A sea surface wind speed retrieval method, system, device and medium combining the advantages of simulation model and satellite data

Abstract

A method, system, device, and medium for sea surface wind speed inversion combining the advantages of simulation models and satellite data are disclosed. The method generates simulated time-delay Doppler (DDM) images by inputting parameters such as satellite and receiver positions, velocities, and receiving antenna gain. The simulated images are then combined with satellite measured data to construct a joint measured-simulated dataset. Leading edge and trailing edge slopes are extracted from this dataset, and the leading-and trailing edge composite feature value is calculated to invert sea surface wind speed values. The system, device, and medium are used to implement this method. This invention uses the position and velocity of the satellite and receiver, wind speed, GNSS signal frequency, and receiving antenna gain parameters to simulate time-delay Doppler images. It can expand data samples even when measured data samples are scarce in specific sea areas or sea states, offering advantages such as wide applicability and relatively accurate sea surface wind speed inversion even with scarce samples.

Description

Technical Field

[0001] This invention belongs to the field of marine data processing technology, and specifically relates to a method, system, equipment and medium for sea surface wind speed inversion that combines the advantages of simulation models and satellite data. Background Technology

[0002] Traditional methods for monitoring sea surface wind speed (such as buoy measurements, radar measurements, and weather station observations) are difficult to implement over large areas due to high costs and limited observation ranges. Global Navigation Satellite System-Reflectometry (GNSS-R), with its advantages of low cost, all-weather operation, and global coverage, has become an ideal means of monitoring sea surface wind speed.

[0003] Patent application CN114861537A discloses a GNSS-R sea surface wind speed inversion method and system based on CNN multi-information fusion. The method uses a deep learning approach of CNN multi-information fusion to invert sea surface wind speed. However, due to the large amount of data required by the deep learning inversion method, the accuracy of sea surface wind speed inversion decreases when studying specific sea areas or sea conditions due to the scarcity of measured data samples.

[0004] Patent application CN117872424B discloses a method and apparatus for generating GNSS-R sea surface DDM data. It uses a flow model to mine the transformation relationship between the distribution of measured DDM data and the normal distribution, and uses random sampling of the normal distribution to obtain variables to generate DDM data. However, since the generated DDM data lacks physical basis, it has the problem of relying on the input measured data and lacking real sample details. Summary of the Invention

[0005] To overcome the shortcomings of the prior art, the present invention aims to provide a method, system, device, and medium for sea surface wind speed inversion that combines the advantages of simulation models and satellite data. By simulating the time delay Doppler map using the position and velocity of the satellite and receiver, wind speed, GNSS signal frequency, and receiving antenna gain parameters, the invention can expand the data sample even when measured data samples are scarce in a specific sea area or sea state. It has the advantages of wide applicability and relatively accurate sea surface wind speed inversion even when samples are scarce.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0007] A method for retrieving sea surface wind speed that combines the advantages of simulation models and satellite data includes the following steps:

[0008] Step 1: Construct a time-delay Doppler graph simulation model;

[0009] Step 2: Perform data processing and spatiotemporal matching on satellite data and buoy wind speed data to initially construct a GNSS-R measured dataset. Based on the time delay Doppler diagram simulation model constructed in Step 1, input the position and velocity of the satellite and receiver, wind speed, GNSS signal frequency, and receiving antenna gain parameters to perform simulation and construct simulated GNSS-R data. Combine the measured GNSS-R dataset with the simulated GNSS-R data to construct a joint measured-simulation dataset.

[0010] Step 3: Extract wind speed-related feature values ​​from the time-delay Doppler plot in the measured-simulation joint dataset constructed in Step 2. The wind speed-related feature values ​​include: leading edge slope, trailing edge slope, and time-delay Doppler mean. Combine the leading edge slope and trailing edge slope to calculate the leading edge-trailing edge composite feature value, and obtain the measured-simulation joint dataset containing the feature values.

[0011] Step 4: Based on a deep neural network, construct a deep neural network model that includes a time-delay Doppler image, eigenvalues, and other parameters such as signal-to-noise ratio, receiving antenna gain, incident angle, latitude and longitude coordinates of the specular reflection point, and the positions of the satellite and receiver. Train the measured-simulated joint dataset containing eigenvalues ​​constructed in Step 3 to obtain a trained sea surface wind speed inversion model. Input the time-delay Doppler image, eigenvalues, and other parameters such as signal-to-noise ratio, receiving antenna gain, incident angle, latitude and longitude coordinates of the specular reflection point, and the positions of the satellite and receiver to realize sea surface wind speed inversion.

[0012] Step 1 specifically includes:

[0013] Step 1.1: Calculate the coordinates of the reflection point on the mirror.

[0014] Given the positions and velocities of the satellite and receiver in the geocentric-ground-fixed coordinate system, and based on the characteristics that the mirror reflection point is located on the Earth's surface, the shortest path from the satellite to the receiver is through the mirror reflection point, and the incident angle at the mirror reflection point is equal to the reflection angle, the coordinates of the mirror reflection point are iteratively calculated using the shortest path method.

[0015] Step 1.2: Based on the coordinates of the mirror reflection point calculated in Step 1.1, construct a local coordinate system in the YOZ plane with the mirror reflection point as the center;

[0016] First, the position in the geocentric-ground-fixed coordinate system is multiplied by a transformation matrix to transform the geocentric-ground-fixed coordinate system to the northeast-sky coordinate system;

[0017] Then, transform the northeast-sky coordinate system to the local coordinate system centered at the specular reflection point; first calculate the angle between the y-axis of the northeast-sky coordinate system and the local coordinate system, and then calculate the projection of any point m(x,y,z) in the northeast-sky coordinate system onto the XOY plane. The formula for calculating the coordinates of the point in the local coordinate system after transformation is as follows:

[0018] m srf =[|m xy |sin(Δθ),|m xy |cos(Δθ),z]

[0019] Δθ=θ1-θ0

[0020] Where θ1 is the angle between the point and the y-axis of the northeast celestial coordinate system, θ0 is the angle between the point and the y-axis of the local coordinate system, and Δθ is the difference between θ1 and θ0.

[0021] Step 1.3: Construct a geometric model that closely resembles the real sea surface in the local coordinate system established in Step 1.2, and calculate the time delay Doppler value of each surface element;

[0022] In the constructed sea surface model, any scattering surface element S x,y The time delay and Doppler are expressed as:

[0023]

[0024] Where T and R are the positions of the satellite and receiver, c is the speed of light, and V is the speed of light. T and V R It is the velocity vector of the satellite and the receiver, e i It is the incident signal vector, e s It is the scattered signal vector;

[0025] Step 1.4: Calculate the scattering coefficient of each surface element based on the input GNSS signal frequency and satellite and receiver position parameters;

[0026] Step 1.5: Combining the receiving antenna gain, transmitted signal wavelength, and coherent integration time parameters, and based on the time delay Doppler value obtained in Step 1.3 and the scattering coefficient obtained in Step 1.4, calculate the simulated time delay Doppler diagram using the bistatic scattering signal correlation power model; the bistatic scattering signal correlation power model is expressed as:

[0027]

[0028] Among them, G R It is the receiving antenna gain, T i It is the coherent integration time, P T σ is the transmitted signal power, λ is the transmitted signal wavelength, and σ is the transmitted signal wavelength. 0 It is the scattering coefficient, GT It is the transmit antenna gain, R t R is the distance from the surface element to the satellite. r It is the distance from the surface element to the receiver, Λ 2 (τ-τ x,y ) is the PRN code coherence function, |S(f c -f x,y )| 2 It is the Doppler spread function.

[0029] Step 2 specifically includes:

[0030] Step 2.1: Filter the satellite data and buoy wind speed data, remove outliers such as missing or negative wind speeds in the buoy wind speed data, filter the satellite data whose specular reflection points are located in non-land and near-shore areas, and select satellite data with a peak signal-to-noise ratio greater than 3dB, a receiving antenna gain greater than 3dB, and an RCG value greater than 5.

[0031] Step 2.2: Match the satellite data obtained in Step 2.1 with the buoy wind speed data to obtain satellite-buoy wind speed matching data;

[0032] During the matching process, the spatial distance between the specular reflection point of the satellite data and the buoy station location of the buoy wind speed data is less than 7.5 kilometers, and satellite data with an observation time difference of less than 15 minutes are selected for matching with the buoy wind speed data;

[0033] Step 2.3: The satellite-buoy wind speed data obtained in Step 2.2 is denoised using the 3D matched filtering method to obtain satellite-buoy wind speed matched denoised data;

[0034] Step 2.4: Normalize the satellite-buoy wind speed matching and denoising data obtained in Step 2.3 to initially construct the GNSS-R measured dataset;

[0035] Step 2.5: Based on the time delay Doppler image simulation model constructed in Step 1, input the position and velocity of the satellite and receiver, wind speed, GNSS signal frequency and receiving antenna gain parameters to perform simulation and obtain simulated GNSS-R data. Combine the measured GNSS-R dataset with the simulated GNSS-R data to construct a joint measured-simulation dataset.

[0036] The specific method for step 3 includes:

[0037] Step 3.1: Extract the leading edge slope (LES), trailing edge slope (TES), and time delay Doppler mean eigenvalue (DDMA) from the time delay Doppler image in the measured-simulated joint dataset constructed in Step 2. The calculation formula is as follows:

[0038]

[0039] in, and The leading and trailing edge waveforms are represented using a time-delay Doppler plot at the center time delay waveform at a Doppler frequency of 0 Hz, obtained by taking time delays of -1.25 to 0 chips and 0 to 1.25 chips respectively; P DDM (i,j) represents the value of the time-delay Doppler graph at position (i,j); N represents the number of time-delay Doppler graph values ​​taken, n represents the number of time-delay waveform values ​​taken, and τ i It is the delay value at position i;

[0040] Step 3.2: Combining the leading edge slope (LES) and trailing edge slope (TES) extracted in Step 3.1, calculate the leading edge-trailing edge composite eigenvalue (Eigen). L-T The calculation formula is as follows:

[0041]

[0042] Where LES represents the leading edge slope and TES represents the trailing edge slope;

[0043] Combine the leading edge slope (LES), trailing edge slope (TES), and time delay Doppler mean (DDMA) calculated in step 3.1 with the leading edge-trailing edge composite eigenvalue (Eigen) calculated in step 3.2. L-T Add it to the test-simulation joint dataset constructed in step 2 to obtain the test-simulation joint dataset containing feature values.

[0044] The specific method for step 4 includes:

[0045] Step 4.1: Based on CNN and CNN_1D models, construct a deep neural network model that integrates time-delay Doppler images, eigenvalues, and other parameters such as signal-to-noise ratio, receiving antenna gain, incident angle, latitude and longitude coordinates of the specular reflection point, and the joint input of satellite and receiver positions. The time-delay Doppler image is input into a neural network consisting of three 3×3 convolutional kernels for training. The kernel size is 3×3, and the number of kernels is 256, 128, and 64 respectively. The training result is then flattened and passed through two FCN layers to obtain a 128×1 kernel. The first set of data; feature values ​​and other parameters, namely signal-to-noise ratio, receiving antenna gain, incident angle, latitude and longitude coordinates of the specular reflection point, and satellite and receiver positions, are input into a neural network composed of three layers of one-dimensional convolutional kernels for training. The convolutional kernel size is 3×1, and the number of convolutional kernels are 512, 256, and 128, respectively. After the training results are flattened and passed through two FCN layers, a second set of data of 128×1 is obtained. The two sets of data are fused based on an attention mechanism, and finally passed through two FCN layers to obtain the sea surface wind speed inversion result.

[0046] Step 4.2: Input the measured-simulation joint dataset containing feature values ​​constructed in Step 3 into the deep neural network model constructed in Step 4.1 for training to obtain the sea surface wind speed inversion model. Input the time delay Doppler image, feature values ​​and other parameters, namely signal-to-noise ratio, receiving antenna gain, incident angle, latitude and longitude coordinates of the specular reflection point, and the positions of the satellite and receiver, to realize the sea surface wind speed inversion. Finally, evaluate the sea surface wind speed inversion accuracy of the trained sea surface wind speed inversion model.

[0047] This invention also provides a sea surface wind speed inversion system that combines the advantages of simulation models and satellite data, comprising:

[0048] The model building module is used to build time-delay Doppler plot simulation models.

[0049] The measured-simulation joint dataset construction module is used to process and match satellite data and buoy wind speed data in time and space, initially constructing a GNSS-R measured dataset. Based on the time delay Doppler diagram simulation model, the module inputs the position and velocity of the satellite and receiver, wind speed, GNSS signal frequency, and receiving antenna gain parameters to perform simulation and construct simulated GNSS-R data. The measured GNSS-R dataset is then combined with the simulated GNSS-R data to construct the measured-simulation joint dataset.

[0050] The measured-simulation joint dataset construction module, which includes feature values, is used to extract wind speed-related feature values ​​from the time-delay Doppler image in the measured-simulation joint dataset. The wind speed-related feature values ​​include: leading edge slope, trailing edge slope, and time-delay Doppler mean. The leading edge slope and trailing edge slope are combined to calculate the leading edge-trailing edge composite feature value, thus obtaining the measured-simulation joint dataset containing feature values.

[0051] The sea surface wind speed inversion module is used to implement a deep neural network model based on a time-delay Doppler image, eigenvalues ​​and other parameters such as signal-to-noise ratio, receiving antenna gain, incident angle, latitude and longitude coordinates of the specular reflection point, and the positions of the satellite and receiver. The model is trained on a joint measured-simulated dataset containing eigenvalues ​​to obtain a trained sea surface wind speed inversion model. By inputting the time-delay Doppler image, eigenvalues ​​and other parameters such as signal-to-noise ratio, receiving antenna gain, incident angle, latitude and longitude coordinates of the specular reflection point, and the positions of the satellite and receiver, the sea surface wind speed can be inverted.

[0052] This invention also provides a sea surface wind speed inversion device that combines the advantages of simulation models and satellite data, comprising:

[0053] Memory: A computer program that stores the above-mentioned method for retrieving sea surface wind speed by combining the advantages of simulation models and satellite data, and is a computer-readable device;

[0054] Processor: Used to implement the sea surface wind speed inversion method that combines the advantages of simulation models and satellite data when executing the computer program.

[0055] The present invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, enables the implementation of the aforementioned method for inverting sea surface wind speed by combining the advantages of simulation models and satellite data.

[0056] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0057] 1. This invention simulates time-delay Doppler maps using physical parameters such as satellite and receiver position and velocity, receiving antenna gain, transmitted signal wavelength, and coherence integration time. Combined with a theoretical model, it simulates GNSS-R echo signals under different wind speeds and observation conditions. This method addresses the problem of scarce measured data samples when studying sea surface wind speed inversion based on GNSS-R under specific sea areas or sea states by expanding the data sample. Compared to traditional methods that rely on a large number of measured samples, this invention has good versatility and adaptability, and can still accurately invert sea surface wind speed even with limited measured samples, improving the reliability and application range of the model.

[0058] 2. Based on the study of time-delay waveforms, this invention extracts the leading edge slope, trailing edge slope, and time-delay Doppler mean characteristic values. It uses the combined influence of wind speed on the shape of the leading and trailing edges of the time-delay waveform as a key inversion feature, introducing a leading-and trailing edge composite characteristic value parameter. This composite characteristic value not only enhances the ability of wind speed changes to express the overall waveform shape but also effectively reduces errors caused by fluctuations in a single feature. By constructing the composite characteristic value, the modulation effect of wind speed changes on scattering characteristics can be more accurately characterized, thereby improving the sensitivity and stability of sea surface wind speed inversion.

[0059] In summary, this invention simulates time-delay Doppler images using physical parameters such as satellite and receiver position and velocity, and constructs eigenvalue combinations by combining leading-edge slope and trailing-edge slope. This can expand the data sample when measured data samples are scarce in specific sea areas or sea states, and better reflect the influence of wind speed. It has the advantages of wide applicability and relatively accurate sea surface wind speed inversion even when samples are scarce. Attached Figure Description

[0060] Figure 1 This is a flowchart of the sea surface wind speed inversion method according to an embodiment of the present invention.

[0061] Figure 2 This is a flowchart of the simulation delay Doppler plot of an embodiment of the present invention.

[0062] Figure 3This is a graph showing the DDM calculation results for different wind speeds according to an embodiment of the present invention. Figure 3 (a) is the DDM diagram for a wind speed of 5 m / s. Figure 3 (b) is the DDM with a wind speed of 20 m / s. Figure 3 (c) shows the time delay waveforms of the two DDMs. Figure 3 (d) is the normalized time delay waveform.

[0063] Figure 4 This is a flowchart of the measured data matching process according to an embodiment of the present invention.

[0064] Figure 5 These are comparison charts of measured and simulated DDM parameters for different embodiments of the present invention, wherein... Figure 5 (a) is the measured image of sample DDM1. Figure 5 (b) is a simulation diagram of sample DDM1. Figure 5 (c) is the measured image of sample DDM2. Figure 5 (d) is the simulation diagram of sample DDM2.

[0065] Figure 6 This is a comparison diagram of the measured and simulated normalized integral time delay waveforms according to an embodiment of the present invention, wherein... Figure 6 (a) shows the normalized integral time delay waveforms of the sample DDM1, both measured and simulated. Figure 6 (b) is the normalized integral time delay waveform of the sample DDM2, both measured and simulated.

[0066] Figure 7 This is a comparison chart of simulated and measured feature values ​​of an embodiment of the present invention, wherein, Figure 7 (a) is a distribution map of the measured front-back composite eigenvalues ​​at different wind speeds. Figure 7 (b) is a distribution diagram of the composite eigenvalues ​​of the leading and trailing edges in the simulation data at different wind speeds. Figure 7 (c) is a comparison of the leading-backward composite eigenvalues ​​of the measured data and the simulation data.

[0067] Figure 8 This is a comparison chart of inverted wind speeds and measured wind speeds from buoys for different datasets according to embodiments of the present invention. Figure 8 (a) is a comparison chart of wind speed retrieved by the model trained using measured data and wind speed on the buoy. Figure 8 (b) is a comparison chart of wind speed inversion from the model trained using measured and simulated data and buoy wind speed.

[0068] Figure 9 This is a graph showing the root mean square error and mean absolute error for different wind speed ranges according to an embodiment of the present invention. Figure 9 (a) is a root mean square error plot for different wind speed ranges. Figure 9 (b) is a graph of the average absolute error for different wind speed ranges. Detailed Implementation

[0069] The present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0070] This invention mainly studies the sea surface wind speed inversion method based on measured data and simulated GNSS-R data collected by Fengyun-3E satellite (FY-3E), providing theoretical and methodological support for applications such as marine environmental monitoring and marine target detection, and has important applied research significance.

[0071] This invention first starts with sea surface geometry and electromagnetic scattering theory to achieve GNSS-R simulation modeling that considers satellite and receiver positions and velocities, wind speed, GNSS signal frequency, and receiving antenna gain parameters, obtaining time-delay Doppler map calculation results that more closely resemble real-world scenarios. Data processing is performed on FY-3E satellite data and buoy wind speed data, followed by rigorous spatiotemporal matching to initially construct a GNSS-R measured dataset. Addressing the real-world problem of scarce measured data samples for specific sea areas or sea states, a combined measured-simulation dataset is constructed by combining measured and simulated data. Then, wind speed-related feature values ​​(leading edge slope, trailing edge slope, and time-delay Doppler mean) are extracted from the time-delay Doppler map, and the leading-and trailing edge composite feature value is calculated by combining the leading and trailing edge slopes. Finally, a deep neural network is used to train the combined measured-simulation dataset to obtain a sea surface wind speed inversion model, achieving efficient and widely applicable sea surface wind speed inversion.

[0072] See Figure 1 A method for retrieving sea surface wind speed that combines the advantages of simulation models and satellite data includes the following steps:

[0073] Step 1, see Figure 2 Construct a time-delay Doppler graph simulation model

[0074] Step 1 specifically includes:

[0075] Step 1.1: Calculate the coordinates of the reflection point on the mirror.

[0076] Given the positions and velocities of the satellite and receiver in the Earth-centered, Earth-fixed coordinate system, and based on the characteristics that the specular reflection point is located on the Earth's surface, the shortest path from the satellite to the receiver is through the specular reflection point, and the incident angle at the specular reflection point equals the reflection angle, the coordinates of the specular reflection point are iteratively calculated using the shortest path method. The formula for calculating the path length from the satellite to the receiver through the reflection point is as follows:

[0077] f(S) = |T ecef -S|+|R ecef -S|

[0078]

[0079] Among them, T ecef R ecef Let S and S represent the positions of the satellite, receiver, and reflection point in the Earth-centered Earth-fixed coordinate system, respectively. The shortest path f(S) can be obtained through iteration. sp )=|T ecef -S sp |+|R ecef -S sp | The coordinates of the reflection point S corresponding to the shortest path sp These are the coordinates of the point of reflection on the mirror.

[0080] Step 1.2: Based on the coordinates of the mirror reflection point calculated in Step 1.1, construct a local coordinate system in the YOZ plane with the mirror reflection point as the center;

[0081] First, the position in the geocentric-ground-fixed coordinate system is multiplied by a transformation matrix to transform the geocentric-ground-fixed coordinate system to the northeast-sky coordinate system. The transformation process is as follows:

[0082]

[0083] Where, λ s It is the longitude of the point of reflection of the mirror. It is the latitude of the point of mirror reflection;

[0084] Then, the northeast-sky coordinate system is transformed into a local coordinate system centered on the specular reflection point. The difference between the northeast-sky and local coordinate systems lies in the directions of the x and y axes. The y-axis of the local coordinate system is in the same direction as the projection of the receiver-to-satellite vector onto the XOY plane. The transformation process requires first calculating the angle between the y-axis of the northeast-sky and local coordinate systems, and the projection of any point m(x,y,z) in the northeast-sky coordinate system onto the XOY plane. The formula for calculating the coordinates of the point in the local coordinate system after transformation is as follows:

[0085] m srf =[|m xy |sin(Δθ),|m xy |cos(Δθ),z]

[0086] Δθ=θ1-θ0

[0087] Where θ1 is the angle between the point and the y-axis of the northeast celestial coordinate system, θ0 is the angle between the point and the y-axis of the local coordinate system, and Δθ is the difference between θ1 and θ0.

[0088] Step 1.3: Construct a geometric model that closely resembles the real sea surface in the local coordinate system established in Step 1.2, and calculate the time delay Doppler value of each surface element;

[0089] The linear filtering method is used to construct the geometric model of the sea surface. The formula for the linear filtering method is as follows;

[0090]

[0091] Among them, S(k) is the sea surface spectral function, which is used to construct sea surface models for different sea conditions. It is the direction function, k is the wave number, δk x and δk y ω is the discrete interval of the wavenumber in the x and y directions, ω is the angular frequency of the incident electromagnetic wave, and t is time;

[0092] In the constructed sea surface model, any scattering surface element S x,y The time delay and Doppler are expressed as:

[0093]

[0094] Where T and R are the positions of the satellite and receiver, c is the speed of light, and V is the speed of light. T and V R It is the velocity vector of the satellite and the receiver, e i It is the incident signal vector, e s It is the scattered signal vector; steps

[0095] 1.4. Based on the input GNSS signal frequency and satellite and receiver position parameters, calculate the scattering coefficient of each surface element; the formula for calculating the scattering coefficient is as follows:

[0096]

[0097] Where q is the scattering vector, q z It is the z-direction component of the scattering vector, q ⊥ It is the perpendicular component of the scattering vector, and π is pi. It is the Fresnel scattering coefficient. It is the probability density function, calculated using the sea surface geometry model constructed in step 1.3. The calculation formula is as follows:

[0098]

[0099]

[0100] c 21 =0.01-0.0086·u 10 ,c 03 =0.04-0.0330·u 10

[0101] c 40 =0.40,c 22 =0.12,c04 =0.23

[0102] Where, σ u and σ c These are the mean square slopes of the sea surface under headwind and crosswind conditions, u 10 The wind speed at a height of 10 meters above sea level, q x q y and q z These are the components of the scattering vector in the x, y, and z directions;

[0103] Step 1.5: Combining the receiving antenna gain, transmitted signal wavelength, and coherent integration time parameters, and based on the time delay Doppler value obtained in Step 1.3 and the scattering coefficient obtained in Step 1.4, calculate the simulated time delay Doppler diagram using the bistatic scattering signal correlation power model; the bistatic scattering signal correlation power model is expressed as:

[0104]

[0105] Among them, G R It is the receiving antenna gain, T i It is the coherent integration time, P T σ is the transmitted signal power, λ is the transmitted signal wavelength, and σ is the transmitted signal wavelength. 0 It is the scattering coefficient, G T It is the transmit antenna gain, R t R is the distance from the surface element to the satellite. r It is the distance from the surface element to the receiver, Λ 2 (τ-τ x,y ) is the PRN code coherence function, |S(f c -f x,y )| 2 It is the Doppler spread function.

[0106] Step 2, see Figure 4 Data processing and spatiotemporal matching were performed on FY-3E satellite data and buoy wind speed data to initially construct a GNSS-R measured dataset. Based on the time-delay Doppler diagram simulation model constructed in step 1, the positions and velocities of the satellite and receiver, wind speed, GNSS signal frequency, and receiving antenna gain parameters were input for simulation to construct simulated GNSS-R data. The measured GNSS-R dataset was then combined with the simulated GNSS-R data to construct a joint measured-simulation dataset.

[0107] Step 2 specifically includes:

[0108] Step 2.1: Filter the FY-3E satellite data and buoy wind speed data, remove outliers such as missing or negative wind speeds in the buoy wind speed data, filter the FY-3E satellite data where the specular reflection point is located in non-land and near-shore areas, and select FY-3E satellite data with a peak signal-to-noise ratio greater than 3dB, a receiving antenna gain greater than 3dB, and an RCG value greater than 5.

[0109] Step 2.2: Match the FY-3E satellite data obtained in Step 2.1 with the buoy wind speed data to obtain satellite-buoy wind speed matching data;

[0110] During the matching process, it is necessary to ensure that the spatial distance between the specular reflection point of the FY-3E satellite data and the buoy station location of the buoy wind speed data is less than 7.5 kilometers. Then, select FY-3E satellite data with an observation time difference of less than 15 minutes to match the buoy wind speed data. The spatial distance and observation time difference can be determined based on the data characteristics (such as satellite and receiver speed, coherence integration time) and subsequent application requirements. Typically, the spatial distance is within 7 to 10 kilometers and the time difference is within 10 to 60 minutes. The specific thresholds can be adjusted according to the actual situation or determined through experiments.

[0111] The time delay Doppler images in the satellite-buoy wind speed matching data obtained in steps 2.3 and 2.2 include noise from the equipment and the environment. The satellite-buoy wind speed data obtained in step 2.2 is denoised using the 3D matched filtering method to obtain the satellite-buoy wind speed matching denoised data.

[0112] Step 2.4: In order to carry out deep learning in the future, the satellite-buoy wind speed matching and denoising data obtained in Step 2.3 is normalized to initially construct the GNSS-R measured dataset;

[0113] Step 2.5: Based on the time delay Doppler image simulation model constructed in Step 1, input the position and velocity of the satellite and receiver, wind speed, GNSS signal frequency and receiving antenna gain parameters to perform simulation and obtain simulated GNSS-R data. Combine the measured GNSS-R dataset with the simulated GNSS-R data to construct a joint measured-simulation dataset.

[0114] Flowchart as follows Figure 4 As shown:

[0115] Step 3: Extract wind speed-related feature values ​​from the time-delay Doppler plot in the measured-simulation joint dataset constructed in Step 2. The wind speed-related feature values ​​include: leading edge slope, trailing edge slope, and time-delay Doppler mean. Combine the leading edge slope and trailing edge slope to calculate the leading edge-trailing edge composite feature value, and obtain the measured-simulation joint dataset containing the feature values.

[0116] The specific method for step 3 includes:

[0117] Step 3.1: Extract the leading edge slope (LES), trailing edge slope (TES), and time delay Doppler mean eigenvalue (DDMA) from the time delay Doppler image in the measured-simulated joint dataset constructed in Step 2. The calculation formula is as follows:

[0118]

[0119]

[0120] in, and The leading and trailing edge waveforms are represented using a time-delay Doppler plot at the center time delay waveform at a Doppler frequency of 0 Hz, obtained by taking time delays of -1.25 to 0 chips and 0 to 1.25 chips respectively; P DDM (i,j) represents the value of the time-delay Doppler graph at position (i,j); N represents the number of time-delay Doppler graph values ​​taken, n represents the number of time-delay waveform values ​​taken, and τ i It is the delay value at position i;

[0121] Step 3.2: Combining the leading edge slope (LES) and trailing edge slope (TES) extracted in Step 3.1, calculate the leading edge-trailing edge composite eigenvalue; the calculation formula is as follows:

[0122]

[0123] Where LES represents the leading edge slope and TES represents the trailing edge slope;

[0124] The leading edge slope (LES), trailing edge slope (TES), time delay Doppler mean (DDMA) calculated in step 3.1, and the leading edge-trailing edge composite eigenvalues ​​calculated in step 3.2 are added to the experimental-simulation joint dataset constructed in step 2 to obtain the experimental-simulation joint dataset containing eigenvalues.

[0125] Step 4: Based on a deep neural network, construct a deep neural network model that includes a time-delay Doppler image, eigenvalues, and other parameters such as signal-to-noise ratio, receiving antenna gain, incident angle, latitude and longitude coordinates of the specular reflection point, and the positions of the satellite and receiver. Train the measured-simulated joint dataset containing eigenvalues ​​constructed in Step 3 to obtain a trained sea surface wind speed inversion model. Input the time-delay Doppler image, eigenvalues, and other parameters such as signal-to-noise ratio, receiving antenna gain, incident angle, latitude and longitude coordinates of the specular reflection point, and the positions of the satellite and receiver to realize sea surface wind speed inversion.

[0126] The specific method for step 4 includes:

[0127] Step 4.1: Based on CNN and CNN_1D models, construct a deep neural network model that integrates time-delay Doppler images, eigenvalues, and other parameters such as signal-to-noise ratio, receiving antenna gain, incident angle, latitude and longitude coordinates of the specular reflection point, and the joint input of satellite and receiver positions. The time-delay Doppler image is input into a neural network consisting of three 3×3 convolutional kernels for training. The kernel size is 3×3, and the number of kernels is 256, 128, and 64 respectively. The training result is then flattened and passed through two FCN layers to obtain a 128×1 kernel. The first set of data; feature values ​​and other parameters, namely signal-to-noise ratio, receiving antenna gain, incident angle, latitude and longitude coordinates of the specular reflection point, and satellite and receiver positions, are input into a neural network composed of three layers of one-dimensional convolutional kernels for training. The convolutional kernel size is 3×1, and the number of convolutional kernels are 512, 256, and 128, respectively. After the training results are flattened and passed through two FCN layers, a second set of data of 128×1 is obtained. The two sets of data are fused based on an attention mechanism, and finally passed through two FCN layers to obtain the sea surface wind speed inversion result.

[0128] During training, the initial learning rate was set to 0.001, and the learning rate was dynamically adjusted as training progressed. The random deactivation rate was set to 0.1 to improve the model's generalization ability.

[0129] Step 4.2: Input the measured-simulation joint dataset containing feature values ​​constructed in Step 3 into the deep neural network model constructed in Step 4.1 for training to obtain the sea surface wind speed inversion model. Input the time delay Doppler image, feature values ​​and other parameters, namely signal-to-noise ratio, receiving antenna gain, incident angle, latitude and longitude coordinates of the specular reflection point, and the positions of the satellite and receiver, to realize the sea surface wind speed inversion. Finally, evaluate the sea surface wind speed inversion accuracy of the trained sea surface wind speed inversion model.

[0130] Example

[0131] A method for retrieving sea surface wind speed that combines the advantages of simulation models and satellite data includes the following steps:

[0132] Step 1: Construct a time-delay Doppler graph simulation model

[0133] Time-delay Doppler images are crucial for GNSS-R research, containing rich remote sensing information about the Earth's surface. They can be used for sea surface wind speed retrieval, sea ice detection, soil moisture detection, forest fire early warning, and more. Therefore, a key aspect of GNSS-R technology research is achieving simulation of time-delay Doppler images under different parameter conditions. This invention, based on sea surface geometry and electromagnetic scattering theory, realizes time-delay Doppler image simulation modeling considering parameters such as satellite and receiver position, velocity, and antenna gain.

[0134] Real-world ocean waves are complex and variable. This invention uses a sea surface spectral function combined with sea surface parameters to perform geometric modeling of the sea surface. Based on the frequency domain distribution characteristics of ocean waves, a Fourier transform is used to obtain a sea surface model that more closely resembles the real-world scenario. When studying time-delay Doppler maps of special sea conditions such as mixed wind-wave and swell sea surfaces, modifications to the sea surface simulation can be made to conduct research on specific sea surface conditions, effectively overcoming the limitations caused by the difficulty in obtaining satellite measurement data for various specific sea conditions.

[0135] Commonly used theoretical models for sea surface electromagnetic scattering mainly include analytical approximation methods and numerical methods. Numerical methods offer high accuracy but are computationally complex and inefficient, consuming significant computational resources when dealing with large-scale sea surface electromagnetic scattering problems. Analytical approximation methods, using physical approximations, reduce computational resource consumption and consider key physical mechanisms for specific situations during the approximation process, resulting in good modeling accuracy. This invention employs the Kirchhoff approximation-geometric optics (KA-GO) method within analytical approximation for sea surface electromagnetic scattering modeling.

[0136] In GNSS-R data obtained from satellite measurements, the positions of the satellite and receiver are typically given in a geocentric-fixed (ECEF) coordinate system. Therefore, to compare the simulated time-delay Doppler map results with the measured data, it is necessary to start from the satellite and receiver positions in the ECEF coordinate system, calculate the coordinates of the specular reflection point, and construct a local coordinate system centered on the specular reflection point. Within the local coordinate system, a geometric model that more closely resembles the real sea surface is constructed, and the time-delay Doppler value for each surface element is calculated. Then, the scattering coefficient of each surface element is calculated. Combining the receiving antenna gain, transmitted signal wavelength, and coherent integration time parameters, a bistatic scattering signal correlation power model is used to calculate the simulated time-delay Doppler map. The process for simulating the time-delay Doppler map is as follows: Figure 2 As shown.

[0137] The following simulation uses GNSS-R parameters with an incident angle of 20° and a receiver altitude of 500km as an example to simulate time-delay Doppler plots for different wind speeds. The simulated wind speeds at a height of 10m above the sea surface are 5m / s and 20m / s, and the GNSS signal frequency is 1.57GHz. Figure 3 The results of time delay Doppler plots and time delay waveform calculations under different wind speeds are presented. Figure 3 (a) and (b) are two-dimensional top views of the time-delay Doppler plots for wind speeds of 5 m / s and 20 m / s, respectively. By comparison... Figure 3 As can be seen from (a) and (b), the higher the wind speed, the lower the overall power of the time-delay Doppler graph, and the smaller the proportion of power near the peak to the total power. This is because as the wind speed increases, the roughness of the sea surface increases, diffuse reflection intensifies, and the intensity of the reflected signal near the specular reflection point decreases. Figure 3(c) and (d) are the time delay waveforms of the time delay Doppler graph at a Doppler frequency of 0Hz, respectively. Figure 3 (d) shows the normalized time delay waveform. It can be seen that the lower the wind speed, the faster the normalized time delay waveform decreases. If the wind speed drops to 0 m / s, the sea surface will act like a mirror, and only the reflection point will reflect the signal back.

[0138] Step 2: Perform data processing and spatiotemporal matching on FY-3E satellite data and buoy wind speed data to initially construct a GNSS-R measured dataset. Based on the time delay Doppler diagram simulation model constructed in Step 1, input the positions and velocities of the satellite and receiver, wind speed, GNSS signal frequency, and receiving antenna gain parameters to perform simulation and construct simulated GNSS-R data. Combine the measured GNSS-R dataset with the simulated GNSS-R data to construct a joint measured-simulation dataset.

[0139] The steps for constructing the GNSS-R measured dataset include: (1) Data filtering. Outliers and invalid data points are removed, and quality control is performed; (2) Spatiotemporal matching. FY-3E satellite observation data is matched with buoy wind speed data to ensure that the time difference and spatial error between the two are within a reasonable range; (3) Data denoising; (4) Data normalization. The flowchart is as follows. Figure 4 As shown.

[0140] To address the issue of scarce and difficult-to-obtain measured data samples in certain research scenarios (such as specific regions, times, and sea states), simulations were considered to expand the dataset. Therefore, further verification of the accuracy of the simulation data is necessary. Because the Fengyun-3E satellite data is non-uniformly sampled in the time delay direction, the time delay resolution near the specular reflection point is 0.125 chips, while in regions far from the specular reflection point it is 0.25 chips. That is, the time delay interval in the range of -2.875 to 2.875 chips around the specular reflection point is 0.125 chips, and in other intervals it is 0.25 chips. Therefore, to compare the simulated data with the measured data, the simulated data first needs to be resampled according to the time delay interval of the Fengyun satellite data. Two DDMs were selected for comparison. The measured wind speeds of these two DDMs were 6.60 and 17.15 m / s, respectively, representing a wide wind speed range and thus being relatively representative. The simulation parameters were derived from the measured data. Table 1 shows the specific parameter settings, such as the data detection time, the position and speed of the launching satellite and receiver, and the wind speed in the observation area.

[0141] Table 1 Simulation Calculation Parameter Settings

[0142]

[0143] To facilitate the evaluation of the accuracy of the simulation results, the root mean square error (RMSE) and structural similarity (SSIM) of the simulated and measured data near the mirror reflection point in the time-delay Doppler image are used to assess the accuracy of the simulation results. The calculation formulas are as follows:

[0144]

[0145] The smaller the root mean square error, the closer the structural similarity value is to 100%, indicating that the real time-delay Doppler image and the simulated time-delay Doppler image tend to be consistent.

[0146] Table 2 presents the comparison results between simulated and measured time-delay Doppler plots. Here, the simulated time-delay Doppler plot results are compared with the denoised measured satellite data. It can be seen that the RMSE between the simulated and measured data is less than 0.06, and the SSIM is greater than 80%, indicating that the difference between the two is not significant. Comparing the leading edge slopes of the two datasets reveals that the LES values ​​are basically consistent and not significantly different. Figure 5 These are comparison charts of measured and simulated DDM with different parameters. Figure 5 (a) and (b) are the measured and simulated graphs of sample DDM1. Figure 5 (c) and (d) are the measured and simulated figures of sample DDM2. Figure 6 This is a comparison chart of the measured and simulated normalized integral time delay waveforms of two DDM samples. Figure 6 (a) shows the normalized integral time delay waveforms of the sample DDM1, both measured and simulated. Figure 6 (b) shows the normalized integral time delay waveforms of the sample DDM2, both measured and simulated. (Comparison) Figure 5 (a) Figure 5 (b) Figure 5 (c) and Figure 5 (d) It can be seen that, Figure 5 (a) and Figure 5 (c) The measured DDM data has a lot of noise interference, but the overall trends, including the peak point, Doppler broadening, and power value decrease trend, are consistent with... Figure 5 (b) and Figure 5 The simulation data in (d) are consistent. The simulation results and the measured data have good consistency, and the time delay Doppler map can be calculated using simulation to construct a joint measured-simulation dataset.

[0147] Table 2 Comparison of simulation results and measured data

[0148]

[0149] Step 3: Extract wind speed-related feature values ​​from the time-delay Doppler plot in the measured-simulation joint dataset constructed in Step 2. The wind speed-related feature values ​​include: leading edge slope, trailing edge slope, and time-delay Doppler mean. Combine the leading edge slope and trailing edge slope to calculate the leading edge-trailing edge composite feature value, and obtain the measured-simulation joint dataset containing the feature values.

[0150] Further calculations were performed on the time-delay Doppler image to extract the leading-edge slope, trailing-edge slope, and the mean time-delay Doppler value. These were then combined with the leading-edge and trailing-edge slopes to calculate the leading-edge-trailing-edge composite eigenvalue. The calculation formula is as follows:

[0151]

[0152]

[0153] Figure 7 A comparison of the leading-tailed composite eigenvalues ​​calculated through simulation and field measurements is presented. Figure 7 (a) is a distribution map of the measured front-back composite eigenvalues ​​at different wind speeds. Figure 7 (b) is a distribution diagram of the composite eigenvalues ​​of the leading and trailing edges in the simulation data at different wind speeds. (Comparison) Figure 7 (a) and Figure 7 (b) It can be seen that the distribution trend of the front slope of the simulation and the measured data is the same, both decreasing as the wind speed increases. The simulation data is more concentrated, while the measured data is more dispersed due to noise and other factors. However, they are all distributed in the same area. Figure 7 (c) is a comparison of the leading-backward composite eigenvalues ​​of the simulation and measured data. It can be seen that the data points are basically distributed near the contour lines, which further illustrates that the simulation data and the measured data are consistent. The simulation data can be used to expand the sample when the measured data is insufficient.

[0154] The calculated leading edge slope, trailing edge slope, time delay Doppler mean, and leading edge-trailing edge composite eigenvalue are added to the measured-simulation joint dataset constructed in step 2 to obtain the measured-simulation joint dataset containing eigenvalues.

[0155] Step 4: Based on a deep neural network, construct a deep neural network model that includes a time-delay Doppler image, eigenvalues, and other parameters such as signal-to-noise ratio, receiving antenna gain, incident angle, latitude and longitude coordinates of the specular reflection point, and the positions of the satellite and receiver. Train the measured-simulated joint dataset containing eigenvalues ​​constructed in Step 3 to obtain a trained sea surface wind speed inversion model. Input the time-delay Doppler image, eigenvalues, and other parameters such as signal-to-noise ratio, receiving antenna gain, incident angle, latitude and longitude coordinates of the specular reflection point, and the positions of the satellite and receiver to realize sea surface wind speed inversion.

[0156] GNNS-R, due to its low cost, all-weather capability, and global coverage, is suitable for global wind speed monitoring. Its application in both routine and military observations can effectively save manpower and resources. Therefore, retrieving actual sea surface wind speeds from time-delayed Doppler images has significant research value. Traditional wind speed retrieval methods typically rely on satellite-measured data, requiring large datasets, which are often difficult to obtain. Based on the measured-simulation joint dataset containing feature values ​​constructed in step 3, a deep learning neural network model is introduced to propose a sea surface wind speed retrieval method with fewer limitations and applicable to situations with limited observation samples.

[0157] This invention constructs a Ddm-Eigen model based on CNN and CNN_1D models, using time-delay Doppler maps, eigenvalues, and other parameters such as signal-to-noise ratio, receiving antenna gain, incident angle, latitude and longitude coordinates of the specular reflection point, and the joint input of satellite and receiver positions. The Ddm-Eigen model uses two-dimensional convolutional kernels to extract spatial features of the DDM power map and one-dimensional convolutional kernels to process eigenvalues ​​and other parameters. Both are trained separately, and the training results are flattened and passed through two FCN layers to obtain two sets of data of the same dimension. These two sets of data are then fused using an attention mechanism, and finally, the sea surface wind speed inversion result is calculated through two FCN layers.

[0158] First, the model was trained using a small sample of measured data to study its effectiveness in retrieving sea surface wind speed when limited measured data was available. Then, simulated data was added to the training set, without altering the test and validation sets, and the model was trained again. Table 3 presents a statistical comparison of the accuracy of the model in retrieving wind speed using only measured data and using simulated data to augment the dataset. Figure 8 This is a comparison chart of the wind speed derived from the model and the wind speed measured by the buoy. Figure 8 (a) is a comparison chart of wind speed retrieved by the model trained using measured data and wind speed on the buoy. Figure 8 (b) is a comparison chart of wind speed retrieved from the model and buoy wind speed trained using measured and simulated data. As shown in Table 3, after adding simulated data to expand the dataset, the model improved the accuracy of sea surface wind speed retrieval, with both RMSE and MAE decreasing. Furthermore, the proportion of samples with absolute errors of less than 1 m / s and 2 m / s between the retrieved and actual wind speeds increased. Figure 8 As can be seen from (a) and (b), after adding simulation data, the wind speed retrieved by the model is more distributed near the y=x reference line, and the accuracy of the retrieved wind speed is improved.

[0159] Table 3. Accuracy of wind speed retrieval from different datasets

[0160]

[0161] To compare the inversion accuracy of the model across different wind speed ranges, Figure 9 The root mean square error (RMSE) and mean absolute error (MAE) for different wind speed ranges are given. Figure 9 (a) is a root mean square error plot for different wind speed ranges. Figure 9 (b) is a graph showing the average absolute error across different wind speed ranges. From Figure 9 As can be seen from (a) and (b), after using the simulation-expanded dataset, the root mean square error and mean absolute error of the model in all wind speed ranges were reduced, which indicates that the inversion effect was improved, especially the inversion accuracy in the high wind speed range (15-20 m / s) was improved most significantly.

[0162] This invention proposes a sea surface wind speed inversion method that combines the advantages of simulation models and satellite data. It introduces simulated time-delay Doppler images to expand the measured samples, thereby improving the limitations of traditional wind speed inversion methods based on measured GNSS-R data, which are limited by the scarcity of measured data samples for specific sea areas or sea conditions.

[0163] The sea surface wind speed inversion method implemented in this invention uses eigenvalues ​​that differ from the leading edge slope eigenvalues ​​used in traditional methods. It considers the combined effect of wind speed on the leading and trailing edges of the time delay waveform, and constructs a eigenvalue combination by combining the leading edge slope and the trailing edge slope, which can better reflect the influence of wind speed.

[0164] This invention also provides a sea surface wind speed inversion system that combines the advantages of simulation models and satellite data, comprising:

[0165] The model building module is used to build the time-delay Doppler diagram simulation model in step 1;

[0166] The measured-simulation joint dataset construction module is used to perform data processing and spatiotemporal matching on FY-3E satellite data and buoy wind speed data in step 2, and initially construct the GNSS-R measured dataset. Based on the time delay Doppler diagram simulation model constructed in step 1, the module inputs the position and velocity of the satellite and receiver, wind speed, GNSS signal frequency and receiving antenna gain parameters to perform simulation, construct simulated GNSS-R data, and combine the measured GNSS-R dataset with the simulated GNSS-R data to construct the measured-simulation joint dataset.

[0167] The measured-simulation joint dataset construction module containing feature values ​​is used to extract wind speed-related feature values ​​from the time-delay Doppler image in the measured-simulation joint dataset constructed in step 2 in step 3. The wind speed-related feature values ​​include: leading edge slope, trailing edge slope, and time-delay Doppler mean. Combining the leading edge slope and trailing edge slope, the leading edge-trailing edge composite feature value is calculated to obtain the measured-simulation joint dataset containing feature values.

[0168] The sea surface wind speed inversion module is used to implement the deep neural network model constructed in step 4, which is based on a deep neural network and includes a time delay Doppler image, eigenvalues, and other parameters such as signal-to-noise ratio, receiving antenna gain, incident angle, latitude and longitude coordinates of the specular reflection point, and the positions of the satellite and receiver. The module trains the measured-simulation joint dataset containing eigenvalues ​​constructed in step 3 to obtain a trained sea surface wind speed inversion model. The module inputs the time delay Doppler image, eigenvalues, and other parameters such as signal-to-noise ratio, receiving antenna gain, incident angle, latitude and longitude coordinates of the specular reflection point, and the positions of the satellite and receiver to realize the sea surface wind speed inversion.

[0169] This invention also provides a sea surface wind speed inversion device that combines the advantages of simulation models and satellite data, comprising:

[0170] Memory: A computer program that stores the above-mentioned method for retrieving sea surface wind speed by combining the advantages of simulation models and satellite data, and is a computer-readable device;

[0171] Processor: Used to implement the sea surface wind speed inversion method that combines the advantages of simulation models and satellite data when executing the computer program.

[0172] The present invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, enables the implementation of the aforementioned method for inverting sea surface wind speed by combining the advantages of simulation models and satellite data.

Claims

1. A method for retrieving sea surface wind speed by combining the advantages of simulation models and satellite data, characterized in that, Includes the following steps: Step 1: Construct a time-delay Doppler graph simulation model; Step 2: Perform data processing and spatiotemporal matching on satellite data and buoy wind speed data to initially construct a GNSS-R measured dataset. Based on the time delay Doppler diagram simulation model constructed in Step 1, input the position and velocity of the satellite and receiver, wind speed, GNSS signal frequency, and receiving antenna gain parameters to perform simulation and construct simulated GNSS-R data. Combine the measured GNSS-R dataset with the simulated GNSS-R data to construct a joint measured-simulation dataset. Step 3: Extract wind speed-related feature values ​​from the time-delay Doppler plot in the measured-simulation joint dataset constructed in Step 2. The wind speed-related feature values ​​include: leading edge slope, trailing edge slope, and time-delay Doppler mean. Combine the leading edge slope and trailing edge slope to calculate the leading edge-trailing edge composite feature value, and obtain the measured-simulation joint dataset containing the feature values. The specific method for step 3 includes: Step 3.1: Extract the leading edge slope (LES), trailing edge slope (TES), and time delay Doppler mean eigenvalue (DDMA) from the time delay Doppler image in the measured-simulated joint dataset constructed in Step 2. The calculation formula is as follows: in, and The leading edge and trailing edge waveforms are represented by the time-delay Doppler plot at the center time delay waveform at a Doppler frequency of 0 Hz, obtained by taking time delays of -1.25~0 chip and 0~1.25 chip respectively; This represents the value of the time-delay Doppler graph at position (i,j); N represents the number of time-delay Doppler graph values ​​taken. This indicates the number of time-delay waveform values ​​taken. It is the delay value at position i; Step 3.2: Combining the leading edge slope (LES) and trailing edge slope (TES) extracted in Step 3.1, calculate the leading edge-trailing edge composite eigenvalue. The calculation formula is as follows: Where LES represents the leading edge slope and TES represents the trailing edge slope; Combine the leading edge slope (LES), trailing edge slope (TES), and time delay Doppler mean (DDMA) calculated in step 3.1 with the leading edge-trailing edge composite eigenvalue calculated in step 3.

2. Add it to the test-simulation joint dataset constructed in step 2 to obtain the test-simulation joint dataset containing feature values; Step 4: Based on a deep neural network, construct a deep neural network model that includes a time-delay Doppler image, eigenvalues, and other parameters such as signal-to-noise ratio, receiving antenna gain, incident angle, latitude and longitude coordinates of the specular reflection point, and the positions of the satellite and receiver. Train the measured-simulated joint dataset containing eigenvalues ​​constructed in Step 3 to obtain a trained sea surface wind speed inversion model. Input the time-delay Doppler image, eigenvalues, and other parameters such as signal-to-noise ratio, receiving antenna gain, incident angle, latitude and longitude coordinates of the specular reflection point, and the positions of the satellite and receiver to realize sea surface wind speed inversion.

2. The sea surface wind speed inversion method combining the advantages of simulation models and satellite data according to claim 1, characterized in that, Step 1 specifically includes: Step 1.1: Calculate the coordinates of the reflection point on the mirror. Given the positions and velocities of the satellite and receiver in the geocentric-ground-fixed coordinate system, and based on the characteristics that the mirror reflection point is located on the Earth's surface, the shortest path from the satellite to the receiver is through the mirror reflection point, and the incident angle at the mirror reflection point is equal to the reflection angle, the coordinates of the mirror reflection point are iteratively calculated using the shortest path method. Step 1.2: Based on the coordinates of the mirror reflection point calculated in Step 1.1, construct a local coordinate system in the YOZ plane with the mirror reflection point as the center; First, the position in the geocentric-ground-fixed coordinate system is multiplied by a transformation matrix to transform the geocentric-ground-fixed coordinate system to the northeast-sky coordinate system; Then, transform the northeast-sky coordinate system to the local coordinate system centered at the mirror reflection point; first calculate the angle between the y-axis of the northeast-sky coordinate system and the local coordinate system, and then calculate the angle between the y-axis of any point in the northeast-sky coordinate system. Projection on the XOY plane The formula for calculating the coordinates of this point in the local coordinate system after transformation is as follows: in, It is the angle between the point and the y-axis of the northeast celestial coordinate system. It is the angle between the point and the y-axis of the local coordinate system. yes and The difference; Step 1.3: Construct a geometric model that closely resembles the real sea surface in the local coordinate system established in Step 1.2, and calculate the time delay Doppler value of each surface element; In the constructed sea surface model, any scattering surface element The time delay and Doppler are expressed as: in, and It refers to the location of the satellite and receiver. It's the speed of light. and It is the velocity vector of the satellite and the receiver. It is the incident signal vector. It is the scattered signal vector; Step 1.4: Calculate the scattering coefficient of each surface element based on the input GNSS signal frequency and satellite and receiver position parameters; Step 1.5: Combining the receiving antenna gain, transmitted signal wavelength, and coherent integration time parameters, and based on the time delay Doppler value obtained in Step 1.3 and the scattering coefficient obtained in Step 1.4, calculate the simulated time delay Doppler diagram using the bistatic scattering signal correlation power model; the bistatic scattering signal correlation power model is expressed as: in, It is the receiving antenna gain. It is the coherent integration time. It is the transmitted signal power. It is the wavelength of the transmitted signal. It is the scattering coefficient. It is the transmit antenna gain. It is the distance from the surface element to the satellite. It is the distance from the surface element to the receiver. It is a PRN code coherence function. It is the Doppler spread function.

3. The sea surface wind speed inversion method combining the advantages of simulation models and satellite data according to claim 1, characterized in that, Step 2 specifically includes: Step 2.1: Filter the satellite data and buoy wind speed data, remove outliers such as missing or negative wind speeds in the buoy wind speed data, filter the satellite data whose specular reflection points are located in non-land and near-shore areas, and select satellite data with a peak signal-to-noise ratio greater than 3dB, a receiving antenna gain greater than 3dB, and an RCG value greater than 5. Step 2.2: Match the satellite data obtained in Step 2.1 with the buoy wind speed data to obtain satellite-buoy wind speed matching data; During the matching process, the spatial distance between the specular reflection point of the satellite data and the buoy station location of the buoy wind speed data is less than 7.5 kilometers, and satellite data with an observation time difference of less than 15 minutes are selected for matching with the buoy wind speed data; Step 2.3: The satellite-buoy wind speed data obtained in Step 2.2 is denoised using the 3D matched filtering method to obtain satellite-buoy wind speed matched denoised data; Step 2.4: Normalize the satellite-buoy wind speed matching and denoising data obtained in Step 2.3 to initially construct the GNSS-R measured dataset; Step 2.5: Based on the time delay Doppler image simulation model constructed in Step 1, input the position and velocity of the satellite and receiver, wind speed, GNSS signal frequency and receiving antenna gain parameters to perform simulation and obtain simulated GNSS-R data. Combine the measured GNSS-R dataset with the simulated GNSS-R data to construct a joint measured-simulation dataset.

4. The sea surface wind speed inversion method combining the advantages of simulation models and satellite data according to claim 1, characterized in that, The specific method for step 4 includes: Step 4.1: Based on CNN and CNN_1D models, construct a deep neural network model that integrates time-delay Doppler images, eigenvalues, and other parameters such as signal-to-noise ratio, receiving antenna gain, incident angle, latitude and longitude coordinates of the specular reflection point, and the joint input of satellite and receiver positions. The time-delay Doppler image is input into a neural network consisting of three 3×3 convolutional kernels for training. The kernel size is 3×3, and the number of kernels is 256, 128, and 64 respectively. The training result is then flattened and passed through two FCN layers to obtain a 128×1 kernel. The first set of data; feature values ​​and other parameters, namely signal-to-noise ratio, receiving antenna gain, incident angle, latitude and longitude coordinates of the specular reflection point, and satellite and receiver positions, are input into a neural network composed of three layers of one-dimensional convolutional kernels for training. The convolutional kernel size is 3×1, and the number of convolutional kernels are 512, 256, and 128, respectively. After the training results are flattened and passed through two FCN layers, a second set of data of 128×1 is obtained. The two sets of data are fused based on an attention mechanism, and finally passed through two FCN layers to obtain the sea surface wind speed inversion result. Step 4.2: Input the measured-simulation joint dataset containing feature values ​​constructed in Step 3 into the deep neural network model constructed in Step 4.1 for training to obtain the sea surface wind speed inversion model. Input the time delay Doppler image, feature values ​​and other parameters, namely signal-to-noise ratio, receiving antenna gain, incident angle, latitude and longitude coordinates of the specular reflection point, and the positions of the satellite and receiver, to realize the sea surface wind speed inversion. Finally, evaluate the sea surface wind speed inversion accuracy of the trained sea surface wind speed inversion model.

5. A sea surface wind speed inversion system combining the advantages of simulation models and satellite data, based on the method described in any one of claims 1 to 4, characterized in that, include: The model building module is used to build time-delay Doppler plot simulation models. The measured-simulation joint dataset construction module is used to process and match satellite data and buoy wind speed data in time and space, initially constructing a GNSS-R measured dataset. Based on the time delay Doppler diagram simulation model, the module inputs the position and velocity of the satellite and receiver, wind speed, GNSS signal frequency, and receiving antenna gain parameters to perform simulation and construct simulated GNSS-R data. The measured GNSS-R dataset is then combined with the simulated GNSS-R data to construct the measured-simulation joint dataset. A measured-simulation joint dataset construction module containing feature values ​​is used to extract wind speed-related feature values ​​from the time-delay Doppler plot in the measured-simulation joint dataset. The wind speed-related feature values ​​include: leading edge slope, trailing edge slope, and time-delay Doppler mean. By combining the leading edge slope and the trailing edge slope, the leading edge-trailing edge composite eigenvalue is calculated to obtain a joint measured-simulated dataset containing the eigenvalues. The sea surface wind speed inversion module is used to implement a deep neural network model based on a time-delay Doppler image, eigenvalues ​​and other parameters such as signal-to-noise ratio, receiving antenna gain, incident angle, latitude and longitude coordinates of the specular reflection point, and the positions of the satellite and receiver. The model is trained on a joint measured-simulated dataset containing eigenvalues ​​to obtain a trained sea surface wind speed inversion model. By inputting the time-delay Doppler image, eigenvalues ​​and other parameters such as signal-to-noise ratio, receiving antenna gain, incident angle, latitude and longitude coordinates of the specular reflection point, and the positions of the satellite and receiver, the sea surface wind speed can be inverted.

6. A sea surface wind speed inversion device that combines the advantages of simulation models and satellite data, characterized in that, include: Memory: A computer program for a sea surface wind speed inversion method combining the advantages of simulation models and satellite data as described in any one of claims 1-4, and is a computer-readable device; Processor: Used to implement the sea surface wind speed inversion method that combines the advantages of simulation models and satellite data as described in any one of claims 1-4 when executing the computer program.

7. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, enables the implementation of the sea surface wind speed inversion method according to any one of claims 1-4, which combines the advantages of simulation models and satellite data.

Images