A shipboard radar sea wave perception domain expansion method and system based on deep learning

Abstract

The application belongs to the technical field of marine environment expansion, and discloses a shipborne radar sea wave perception domain expansion method and system based on deep learning, which constructs a sea wave perception domain step-by-step expansion model, uses a sea wave perception domain expansion submodule to realize spatial feature modeling and limited range expansion of local observation data, and gradually deduces larger range sea wave field information through a step-by-step expansion and residual connection mode; meanwhile, a sea wave continuity constraint mechanism is introduced to apply continuity constraints at the junction of the expanded sea area and the original observation sea area, so that the overall wave field can maintain high accuracy and spatial continuity in a large range. Through the shipborne radar sea wave perception domain expansion method based on deep learning, the real-time perception ability of the shipborne radar to a large range of sea areas in the local environment can be significantly improved, the independent operation of the ship in the marine environment is supported, and the real-time sea wave perception ability of the ship navigation safety and the marine engineering application is significantly improved.

Description

Technical Field

[0001] This invention belongs to the field of marine environment extension technology, and particularly relates to a method and system for extending the wave perception domain of shipborne radar based on deep learning. Background Technology

[0002] With the continuous development of global maritime transport, fisheries production, and marine engineering activities, real-time and accurate acquisition of wave field information over large areas of sea has become a crucial foundation for ensuring ship navigation safety, formulating route plans, and conducting offshore operations. Waves are a key environmental factor affecting ship motion and structural safety, and their spatial distribution characteristics directly relate to navigational risk avoidance, selection of operational windows, and disaster early warning. Therefore, timely understanding of large-scale wave fields not only helps improve the safety and efficiency of autonomous ship navigation but also has significant implications for marine engineering and disaster prevention and mitigation.

[0003] Currently, shipborne radar, as a commonly used observation method, can detect wave fields within a certain range near a ship and extract key parameters such as significant wave height, spectral peak period, and dominant wave direction. This type of data is very useful for local situational awareness of ships, but its coverage is limited, generally only reflecting local wave conditions within a few tens of kilometers of the ship. Therefore, how to extrapolate and obtain wave field information over a larger area based on limited local observation data has become a critical technical problem that urgently needs to be solved for current ship navigation safety and marine engineering applications.

[0004] Existing methods for expanding the range of wave sensing mainly fall into the following categories:

[0005] (1) Numerical model calculation and data assimilation: This type of method uses large-scale ocean wave numerical models such as WAVEWATCH III and SWAN, and combines buoy, satellite or ship observation data for assimilation correction to obtain large-scale wave field information.

[0006] (2) Sea area interpolation based on measured data: This type of method extends the local wave information measured by shipborne torpedoes to the adjacent area through spatial interpolation or statistical models, thereby obtaining wave estimates within a certain range.

[0007] Based on the above analysis, the problems and shortcomings of the existing technology are as follows:

[0008] (1) Insufficient real-time performance: Existing large-scale wave field acquisition mainly relies on numerical models and data assimilation methods. These methods involve large computational loads and complex processes, often requiring several hours or even longer computation time, which is difficult to meet the needs of ships for real-time wave field information during navigation.

[0009] (2) Lack of local independence: Numerical model results usually rely on shore-based supercomputing platforms for generation, and ships need to obtain them through satellite communication or other remote means. There are data transmission dependencies and delays, making it difficult to achieve local autonomous operation of a single ship in the marine environment.

[0010] (3) Limited computational accuracy: Some methods extend local observation data to neighboring areas through interpolation or statistical models, but the results are highly dependent on the number and distribution of observation points, making it difficult to accurately reflect wave propagation and nonlinear evolution characteristics, resulting in limited prediction accuracy. Summary of the Invention

[0011] To overcome the problems existing in related technologies, the present invention discloses an embodiment of a method and system for extending the wave perception domain of shipborne radar based on deep learning, the technical solution of which is as follows:

[0012] This invention is implemented as follows: a deep learning-based method for extending the wave perception domain of shipborne radar, comprising the following steps:

[0013] S1, Model Dataset Construction: Obtain wave field data observed by shipborne radar, as well as wave field analysis data that covers and exceeds the observation range of shipborne radar. Through spatiotemporal matching and sea area range interception operations, generate a dataset that matches the shipborne radar observed wave field with the target extended wave field.

[0014] S2, Construction and Training of a Stepwise Expanding Wave Perception Domain Model: A stepwise expanding wave perception domain model is constructed. A local wave field is expanded within a limited range using a wave perception domain expansion submodule, and the expanded field is input into a visual converter network for spatial feature extraction and modeling. The local wave field is gradually expanded in stages using the stepwise expanding wave perception domain module, with residual connections between each submodule. A wave continuity constraint loss function is constructed, which comprehensively considers the overall accuracy of the expansion results and the physical continuity at the boundary between the expanded region and the original observation region. The model is then trained using the obtained model dataset.

[0015] S3, Wave Extension and Validation: The local wave field observed by shipborne radar is extended using the trained wave perception domain stepwise extension model, and the reliability of the extension results is verified using accuracy indicators.

[0016] In step S1, the model dataset construction includes:

[0017] Acquire the measured wave field data and regional effective wave height analysis data of the shipborne radar, and match the wave height analysis data of the shipborne radar at the corresponding time. Extract the sea area of ​​the wave height analysis data at the corresponding time, and make the center point of the sea area consistent with the sea area measured by the shipborne radar to complete the construction of the model dataset.

[0018] Among them, the wave field data is based on the inversion of X-band navigation radar data to obtain the significant wave height data of the latitude and longitude gridded area; the wave height analysis data adopts the 1 / 12° resolution significant wave height analysis data of Copernicus Marine Environment Observation Service (CMEMS); the spatiotemporal matching is to perform spatiotemporal interpolation on the significant wave height analysis data to form analysis data with a unified spatial resolution that covers and exceeds the radar observation range, while making it correspond to the timestamp of the radar measured wave height data.

[0019] Furthermore, the length and width of the wave field data measured by the shipborne radar are both... The length and width of the wave height analysis data at the corresponding time are both By constructing a model dataset, the side length of the shipborne radar wave perception domain is expanded to the original value. The area expansion target is [number] times the original [area]. times.

[0020] In step S2, the construction of the wave sensing domain extension submodule includes:

[0021] For the first-layer wave sensing domain expansion submodule in the hierarchical expansion model of the wave sensing domain, the local wave field data obtained from shipborne radar observations are processed using a boundary expansion function to supplement the outer edge, thereby generating an expanded region with a wider width outside the original observation range, and increasing the side length of the sea area grid from... Expand to The expanded wave field data is input into a visual converter network for spatial feature encoding and modeling. In this network, each wave perception domain extension submodule uses a 4-layer encoder (ViT), with 4 attention heads and 256 hidden layer dimensions. This yields a wave field feature representation that simultaneously includes local observations and extended regions, providing the input basis for subsequent progressive expansion. The expression is:

[0022] ;

[0023] In the formula, The feature vector output by the wave sensing domain extension submodule. For visual converter networks, For the row and column positions of the grid points, The number of grid squares in the length and width of the measured wave field data from shipborne radar. For shipborne radar observation wave height input Grid point data, The interpolation weights are used for the boundary expansion function of the first-layer wave sensing domain extension submodule. These are the network parameters for the visual converter of the first-layer wave perception domain extension submodule.

[0024] Furthermore, the wave sensing domain progressively expanding module consists of... It consists of a series of wave sensing domain extension sub-modules, used to realize the gradual expansion of wave field data from the initial range to the target range.

[0025] Furthermore, the construction of the wave sensing domain progressively expanding module includes:

[0026] The input wave field data from the shipborne radar's local observations is fed into the first wave sensing domain extension submodule. This submodule performs a limited-range boundary extension and extracts spatial features, resulting in a grid with a side length larger than the original range. wave field The extended result is used as the input to the next wave sensing domain extension submodule, and the process is repeated iteratively. Next, the initial small-scale wave field is gradually pushed outward to a grid with a side length of... This module covers a wide range of sea areas; it also adds residual links between sub-modules, and the module ultimately outputs feature vectors. ;

[0027] ;

[0028] In the formula, For the first The feature vector output by the extended submodule of the layered wave sensing domain. For shipborne radar observation wave height input Grid point data, For the number of iterations, For the first In the feature vector output by the extended submodule of the wave sensing domain Grid point data, For the first Interpolation weights for the boundary extension function of the extended submodule of the layered wave sensing domain. For the first The feature vector output by the extended submodule of the layered wave sensing domain. To expand the number of submodules in the wave sensing domain, For the first Visual converter network parameters of the extended submodule of the layered wave perception domain. Input for shipborne radar observation wave height.

[0029] In step S2, the construction of the wave continuity constraint loss function includes:

[0030] Error term of the step-by-step expansion module: Step-by-step expansion of the wave sensing domain module After this, the output feature vector Corresponding actual observed wave field Error calculations are performed across the entire sea area to constrain the overall prediction results of the wave sensing domain progressively expanding modules;

[0031] Continuity constraint error term: for the boundary between the extended region and the original observation region. The first derivative of the wave field output by the wave sensing domain progressively expanded model and the observed wave field at the boundary is calculated. With the second derivative ; Calculate the first derivative of the observed wave field at the inner boundary. With the second derivative and calculate and ,as well as and To minimize errors between them and ensure the continuity of the wave field at the boundary;

[0032] The error of the progressively expanding module and the error of the continuity constraint are weighted and combined to form the total loss function;

[0033] ;

[0034] In the formula, For the total loss, These are all weighting coefficients for the first and second derivative losses in the error terms of the progressively expanding module of the wave sensing domain and the continuity constraint error terms. For the sample size, For the first Number of samples The first Model output data and actual observed wave field data for each sample. The first The first and second derivatives of the wave field output by the sample model and the observed wave field at the boundary. The first The first and second derivatives of the observed wave field at the inner boundary of each sample.

[0035] In step S3, the training of the wave perception domain progressively expanding model includes:

[0036] Wave field data observed by shipborne radar is input into the wave sensing domain progressively expanding model for training. The model is trained for a total of 100 rounds, with 64 sets of data input in each batch. Simultaneously, a cosine annealing algorithm is introduced to dynamically schedule the learning rate, with an initial learning rate of... Eventually decayed to ;

[0037] ;

[0038] In the formula, For training the first The learning rate of the step. These are the minimum and maximum values ​​of the learning rate, respectively. This represents the current iteration step. This represents the total number of iterations.

[0039] In step S3, the wave expansion is followed by wave expansion result accuracy verification;

[0040] Using root mean square error (RMSE) and correlation coefficient As an accuracy evaluation index, the wave perception domain expansion effect of the wave perception domain progressive expansion model is evaluated.

[0041] Another objective of this invention is to provide a deep learning-based shipborne radar wave perception domain extension system, which is implemented using the aforementioned deep learning-based shipborne radar wave perception domain extension method. The system includes:

[0042] The model dataset construction module is configured to acquire wave field data observed by shipborne radar and wave field analysis data that covers and exceeds the observation range of the shipborne radar, and generate a dataset that matches the shipborne radar observed wave field with the target extended wave field through spatiotemporal matching and sea area truncation operations.

[0043] The wave perception domain progressive expansion model construction and training module is configured to construct a wave perception domain progressive expansion model. It expands the local wave field to a limited range through a wave perception domain expansion sub-module and inputs the data into a visual converter network for spatial feature extraction and modeling. The local wave field is progressively expanded in stages through the wave perception domain progressive expansion module, with residual connections between each sub-module. A wave continuity constraint loss function is constructed, which comprehensively considers the overall accuracy of the expansion result and the physical continuity at the boundary between the expanded region and the original observation region. The model is then trained using the constructed dataset.

[0044] The wave extension and verification module is configured to extend the local wave field observed by shipborne radar using a trained wave perception domain hierarchical extension model, and to verify the reliability of the extension results using accuracy indicators.

[0045] Among them, any one or more of the model dataset construction module, the wave perception domain progressively expanded model construction and training module, and the wave expansion and verification module are implemented by the processor executing computer program instructions stored in the memory.

[0046] Another objective of this invention is to provide a deep learning-based shipborne radar wave perception domain extension system, the system comprising:

[0047] The model dataset construction module is configured to acquire wave field data observed by shipborne radar and wave field analysis data that covers and exceeds the observation range of the shipborne radar, and generate a dataset that matches the shipborne radar observed wave field with the target extended wave field through spatiotemporal matching and sea area truncation operations.

[0048] The wave perception domain progressive expansion model construction and training module is configured to construct a wave perception domain progressive expansion model. It expands the local wave field to a limited range through a wave perception domain expansion sub-module and inputs the data into a visual converter network for spatial feature extraction and modeling. The local wave field is progressively expanded in stages through the wave perception domain progressive expansion module, with residual connections between each sub-module. A wave continuity constraint loss function is constructed, which comprehensively considers the overall accuracy of the expansion result and the physical continuity at the boundary between the expanded region and the original observation region. The model is then trained using the constructed dataset.

[0049] The wave extension and verification module is configured to extend the local wave field observed by shipborne radar using a trained wave perception domain hierarchical extension model, and to verify the reliability of the extension results using accuracy indicators.

[0050] Among them, any one or more of the model dataset construction module, the wave perception domain progressively expanded model construction and training module, and the wave expansion and verification module are implemented by the processor executing computer program instructions stored in the memory.

[0051] Combining all the above technical solutions, the beneficial effects of this invention are as follows:

[0052] First, this invention aims to overcome the limitations of existing shipborne radar wave sensing range, insufficient real-time performance due to reliance on numerical models, and limited accuracy of interpolation methods. It proposes a deep learning-based method for expanding the wave sensing domain of shipborne radar. This method constructs a hierarchical expansion model of the wave sensing domain, utilizing a wave sensing domain expansion submodule to model the spatial characteristics of local observation data and expand it within a limited range. The hierarchical expansion operation avoids the model directly learning the complex mapping from local to global in a single calculation, reducing model training difficulty and enabling progressively high-precision extrapolation of wave field information over a larger area. Simultaneously, this invention introduces a wave continuity constraint mechanism, introducing multi-order differential continuity constraints at the boundary between the expanded and original observation areas. The first derivative ensures a smooth transition of the wave surface slope at the boundary, while the second derivative ensures the natural extension of the wave surface curvature, thus avoiding abrupt changes or broken line shapes and improving the physical reliability of the expansion results. Finally, by weighting the overall error term and the continuity error term, a balance between overall accuracy and physical continuity is achieved, simultaneously improving the expansion accuracy and the physical rationality of the expanded wave field. The method of this invention can significantly improve the real-time perception capability of shipborne radar over a large area of ​​sea in a local environment, support the independent operation of ships in the marine environment, and thus improve navigation safety and the efficiency of marine engineering applications.

[0053] Secondly, this invention improves the wave field perception range and accuracy of shipborne radar. By using a progressively expanding wave perception domain module, this invention gradually extends the local wave field data observed by shipborne radar to a larger sea area. At the same time, it combines a wave continuity constraint loss function to ensure the continuity of the wave field at the boundary, so that the expanded wave field features not only cover a large sea area but also maintain high-precision spatial features, significantly improving the real-time wave perception capability for ship navigation safety and marine engineering applications.

[0054] Third, this invention supports the localized and rapid operation of shipborne radar. Unlike traditional methods that rely on large-scale numerical models and remote data assimilation, this invention uses deep learning models to extrapolate wave fields, which can quickly perform wave field feature extrapolation over a large area in the ship's local environment without relying on shore-based supercomputing platforms or satellite communications, thus enabling independent real-time wave perception and application for ships.

[0055] Fourth, existing methods mainly rely on large-scale numerical models and data assimilation or interpolation-based extension methods, which are difficult to achieve real-time, independent, large-scale wave field perception in the local environment of ships. This invention is the first to realize the localized, rapid, and high-precision extension of shipborne radar local observation data to large-scale wave fields, breaking through the limitations of traditional numerical models and interpolation methods in terms of real-time performance and accuracy. Attached Figure Description

[0056] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure;

[0057] Figure 1 This is a flowchart of the deep learning-based method for extending the wave perception domain of shipborne radar provided in an embodiment of the present invention;

[0058] Figure 2 This is a structural diagram of the wave sensing domain progressively expanding model provided in an embodiment of the present invention;

[0059] Figure 3 This is a comparison chart of the extended effects of different models provided in the embodiments of the present invention;

[0060] Figure 4 This is a comparative model 1 sea area wave height spread RMSE error diagram provided in the embodiments of the present invention;

[0061] Figure 5 This is the RMSE error diagram of wave height spread in the sea area provided by the comparative model two in this embodiment of the invention;

[0062] Figure 6 This is a comparative model of wave height spread RMSE error in three sea areas provided in this embodiment of the invention;

[0063] Figure 7 This is the wave height spread RMSE error diagram of the model sea area provided in the embodiment of the present invention. Detailed Implementation

[0064] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Many specific details are set forth in the following description to provide a thorough understanding of the present invention. However, the present invention can be practiced in many other ways different from those described herein, and those skilled in the art can make similar modifications without departing from the spirit of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0065] This invention proposes a deep learning-based method for extending the wave perception domain of shipborne radar. It utilizes a visual converter network to perform feature modeling and finite extension of local wave observations. Combining a hierarchical extension module with residual connections, it achieves high-precision extrapolation from local to large-scale data. Furthermore, it introduces multi-order differential continuity constraints to ensure the spatial continuity and physical rationality of the extension boundary, thereby enabling rapid, high-precision, and continuous local extension of shipborne radar observation data to a large-scale wave field. Building upon traditional error constraints, this invention proposes a wave continuity constraint loss to address the discontinuity problem at the boundary between the extended and observed regions. The first derivative ensures a smooth transition of the wave surface slope at the boundary, while the second derivative ensures the natural extension of the wave surface curvature, thus avoiding abrupt changes or broken line shapes and improving the physical reliability of the extension results.

[0066] The innovations of this invention are specifically reflected in the following aspects:

[0067] (1) Wave perception domain progressive expansion module: This invention proposes a multi-stage expansion strategy, which extends the local wave data observed by shipborne radar to a larger sea area through multi-stage extrapolation, reducing the difficulty of the model to learn the global mapping at one time and achieving high-precision progressive extrapolation.

[0068] (2) Wave continuity constraint mechanism: multi-order differential continuity constraint is introduced at the boundary between the extended sea area and the original observation sea area, and error calculation of continuity is added to the loss function to ensure the continuity between the boundary and the inner sea area and solve the spatial discontinuity problem that may occur during the expansion process.

[0069] (3) Rapid expansion of shipborne radar wave field data based on deep learning models: This invention utilizes boundary spread functions and visual converter networks to spatially expand and encode spatial features of shipborne radar wave field data, achieving efficient expansion of local observation information to a large sea area. Compared with traditional numerical modeling methods, this invention, based on deep learning methods, significantly improves the wave field extrapolation speed while ensuring prediction accuracy.

[0070] Example 1, such as Figure 1 As shown, the deep learning-based method for extending the wave perception domain of shipborne radar provided in this embodiment of the invention includes the following steps:

[0071] S1, Model Dataset Construction: Obtain wave field data observed by shipborne radar, as well as wave field analysis data that covers and exceeds the observation range of shipborne radar. Through spatiotemporal matching and sea area range interception operations, generate a dataset that matches the shipborne radar observed wave field with the target extended wave field.

[0072] Acquire the measured wave field data and regional effective wave height analysis data of the shipborne radar, and match the wave height analysis data of the shipborne radar at the corresponding time. Extract the sea area of ​​the wave height analysis data at the corresponding time, and make the center point of the sea area consistent with the sea area measured by the shipborne radar to complete the construction of the model dataset.

[0073] Among them, the wave field data is based on the inversion of X-band navigation radar data to obtain the significant wave height data of the latitude and longitude gridded area; the wave height analysis data adopts the 1 / 12° resolution significant wave height analysis data of Copernicus Marine Environment Observation Service (CMEMS); the spatiotemporal matching is to perform spatiotemporal interpolation on the significant wave height analysis data to form analysis data with a unified spatial resolution that covers and exceeds the radar observation range, while making it correspond to the timestamp of the radar measured wave height data.

[0074] The spatiotemporal matching and sea area interception operations align the center point of the sea area between the numerically simulated wave height data and the radar-measured wave height data, and set the side length to the radar-measured wave height data. Set the intercept range to a size equal to a multiple of the sea area.

[0075] Acquire measured wave field data and regional significant wave height analysis data from shipborne radar, and match the wave height analysis data at the corresponding time points from the measured wave field data from shipborne radar. The length and width of the measured wave field data from shipborne radar are both [missing information]. The length and width of the wave height analysis data at the corresponding time are both By constructing a model dataset, the side length of the shipborne radar wave perception domain is expanded to the original value. The area expansion target is [number] times the original [area]. times.

[0076] S2, Construction and Training of a Stepwise Expanding Wave Perception Domain Model: A stepwise expanding wave perception domain model is constructed. A local wave field is expanded within a limited range using a wave perception domain expansion submodule, and the expanded field is input into a visual converter network for spatial feature extraction and modeling. The local wave field is gradually expanded in stages using the stepwise expanding wave perception domain module, with residual connections between each submodule. A wave continuity constraint loss function is constructed, which comprehensively considers the overall accuracy of the expansion results and the physical continuity at the boundary between the expanded region and the original observation region. The model is then trained using the obtained model dataset.

[0077] like Figure 2As shown, using the constructed dataset, the local wave field is first expanded to a limited range through the wave perception domain extension submodule, and then input into the visual converter network for spatial feature encoding and modeling to obtain the expanded wave field feature vector. Secondly, through a hierarchical expansion module, the local wave field is gradually expanded to a larger sea area in stages, and residual connections are added between each submodule. Finally, by constructing a wave continuity constraint loss function, the model's performance in terms of overall accuracy and continuity at the boundary of the expanded sea area is constrained, resulting in high-precision, large-scale wave field features. Specifically, this includes:

[0078] (1) Constructing a wave sensing domain extension submodule; Taking the first layer of the wave sensing domain extension submodule in the model as an example, the local wave field data obtained by shipborne radar observation is first processed by the boundary expansion function to supplement the outer edge, so as to generate an extended area with a wider width outside the original observation range, and the side length of the sea area grid is changed from Expand to Subsequently, the expanded wave field data is input into the Vision Transformer network to encode and model its spatial features, thereby obtaining a wave field feature representation that simultaneously includes local observations and extended regions, providing an input basis for subsequent step-by-step expansion.

[0079] ;

[0080] In the formula, The feature vector output by the wave sensing domain extension submodule. For visual converter networks, For the row and column positions of the grid points, The number of grid squares in the length and width of the measured wave field data from shipborne radar. For shipborne radar observation wave height input Grid point data, The interpolation weights are used for the boundary expansion function of the first-layer wave sensing domain extension submodule. These are the network parameters for the visual converter of the first-layer wave perception domain extension submodule.

[0081] (2) Construct a progressively expanding module for the wave sensing domain; the progressively expanding module for the wave sensing domain consists of... The system consists of several wave sensing domain extension submodules connected in series. First, the input wave field data from the shipborne radar's local observations is fed into the first wave sensing domain extension module. This module extends the boundary of the wave sensing domain within a limited range and extracts spatial features, resulting in a grid with a side length larger than the original range. wave field The extended result is then used as the input for the next wave sensing domain extension module, and this process is repeated iteratively. This approach avoids the model directly learning the complex mapping from local to global in a single computation, reducing training difficulty. It gradually extrapolates the initial small-scale wave field to a grid with a side length of [missing information]. This module covers a wider sea area. Simultaneously, it adds residual links between sub-modules to ensure stable model convergence. The module ultimately outputs feature vectors. .

[0082] ;

[0083] In the formula, For the first The feature vector output by the extended submodule of the layered wave sensing domain. For shipborne radar observation wave height input Grid point data, For the number of iterations, For the first In the feature vector output by the extended submodule of the wave sensing domain Grid point data, For the first Interpolation weights for the boundary extension function of the extended submodule of the layered wave sensing domain. For the first The feature vector output by the extended submodule of the layered wave sensing domain. To expand the number of submodules in the wave sensing domain, For the first Visual converter network parameters of the extended submodule of the layered wave perception domain. Input for shipborne radar observation wave height.

[0084] (3) Construct the wave continuity constraint loss function;

[0085] In this embodiment, the calculation of the wave continuity constraint loss function includes two parts:

[0086] Error term of the step-by-step expansion module: Step-by-step expansion of the wave sensing domain module After this, the output feature vector Corresponding actual observed wave field Error calculations are performed across the entire sea area to constrain the overall prediction results of the extended module.

[0087] Continuity constraint error term: for the boundary between the extended region and the original observation region. The first derivative of the wave field output by the wave sensing domain progressively expanded model and the observed wave field at the boundary is calculated. With the second derivative ; Calculate the first derivative of the observed wave field at the inner boundary. With the second derivative and calculate and ,as well as and To minimize errors between them and ensure the continuity of the wave field at the boundary;

[0088] The error of the progressively expanding module and the error of the continuity constraint are weighted and combined to form the total loss function;

[0089] ;

[0090] In the formula, For the total loss, These are all weighting coefficients for the first and second derivative losses in the error terms of the progressively expanding module of the wave sensing domain and the continuity constraint error terms. For the sample size, For the first Number of samples The first Model output data and actual observed wave field data for each sample. The first The first and second derivatives of the wave field output by the sample model and the observed wave field at the boundary. The first The first and second derivatives of the observed wave field at the inner boundary of each sample.

[0091] S3, conduct accuracy verification of wave extension results.

[0092] After model training is complete, the local wave field observed by shipborne radar is input into the model to expand it, and a cosine annealing algorithm is introduced to dynamically adjust the learning rate to improve training performance; the root mean square error (RMSE) and correlation coefficient are used to measure the learning rate. Indicators such as [list of indicators] are used to compare the extended results with actual observation data, verifying the accuracy and reliability of the model's wave field extension over a large area of ​​sea. Specifically, these include:

[0093] (1) Training the wave sensing domain progressive expansion model; Wave field data observed by shipborne radar were input into the wave sensing domain progressive expansion model for training. The model was trained for a total of 100 rounds, with 64 sets of data input in each batch. At the same time, the model introduced a cosine annealing algorithm to dynamically schedule the learning rate. The initial learning rate was Eventually decayed to ;

[0094] ;

[0095] In the formula, For training the first The learning rate of the step. These are the minimum and maximum values ​​of the learning rate, respectively. This represents the current iteration step. This represents the total number of iterations.

[0096] (2) Verify the wave propagation effect of the model; This invention uses root mean square error (RMSE) and correlation coefficient ( Error evaluation indexes are used to calculate the error between the true value of the sea area and the value of the sea area expansion.

[0097] ;

[0098] ;

[0099] In the formula, For the number of test samples, and The first True and extended values ​​of wave height for each sample This represents the average of the true wave height values.

[0100] Example 2: The deep learning-based shipborne radar wave perception domain extension system provided in this embodiment of the invention includes:

[0101] The model dataset construction module is configured to acquire wave field data observed by shipborne radar and wave field analysis data that covers and exceeds the observation range of the shipborne radar, and generate a dataset that matches the shipborne radar observed wave field with the target extended wave field through spatiotemporal matching and sea area truncation operations.

[0102] The wave perception domain progressive expansion model construction and training module is configured to construct a wave perception domain progressive expansion model. It expands the local wave field to a limited range through a wave perception domain expansion sub-module and inputs the data into a visual converter network for spatial feature extraction and modeling. The local wave field is progressively expanded in stages through the wave perception domain progressive expansion module, with residual connections between each sub-module. A wave continuity constraint loss function is constructed, which comprehensively considers the overall accuracy of the expansion result and the physical continuity at the boundary between the expanded region and the original observation region. The model is then trained using the constructed dataset.

[0103] The wave extension and verification module is configured to extend the local wave field observed by shipborne radar using a trained wave perception domain hierarchical extension model, and to verify the reliability of the extension results using accuracy indicators.

[0104] Among them, any one or more of the model dataset construction module, the wave perception domain progressively expanded model construction and training module, and the wave expansion and verification module are implemented by the processor executing computer program instructions stored in the memory.

[0105] To further demonstrate the positive effects of the above embodiments, the present invention conducts the following experiments based on the above technical solution: The model of the present invention is compared with a bilinear marginal extrapolation interpolation model (Comparison Model 1), the model of the present invention without using a stepwise expansion strategy (Comparison Model 2), and the model of the present invention without adding a continuity constraint error term (Comparison Model 3). The wave field expansion effects of different models are compared. (See...) Figure 3 .

[0106] The experimental results are shown in Table 1. The model of this invention exhibits lower error and higher correlation in the comparative experiments. Compared to comparative model two, the model of this invention adds a step-by-step expansion strategy, demonstrating that this strategy effectively reduces the training difficulty and stability of the model, achieving high-precision expansion of the wave field. Compared to comparative model three, the model of this invention adds a continuity constraint error term, demonstrating that this continuity constraint error term provides a high-precision representation of the wave continuity characteristics of the model of this invention, further improving the accuracy and reliability of the model in sensing the sea area expansion. The RMSE error diagrams for different models' sea area wave height expansion are shown below. Figures 4-7 .

[0107] Table 1 Comparison of reconstruction errors of different models

[0108]

[0109] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications, equivalent substitutions and improvements made by those skilled in the art within the scope of the technology disclosed in the present invention and within the spirit and principles of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A method for extending the wave perception domain of shipborne radar based on deep learning, characterized in that, The method includes the following steps: S1, Model Dataset Construction: Obtain wave field data observed by shipborne radar, as well as wave field analysis data that covers and exceeds the observation range of shipborne radar. Through spatiotemporal matching and sea area range interception operations, generate a dataset that matches the shipborne radar observed wave field with the target extended wave field. S2, Construction and Training of a Stepwise Expanding Wave Perception Domain Model: A stepwise expanding wave perception domain model is constructed. A local wave field is expanded within a limited range using a wave perception domain expansion submodule, and the expanded field is input into a visual converter network for spatial feature extraction and modeling. The local wave field is gradually expanded in stages using the stepwise expanding wave perception domain module, with residual connections between each submodule. A wave continuity constraint loss function is constructed, which comprehensively considers the overall accuracy of the expansion results and the physical continuity at the boundary between the expanded region and the original observation region. The model is then trained using the obtained model dataset. S3, Wave Extension and Validation: The local wave field observed by shipborne radar is extended using the trained wave perception domain stepwise extension model, and the reliability of the extension results is verified using accuracy indicators. In step S2, the construction of the wave sensing domain extension submodule includes: For the first-layer wave sensing domain expansion submodule in the hierarchical expansion model of the wave sensing domain, the local wave field data obtained from shipborne radar observations are processed using a boundary expansion function to supplement the outer edge, thereby generating an expanded region with a wider width outside the original observation range, and increasing the side length of the sea area grid from... Expand to The expanded wave field data is input into a visual converter network for spatial feature encoding and modeling. In this network, each wave perception domain extension submodule uses a 4-layer encoder (ViT), with 4 attention heads and 256 hidden layer dimensions. This yields a wave field feature representation that simultaneously includes local observations and extended regions, providing the input basis for subsequent progressive expansion. The expression is: In the formula, The feature vector output by the wave sensing domain extension submodule. For visual converter networks, For the row and column positions of the grid points, The number of grid squares in the length and width of the measured wave field data from shipborne radar. For shipborne radar observation wave height input Grid point data, The interpolation weights are used for the boundary expansion function of the first-layer wave sensing domain extension submodule. The network parameters for the visual converter of the first-layer wave perception domain extension submodule; The wave sensing domain expansion module consists of... It consists of a series of wave sensing domain extension sub-modules, used to realize the gradual expansion of wave field data from the initial range to the target range.

2. The deep learning-based method for extending the wave perception domain of shipborne radar according to claim 1, characterized in that, In step S1, the model dataset construction includes: Acquire the measured wave field data and regional effective wave height analysis data of the shipborne radar, and match the wave height analysis data of the shipborne radar at the corresponding time. Extract the sea area of ​​the wave height analysis data at the corresponding time, and make the center point of the sea area consistent with the sea area measured by the shipborne radar to complete the construction of the model dataset. Among them, the wave field data is based on the inversion of X-band navigation radar data to obtain the significant wave height data of the latitude and longitude gridded area; the wave height analysis data adopts the 1 / 12° resolution significant wave height analysis data of Copernicus Marine Environment Observation Service (CMEMS); the spatiotemporal matching is to perform spatiotemporal interpolation on the significant wave height analysis data to form analysis data with a unified spatial resolution that covers and exceeds the radar observation range, while making it correspond to the timestamp of the radar measured wave height data. The spatiotemporal matching and sea area interception operation aligns the center point of the sea area between the numerical simulation wave height data and the radar measured wave height data, and sets the sea area size with a side length that is a multiple of the radar measured wave height data as the interception range.

3. The deep learning-based method for extending the wave perception domain of shipborne radar according to claim 2, characterized in that, The length and width of the wave field data measured by shipborne radar are both The length and width of the wave height analysis data at the corresponding time are both By constructing a model dataset, the side length of the shipborne radar wave perception domain is expanded to the original value. The area expansion target is [number] times the original [area]. times.

4. The deep learning-based method for extending the wave perception domain of shipborne radar according to claim 1, characterized in that, The construction of the wave sensing domain progressively expanding module includes: The input wave field data from the shipborne radar's local observations is fed into the first wave sensing domain extension submodule. This submodule performs a limited-range boundary extension and extracts spatial features, resulting in a grid with a side length larger than the original range. wave field The extended result is used as the input to the next wave sensing domain extension submodule, and the process is repeated iteratively. Next, the initial small-scale wave field is gradually pushed outward to a grid with a side length of... The wave perception domain covers a wide area of ​​sea; at the same time, the progressively expanding module adds residual links between each sub-module, and the module finally outputs a feature vector. ; ； In the formula, For the first The feature vector output by the extended submodule of the layered wave sensing domain. For the number of iterations, The weighting coefficients for the first derivative loss in the error terms of the progressively expanding modules of the wave sensing domain and the continuity constraint error terms are given. The weighting coefficients for the second derivative loss in the error terms of the progressively expanding modules of the wave sensing domain and the continuity constraint error terms are given. For the first In the feature vector output by the extended submodule of the wave sensing domain Grid point data, For the first Interpolation weights for the boundary extension function of the extended submodule of the layered wave sensing domain. For the first The feature vector output by the extended submodule of the layered wave sensing domain. To expand the number of submodules in the wave sensing domain, For the first Visual converter network parameters of the extended submodule of the layered wave perception domain. Input for shipborne radar observation wave height.

5. The deep learning-based method for extending the wave perception domain of shipborne radar according to claim 1, characterized in that, In step S2, the construction of the wave continuity constraint loss function includes: Error term of the step-by-step expansion module: Step-by-step expansion of the wave sensing domain module After this, the output feature vector Corresponding actual observed wave field Error calculations are performed across the entire sea area to constrain the overall prediction results of the wave sensing domain progressively expanding modules; Continuity constraint error term: for the boundary between the extended region and the original observation region. The first derivative of the wave field output by the wave sensing domain progressively expanded model and the observed wave field at the boundary is calculated. With the second derivative ; Calculate the first derivative of the observed wave field at the inner boundary. With the second derivative and calculate and ,as well as and To minimize errors between them and ensure the continuity of the wave field at the boundary; The error of the progressively expanding module and the error of the continuity constraint are weighted and combined to form the total loss function; ； In the formula, For the total loss, For the sample size, For the first Number of samples The first Model output data and actual observed wave field data for each sample. The first The first and second derivatives of the wave field output by the sample model and the observed wave field at the boundary. The first The first and second derivatives of the observed wave field at the inner boundary of each sample.

6. The deep learning-based method for extending the wave perception domain of shipborne radar according to claim 1, characterized in that, In step S3, the training of the wave perception domain progressively expanding model includes: Wave field data observed by shipborne radar is input into the wave sensing domain progressively expanding model for training. The model is trained for a total of 100 rounds, with 64 sets of data input in each batch. Simultaneously, a cosine annealing algorithm is introduced to dynamically schedule the learning rate, with an initial learning rate of... Eventually decayed to ; In the formula, For training the first The learning rate of the step. These are the minimum and maximum values ​​of the learning rate, respectively. This represents the current iteration step. This represents the total number of iterations.

7. The deep learning-based method for extending the wave perception domain of shipborne radar according to claim 1, characterized in that, In step S3, the wave expansion is followed by wave expansion result accuracy verification; Using root mean square error (RMSE) and correlation coefficient As an accuracy evaluation index, the wave perception domain expansion effect of the wave perception domain progressive expansion model is evaluated.

8. A deep learning-based shipborne radar wave perception domain extension system, characterized in that, The system is implemented using the deep learning-based shipborne radar wave perception domain extension method as described in any one of claims 1-7, and the system includes: The model dataset construction module is configured to acquire wave field data observed by shipborne radar and wave field analysis data that covers and exceeds the observation range of the shipborne radar, and generate a dataset that matches the shipborne radar observed wave field with the target extended wave field through spatiotemporal matching and sea area truncation operations. The wave perception domain progressive expansion model construction and training module is configured to construct a wave perception domain progressive expansion model. It expands the local wave field to a limited range through a wave perception domain expansion sub-module and inputs the data into a visual converter network for spatial feature extraction and modeling. The local wave field is progressively expanded in stages through the wave perception domain progressive expansion module, with residual connections between each sub-module. A wave continuity constraint loss function is constructed, which comprehensively considers the overall accuracy of the expansion result and the physical continuity at the boundary between the expanded region and the original observation region. The model is then trained using the constructed dataset. The wave extension and verification module is configured to extend the local wave field observed by shipborne radar using a trained wave perception domain hierarchical extension model, and to verify the reliability of the extension results using accuracy indicators. Among them, any one or more of the model dataset construction module, the wave perception domain progressively expanded model construction and training module, and the wave expansion and verification module are implemented by the processor executing computer program instructions stored in the memory.

Images