Multi-source and multi-physics field gravity data super-resolution reconstruction method

Abstract

The invention relates to the technical field of ocean geophysical exploration and surveying and mapping, and provides a multi-source and multi-physics field gravity data super-resolution reconstruction method. The method comprises the following steps: acquiring preprocessed low-resolution satellite altimetry gravity data and preprocessed high-resolution ship load data, and constructing a training data set; constructing a dual regression deep learning model; training the dual regression deep learning model by using the training data set to obtain a trained dual regression deep learning model; and inputting preprocessed low-resolution satellite gravity data of a to-be-reconstructed target area into the trained dual regression deep learning model to obtain super-resolution reconstructed gravity data of the reconstructed target area. Through dual network closed-loop constraint and adaptive degradation modeling, dependence on high-resolution shipborne load data is significantly reduced, and reconstruction precision and generalization ability are improved at the same time.

Description

Technical Field

[0001] The present application relates to the field of marine geophysical exploration and mapping technology, and in particular to a method for super-resolution reconstruction of multi-source, multi-physical field gravity data. Background Art

[0002] High-resolution, high-precision ocean gravity field data are of great significance in marine resource exploration, seabed topography mapping, and marine science research. Currently, ocean gravity data mainly relies on two observation methods: one is satellite altimetry technology, which has global coverage and is suitable for data acquisition over large sea areas; the other is shipborne gravity measurement, which can provide data with high spatial resolution and is suitable for fine measurements in local areas. However, the spatial resolution of satellite altimetry gravity data is limited, making it difficult to meet the needs of high-precision research on local complex structures or small-scale features; and although shipborne measurements have high resolution, they are limited by high costs, long measurement cycles, and complex geographical environments, and their spatial coverage is usually sparse and discontinuous. Therefore, how to integrate these two complementary data sources, taking into account both high resolution and large-scale coverage, and construct a high-precision, high-resolution ocean gravity field is one of the key challenges in current marine geophysical research.

[0003] Traditional data fusion methods often face challenges when processing multi-source gravity data, including complex modeling, high computational overhead, and difficulty effectively handling boundary effects and complex terrain interference. These issues lead to low fusion accuracy and difficulty generalizing to large ocean areas. Furthermore, these methods often rely on extensive prior knowledge and manual parameter tuning in practical applications, making them difficult to adapt to the demands of large-scale, data-driven research. In recent years, with the rapid development of deep learning technology, image super-resolution reconstruction methods have demonstrated powerful data reconstruction capabilities in numerous fields, providing a new technical path for high-resolution and high-precision reconstruction of ocean gravity fields. However, ocean gravity data exhibits spatial distribution patterns and physical properties that differ significantly from those of natural images, making it difficult to directly apply traditional super-resolution network models to achieve ideal results. On the one hand, most existing methods ignore the physical consistency between geophysical laws and data, which can lead to unrealistic results. On the other hand, training deep models typically relies on large-scale, high-resolution annotated samples. High-resolution shipborne gravity data is expensive to acquire and limited in quantity, further complicating model training. Furthermore, low-resolution satellite altimetry gravity maps generated from low-resolution satellite altimetry gravity data not only have large grid spacing, but also have inaccurate pixel values ​​due to the influence of the satellite acquisition method. Therefore, converting low-resolution to high-resolution gravity data requires not only predicting the pixel values ​​corresponding to missing grids, but also correcting the pixel values ​​of existing grids in the low-resolution image. Accordingly, the process of converting high-resolution to low-resolution gravity data is not a simple downsampling process; it also requires considering the effects of noise and filtering contained in the low-resolution image.

[0004] Therefore, how to fully explore the potential information of a large amount of unlabeled satellite gravity data under the conditions of limited high-resolution shipborne data and construct a multi-source gravity data fusion and reconstruction framework with semi-supervisory capabilities that can simultaneously capture physical consistency and statistical characteristics is a core issue that needs to be solved urgently. Summary of the Invention

[0005] In response to the problems existing in the prior art, this application proposes a multi-source, multi-physics field gravity data super-resolution reconstruction method to solve the problems existing in the prior art.

[0006] This application provides a multi-source, multi-physics field ocean gravity data super-resolution reconstruction method, including: Step S101: Acquire gravity data, where the gravity data includes low-resolution satellite altimetry gravity data and high-resolution shipborne gravity data; Step S102: Preprocessing the low-resolution satellite altimetry gravity data and the high-resolution shipborne gravity data to obtain preprocessed low-resolution satellite altimetry gravity data and preprocessed high-resolution shipborne gravity data, constructing a paired sample dataset and an unpaired sample dataset based on the preprocessed low-resolution satellite altimetry gravity data and the preprocessed high-resolution shipborne gravity data; the paired sample dataset and the unpaired sample dataset constitute a training dataset; Step S103: constructing a dual regression deep learning model, wherein the dual regression deep learning model includes: a main regression network P and a dual regression network D; Step S104: using the training data set to train the dual regression deep learning model to obtain a trained dual regression deep learning model; Step S105: inputting the pre-processed low-resolution satellite gravity data of the target area to be reconstructed into the main regression network P in the trained dual regression deep learning model to obtain super-resolution reconstructed gravity data of the target area to be reconstructed.

[0007] Furthermore, the paired sample data set is composed of pre-processed low-resolution satellite altimetry gravity data. and pre-processed high-resolution shipborne gravity data The unpaired sample dataset is composed of preprocessed low-resolution satellite altimetry gravity data. composition; Low-resolution satellite altimetry gravity data in the training dataset Includes preprocessed low-resolution satellite altimetry gravity data in paired sample datasets Preprocessed low-resolution satellite altimetry gravity data in the unpaired sample dataset .

[0008] Furthermore, the preprocessing includes data cleaning, quality control, and outlier detection and elimination.

[0009] Furthermore, the quality control includes performing a first-stage quality control and a second-stage quality control on the high-resolution shipborne gravity data.

[0010] Furthermore, the first stage quality control includes: Calculation of high-resolution shipborne gravity data in The local mean μ and standard deviation σ within the spatial window will satisfy the following formula Culling: , in, is the gravity value of the current data point, μ is the high-resolution shipborne gravity data at The local mean value in the spatial window, σ is the local mean value of high-resolution shipborne gravity data in The standard deviation within the spatial window.

[0011] Furthermore, the second stage quality control includes: Obtain the standard Earth gravity model EGM2008, which satisfies the following formula Culling: , in, is the gravity value of the current data point, is the gravity value of the current data point of the standard earth gravity model EGM2008, is the standard deviation of the EGM2008 model.

[0012] Furthermore, the step S104 includes: Step S1041: The pre-processed low-resolution satellite altimetry gravity data in the training data set is As input data, input the main regression network P to obtain the corresponding super-resolution reconstructed gravity data , the expression is: , Super-resolution reconstruction of gravity data Includes the first super-resolution reconstruction of gravity data and the second super-resolution reconstruction of gravity data , the first super-resolution reconstruction of gravity data for: , in, The pre-processed low-resolution satellite altimetry gravity data in the paired sample dataset; The second super-resolution reconstruction of gravity data for: , in, The pre-processed low-resolution satellite altimetry gravity data in the unpaired sample dataset; Step S1042: reconstructing the gravity data with super resolution , input the dual regression network D to obtain the reconstructed low-resolution gravity data after low-resolution mapping , the expression is: , in, The pre-processed low-resolution satellite altimetry gravity data in the training dataset; Reconstructing low-resolution gravity data Includes the first reconstruction of low-resolution gravity data and the second reconstruction of low-resolution gravity data , the first reconstruction of low-resolution gravity data for: , in, Reconstructing gravity data for the first super-resolution; The pre-processed low-resolution satellite altimetry gravity data in the paired sample dataset; The second reconstruction of low-resolution gravity data for: , in, Reconstruct gravity data for the second super-resolution; The pre-processed low-resolution satellite altimetry gravity data in the unpaired sample dataset; Step S1043: Determine the pre-processed low-resolution satellite altimetry gravity data in the input training data set Whether it belongs to the paired sample data set, if yes, execute steps S1044 to S1047; if not, execute steps S1045 to S1047; Step S1044: Step S1044: The pre-processed high-resolution shipborne gravity data Input the dual regression network D to obtain the shipborne low-resolution gravity data after low-resolution mapping , the expression is: ; Step S1045: constructing an objective function to calculate the errors between all output data and their corresponding label data in the primary regression network P and the dual regression network D; Step S1046: Determine whether the objective function meets the preset conditions. If not, execute step S1047. If so, complete the training of the dual regression deep learning model to obtain the trained dual regression deep learning model. Step S1047: Set the target error Backpropagate to the main regression network P and the dual regression network D, update the learnable parameters of the dual regression deep learning model, and continue to execute steps S1041 to S1046.

[0013] Furthermore, the objective function is: , in, is the target error, is a binary indicator function. If the pre-processed low-resolution satellite altimetry gravity data in the training dataset in step S1043 is Belong to the paired sample data set, then If the low-resolution satellite altimetry gravity data pre-processed in step S1043 It does not belong to the paired sample data set, but to the unpaired sample data set, then ; is the cycle consistency error, which is used to describe the super-resolution reconstruction of gravity data by the dual regression network D Reconstructed low-resolution gravity data obtained by degraded reconstruction The preprocessed low-resolution satellite altimetry gravity data in the training dataset The degree of deviation between pixel and structural consistency; is the supervised reconstruction error, which is used to describe the preprocessed low-resolution satellite altimetry gravity data in the paired sample dataset by the main regression network P. The first super-resolution reconstructed gravity data obtained by super-resolution reconstruction Paired with the preprocessed high-resolution shipborne gravity data in the sample dataset Data and structural consistency errors between; is the dual regression error, which is used to describe the high-resolution shipborne gravity data in the paired sample dataset through the dual regression network D. Shipborne low-resolution gravity data obtained by degraded reconstruction The preprocessed low-resolution satellite altimetry data in the paired sample dataset The degree of error between is the dual consistency error, which is used to describe the first super-resolution reconstruction of gravity data by the dual regression network D The first reconstructed low-resolution gravity data obtained by degenerate reconstruction The high-resolution shipborne gravity data in the paired sample dataset is obtained by the dual regression network D. Shipborne low-resolution gravity data obtained by degraded reconstruction The scalars λ, β, and γ are learnable parameters that control the relative contributions of cycle consistency error, supervised reconstruction error, dual regression error, and dual consistency error.

[0014] Furthermore, the cycle consistency error for: , in, To reconstruct low-resolution gravity data; The pre-processed low-resolution satellite altimetry gravity data in the training dataset; represents pixel L1 loss, SSIM represents structural similarity index metric, is a learnable parameter; The supervised reconstruction error for: , in, For the first super-resolution reconstruction of gravity data, is the preprocessed high-resolution shipborne gravity data in the paired sample dataset; represents pixel L1 loss, SSIM represents structural similarity index metric, is a learnable parameter; The dual regression error for: , in, It is shipborne low-resolution gravity data; It is the pre-processed low-resolution satellite altimetry data in the paired sample dataset; represents pixel L1 loss, SSIM represents structural similarity index metric, is a learnable parameter; The dual consistency error for: , in, Reconstruct low-resolution gravity data for the first time; It is shipborne low-resolution gravity data; represents pixel L1 loss, SSIM represents structural similarity index metric, is a learnable parameter.

[0015] Based on the above invention content, compared with the existing technology, this application has achieved the following technical effects: This application proposes a multi-source, multi-physics field gravity data super-resolution reconstruction method. This method is based on a semi-supervised dual regression deep learning model. It constructs a closed-loop deep learning architecture consisting of a main regression network P (which realizes the mapping of low-resolution gravity field to high-resolution gravity field) and a dual regression network D (which realizes the mapping of high-resolution gravity field to low-resolution gravity field). It jointly utilizes paired sample data sets (including paired pre-processed low-resolution satellite altimetry gravity data and its corresponding pre-processed high-resolution shipborne gravity data) and unpaired sample data sets (including only pre-processed low-resolution satellite altimetry gravity data). The closed-loop architecture reconstructs gravity data by calculating the first super-resolution Preprocessed high-resolution shipborne gravity data in the paired sample dataset The supervised reconstruction error between the low-resolution shipborne gravity data after low-resolution mapping and preprocessed low-resolution satellite altimetry data in the paired sample dataset The dual regression error between the low-resolution gravity data and the first reconstruction after low-resolution mapping and shipborne low-resolution gravity data after low-resolution mapping The dual consistency error constrains the learning of paired sample data and reconstructs low-resolution gravity data after calculating the low-resolution mapping. and the preprocessed low-resolution satellite altimetry gravity data in the training dataset The cyclic consistency error between them is used to constrain the paired sample data and the unpaired sample data respectively; the low-resolution satellite data covering a large area of ​​sea area is fully utilized through the above-mentioned semi-supervised dual regression learning mechanism.

[0016] This application simulates the degradation process from high-resolution gravity data to low-resolution gravity data through a dual regression network D, and adaptively models key factors in the degradation process (such as noise and filtering effects); this mechanism significantly enhances the model's ability to capture regional degradation characteristics and improves the reliability of the model's knowledge transfer between labeled and unlabeled regions, thereby greatly improving the generalization ability of the reconstruction results for data from different regions.

[0017] This application is based on a semi-supervised learning mechanism and adaptive degradation modeling. By inputting low-resolution satellite gravity data of the target area into a trained main regression network P, high-resolution reconstructed gravity data can be obtained. This method can fully tap the potential information of a large amount of unlabeled satellite gravity data under the conditions of limited high-resolution shipborne data, breaking through the bottleneck of the traditional method between accuracy and efficiency, and also providing important support for improving the ability to model the ocean gravity field. In addition, the method of this application effectively improves the reconstruction accuracy and robustness while significantly reducing the dependence on shipborne gravity data, providing high-precision, high-resolution gravity field data support for marine resource exploration and seabed topography mapping. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0019] Figure 1 This is a flow chart of a multi-source, multi-physics field gravity data super-resolution reconstruction method provided in an embodiment of the present application; Figure 2 This is a schematic diagram of the framework for super-resolution reconstruction of gravity data based on the dual regression deep learning model provided in an embodiment of the present application. DETAILED DESCRIPTION

[0020] In order to better understand the technical solution of the present invention, the embodiments of the present application are described in detail below with reference to the accompanying drawings. It should be clear that the embodiments described are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0021] The terms used in the embodiments of the present application are for the purpose of describing specific embodiments only and are not intended to limit the present application. The singular forms "a", "an", "the" and "the" used in the embodiments of the present application and the appended claims are also intended to include plural forms unless the context clearly indicates otherwise.

[0022] It should be understood that the term "and / or" as used herein simply describes a relationship between associated objects, indicating that three possible relationships exist. For example, "A and / or B" can represent: A alone, A and B together, or B alone. Furthermore, the character " / " in this document generally indicates an "or" relationship between the associated objects.

[0023] See also Figure 1 , is a flow chart of a multi-source, multi-physics field gravity data super-resolution reconstruction method provided in an embodiment of the present application. Figure 1 As shown, it mainly includes the following steps.

[0024] Step S101: Acquire gravity data, which includes low-resolution satellite altimetry gravity data and high-resolution shipborne gravity data.

[0025] Step S102: Preprocess the low-resolution satellite altimetry gravity data and the high-resolution shipborne gravity data to obtain the preprocessed low-resolution satellite altimetry gravity data and the preprocessed high-resolution shipborne gravity data, and construct a paired sample data set and an unpaired sample data set based on the preprocessed low-resolution satellite altimetry gravity data and the preprocessed high-resolution shipborne gravity data.

[0026] The paired sample dataset is composed of preprocessed low-resolution satellite altimetry gravity data. and pre-processed high-resolution shipborne gravity data The unpaired sample dataset is composed of preprocessed low-resolution satellite altimetry gravity data. composition.

[0027] The paired sample data set and the unpaired sample data set constitute a training data set. The low-resolution satellite altimetry gravity data in the training data set Includes preprocessed low-resolution satellite altimetry gravity data in paired sample datasets and preprocessed low-resolution satellite altimetry gravity data in the unpaired sample dataset .

[0028] The preprocessing includes data cleaning, quality control, and outlier detection and elimination.

[0029] The quality control comprises performing a first-stage quality control and a second-stage quality control on the high-resolution shipborne gravity data, wherein the first-stage quality control comprises: Calculation of high-resolution shipborne gravity data in The local mean μ and standard deviation σ within the spatial window will satisfy the following formula Culling: , in, is the gravity value of the current data point, μ is the high-resolution shipborne gravity data at The local mean value in the spatial window, σ is the local mean value of high-resolution shipborne gravity data in The standard deviation within the spatial window.

[0030] The second stage of quality control includes: Obtain the standard Earth gravity model EGM2008, which satisfies the following formula Culling: , in, is the gravity value of the current data point, is the gravity value of the current data point of the standard earth gravity model EGM2008, is the standard deviation of the EGM2008 model.

[0031] The preprocessed shipborne gravity data are resampled to a uniform grid using bilinear interpolation to achieve spatial alignment with the preprocessed low-resolution satellite altimetry gravity data, and the paired sample dataset and the unpaired sample dataset are generated using the spatially aligned preprocessed low-resolution satellite altimetry gravity data and the preprocessed high-resolution shipborne gravity data.

[0032] Step S103: Construct a dual regression deep learning model, which includes: a main regression network P and a dual regression network D; the main regression network P and the dual regression network D can be any deep neural network type, including but not limited to the following classic types of networks: convolutional neural network, recurrent neural network, fully connected neural network, Transformer and Mamba.

[0033] The main regression network P is responsible for super-resolution gravity data reconstruction, realizing the mapping between the pre-processed low-resolution satellite altimetry gravity data and the super-resolution reconstructed gravity data, which is recorded as the first mapping and is expressed as: , in, is the low-resolution satellite altimetry gravity data preprocessed in the training dataset. Reconstructing gravity data for super-resolution.

[0034] Super-resolution reconstruction of gravity data Includes the first super-resolution reconstruction of gravity data and the second super-resolution reconstruction of gravity data , the first super-resolution reconstruction of gravity data for: , in, The pre-processed low-resolution satellite altimetry gravity data in the paired sample dataset; The second super-resolution reconstruction of gravity data for: , in, The preprocessed low-resolution satellite altimetry gravity data in the unpaired sample dataset.

[0035] The dual regression network D is responsible for simulating the degradation process, achieving the mapping between super-resolution reconstructed gravity data and reconstructed low-resolution gravity data, as well as the mapping between preprocessed high-resolution shipborne gravity data and shipborne low-resolution gravity data. The mapping between super-resolution reconstructed gravity data and reconstructed low-resolution gravity data is denoted as the second mapping, and the mapping between preprocessed high-resolution shipborne gravity data and shipborne low-resolution gravity data is denoted as the third mapping.

[0036] The expression of the second mapping is: , in, To reconstruct low-resolution gravity data, Reconstructing gravity data for super-resolution.

[0037] Reconstructing low-resolution gravity data Includes the first reconstruction of low-resolution gravity data and the second reconstruction of low-resolution gravity data , the first reconstruction of low-resolution gravity data for: , in, Reconstructing gravity data for the first super-resolution; The second reconstruction of low-resolution gravity data for: , in, Reconstruct gravity data for the second super-resolution.

[0038] The expression of the third mapping is: , in, For shipborne low-resolution gravity data, is the preprocessed high-resolution shipborne gravity data.

[0039] The first mapping is a mapping from low-resolution gravity data LRQ to high-resolution gravity data HRQ, and the second and third mappings are a mapping from high-resolution gravity data HRQ to low-resolution gravity data LRQ. The low-resolution gravity data LRQ includes pre-processed low-resolution satellite altimetry gravity data, reconstructed low-resolution gravity data, and shipborne low-resolution gravity data; the high-resolution gravity data HRQ includes super-resolution reconstructed gravity data and pre-processed high-resolution shipborne gravity data.

[0040] If the training data is a paired sample data set, the dual regression deep learning model can realize the training and learning of the dual regression deep learning model by synchronously constraining the three mapping relationships of the first mapping, the second mapping and the third mapping.

[0041] If the training data is an unpaired sample data set, the dual regression deep learning model realizes the training and learning of the dual regression deep learning model by constraining the two mapping relationships of the first mapping and the second mapping.

[0042] The main regression network P and the dual regression network D form a cycle-consistent mapping relationship from low-resolution gravity data LRQ to high-resolution gravity data HRQ, and then to low-resolution gravity data LRQ (LRQ → HRQ → LRQ). The formed closed-loop structure enables the dual regression deep learning model to fully utilize paired sample datasets and a large number of unpaired sample datasets, significantly reducing the dependence on the high-cost high-resolution shipborne gravity data. By introducing the adaptive degradation modeling of the dual regression network as additional supervision information, the model's ability to capture regional deviations is further enhanced, promoting the model's knowledge transfer between labeled areas (paired sample areas) and unlabeled areas (unpaired sample areas), significantly reducing the dependence on large quantities of shipborne gravity data, and thus enhancing the accuracy, robustness, and generalization ability of the reconstruction results.

[0043] Step S104: Use the training data set to train the dual regression deep learning model to obtain a trained dual regression deep learning model.

[0044] In one embodiment, the preprocessed low-resolution satellite altimetry gravity data in the training data set is input into a primary regression network P, and the preprocessed high-resolution shipborne gravity data in the training data and the super-resolution reconstructed gravity data generated by the primary regression network P are input into a dual regression network D. The primary regression network P and the dual regression network D are trained simultaneously, specifically including: Step S1041: The pre-processed low-resolution satellite altimetry gravity data in the training data set is As input data, input the main regression network P to obtain the corresponding super-resolution reconstructed gravity data , the expression is: , Super-resolution reconstruction of gravity data Includes the first super-resolution reconstruction of gravity data and the second super-resolution reconstruction of gravity data , the first super-resolution reconstruction of gravity data for: , in, The pre-processed low-resolution satellite altimetry gravity data in the paired sample dataset; The second super-resolution reconstruction of gravity data for: , in, The preprocessed low-resolution satellite altimetry gravity data in the unpaired sample dataset.

[0045] Step S1042: reconstructing the gravity data with super resolution , input the dual regression network D to obtain the reconstructed low-resolution gravity data after low-resolution mapping , the expression is: , in, The pre-processed low-resolution satellite altimetry gravity data in the training dataset; Reconstructing low-resolution gravity data Includes the first reconstruction of low-resolution gravity data and the second reconstruction of low-resolution gravity data , the first reconstruction of low-resolution gravity data for: , in, Reconstructing gravity data for the first super-resolution; The pre-processed low-resolution satellite altimetry gravity data in the paired sample dataset; The second reconstruction of low-resolution gravity data for: , in, Reconstruct gravity data for the second super-resolution; The preprocessed low-resolution satellite altimetry gravity data in the unpaired sample dataset.

[0046] Step S1043: Determine the pre-processed low-resolution satellite altimetry gravity data in the input training data set Whether it belongs to the paired sample data set, if yes, execute steps S1044 to S1047; if not, execute steps S1045 to S1047; Step S1044: The pre-processed high-resolution shipborne gravity data Input the dual regression network D to obtain the shipborne low-resolution gravity data after low-resolution mapping , the expression is: , Step S1045: constructing an objective function to calculate the errors between all output data and their corresponding label data in the primary regression network P and the dual regression network D; Taking into account the errors in pixel and structural consistency of the reconstructed gravity data, an error function that integrates local detail fidelity and global perceptual similarity is defined. The expression is as follows: , in, represents pixel L1 loss, SSIM represents structural similarity index metric, is a learnable parameter that dynamically balances the contribution of pixel L1 loss and structural similarity index metric.

[0047] It should be understood that the above error function is only one possibility provided by the embodiment of the present application, and any error function or function combination under the dual regression learning framework proposed in the present application is within the scope of protection of the present application.

[0048] The error function As the basic error metric function, it is used to calculate the cycle consistency error , supervised reconstruction error , dual regression error and dual consistency error .

[0049] The objective function for training the dual regression learning model is: , in, is the target error, is a binary indicator function. If the pre-processed low-resolution satellite altimetry gravity data in the training dataset in step S1043 is Belong to the paired sample data set, then If the low-resolution satellite altimetry gravity data pre-processed in step S1043 It does not belong to the paired sample data set, but to the unpaired sample data set, then ; is the cycle consistency error, which is used to describe the super-resolution reconstruction of gravity data by the dual regression network D Reconstructed low-resolution gravity data obtained by degraded reconstruction The preprocessed low-resolution satellite altimetry gravity data in the training dataset The degree of deviation between pixel and structural consistency; is the supervised reconstruction error, which is used to describe the preprocessed low-resolution satellite altimetry gravity data in the paired sample dataset by the main regression network P. The first super-resolution reconstructed gravity data obtained by super-resolution reconstruction Paired with the preprocessed high-resolution shipborne gravity data in the sample dataset Data and structural consistency errors between; is the dual regression error, which is used to describe the high-resolution shipborne gravity data in the paired sample dataset through the dual regression network D. Shipborne low-resolution gravity data obtained by degraded reconstruction The preprocessed low-resolution satellite altimetry data in the paired sample dataset The degree of error between is the dual consistency error, which is used to describe the first super-resolution reconstruction of gravity data by the dual regression network D The first reconstructed low-resolution gravity data obtained by degenerate reconstruction The high-resolution shipborne gravity data in the paired sample dataset is obtained by the dual regression network D. Shipborne low-resolution gravity data obtained by degraded reconstruction The scalars λ, β, and γ are learnable parameters that control the relative contributions of cycle consistency error, supervised reconstruction error, dual regression error, and dual consistency error.

[0050] The cycle consistency error for: , in, To reconstruct low-resolution gravity data; The pre-processed low-resolution satellite altimetry gravity data in the training dataset; represents pixel L1 loss, SSIM represents structural similarity index metric, is a learnable parameter.

[0051] The supervised reconstruction error for: , in, For the first super-resolution reconstruction of gravity data, is the preprocessed high-resolution shipborne gravity data in the paired sample dataset; represents pixel L1 loss, SSIM represents structural similarity index metric, is a learnable parameter.

[0052] The dual regression error for: , in, It is shipborne low-resolution gravity data; It is the pre-processed low-resolution satellite altimetry data in the paired sample dataset; represents pixel L1 loss, SSIM represents structural similarity index metric, is a learnable parameter.

[0053] The dual consistency error for: , in, Reconstruct low-resolution gravity data for the first time; It is shipborne low-resolution gravity data; represents pixel L1 loss, SSIM represents structural similarity index metric, is a learnable parameter.

[0054] Step S1046: Determine whether the objective function meets the preset conditions. If not, execute step S1047. If so, complete the training of the dual regression deep learning model to obtain the trained dual regression deep learning model. The preset condition includes one of the following criteria: the number of iterations reaches a specified upper limit of the number of iterations; the target error is lower than a specified error limit; the decrease rate of the target error is lower than a set decrease rate threshold; the running time reaches a predetermined time upper limit.

[0055] Step S1047: Set the target error Backpropagate to the main regression network P and the dual regression network D, update the learnable parameters of the dual regression deep learning model, and continue to execute steps S1041 to S1046.

[0056] Figure 2This is a schematic diagram of the framework for super-resolution reconstruction of gravity data based on a dual regression deep learning model, as provided in an embodiment of the present application. The primary regression network P is responsible for mapping the low-resolution gravity field to the high-resolution gravity field (LRQ → HRQ), i.e., super-resolution reconstruction of gravity data; the dual regression network D is responsible for mapping the high-resolution gravity field to the low-resolution gravity field (HRQ → LRQ), i.e., degradation process simulation. By collaboratively constraining the cycle consistency loss, reconstruction loss, dual regression, and dual consistency, the primary regression network P and the dual regression network D are collaboratively optimized. At the same time, the adaptive degradation process simulation implemented by the dual regression network D can better capture the regional biases and resolution loss patterns introduced during satellite data processing, ensuring that the model can effectively learn the laws of data fusion even with only a small amount of labeled data.

[0057] Step S105: inputting the pre-processed low-resolution satellite gravity data of the target area to be reconstructed into the main regression network P in the trained dual regression deep learning model to obtain super-resolution reconstructed gravity data of the target area to be reconstructed.

[0058] This application proposes a multi-source, multi-physics field gravity data super-resolution reconstruction method. This method is based on a semi-supervised dual regression deep learning model. It constructs a closed-loop deep learning architecture consisting of a main regression network P (which realizes the mapping of low-resolution gravity field to high-resolution gravity field) and a dual regression network D (which realizes the mapping of high-resolution gravity field to low-resolution gravity field). It jointly utilizes paired sample data sets (including paired pre-processed low-resolution satellite altimetry gravity data and its corresponding pre-processed high-resolution shipborne gravity data) and unpaired sample data sets (including only pre-processed low-resolution satellite altimetry gravity data). The closed-loop architecture reconstructs gravity data by calculating the first super-resolution Preprocessed high-resolution shipborne gravity data in the paired sample dataset The supervised reconstruction error between the low-resolution shipborne gravity data after low-resolution mapping and preprocessed low-resolution satellite altimetry data in the paired sample dataset The dual regression error between the low-resolution gravity data and the first reconstruction after low-resolution mapping and shipborne low-resolution gravity data after low-resolution mapping The dual consistency error constrains the learning of paired sample data and reconstructs low-resolution gravity data after calculating the low-resolution mapping. and the preprocessed low-resolution satellite altimetry gravity data in the training dataset The cyclic consistency error between them is used to constrain the paired sample data and the unpaired sample data respectively; the low-resolution satellite data covering a large area of ​​sea area is fully utilized through the above-mentioned semi-supervised dual regression learning mechanism.

[0059] The example of this application uses the proposed dual regression network learning mechanism, based on the joint constraints of cycle consistency and dual consistency error, to ensure that the model can still effectively learn the rules of data fusion when there is only a small amount of labeled data. It organically integrates wide-area, massive low-resolution satellite altimetry data with local, sparse, high-resolution shipborne gravity observation data, significantly reducing the dependence on a large number of labeled data sets, improving the efficiency and accuracy of data fusion, and realizing the reconstruction of high-resolution ocean gravity field.

[0060] This application simulates the degradation process from high-resolution gravity data to low-resolution gravity data through a dual regression network D, and adaptively models key factors in the degradation process (such as noise and filtering effects); this mechanism significantly enhances the model's ability to capture regional degradation characteristics and improves the reliability of the model's knowledge transfer between labeled and unlabeled regions, thereby greatly improving the generalization ability of the reconstruction results for data from different regions.

[0061] This application is based on a semi-supervised learning mechanism and adaptive degradation modeling. By inputting low-resolution satellite gravity data of the target area into a trained primary regression network P, high-resolution reconstructed gravity data can be obtained. This method can fully exploit the potential information of a large amount of unlabeled satellite gravity data under the conditions of limited high-resolution shipborne data, breaking through the bottleneck of the traditional method between accuracy and efficiency, and also providing important support for improving the ability to model the ocean gravity field. In addition, the method of this application effectively improves the reconstruction accuracy and robustness while significantly reducing the dependence on shipborne gravity data, providing high-resolution gravity field data support for marine geophysical exploration and seabed topography mapping.

[0062] In the embodiment of the present invention, "at least one" refers to one or more, and "more" refers to two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can represent the existence of A alone, the existence of A and B at the same time, and the existence of B alone. Among them, A and B can be singular or plural. The character " / " generally indicates that the previous and subsequent associated objects are in an "or" relationship. "At least one of the following" and similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one of a, b and c can be represented by: a, b, c, ab, ac, bc or abc, where a, b, c can be single or multiple.

[0063] The above description is only a specific embodiment of the present invention. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, and all of them should be included in the protection scope of the present invention.

Claims

1. A multi-source, multi-physics field gravity data super-resolution reconstruction method, characterized in that: include: Step S101: Acquire gravity data, where the gravity data includes low-resolution satellite altimetry gravity data and high-resolution shipborne gravity data; Step S102: preprocessing the low-resolution satellite altimetry gravity data and the high-resolution shipborne gravity data to obtain the preprocessed low-resolution satellite altimetry gravity data and the preprocessed high-resolution shipborne gravity data, and constructing a paired sample data set and an unpaired sample data set based on the preprocessed low-resolution satellite altimetry gravity data and the preprocessed high-resolution shipborne gravity data; The paired sample data set and the unpaired sample data set constitute a training data set; Step S103: constructing a dual regression deep learning model, wherein the dual regression deep learning model includes: a main regression network P and a dual regression network D; Step S104: using the training data set to train the dual regression deep learning model to obtain a trained dual regression deep learning model; Step S105: inputting the pre-processed low-resolution satellite gravity data of the target area to be reconstructed into the main regression network P in the trained dual regression deep learning model to obtain super-resolution reconstructed gravity data of the target area to be reconstructed.

2. The method according to claim 1, characterized in that The paired sample dataset is composed of preprocessed low-resolution satellite altimetry gravity data. and pre-processed high-resolution shipborne gravity data The unpaired sample dataset is composed of preprocessed low-resolution satellite altimetry gravity data. composition; Low-resolution satellite altimetry gravity data in the training dataset Includes preprocessed low-resolution satellite altimetry gravity data in paired sample datasets Preprocessed low-resolution satellite altimetry gravity data in the unpaired sample dataset .

3. The method according to claim 1, characterized in that The preprocessing includes data cleaning, quality control, and outlier detection and elimination.

4. The method according to claim 3, characterized in that The quality control includes performing a first-stage quality control and a second-stage quality control on the high-resolution shipborne gravity data.

5. The method according to claim 4, characterized in that The first stage of quality control includes: Calculation of high-resolution shipborne gravity data in The local mean μ and standard deviation σ within the spatial window will satisfy the following formula Culling: ， in, is the gravity value of the current data point, μ is the high-resolution shipborne gravity data at The local mean value in the spatial window, σ is the local mean value of high-resolution shipborne gravity data in The standard deviation within the spatial window.

6. The method according to claim 4, characterized in that The second stage of quality control includes: Obtain the standard Earth gravity model EGM2008, which satisfies the following formula Culling: ， in, is the gravity value of the current data point, is the gravity value of the current data point of the standard earth gravity model EGM2008, is the standard deviation of the EGM2008 model.

7. The method according to claim 1, characterized in that The step S104 includes: Step S1041: The pre-processed low-resolution satellite altimetry gravity data in the training data set is As input data, input the main regression network P to obtain the corresponding super-resolution reconstructed gravity data , the expression is: ， Super-resolution reconstruction of gravity data Includes the first super-resolution reconstruction of gravity data and the second super-resolution reconstruction of gravity data , the first super-resolution reconstruction of gravity data for: ， in, The pre-processed low-resolution satellite altimetry gravity data in the paired sample dataset; The second super-resolution reconstruction of gravity data for: ， in, The pre-processed low-resolution satellite altimetry gravity data in the unpaired sample dataset; Step S1042: reconstructing the gravity data with super resolution , input the dual regression network D to obtain the reconstructed low-resolution gravity data after low-resolution mapping , the expression is: ， in, The pre-processed low-resolution satellite altimetry gravity data in the training dataset; Reconstructing low-resolution gravity data Includes the first reconstruction of low-resolution gravity data and the second reconstruction of low-resolution gravity data , the first reconstruction of low-resolution gravity data for: ， in, Reconstructing gravity data for the first super-resolution; The pre-processed low-resolution satellite altimetry gravity data in the paired sample dataset; The second reconstruction of low-resolution gravity data for: ， in, Reconstruct gravity data for the second super-resolution; The pre-processed low-resolution satellite altimetry gravity data in the unpaired sample dataset; Step S1043: Determine the pre-processed low-resolution satellite altimetry gravity data in the input training data set Whether it belongs to the paired sample data set, if yes, execute steps S1044 to S1047; if not, execute steps S1045 to S1047; Step S1044: Step S1044: The pre-processed high-resolution shipborne gravity data Input the dual regression network D to obtain the shipborne low-resolution gravity data after low-resolution mapping , the expression is: ； Step S1045: constructing an objective function to calculate the errors between all output data and their corresponding label data in the primary regression network P and the dual regression network D; Step S1046: Determine whether the objective function meets the preset conditions. If not, execute step S1047. If so, complete the training of the dual regression deep learning model to obtain the trained dual regression deep learning model. Step S1047: Set the target error Backpropagate to the main regression network P and the dual regression network D, update the learnable parameters of the dual regression deep learning model, and continue to execute steps S1041 to S1046.

8. The method according to claim 7, characterized in that The objective function is: ， in, is the target error, is a binary indicator function. If the pre-processed low-resolution satellite altimetry gravity data in the training dataset in step S1043 is Belong to the paired sample data set, then If the low-resolution satellite altimetry gravity data pre-processed in step S1043 It does not belong to the paired sample data set, but to the unpaired sample data set, then ; is the cycle consistency error, which is used to describe the super-resolution reconstruction of gravity data by the dual regression network D Reconstructed low-resolution gravity data obtained by degraded reconstruction The preprocessed low-resolution satellite altimetry gravity data in the training dataset The degree of deviation between pixel and structural consistency; is the supervised reconstruction error, which is used to describe the preprocessed low-resolution satellite altimetry gravity data in the paired sample dataset by the main regression network P. The first super-resolution reconstructed gravity data obtained by super-resolution reconstruction Paired with the preprocessed high-resolution shipborne gravity data in the sample dataset Data and structural consistency errors between; is the dual regression error, which is used to describe the high-resolution shipborne gravity data in the paired sample dataset through the dual regression network D. Shipborne low-resolution gravity data obtained by degraded reconstruction The preprocessed low-resolution satellite altimetry data in the paired sample dataset The degree of error between is the dual consistency error, which is used to describe the first super-resolution reconstruction of gravity data by the dual regression network D The first reconstructed low-resolution gravity data obtained by degenerate reconstruction The high-resolution shipborne gravity data in the paired sample dataset is obtained by the dual regression network D. Shipborne low-resolution gravity data obtained by degraded reconstruction The scalars λ, β, and γ are learnable parameters that control the relative contributions of cycle consistency error, supervised reconstruction error, dual regression error, and dual consistency error.

9. The method according to claim 8, characterized in that The cycle consistency error for: ， in, To reconstruct low-resolution gravity data; The pre-processed low-resolution satellite altimetry gravity data in the training dataset; represents pixel L1 loss, SSIM represents structural similarity index metric, is a learnable parameter; The supervised reconstruction error for: ， in, For the first super-resolution reconstruction of gravity data, is the preprocessed high-resolution shipborne gravity data in the paired sample dataset; represents pixel L1 loss, SSIM represents structural similarity index metric, is a learnable parameter; The dual regression error for: ， in, It is shipborne low-resolution gravity data; It is the pre-processed low-resolution satellite altimetry data in the paired sample dataset; represents pixel L1 loss, SSIM represents structural similarity index metric, is a learnable parameter; The dual consistency error for: ， in, Reconstruct low-resolution gravity data for the first time; It is shipborne low-resolution gravity data; represents pixel L1 loss, SSIM represents structural similarity index metric, is a learnable parameter.

Images