Artificial intelligence inversion system and method for seafloor topography using gravity field spatial spectral data

Abstract

The application provides an artificial intelligence seabed topography inversion system and method using gravity field spatial spectrum data, the system comprising: a gravity field spatial spectrum calculation module: using several band-pass filters to intercept gravity data at different wavelengths. A gravity-water depth mapping relationship establishment module: used for processing gravity and water depth data and establishing their correlation. Obtain the gravity field spatial spectrum of the sparse ship survey water depth position, and pair it with the water depth data. Divide the data set into a training set and a test set in proportion. Design an artificial intelligence model suitable for regression tasks to train and predict water depth. An artificial intelligence seabed topography inversion module: uses the trained model to input the gravity field spatial spectrum of the test set, predicts the water depth value, and evaluates the prediction effect through the accuracy index. The advantage of the application is that the calculation of the gravity field spatial spectrum enriches the characteristics of the gravity data, and considers the different degrees of influence of gravity on topography inversion.

Description

Technical Field

[0001] This application belongs to the field of marine remote sensing and mapping, specifically relating to an artificial intelligence inversion system and method for seabed topography using gravity field spatial spectrum data. Background Technology

[0002] Submarine topographic mapping helps humanity gain a deeper understanding and develop marine resources. In scientific research, submarine topography is closely related to important studies such as ocean circulation, seabed geology, and tsunamis; in the field of economic development, submarine topographic mapping provides crucial support and assurance for the construction of marine facilities and navigation safety.

[0003] Satellite altimetry can rapidly and cost-effectively acquire high-precision gravity field data for global ocean areas. Gravity anomalies and seafloor topography exhibit a strong correlation within certain wavebands, making this method a fast and cost-effective way to obtain global seafloor topography. Currently, the main methods for retrieving seafloor topography include gravity geology and the SAS method. Dixon et al. (1983) proposed the admittance function method, considering flexural equilibrium theory based on Parker's formula. Smith and Sandwell (1994) proposed the S&S method, constructing a functional relationship between gravity and seafloor topography in relevant wavebands, and obtaining the scaling factor between gravity data and seafloor topography using known sparse points of water depth and gravity. Tozer et al. (2019) updated the overall workflow of the S&S method and established a 15-second global seafloor topography model, STRM15+V2.0. Abukatijiang (2019) adjusted the filtering and slope estimation of the S&S method for estimation in the Arctic region. Fan Diao (2020) used the Oldenburg perturbation method to invert fourth-order seafloor topography based on the Parker formula. To simplify the calculation process and improve the practical application value of the inversion procedure, these traditional methods consider relatively simple physical parameters, focusing only on the most important variables affecting the relationship between gravity and water depth, ignoring the influence of other details, thus limiting the accuracy of the inversion. Gravity geology methods require ship-based depth data points to calculate shortwave information; if there are insufficient ship-based depth data points, the shortwave information will be reduced. The SAS method, on the other hand, relies on ship-based depth data points to calculate the scale factor, and then uses the scale factor to generate a grid to obtain the spatial mapping relationship between gravity and water depth. Insufficient ship-based depth data points will also affect the accuracy of the scale factor. Therefore, both methods require a sufficient amount of ship-based depth data and cannot utilize abundant ship-based depth data from other regions to assist in the inversion.

[0004] With the development of artificial intelligence (AI) technology, AI has been widely applied in data processing algorithms in geosciences such as seismology, geophysics, and geochemistry, achieving significant results. Many scholars have attempted to apply AI technology to seafloor topography inversion based on altimeter-based ocean gravity data, with initial success. For example, Anna and Wan (2022) and Wan Xiaoyun et al. (2023) used neural networks with different structures to predict absolute water depth, such as inferring vertical deflection and gravity gradient deflection from gravity and its related quantities. Moran et al. (2022) tested various machine learning algorithms, hoping to predict water depth using a series of geophysical and oceanographic features. Yang Lei (2023) used shortwave gravity anomalies and other geophysical information as input, employing fully connected and convolutional networks for prediction, and evaluated the performance of different input and model combinations. Sun Yongjin (2023) improved the accuracy of water depth measurements based on altimeter data by combining neural networks and gravity wavelet decomposition, achieving a 12.45% improvement compared to using neural networks alone. The powerful computing capabilities of artificial intelligence can uncover the characteristic relationships between spatial data. Compared to traditional physical models, its models are more complex enough to account for perturbations beyond the main influencing variables. Artificial intelligence methods can establish a mapping between gravity and water depth in areas with dense ship-gauge depth data, and in areas lacking such data, they can be fine-tuned using a small amount of sparse ship-gauge depth data, thus improving the efficiency of ship-gauge depth data utilization.

[0005] The advantage of artificial intelligence (AI) methods over traditional methods lies in their ability to improve accuracy through model complexity and their applicability to areas lacking ship-based depth data. However, current AI methods for estimating seabed topography using ocean gravity fields face the following challenges: ① AI methods extract features between gravity and ocean depth through a large number of parameters to fit the relationship between them, but insufficient data limits the effectiveness of feature extraction. ② There are numerous types of gravity-related parameters, and different seabed physical environments are affected differently by these parameters. It is impossible to accurately assign corresponding weights to different parameters, thus impacting the accuracy of the generated seabed topography. Summary of the Invention

[0006] The purpose of this application is to overcome the shortcomings of the existing technology in predicting seabed topography, which has low accuracy.

[0007] To achieve the above objectives, this application proposes an artificial intelligence inversion system for seabed topography using gravity field spatial spectrum data, the system comprising:

[0008] The gravity field spatial spectrum calculation module is used to extract gravity data at different wavelengths using several bandpass filters to obtain the gravity field spatial spectrum; each band represents the gravity information at the corresponding wavelength.

[0009] The AI-based seabed topography inversion module is used to input the spatial spectrum of the gravity field into a trained gravity-depth mapping model to generate predicted depth values.

[0010] As an improvement to the above system, the processing procedure of the gravity field spatial spectrum calculation module includes:

[0011] Calculate the gravity signal in the frequency domain;

[0012] Set up wavelength nodes for bandpass filtering; establish k-1 bandpass filters, each filter processing the wavelength range between two nodes; k is the number of wavelength nodes;

[0013] The gravity signal is multiplied by k-1 bandpass filters, and an inverse Fourier transform is performed on each filtered frequency band.

[0014] The two-dimensional gravity signals processed by each filter are superimposed layer by layer to obtain a gravity spatial spectrum with k-1 bands in three dimensions.

[0015] As an improvement to the above system, the gravity signal is a gravity anomaly or a vertical gravity gradient.

[0016] As an improvement to the above system, the system also includes a gravity-water depth mapping relationship establishment module, which is used to train a machine learning model or a neural network model using the gravity field spatial spectrum obtained by the gravity field spatial spectrum calculation module, so as to obtain a trained gravity-water depth mapping model.

[0017] As an improvement to the above system, the processing procedure of the gravity-depth mapping relationship establishment module includes:

[0018] Obtain the spatial spectrum of the gravity field at the location corresponding to the water depth data of the sparse ship, and form a data pair of gravity field spatial spectrum and water depth at each location;

[0019] All data pairs are divided into training and testing sets according to a set ratio;

[0020] Based on the model's requirements for input data format, decide whether to convert the training and test sets into tensors; normalize the data and water depth for each band of the gravity field spatial spectrum;

[0021] The normalized training set data is input into a machine learning model or a neural network model for training, resulting in a trained gravity-depth mapping model.

[0022] As an improvement to the above system, the processing procedure of the artificial intelligence-based seabed topography inversion module includes:

[0023] Based on the requirements of the artificial intelligence model for the input data format, choose whether to convert the gravity field spatial spectrum data into tensors and normalize the data of each band of the gravity field spatial spectrum;

[0024] Normalized gravity spatial spectrum data are input into a trained gravity-depth mapping model to obtain seabed topography. The model's prediction results are compared with actual ship-measured depths to evaluate accuracy, and the prediction accuracy is obtained based on accuracy evaluation indicators.

[0025] This application also provides an artificial intelligence-based method for retrieving seabed topography using gravity field spatial spectrum data, implemented by the aforementioned system, the method comprising:

[0026] The gravity field spatial spectrum calculation module extracts gravity data at different wavelengths to obtain the gravity field spatial spectrum;

[0027] The gravity field spatial spectrum is input into the gravity-depth mapping model trained in the artificial intelligence inversion seabed topography module to generate water depth prediction values.

[0028] As an improvement to the above method, the method further includes:

[0029] The gravity field spatial spectrum obtained by the gravity field spatial spectrum calculation module is used to train a machine learning model or a neural network model to obtain a trained gravity-water depth mapping model.

[0030] As an improvement to the above method, the method further includes:

[0031] The accuracy of the model's predictions is evaluated by comparing them with actual ship-measured water depths, and the accuracy of the predictions is obtained based on the accuracy evaluation indicators.

[0032] Compared with existing technologies, the advantages of this application are:

[0033] 1. Gravity field spatial spectrum, compared with directly using unprocessed full-band gravity anomaly data for water depth prediction, can explore the characteristics of seafloor topography more deeply from multiple scales and perspectives. This method not only improves the identification ability of small-scale topographic features, but also makes the analysis of the complex relationship between gravity and topography more detailed and accurate.

[0034] 2. The spatial spectrum of the gravity field also enhances the model's ability to handle the complex relationship between terrain and gravity information. The response relationship between gravity and terrain varies at different wavelengths, allowing machine learning and deep learning technologies to better leverage their "nonlinear" capabilities. Attached Figure Description

[0035] Figure 1 The diagram shows the structure of an artificial intelligence-based system for retrieving seabed topography using gravity field spatial spectrum data.

[0036] Figure 2 The diagram shows the flowchart of an artificial intelligence method for retrieving seabed topography using gravity field spatial spectrum data. Detailed Implementation

[0037] The technical solution of this application will be described in detail below with reference to the accompanying drawings.

[0038] Existing AI-based gravity-based seabed topography retrieval technologies struggle to uncover characteristic relationships when data is insufficient, especially given that different physical environments are affected differently by varying gravity parameters. Simply accumulating a large number of parameters is insufficient for retrieving various types of terrain. The applicant's research reveals that different wavelengths of gravity influence gravity to varying degrees. Therefore, the applicant proposes enriching gravity data through the calculation of the gravity field's spatial spectrum to improve the effectiveness of AI-based seabed topography retrieval. Different wavelengths of gravity can be adaptively weighted, taking into account their varying degrees of influence on topography retrieval, thereby enhancing the accuracy of gravity-based seabed topography retrieval.

[0039] This application proposes an artificial intelligence-based system for retrieving seabed topography using gravity field spatial spectrum data, the system comprising:

[0040] Gravity field spatial spectrum calculation module: Uses several bandpass filters to extract gravity data at different wavelengths (such as gravity anomalies or vertical gravity gradients, etc.), with each band representing gravity information at the corresponding wavelength.

[0041] The gravity-depth mapping module processes gravity and depth data and establishes their correlation. This module acquires the spatial spectrum of the gravity field at locations corresponding to sparse ship depth measurements and pairs the spatial spectrum of the gravity field at each location with the depth data. Next, the dataset is divided into training and testing sets according to an appropriate ratio. Based on the requirements of the AI ​​model for the input data format, it is selected whether to convert the data to tensors and perform normalization. Then, an AI model suitable for the regression task is designed. The model can be a machine learning method (such as random forest or support vector machine) or a neural network method (such as fully connected networks, convolutional neural networks, Transformers, etc.) for training and predicting depth.

[0042] The AI-powered seafloor topography inversion module is used to evaluate the accuracy of seafloor topography and to invert the seafloor topography of the inverted area. It takes the gravity field spatial spectrum from the test data, inputs it into a trained gravity-depth mapping model, generates predicted depth values, and evaluates the model's predictive performance by calculating accuracy metrics (such as standard deviation, root mean square error, correlation coefficient, etc.). Then, based on the AI ​​model's requirements for input data format, it selects whether to convert the gravity field spatial spectrum data of the study area into tensors and normalizes each band. Finally, it inputs the normalized gravity field spatial spectrum data into the model to generate a seafloor topography map of the entire area.

[0043] The processing steps of the gravity field spatial spectrum calculation module include:

[0044] Step A1: Two-dimensional Fourier transform:

[0045]

[0046] f x and f y These are coordinates in the frequency domain, with values ​​ranging from 0 to M-1 and 0 to N-1, respectively. It is a complex exponential term that describes the transformation of a signal in the frequency domain.

[0047] Step A2: Set band nodes and apply bandpass filtering. Perform bandpass filtering on the gravity data according to the set band nodes. Define filter W (Band Notes). i

[0048] W (Band Notes) i ={1, if Wavelenth∈(Band Notes i Band Notes i+1 ],0,otherwise}

[0049] The Band Notes represent the wavelengths at the set band nodes. The number of Band Notes can be arbitrary, and can be adjusted according to the characteristics of the study area. For example, suppose the set band nodes are:

[0050] Band Notes = [0, 3, 6, 10, 15, 30, 60, 100, 140, 180, 300, 1000] (unit: km)

[0051] Step A3: Calculate the gravity for each band.

[0052] G0Spe{i}(f x f y )=G0(f x f y )·W(Band Notes{i})

[0053] Inverse Fourier transform back to the spatial domain:

[0054]

[0055] Step A4: Splicing Gravity Spatial Spectrum Bands

[0056] All processed gravity space spectrum bands are stitched together along the third dimension:

[0057] g0Spe(f x f y = stack[g0Spe{i}(f x f y ), ..., g0Spe{n}(f x f y )]

[0058] The processing steps of the gravity-depth mapping relationship establishment module include:

[0059] Step B1: Obtain the spatial spectrum of the gravity field at the location corresponding to the sparse ship depth measurement data. For the spatial spectrum of the gravity field at each location, extract the corresponding depth data to form a data pair, that is, the spatial spectrum of the gravity field at each location is paired with the corresponding depth data for subsequent model training and prediction.

[0060] Step B2: Divide all acquired data pairs into training and test sets according to a certain ratio (e.g., 7:3, 8:2, or 9:1). The training set is used to train the model, and the test set is used to evaluate the model's performance. The data can be divided randomly or reasonably grouped according to features such as spatial location to ensure that the training and test data are representative in distribution.

[0061] Step B3: Based on the requirements of the AI ​​model for input data format, choose whether to convert the training and test set data into tensors to facilitate parallel computing and model training. Next, normalize the data for each band of the gravity field spatial spectrum and the corresponding water depth data, scaling them to the range of [0,1] or [-1,1] to prevent the magnitude differences of different features from adversely affecting model training and improve the training efficiency and stability of the model.

[0062] Step B4: Design an AI model suitable for this regression task. The model can be a traditional machine learning method, such as random forest or support vector machine, or a deep learning method based on neural networks, such as fully connected neural networks, convolutional neural networks, or Transformers.

[0063] Step B5: Define a suitable optimizer (e.g., SGD or Adam) to guide model parameter updates. Define a suitable loss function (e.g., Mean Squared Error Loss) to measure the difference between the model's predicted values ​​and the true values.

[0064] Step B6: Set several epochs. In each epoch, the model loads training set data in batches using a data loader, calculates the loss batch by batch, and updates the model parameters using an optimizer. Training continues until the loss function converges, meaning the model's error decreases to a low level, or the set number of training epochs is reached. After training, save the final regression model to obtain the mapping relationship between the gravity field and water depth, which can be used for subsequent water depth prediction tasks.

[0065] The processing steps of the AI-based seabed topography inversion module include:

[0066] Step C1: Extract the spatial gravity field from the test data obtained in Step B3 and input it into the gravity-depth mapping model to generate the corresponding predicted depth value. Compare the predicted depth value with the shipborne depth sounding data, and quantify the model's predictive performance by calculating a series of accuracy evaluation indicators (such as standard deviation, root mean square error RMSE, correlation coefficient, etc.) to evaluate the model's accuracy and reliability.

[0067] Step C2: Based on the input data format requirements of the artificial intelligence model, choose whether to integrate and convert the spatial spectrum data of all gravity fields within the study area into tensor form for further data processing. Perform data normalization for each band of the gravity field spatial spectrum to ensure data scale consistency and prepare for subsequent model input.

[0068] Step C3: Input the gravity field spatial spectrum processed in step C2 into the gravity-depth mapping relationship model, and generate seabed topography data for the entire study area through model calculation.

[0069] This application also provides an artificial intelligence-based method for retrieving seabed topography using gravity field spatial spectrum data, implemented based on the aforementioned system. The method includes:

[0070] Step S1: First, calculate the gravity signal in the frequency domain, such as gravity anomalies or vertical gravity gradients. Next, by setting band nodes for the gravity field spatial spectrum and performing truncated bandpass filtering, information within a specific wavelength range is preserved; the node positions can be adjusted according to the characteristics of the study area. Then, perform an inverse Fourier transform on the filtered frequency-domain gravity for each band to convert it into spatial domain data. Finally, stack all bands layer by layer to form a multi-band gravity field spatial spectrum with three dimensions.

[0071] Step S2: First, obtain the spatial spectrum of the gravity field at the location corresponding to the sparse ship depth measurement data, and pair the spatial spectrum of the gravity field at each location with the depth data. Next, divide the dataset into training and testing sets according to a certain ratio. Based on the requirements of the artificial intelligence model for the input data format, choose whether to convert it to a tensor, and normalize the spatial spectrum of the gravity field and the depth data respectively. Subsequently, design and train an artificial intelligence model (such as random forest, support vector machine, fully connected network, etc.), and use an optimizer (such as SGD or Adam) and a loss function (such as MSE Loss) to update the model parameters until the loss function converges. After training, save the regression model, and finally obtain the mapping relationship model between gravity field and water depth.

[0072] Step S3: First, take the gravity field spatial spectrum from the test set obtained in Step S2 and input it into the trained gravity-depth mapping model to obtain the predicted depth value. Then, compare the predicted result with the actual ship-measured depth to evaluate the accuracy, calculating accuracy evaluation indicators (e.g., standard deviation, root mean square error, and correlation coefficient) to measure the model's performance. Next, based on the AI ​​model's requirements for input data format, choose whether to convert the gravity field spatial spectrum data within the study area into tensors and normalize each band. Finally, input the normalized gravity spatial spectrum data into the gravity-depth mapping model to generate the seabed topography of the entire area.

[0073] Example 1

[0074] This application proposes an artificial intelligence inversion system for seabed topography using gravity field spatial spectrum data, including a gravity field spatial spectrum calculation module, a gravity-depth mapping relationship establishment module, and an artificial intelligence inversion module for seabed topography.

[0075] The design of each module is as follows:

[0076] 1. Gravity field spatial spectrum calculation module

[0077] By using multiple bandpass filters, gravity data (such as gravity anomalies or vertical gravity gradients) is extracted from different wavelengths, with each band corresponding to gravity information at a specific wavelength.

[0078] The processing steps of the gravity field spatial spectrum calculation module include:

[0079] Step 1: Calculate the gravity signal in the frequency domain (such as gravity anomaly or vertical gravity gradient).

[0080] Step 2: Set up wavelength nodes. These nodes are used for bandpass filtering. You can choose any number of nodes, and their specific locations can be adjusted according to the characteristics of the study area. If there are k nodes, there will be k-1 bandpass filters, and each filter processes the wavelength range between two nodes.

[0081] Step 3: Multiply the gravity signal by k-1 bandpass filters. Perform an inverse Fourier transform on each filtered frequency band.

[0082] Step 4: Stack the two-dimensional gravity signals processed by each filter layer to obtain a gravity field spatial spectrum with k-1 bands in three dimensions.

[0083] 2. Gravity-water-depth mapping relationship establishment module:

[0084] This module processes gravity and water depth data and establishes the relationship between them. It acquires the spatial spectrum of the gravity field at locations corresponding to sparse ship depth measurements and pairs this spectrum with the corresponding water depth data. The dataset is divided into training and testing sets according to an appropriate ratio. Then, based on the input data format requirements of the AI ​​model, it selects whether to convert the data to tensors and performs normalization. An AI model suitable for regression tasks is designed. This model can employ machine learning methods (e.g., random forests or support vector machines) or neural network methods (e.g., fully connected networks, convolutional neural networks, or Transformers) to train the data and predict water depth values.

[0085] Step 1: Obtain the spatial spectrum of the gravity field at the location corresponding to the sparse ship depth measurement data, and form a data pair of gravity field spatial spectrum and water depth at each location.

[0086] Step 2: Divide all data into training and test sets, and divide the data in the training and test sets in an appropriate ratio (e.g., 7:3, 8:2 or 9:1) and manner.

[0087] Step 3: Based on the AI ​​model's requirements for input data format, choose whether to convert the training and test sets into tensors. Normalize the data for each band of the gravity field spatial spectrum and the water depth separately.

[0088] Step 4: Design an artificial intelligence model, which can be a machine learning model (e.g., random forest, support vector machine, etc.) or a neural network model (e.g., fully connected network, convolutional neural network, Transformer, etc., which can be used for regression tasks).

[0089] Step 5: Define the optimizer (e.g., SGD or Adam) and the loss function (e.g., MSE Loss).

[0090] Step 6: The model loads data batches using the data loader from Step 2 over several epochs, calculates the loss and updates the model parameters until the loss converges, training ends, the regression model is saved, and the gravity-water depth mapping relationship model is obtained.

[0091] 3. Artificial Intelligence-Based Seafloor Topography Inversion Module: This module is used to evaluate the accuracy of seafloor topography and invert the seafloor topography of the inverted area. First, it extracts the spatial spectrum of the gravity field from the test data and inputs it into a trained gravity-depth mapping model to generate predicted depth values. The model performance is evaluated by calculating accuracy metrics such as standard deviation, root mean square error, and correlation coefficient. The normalized data is then input into the model to generate the seafloor topography for the entire area.

[0092] Step 1: Take the gravity field spatial spectrum from the test data obtained in the previous module, input it into the gravity-water depth mapping relationship model to obtain the water depth prediction value, and calculate the accuracy index (e.g., standard deviation, root mean square error, correlation coefficient, etc.) of the prediction value with the corresponding ship-measured water depth to evaluate the performance of the predicted water depth.

[0093] Step 2: Based on the requirements of the artificial intelligence model for the input data format, choose whether to convert the gravity spatial spectrum data of the entire study area into tensors and normalize the data of each band of the gravity field spatial spectrum.

[0094] Step 3: Input the normalized gravity spatial spectrum data from the previous step into the gravity-depth mapping model to obtain the seabed topography of the entire region.

[0095] Example 2

[0096] like Figure 2 As shown, this application also proposes an artificial intelligence inversion method for seabed topography using gravity field spatial spectrum data, implemented based on the above system, including:

[0097] Step 1: Calculate the gravity signal in the frequency domain (such as gravity anomaly or vertical gravity gradient).

[0098] Step 2: Set up wavelength nodes. These nodes are used for bandpass filtering. You can choose any number of nodes, and their specific locations can be adjusted according to the characteristics of the study area. If there are k nodes, there will be k-1 bandpass filters, and each filter processes the wavelength range between two nodes.

[0099] Step 3: Multiply the gravity signal by k-1 bandpass filters. Perform an inverse Fourier transform on each filtered frequency band.

[0100] Step 4: Stack the two-dimensional gravity signals processed by each filter layer to obtain a gravity field spatial spectrum with k-1 bands in three dimensions.

[0101] Step 5: Obtain the spatial spectrum of the gravity field at the location corresponding to the sparse ship depth measurement data, and pair the gravity field spectrum at each location with the depth value to form a data pair.

[0102] Step 6: Divide all data pairs into training and test sets in an appropriate ratio (e.g., 7:3, 8:2, or 9:1) and use a suitable splitting method.

[0103] Step 7: Based on the AI ​​model's requirements for input data format, choose whether to convert the training and test sets to tensor format. Normalize the data for each band of the gravity field spectrum and the water depth data respectively.

[0104] Step 8: Design an artificial intelligence model. The model can be a machine learning method (such as random forest, support vector machine, etc.) or a neural network (such as fully connected neural network, convolutional neural network, Transformer, etc., which are suitable for regression tasks).

[0105] Step 9: Select the optimizer (such as SGD or Adam) and the loss function (such as the mean squared error loss function MSELoss).

[0106] Step 10: During model training, data is loaded in batches using the data loader from Step 2 over several epochs. The loss is calculated, and the model parameters are continuously updated until the loss function converges. After training, the regression model, i.e., the gravity-depth mapping model, is saved.

[0107] Step 11: Apply the model trained in the previous step to the gravity field spatial spectrum in the test data to predict the water depth value. By comparing it with the water depth measured by the ship, calculate the accuracy indicators such as standard deviation, root mean square error, and correlation coefficient to evaluate the prediction effect of the model.

[0108] Step 12: Based on the requirements of the artificial intelligence model for the input data format, select whether to convert the gravity field spectrum data of the entire study area into tensors, and normalize the data of each spectrum band.

[0109] Step 13: Input the gravity field spectrum data processed in Step 12 into the trained gravity-depth mapping model to generate a seabed topographic map of the entire region.

[0110] Table 1 summarizes the validation results of ship-based depth measurement data, comparing the accuracy of depth retrieval using AI methods with and without gravity field spatial spectra, as well as traditional methods. It can be seen that the root mean square error (RMSE) is significantly reduced when using gravity field spatial spectra for both random forests and fully connected networks. Furthermore, the S&S error using the fully connected network with gravity field spatial spectra is reduced from 105 meters to 78 meters compared to the traditional method. This demonstrates the superiority of gravity field spatial spectra for seabed topography retrieval.

[0111] Table 1. Verification results of shipboard water depth measurement (unit: m)

[0112]

[0113] This application may also provide a computer device, including: at least one processor, memory, at least one network interface, and a user interface. The various components in this device are coupled together via a bus system. It is understood that the bus system is used to implement communication between these components. In addition to a data bus, the bus system also includes a power bus, a control bus, and a status signal bus.

[0114] The user interface can include a display, keyboard, or clicking device. Examples include a mouse, trackball, touchpad, or touchscreen.

[0115] It is understood that the memory in the embodiments disclosed in this application may be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory may be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory may be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as Static Random Access Memory (SRAM), Dynamic Random Access Memory (DRAM), Synchronous DRAM (SDRAM), Double Data Rate Synchronous DRAM (DDRSDRAM), Enhanced Synchronous DRAM (ESDRAM), Synchlink DRAM (SLDRAM), and Direct Rambus RAM (DRRAM). The memories described herein are intended to include, but are not limited to, these and any other suitable types of memory.

[0116] In some implementations, the memory stores elements such as executable modules or data structures, or subsets thereof, or extended sets thereof: operating systems and applications.

[0117] The operating system includes various system programs, such as the framework layer, core library layer, and driver layer, used to implement various basic business functions and handle hardware-based tasks. The application programs include various applications, such as media players and browsers, used to implement various application functions. Programs implementing the methods of the embodiments of this disclosure can be included in the application programs.

[0118] In the above embodiments, the processor can also invoke programs or instructions stored in memory, specifically programs or instructions stored in an application program, for the following purposes:

[0119] Follow the steps described above.

[0120] The above methods can be applied to or implemented by a processor. The processor may be an integrated circuit chip with signal processing capabilities. During implementation, each step of the above methods can be completed by integrated logic circuits in the processor's hardware or by instructions in software form. The processor can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the disclosed methods, steps, and logic block diagrams. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the disclosed methods can be directly implemented by a hardware decoding processor, or implemented by a combination of hardware and software modules in the decoding processor. The software modules can reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. This storage medium is located in memory; the processor reads information from the memory and, in conjunction with its hardware, completes the steps of the above methods.

[0121] It is understood that the embodiments described in this application can be implemented using hardware, software, firmware, middleware, microcode, or a combination thereof. For hardware implementation, the processing unit can be implemented in one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), general-purpose processors, controllers, microcontrollers, microprocessors, other electronic units for performing the functions described in this application, or combinations thereof.

[0122] For software implementation, the technology of this application can be implemented by executing the functional modules (e.g., procedures, functions, etc.) of this application. The software code can be stored in memory and executed by a processor. The memory can be implemented in the processor or externally.

[0123] This application may also provide a non-volatile storage medium for storing a computer program. When the computer program is executed by a processor, it can implement the steps in the above method embodiments.

[0124] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application and are not intended to limit it. Although this application has been described in detail with reference to the embodiments, those skilled in the art should understand that modifications or equivalent substitutions to the technical solutions of this application do not depart from the spirit and scope of the technical solutions of this application, and should all be covered within the scope of the claims of this application.

Claims

1. An artificial intelligence-based system for retrieving seabed topography using gravity field spatial spectrum data, characterized in that, The system includes: The gravity field spatial spectrum calculation module is used to extract gravity data at different wavelengths using several bandpass filters to obtain the gravity field spatial spectrum; each band represents the gravity information at the corresponding wavelength; and The AI-based seabed topography inversion module is used to input the spatial spectrum of the gravity field into a trained gravity-depth mapping model to generate predicted depth values. The processing steps of the gravity field spatial spectrum calculation module include: Calculate the gravity signal in the frequency domain; Set up wavelength nodes for bandpass filtering; establish k-1 bandpass filters, each filter processing the wavelength range between two nodes; k is the number of wavelength nodes; The gravity signal is multiplied by k-1 bandpass filters, and an inverse Fourier transform is performed on each filtered frequency band. The two-dimensional gravity signals processed by each filter are superimposed layer by layer to obtain a gravity spatial spectrum with k-1 bands in three dimensions.

2. The artificial intelligence inversion system for seabed topography using gravity field spatial spectrum data according to claim 1, characterized in that, The gravity signal is a gravity anomaly or a vertical gravity gradient.

3. The artificial intelligence inversion system for seabed topography using gravity field spatial spectrum data according to claim 1, characterized in that, The system also includes: The gravity-water-depth mapping relationship establishment module is used to train machine learning models or neural network models using the gravity field spatial spectrum obtained by the gravity field spatial spectrum calculation module, so as to obtain a trained gravity-water-depth mapping model.

4. The artificial intelligence inversion system for seabed topography using gravity field spatial spectrum data according to claim 3, characterized in that, The processing steps of the gravity-depth mapping relationship establishment module include: Obtain the spatial spectrum of the gravity field at the location corresponding to the water depth data of the sparse ship, and form a data pair of gravity field spatial spectrum and water depth at each location; All data pairs are divided into training and testing sets according to a set ratio; Based on the model's requirements for input data format, decide whether to convert the training and test sets into tensors; normalize the data and water depth for each band of the gravity field spatial spectrum; The normalized training set data is input into a machine learning model or a neural network model for training, resulting in a trained gravity-depth mapping model.

5. The artificial intelligence inversion system for seabed topography using gravity field spatial spectrum data according to claim 1, characterized in that, The processing steps of the AI-based seabed topography inversion module include: Based on the requirements of the artificial intelligence model for the input data format, we can choose whether to convert the gravity spatial spectrum data into tensors and normalize the data of each band of the gravity field spatial spectrum. The processed gravity spatial spectrum data is input into the trained gravity-depth mapping model to obtain the seabed topography.

6. The artificial intelligence inversion system for seabed topography using gravity field spatial spectrum data according to claim 5, characterized in that, The processing steps of the AI-based seabed topography inversion module also include: The accuracy of the model's predictions is evaluated by comparing them with actual ship-measured water depths, and the accuracy of the predicted seabed topography is obtained based on the accuracy evaluation index.

7. An artificial intelligence method for retrieving seafloor topography using gravity field spatial spectrum data, implemented based on the system described in any one of claims 1-6, the method comprising: The gravity field spatial spectrum calculation module extracts gravity data at different wavelengths to obtain the gravity field spatial spectrum; The gravity field spatial spectrum is input into the gravity-depth mapping model trained in the artificial intelligence inversion seabed topography module to generate water depth prediction values.

8. The artificial intelligence inversion method for seabed topography using gravity field spatial spectrum data according to claim 7, characterized in that, The method further includes: The gravity field spatial spectrum obtained by the gravity field spatial spectrum calculation module is used to train a machine learning model or a neural network model to obtain a trained gravity-water depth mapping model.

9. The artificial intelligence inversion method for seabed topography using gravity field spatial spectrum data according to claim 8, characterized in that, The method further includes: The gravity-depth mapping model is used to predict seabed topography. The accuracy of the model's prediction results is evaluated by comparing them with the actual ship-measured depth. The accuracy of the prediction is obtained based on the accuracy evaluation index.

Images