nor is it intended to limit the scope of the invention. Its purpose is merely to present certain concepts in a simplified form as a prelude to the more detailed description that follows.

[0007] In view of this, in order to solve the problem of low accuracy of traditional deep-sea underwater topography imaging and inversion methods in the prior art, this invention provides a novel GEO SAR deep-sea underwater topography imaging and inversion method.

[0008] The technical solution is as follows: A novel GEO SAR deep-sea underwater topography imaging and inversion method, comprising the following steps:

[0009] S1. Deep-sea underwater dynamics model is constructed by using a layered fully nonlinear hydrodynamic equation set;

[0010] S2. Acquire and preprocess the GEO SAR sea surface echo signal to obtain the processed GEO SAR sea surface echo signal;

[0011] S3. The processed GEO SAR sea surface echo signal is further processed and focused to obtain a focused complex image of the sea surface wave;

[0012] S4. For the focused complex image of sea surface waves, extract the multidimensional parameters of sea surface waves based on the deep-sea underwater dynamics model to obtain the sea surface elevation and radial velocity contribution of the current mode;

[0013] S5. Based on the sea surface elevation and radial velocity contribution of the current mode, perform two-dimensional surface flow field inversion, and obtain the corrected two-dimensional flow field according to the observation equation and the set two-dimensional flow field;

[0014] S6. Construct a multi-frame sea surface state time series input and network structure based on the parameters of step S5, train the network structure, and output the deep-sea underwater topography inversion through the obtained trained deep-sea underwater topography inversion network model.

[0015] Furthermore, in S1, in the target deep-sea area, a deep-sea underwater dynamics model is established by using a layered fully nonlinear hydrodynamic equation set to model the interaction between sea surface waves and underwater topography. The specific settings of the layered fully nonlinear hydrodynamic equation set are as follows:

[0016] Horizontal velocity vector Represented as:

[0017]

[0018] in, express Horizontal velocity component in the direction, express Horizontal velocity component in the direction, Indicates matrix transpose;

[0019] The vertical velocity component is expressed as ;

[0020] Seabed topography Represented as:

[0021]

[0022] in, Indicates underwater topographic relief;

[0023] Under the Boussinesq approximation, the three-dimensional layered flow field satisfies the continuity equation as follows:

[0024]

[0025] in, This represents the gradient operator with respect to the horizontal coordinate;

[0026] Under the Boussinesq approximation, the momentum equation with nonlinear convection terms is expressed as:

[0027]

[0028] in, For reference density, For pressure, The resultant force includes gravity, Coriolis force, and viscous term;

[0029] Free sea surface To satisfy the kinematic conditions, a coupled equation for sea surface elevation change and surface current velocity was constructed.

[0030] The coupling equation between sea surface elevation variation and surface current velocity is expressed as:

[0031]

[0032] in, and These are the horizontal and vertical velocities at sea level, respectively. This represents the eastward velocity component of the sea surface. This represents the northward velocity component of the sea surface;

[0033] By solving the above-mentioned stratified fully nonlinear hydrodynamic equations, under a given underwater topography... Under density stratification background conditions, the sea surface height field evolving over time was obtained. and the two-dimensional flow field on the sea surface ;

[0034] Two-dimensional flow field on sea surface Represented as:

[0035]

[0036] in, This represents the eastward velocity component of the sea surface. This represents the northward velocity component of the sea surface.

[0037] Furthermore, in S2, the GEO SAR satellite orbit parameters are set according to the approximately circular nadir ground trajectory, and small perturbations are applied to the orbital elements such as semi-major axis, inclination, eccentricity, and perigee argument to form a hybrid baseline satellite constellation that combines inter-orbital and in-orbital baselines. Based on the spatial coverage and temporal resolution requirements of the target deep-sea area, multiple frames of radar echo data covering the same sea area are acquired. For each frame of observation, the corresponding satellite-ground geometric parameters, incident angle, baseline length, and observation time information are recorded. The original echo signal is compressed in the range direction to obtain range-compressed echo data organized according to fast and slow time, which is the processed GEO SAR sea surface echo signal.

[0038] Furthermore, step S3 specifically includes the following steps:

[0039] S31. Based on the parameters in step S1, construct a nonlinear, space-variable sea surface wave echo signal model:

[0040] In S31, the free sea surface is discretized into regular grid cells, a nonlinear air-varying sea surface wave echo signal model is established, the influence of the relative velocity and acceleration of the target resolution cell in the radial and along-track directions on the echo phase is explicitly characterized, a distance history expression including radial velocity and radial acceleration terms is constructed, and the additional terms of sea surface motion on the echo signal phase are determined.

[0041] S32. Use dual-tree complex wavelet transform (DTCWT) to decompose the additional terms of the echo phase caused by the nonlinear spatial motion of the sea surface in the processed GEO SAR sea surface echo signal.

[0042] In S32, for the residuals in the horizontal and vertical directions, i.e., x and y, a finite number of third-order dual-tree complex wavelet transform low-pass scaling functions and wavelet basis functions in the dominant direction are selected, which only vary along a single spatial coordinate, and a small number of wavelet coefficients are used to describe the slow bending and unidirectional oscillating structure; for the residuals in the cross direction, a simplified two-dimensional dual-tree complex wavelet transform is used, and only the coefficients of the third dominant direction are retained to obtain the signal decomposition result;

[0043] S32. Based on the signal decomposition results, construct the phase compensation kernel function related to the signal decomposition, and use an optimization strategy to achieve imaging. That is, optimize the parameters in the phase compensation kernel function of the spatial correlation processing of dual-tree complex wavelet transform using a genetic algorithm to correct anisotropic and vortex-like nonlinear spatially variable motion, and improve the spatial correlation of sea surface scattering under long synthetic aperture conditions.

[0044] S34. Generate a focused image based on the parameters corrected and improved in step S32, that is, perform coherent integration on the range-compressed echo data in the slow time dimension to obtain a focused complex image of sea surface waves.

[0045] Furthermore, step S4 specifically includes the following steps:

[0046] S41. Construct the hybrid basis phase observation vector and the hybrid basis matrix;

[0047] In S41, for each ground resolution unit Its position is obtained by continuous imaging. The unwrapped hybrid baseline phase at time t is denoted as the hybrid baseline phase observation vector. ;

[0048] Mixed basis phase observation vector Represented as:

[0049]

[0050] in, Let be the observation vector at time 1. For a moment The observation vector;

[0051] Based on linear wave theory, a height basis function corresponding to the frequency and phase of sea surface waves is constructed. and velocity basis functions And combined with phase sensitivity coefficient , At every moment Form row vectors The result is a set of row vectors. , , ;

[0052] row vector Represented as:

[0053]

[0054] row vector set Represented as:

[0055]

[0056] Stacking the row vectors at all time points sequentially yields the mixing basis matrix of the resolving unit. ;

[0057] Mixed basis matrix of the resolving unit Represented as:

[0058]

[0059] S42. The mixed basis coefficient vector is estimated using the regularized least squares method. ;

[0060] In S42, based on the hybrid basis phase observation vector The hybrid basis matrix of the resolving unit Establish a linear observation model;

[0061] The linear observation model is expressed as:

[0062]

[0063] in, The vector of coefficients to be estimated;

[0064] S43. Aligning the mixing basis coefficient vector with modes and physical quantities Grouping is performed to obtain the mixing basis coefficients for each mode;

[0065] In S43, the mixed basis coefficient vector Divided into Each modal sub-block has four coefficients, two for height and two for velocity.

[0066] Mixed basis coefficient vector The grouping is represented as:

[0067]

[0068] in, For the first High correlation coefficient of each mode For the first Radial velocity correlation coefficient for each mode;

[0069] S44. Reconstruct the time series of height and radial velocity for each mode to obtain the sea surface elevation and radial velocity contribution of the current mode;

[0070] In S44, for each mode At every moment By linearly combining the height basis function and the velocity basis function, the sea surface elevation and radial velocity contribution of the current mode are obtained;

[0071] The contributions of sea surface elevation and radial velocity to the current mode are expressed as follows:

[0072]

[0073] in, For modality radial velocity field, Representing modes The sea surface height field.

[0074] Furthermore, step S5 specifically includes the following steps:

[0075] S51. By constructing a two-dimensional Kalman filter framework, establishing the observation equations, and defining the two-dimensional flow field;

[0076] In S51, the two-dimensional Kalman filter framework refers to the line-of-sight radial velocity field. As a constraint for high-resolution observations, external ocean current field products are introduced as low-resolution background current fields. The preset radar line-of-sight unit vector is Establish observation equations ;

[0077] Observation equations Represented as:

[0078]

[0079] Treating radial velocity as a two-dimensional flow field Projection observation along the line of sight;

[0080] S52. Perform two-dimensional Kalman filtering and multipath fusion on the two-dimensional flow field to obtain the corrected two-dimensional flow field;

[0081] In step S52, a filter strategy that follows the streamline is adopted to filter the background flow field. For the prediction term, use the observation equation For the observation term, a filter path is constructed along the local flow direction. Based on the observation equation, the two-dimensional flow field is... A progressive correction is performed, meaning that for multiple velocity estimates given by multiple streamlines for the same grid node, a covariance cross-fusion method is used to synthesize them to obtain a statistically consistent two-dimensional surface total flow field estimate, and the output is a two-dimensional flow field corrected by high-resolution radial velocity constraints. .

[0082] Furthermore, step S6 specifically includes the following steps:

[0083] S61. Construct a multi-frame sea surface state time series;

[0084] In S61, the sea surface height field is obtained through continuous observation by a geosynchronous orbit synthetic aperture radar. and two-dimensional flow field Based on this, a time-series sequence of sea surface states is constructed. In each time frame, sea surface height, east-west current velocity, and north-south current velocity are combined into multi-channel raster data. Forming a time tensor ;

[0085] Multichannel raster data Represented as:

[0086]

[0087] in, Indicates the height of the sea surface;

[0088] S62. Construct a network structure that includes an encoder, a fusion module, a temporal modeling module, and a decoder connected in sequence;

[0089] In S62, the encoder uses convolutional blocks to extract multi-scale features of the sea surface state for each time frame.

[0090] The fusion module performs feature fusion through a gated attention mechanism;

[0091] The temporal modeling module uses a Transformer encoder to perform multi-head attention modeling on the fused feature sequences of different time frames, capturing the propagation delay and interference effect of deep-sea topography-induced internal waves in time.

[0092] The decoder maps the temporally aggregated global features into a two-dimensional depth raster and outputs an underwater topography estimate of the target sea area. ;

[0093] S63. Train the network structure to obtain the trained deep-sea underwater terrain inversion network model, and output the deep-sea underwater terrain inversion;

[0094] In S63, a loss function based on the mean square error of water depth is used to train the network structure end-to-end. Multiple frames of sea surface state time series are input into the trained deep-sea underwater topography inversion network model, and the corresponding deep-sea underwater topography distribution is output, realizing high-precision inversion of deep-sea underwater topography by geostationary orbit synthetic aperture radar.

[0095] The beneficial effects of this invention are as follows: Addressing the characteristics of indistinct texture features and weak contrast of sea surface waves under geosynchronous orbit spatial correlation SAR background, this invention innovatively designs a spatial correlation evaluation factor by combining the unique wave spectrum and image intensity of sea surface waves. Based on this spatial correlation evaluation factor, the quality of sea surface wave imaging by the geosynchronous orbit spatial correlation SAR algorithm is accurately evaluated. Compared with existing evaluation factors, the spatial correlation evaluation factor in this invention can significantly improve the accuracy of sea surface wave imaging quality evaluation by geosynchronous orbit spatial correlation SAR. Attached Figure Description

[0096] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this invention, illustrate exemplary embodiments of the invention and are used to explain the invention, but do not constitute an undue limitation of the invention. In the drawings:

[0097] Figure 1 This is a flowchart illustrating a novel GEO SAR deep-sea underwater topography imaging and inversion method.

[0098] Figure 2 This is a schematic diagram illustrating an embodiment of a novel GEO SAR deep-sea underwater topography imaging and inversion method.

[0099] Figure 3 shows the simulated backscattering distribution of sea surface waves and the imaging results using the traditional method and the present invention. In Figure 3(a), the simulated backscattering distribution of sea surface waves is shown; in Figure 3(b), the imaging result using the traditional method is shown; and in Figure 3(c), the imaging result using the present invention is shown.

[0100] Figure 4 shows the underwater topography inversion results and a comparison diagram of the inversion results of the present invention with those of traditional methods 1 and 2. In Figure 4(a), the actual underwater topography is shown; in Figure 4(b), the inversion results of the present invention are shown; in Figure 4(c), the inversion results of traditional method 1 are shown; and in Figure 4(d), the inversion results of traditional method 2 are shown. Detailed Implementation

[0101] To make the technical solutions and advantages of the embodiments of the present invention clearer, the exemplary embodiments of the present invention will be further 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 an exhaustive list of all embodiments. It should be noted that, unless otherwise specified, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0102] refer to Figure 1 Figure 4 details this embodiment, illustrating a novel GEO SAR deep-sea underwater topography imaging and inversion method, which includes the following steps:

[0103] S1. Deep-sea underwater dynamics model is constructed by using a layered fully nonlinear hydrodynamic equation set;

[0104] S2. Acquire and preprocess the GEO SAR sea surface echo signal to obtain the processed GEO SAR sea surface echo signal;

[0105] S3. The processed GEO SAR sea surface echo signal is further processed and focused to obtain a focused complex image of the sea surface wave;

[0106] S4. For the focused complex image of sea surface waves, extract the multidimensional parameters of sea surface waves based on the deep-sea underwater dynamics model to obtain the sea surface elevation and radial velocity contribution of the current mode;

[0107] S5. Based on the sea surface elevation and radial velocity contribution of the current mode, perform two-dimensional surface flow field inversion, and obtain the corrected two-dimensional flow field according to the observation equation and the set two-dimensional flow field;

[0108] S6. Construct a multi-frame sea surface state time series input and network structure based on the parameters of step S5, train the network structure, and output the deep-sea underwater topography inversion through the obtained trained deep-sea underwater topography inversion network model.

[0109] Furthermore, in S1, in the target deep-sea area, a deep-sea underwater dynamic model is established by using a layered fully nonlinear hydrodynamic equation set to model the interaction between sea surface waves and underwater topography. This model describes the propagation process of deep-sea underwater topographic waves in the stratified water body and their joint modulation of sea surface height and surface current field. The specific settings of the layered fully nonlinear hydrodynamic equation set are as follows:

[0110] Horizontal velocity vector Represented as:

[0111]

[0112] in, express Horizontal velocity component in the direction, express Horizontal velocity component in the direction, Indicates matrix transpose;

[0113] The vertical velocity component is expressed as ;

[0114] Seabed topography Represented as:

[0115]

[0116] in, Indicates underwater topographic relief;

[0117] Under the Boussinesq approximation, the three-dimensional layered flow field satisfies the continuity equation as follows:

[0118]

[0119] in, This represents the gradient operator with respect to the horizontal coordinate;

[0120] Under the Boussinesq approximation, the momentum equation with nonlinear convection terms is expressed as:

[0121]

[0122] in, For reference density, For pressure, The resultant force includes gravity, Coriolis force, and viscous term;

[0123] Free sea surface To satisfy the kinematic conditions, a coupled equation for sea surface elevation change and surface current velocity was constructed.

[0124] The coupling equation between sea surface elevation variation and surface current velocity is expressed as:

[0125]

[0126] in, and These are the horizontal and vertical velocities at sea level, respectively. This represents the eastward velocity component of the sea surface. This represents the northward velocity component of the sea surface;

[0127] By solving the above-mentioned stratified fully nonlinear hydrodynamic equations, under a given underwater topography... Under density stratification background conditions, the sea surface height field evolving over time was obtained. and the two-dimensional flow field on the sea surface ;

[0128] Two-dimensional flow field on sea surface Represented as:

[0129]

[0130] in, This represents the eastward velocity component of the sea surface. This represents the northward velocity component of the sea surface;

[0131] Step S1 quantitatively characterizes the modulation relationship between "underwater topography – stratification propagation – sea surface elevation and current field", providing physical priors for subsequent GEO SAR echo signal acquisition, imaging and inversion.

[0132] Furthermore, in S2, the GEO SAR satellite orbit parameters are set according to the approximately circular nadir ground trajectory, and small perturbations are applied to orbital elements such as semi-major axis, inclination, eccentricity, and perigee argument to form a hybrid baseline satellite constellation that combines inter-orbital and in-orbital baselines. Based on the spatial coverage and temporal resolution requirements of the target deep-sea area, multiple frames of radar echo data covering the same sea area are acquired. For each frame of observation, the corresponding satellite-ground geometric parameters, incident angle, baseline length, and observation time are recorded. The original echo signal is range-compressed to obtain range-compressed echo data organized according to fast and slow time, i.e., the processed GEO SAR sea surface echo signal.

[0133] Furthermore, step S3 specifically includes the following steps:

[0134] S31. Based on the parameters in step S1, construct a nonlinear, space-variable sea surface wave echo signal model:

[0135] In S31, the free sea surface is discretized into regular grid cells, a nonlinear air-varying sea surface wave echo signal model is established, the influence of the relative velocity and acceleration of the target resolution cell in the radial and along-orbit directions on the echo phase is explicitly characterized, a distance history expression including radial velocity and radial acceleration terms is constructed, and the additional terms of sea surface motion on the echo signal phase are determined.

[0136] S32. Use dual-tree complex wavelet transform (DTCWT) to decompose the additional terms of the echo phase caused by the nonlinear spatial motion of the sea surface in the processed GEO SAR sea surface echo signal.

[0137] In S32, for the residuals in the horizontal and vertical directions (x and y), a finite number of third-order dual-tree complex wavelet transform low-pass scaling functions and wavelet basis functions in the dominant direction are selected, which vary only along a single spatial coordinate, and a small number of wavelet coefficients are used to describe the slowly bending and unidirectional oscillating structures; for the residuals in the cross direction, a simplified two-dimensional dual-tree complex wavelet transform is used, retaining only the coefficients of the third dominant direction, which are used to compactly characterize two-dimensional complex structures such as vortex-like structures, to obtain the signal decomposition results;

[0138] S32. Based on the signal decomposition results, construct the phase compensation kernel function related to the signal decomposition, and use an optimization strategy to achieve imaging. That is, optimize the parameters in the phase compensation kernel function of the spatial correlation processing of dual-tree complex wavelet transform using a genetic algorithm to correct anisotropic and vortex-like nonlinear spatially variable motion, and improve the spatial correlation of sea surface scattering under long synthetic aperture conditions.

[0139] S34. Generate a focused image based on the parameters corrected and improved in step S32, that is, perform coherent integration on the range-compressed echo data in the slow time dimension to obtain a focused complex image of the sea surface wave, providing high-definition imaging results for subsequent interferometry and parameter inversion.

[0140] Furthermore, step S4 specifically includes the following steps:

[0141] S41. Construct the hybrid basis phase observation vector and the hybrid basis matrix;

[0142] In S41, for each ground resolution unit Its position is obtained by continuous imaging. The unwrapped hybrid baseline phase at time t is denoted as the hybrid baseline phase observation vector. ;

[0143] Mixed basis phase observation vector Represented as:

[0144]

[0145] in, Let be the observation vector at time 1. For a moment The observation vector;

[0146] Based on linear wave theory, a height basis function corresponding to the frequency and phase of sea surface waves is constructed. and velocity basis functions And combined with phase sensitivity coefficient , At every moment Form row vectors The result is a set of row vectors. , , ;

[0147] row vector Represented as:

[0148]

[0149] row vector set Represented as:

[0150]

[0151] Stacking the row vectors at all time points sequentially yields the mixing basis matrix of the resolving unit. ;

[0152] Mixed basis matrix of the resolving unit Represented as:

[0153]

[0154] S42. The mixed basis coefficient vector is estimated using the regularized least squares method. ;

[0155] In S42, based on the hybrid basis phase observation vector The hybrid basis matrix of the resolving unit Establish a linear observation model;

[0156] The linear observation model is expressed as:

[0157]

[0158] in, The vector of coefficients to be estimated;

[0159] S43. Aligning the mixing basis coefficient vector with modes and physical quantities Grouping is performed to obtain the mixing basis coefficients for each mode;

[0160] In S43, the mixed basis coefficient vector Divided into Each modal sub-block has four coefficients, two for height and two for velocity.

[0161] Mixed basis coefficient vector The grouping is represented as:

[0162]

[0163] in, For the first High correlation coefficient of each mode For the first Radial velocity correlation coefficient for each mode;

[0164] S44. Reconstruct the time series of height and radial velocity for each mode to obtain the sea surface elevation and radial velocity contribution of the current mode;

[0165] In S44, for each mode At every moment By linearly combining the height basis function and the velocity basis function, the sea surface elevation and radial velocity contribution of the current mode are obtained;

[0166] The contributions of sea surface elevation and radial velocity to the current mode are expressed as follows:

[0167]

[0168] in, For modality radial velocity field, Representing modes The sea surface height field.

[0169] Furthermore, step S5 specifically includes the following steps:

[0170] S51. By constructing a two-dimensional Kalman filter framework, establishing the observation equations, and defining the two-dimensional flow field;

[0171] In S51, the two-dimensional Kalman filter framework refers to the line-of-sight radial velocity field. As a constraint for high-resolution observations, external ocean current field products (such as numerical weather prediction or reanalysis data) are introduced as a low-resolution background current field. The preset radar line-of-sight unit vector is Establish observation equations ;

[0172] Observation equations Represented as:

[0173]

[0174] Treating radial velocity as a two-dimensional flow field Projection observation along the line of sight;

[0175] S52. Perform two-dimensional Kalman filtering and multipath fusion on the two-dimensional flow field to obtain the corrected two-dimensional flow field;

[0176] In step S52, a filter strategy that follows the streamline is adopted to filter the background flow field. For the prediction term, use the observation equation For the observation term, a filter path is constructed along the local flow direction. Based on the observation equation, the two-dimensional flow field is... A progressive correction is performed, meaning that for multiple velocity estimates given by multiple streamlines for the same grid node, a covariance cross-fusion method is used to synthesize them to obtain a statistically consistent two-dimensional surface total flow field estimate, and the output is a two-dimensional flow field corrected by high-resolution radial velocity constraints. This refers to the sea surface current vector field that simultaneously contains east-west and north-south components, providing dynamic input features for the StratiFormer deep-sea underwater topography inversion network.

[0177] Furthermore, step S6 specifically includes the following steps:

[0178] S61. Construct multi-frame sea surface state timing input;

[0179] In S61, the sea surface height field is obtained through continuous observation by a geosynchronous orbit synthetic aperture radar. and two-dimensional flow field Based on this, a time-series sequence of sea surface states is constructed. In each time frame, sea surface height, east-west current velocity, and north-south current velocity are combined into multi-channel raster data. Forming a temporal input tensor ;

[0180] Multichannel raster data Represented as:

[0181]

[0182] in, Indicates the height of the sea surface;

[0183] S62. Construct a network structure that includes an encoder, a fusion module, a temporal modeling module, and a decoder connected in sequence;

[0184] In S62, the encoder uses convolutional blocks to extract multi-scale features of the sea surface state for each time frame.

[0185] The fusion module performs feature fusion through a gated attention mechanism;

[0186] The temporal modeling module uses a Transformer encoder to perform multi-head attention modeling on the fused feature sequences of different time frames, capturing the propagation delay and interference effect of deep-sea topography-induced internal waves in time.

[0187] The decoder maps the temporally aggregated global features into a two-dimensional depth raster and outputs an underwater topography estimate of the target sea area. ;

[0188] S63. Train the network structure to obtain the trained deep-sea underwater terrain inversion network model, and output the deep-sea underwater terrain inversion;

[0189] In S63, a loss function based on the mean square error of water depth is used to train the network structure end-to-end. Multiple frames of sea surface state time series are input into the trained deep-sea underwater topography inversion network model, and the corresponding deep-sea underwater topography distribution is output, realizing high-precision inversion of deep-sea underwater topography by geostationary orbit synthetic aperture radar.

[0190] Specifically, in this embodiment, in step S1, deep-sea underwater topographic parameters are set and sea surface state is simulated. Based on the established hydrodynamic equations, kinematic boundary conditions (5) are applied at the free sea surface. The sea surface height field and two-dimensional flow field evolving with time are obtained by using the finite difference method. In step S2, GEO SAR orbit and imaging parameters are configured and echo signals are generated. The orbit parameters of the two GEO SAR satellites are configured so that their ground trajectories are approximately circular. By applying small perturbations to the semi-major axis, inclination, eccentricity, and perigee argument, a hybrid baseline constellation with both inter-orbit and in-orbit baselines is formed. The corresponding orbit and radar parameters are listed in Table 1. Backscattering of the sea surface at various times is generated based on the sea surface height and two-dimensional flow field according to Bragg scattering and tilt modulation. The initial backscattering of the sea surface is shown in Figure 3(a). In step S3, focusing imaging is performed based on sea surface waves. The motion of the sea surface waves is decomposed using DTCWT, and a phase compensation kernel function for spatial correlation processing is constructed to improve the efficiency of the imaging. The spatial correlation between different slow-time echoes was analyzed by optimizing the free parameters in the compensation kernel using a genetic algorithm to maximize imaging contrast. The optimized compensation kernel was then used to perform phase correction on the range-compressed data and accumulate over slow time to generate a series of focused sea surface wave complex images under long synthetic aperture conditions, as shown in Figure 3(c). In step S4, the hybrid baseline interferometric phase construction and sea surface height / radial velocity time-series inversion were performed. Based on the obtained focused complex images, interferometric processing was performed on each time frame to obtain the corresponding hybrid baseline interferogram. The interferometric phase of each pixel was filtered and smoothed to unwrap the image, and the pixel was constructed in the... The mixed baseline phase observation vectors at each time point are used to construct a mixed basis matrix. Based on the linear observation model, a regularized least squares problem is solved for each resolution cell to obtain coefficient vector estimates. The coefficients are grouped according to mode and physical quantity (height, velocity) to obtain the height correlation coefficient and radial velocity correlation coefficient corresponding to each mode. Finally, pixel-level sea surface height time series and radial velocity time series are obtained. In step S5, the two-dimensional surface flow field is inverted using radial velocity constraints. To obtain a complete two-dimensional surface velocity field, a low-resolution background flow field is constructed with an error of 0.2 m / s. For each grid point, the observation equation is constructed using the reconstructed radial velocity: the unknown two-dimensional velocity is projected... The radar line of sight is drawn to obtain the observation relationship. Based on this, according to the constructed two-dimensional Kalman filter framework: using the background flow field as a priori prediction, several filter paths advancing with the streamlines are constructed in the local flow direction; the reconstructed radial velocity is used as a high-resolution observation to progressively correct the prediction field; multiple velocity estimates obtained from multiple streamlines at the same grid point are integrated using methods such as covariance cross-fusion to obtain a statistically consistent two-dimensional flow field estimate; in step S6, a deep learning network is constructed and deep-sea underwater topography is inverted. Using the obtained sea surface height and two-dimensional flow field, the sea surface state of each time frame is organized into three-channel raster data and stacked in chronological order to form a length of [missing information]. For the input sequence, a set of samples is constructed to train the deep learning network: for multiple sets of known underwater topography and their underwater environment, steps S1-S5 are repeated to generate sea surface state sequences, forming a paired dataset of "sea surface state sequence - real water depth map". The loss function with water depth mean square error as the main factor is selected to train the network end-to-end until it converges on the validation set. The sea surface height and current field sequences obtained by decoupling GEO SAR imaging and hybrid baseline interferometry are input into the trained network model, which can output the deep sea underwater topography distribution of the target sea area, completing the integrated processing of GEO SAR deep sea underwater topography imaging and inversion. The deep sea underwater topography inversion result is shown in Figure 3.

[0191] Referring to Figure 4, the root mean square error (RMSE) was used for evaluation during the comparative experiment. The unit is m. According to the parameters of the comparative experiment set in Table 1, the RMSE value of the present invention is 26.04, the RMSE value of the traditional method 1 (CNN-Transformer) is 57.82, and the RMSE value of the traditional method 2 (CNN-GRU) is 28.72.

[0192] Table 1

[0193] Radar parameters numerical values track semi-major axis 42164km Eccentricity 0.1 Perimeter Argument 90° track inclination 10° Angle of incidence 25° carrier frequency 5.3GHz FM bandwidth 5MHz Pulse repetition frequency 200Hz

[0194] Although the invention has been described with reference to a limited number of embodiments, those skilled in the art will understand from the foregoing description that other embodiments are conceivable within the scope of the invention described herein. Furthermore, it should be noted that the language used in this specification has been chosen primarily for readability and instructional purposes, and not for the purpose of interpreting or limiting the subject matter of the invention. Therefore, many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the appended claims. The disclosure of the invention is illustrative and not restrictive, and the scope of the invention is defined by the appended claims.

Claims

1. A novel GEO SAR deep-sea underwater topography imaging and inversion method, characterized in that, Includes the following steps: S1. Deep-sea underwater dynamics model is constructed by using a layered fully nonlinear hydrodynamic equation set; S2. Acquire and preprocess the GEO SAR sea surface echo signal to obtain the processed GEO SAR sea surface echo signal; S3. The processed GEO SAR sea surface echo signal is further processed and focused to obtain a focused complex image of the sea surface wave; S4. For the focused complex image of sea surface waves, extract the multidimensional parameters of sea surface waves based on the deep-sea underwater dynamics model to obtain the sea surface elevation and radial velocity contribution of the current mode; S5. Based on the sea surface elevation and radial velocity contribution of the current mode, a two-dimensional surface flow field inversion is performed. According to the observation equation and the set two-dimensional flow field, the corrected two-dimensional flow field is obtained. S6. Construct a multi-frame sea surface state time series input and network structure based on the parameters of step S5, train the network structure, and output the deep-sea underwater topography inversion through the obtained trained deep-sea underwater topography inversion network model.

2. The novel GEO SAR deep-sea underwater topography imaging and inversion method according to claim 1, characterized in that, In step S1, in the target deep-sea area, a deep-sea underwater dynamics model is established by using a layered fully nonlinear hydrodynamic equation set to model the interaction between sea surface waves and underwater topography. The specific settings of the layered fully nonlinear hydrodynamic equation set are as follows: Horizontal velocity vector Represented as: in, express Horizontal velocity component in the direction, express Horizontal velocity component in the direction, Indicates matrix transpose; The vertical velocity component is expressed as ; Seabed topography Represented as: in, Indicates underwater topographic relief; Under the Boussinesq approximation, the three-dimensional layered flow field satisfies the continuity equation as follows: in, This represents the gradient operator with respect to the horizontal coordinate; Under the Boussinesq approximation, the momentum equation with nonlinear convection terms is expressed as: in, For reference density, For pressure, The resultant force includes gravity, Coriolis force, and viscous term; Free sea surface To satisfy the kinematic conditions, a coupled equation for sea surface elevation change and surface current velocity was constructed. The coupling equation between sea surface elevation variation and surface current velocity is expressed as: in, and These are the horizontal and vertical velocities at sea level, respectively. This represents the eastward velocity component of the sea surface. This represents the northward velocity component of the sea surface; By solving the above-mentioned stratified fully nonlinear hydrodynamic equations, under a given underwater topography... Under density stratification background conditions, the sea surface height field evolving over time was obtained. and the two-dimensional flow field on the sea surface ; Two-dimensional flow field on sea surface Represented as: in, This represents the eastward velocity component of the sea surface. This represents the northward velocity component of the sea surface.

3. The novel GEO SAR deep-sea underwater topography imaging and inversion method according to claim 2, characterized in that, In step S2, GEO SAR satellite orbit parameters are set according to an approximately circular nadir ground trajectory. By applying small perturbations to orbital elements such as semi-major axis, inclination, eccentricity, and perigee argument, a hybrid baseline satellite constellation with both inter-orbit and in-orbit baselines is formed. Based on the spatial coverage and temporal resolution requirements of the target deep-sea area, multiple frames of radar echo data covering the same sea area are acquired. For each frame of observation, the corresponding satellite-ground geometric parameters, incident angle, baseline length, and observation time information are recorded. The original echo signal is range-compressed to obtain range-compressed echo data organized according to fast and slow time, i.e., the processed GEO SAR sea surface echo signal.

4. A novel GEO SAR deep-sea underwater topography imaging and inversion method according to claim 3, characterized in that, S3 specifically includes the following steps: S31. Based on the parameters in step S1, construct a nonlinear, space-variable sea surface wave echo signal model: In S31, the free sea surface is discretized into regular grid cells, a nonlinear air-varying sea surface wave echo signal model is established, the influence of the relative velocity and acceleration of the target resolution cell in the radial and along-track directions on the echo phase is explicitly characterized, a distance history expression including radial velocity and radial acceleration terms is constructed, and the additional terms of sea surface motion on the echo signal phase are determined. S32. Use dual-tree complex wavelet transform (DTCWT) to decompose the additional terms of the echo phase caused by the nonlinear spatial motion of the sea surface in the processed GEO SAR sea surface echo signal. In S32, for the residuals in the horizontal and vertical directions, i.e., x and y, a finite number of third-order dual-tree complex wavelet transform low-pass scaling functions and wavelet basis functions in the dominant direction are selected, which only vary along a single spatial coordinate, and a small number of wavelet coefficients are used to describe the slow bending and unidirectional oscillating structure; for the residuals in the cross direction, a simplified two-dimensional dual-tree complex wavelet transform is used, and only the coefficients of the third dominant direction are retained to obtain the signal decomposition result; S32. Based on the signal decomposition results, construct the phase compensation kernel function related to the signal decomposition, and use an optimization strategy to achieve imaging. That is, optimize the parameters in the phase compensation kernel function of the spatial correlation processing of dual-tree complex wavelet transform using a genetic algorithm to correct anisotropic and vortex-like nonlinear spatially variable motion, and improve the spatial correlation of sea surface scattering under long synthetic aperture conditions. S34. Generate a focused image based on the parameters corrected and improved in step S32, that is, perform coherent integration on the range-compressed echo data in the slow time dimension to obtain a focused complex image of sea surface waves.

5. A novel GEO SAR deep-sea underwater topography imaging and inversion method according to claim 4, characterized in that, S4 specifically includes the following steps: S41. Construct the hybrid basis phase observation vector and the hybrid basis matrix; In S41, for each ground resolution unit Its position is obtained by continuous imaging. The unwrapped hybrid baseline phase at time t is denoted as the hybrid baseline phase observation vector. ; Mixed basis phase observation vector Represented as: in, Let be the observation vector at time 1. For a moment The observation vector; Based on linear wave theory, a height basis function corresponding to the frequency and phase of sea surface waves is constructed. and velocity basis functions And combined with phase sensitivity coefficient , At every moment Form row vectors The result is a set of row vectors. , , ; row vector Represented as: row vector set Represented as: Stacking the row vectors at all time points sequentially yields the mixing basis matrix of the resolving unit. ; Mixed basis matrix of the resolving unit Represented as: S42. The mixed basis coefficient vector is estimated using the regularized least squares method. ; In S42, based on the hybrid basis phase observation vector The hybrid basis matrix of the resolving unit Establish a linear observation model; The linear observation model is expressed as: in, The vector of coefficients to be estimated; S43. Aligning the mixing basis coefficient vector with modes and physical quantities Grouping is performed to obtain the mixing basis coefficients for each mode; In S43, the mixed basis coefficient vector Divided into Each modal sub-block has four coefficients, two for height and two for velocity. Mixed basis coefficient vector The grouping is represented as: in, For the first High correlation coefficient of each mode For the first Radial velocity correlation coefficient for each mode; S44. Reconstruct the time series of height and radial velocity for each mode to obtain the sea surface elevation and radial velocity contribution of the current mode; In S44, for each mode At every moment By linearly combining the height basis function and the velocity basis function, the sea surface elevation and radial velocity contribution of the current mode are obtained; The contributions of sea surface elevation and radial velocity to the current mode are expressed as follows: in, For modality radial velocity field, Representing modes The sea surface height field.

6. A novel GEO SAR deep-sea underwater topography imaging and inversion method according to claim 5, characterized in that, S5 specifically includes the following steps: S51. By constructing a two-dimensional Kalman filter framework, establishing the observation equations, and defining the two-dimensional flow field; In S51, the two-dimensional Kalman filter framework refers to the line-of-sight radial velocity field. As a constraint for high-resolution observations, external ocean current field products are introduced as low-resolution background current fields. The preset radar line-of-sight unit vector is Establish observation equations ; Observation equations Represented as: Treating radial velocity as a two-dimensional flow field Projection observation along the line of sight; S52. Perform two-dimensional Kalman filtering and multipath fusion on the two-dimensional flow field to obtain the corrected two-dimensional flow field; In step S52, a filter strategy that follows the streamline is adopted to filter the background flow field. For the prediction term, use the observation equation For the observation term, a filter path is constructed along the local flow direction. Based on the observation equation, the two-dimensional flow field is... A progressive correction is performed, meaning that for multiple velocity estimates given by multiple streamlines for the same grid node, a covariance cross-fusion method is used to synthesize them to obtain a statistically consistent two-dimensional surface total flow field estimate, and the output is a two-dimensional flow field corrected by high-resolution radial velocity constraints. .

7. A novel GEO SAR deep-sea underwater topography imaging and inversion method according to claim 6, characterized in that, S6 specifically includes the following steps: S61. Construct a multi-frame sea surface state time series; In S61, the sea surface height field is obtained through continuous observation by a geosynchronous orbit synthetic aperture radar. and two-dimensional flow field Based on this, a time-series sequence of sea surface states is constructed. In each time frame, sea surface height, east-west current velocity, and north-south current velocity are combined into multi-channel raster data. Forming a time tensor ; Multichannel raster data Represented as: in, Indicates the height of the sea surface; S62. Construct a network structure that includes an encoder, a fusion module, a temporal modeling module, and a decoder connected in sequence; In S62, the encoder uses convolutional blocks to extract multi-scale features of the sea surface state for each time frame. The fusion module performs feature fusion through a gated attention mechanism; The temporal modeling module uses a Transformer encoder to perform multi-head attention modeling on the fused feature sequences of different time frames, capturing the propagation delay and interference effect of deep-sea topography-induced internal waves in time. The decoder maps the temporally aggregated global features into a two-dimensional depth raster and outputs an underwater topography estimate of the target sea area. ; S63. Train the network structure to obtain the trained deep-sea underwater terrain inversion network model, and output the deep-sea underwater terrain inversion; In S63, a loss function based on the mean square error of water depth is used to train the network structure end-to-end. Multiple frames of sea surface state time series are input into the trained deep-sea underwater topography inversion network model, and the corresponding deep-sea underwater topography distribution is output, realizing high-precision inversion of deep-sea underwater topography by geostationary orbit synthetic aperture radar.

Images