Synthetic aperture radar sea surface flow velocity inversion method and device based on machine learning

Abstract

The invention provides a synthetic aperture radar sea surface flow velocity inversion method and device based on machine learning. The method comprises the following steps: acquiring an original synthetic aperture radar image, and OCN data, mixed coordinate ocean model data and NDBC buoy data corresponding to the original synthetic aperture radar image; calculating a sea wave spectrum of the synthetic aperture radar image by using a PFSM algorithm, and exploring a relationship between the flow velocity and synthetic aperture radar image parameters; a synthetic aperture radar sea surface flow velocity inversion algorithm based on machine learning is constructed; and verifying the accuracy of the inversion algorithm. The invention provides an effective method for SAR sea surface flow velocity inversion.

Description

Technical Field

[0001] The present invention relates to the technical field of synthetic aperture radar, and in particular to a synthetic aperture radar sea surface current velocity inversion method and device based on machine learning. Background Art

[0002] The acceleration of ocean circulation over the past two decades is an important factor affecting global climate and is of great significance for understanding long-term changes in water temperature and salinity. Although observations from drifting buoys are quite reliable, these data are scarce in the high seas. With the development of oceanographic theory and computing technology, numerical models have been developed for ocean dynamics prediction and forecasting, namely WAVEWATCH-III (WW3) and surface waves for simulating nearshore waves (SWAN) and ocean circulation models such as the finite volume model primitive equation community ocean model (FVCOM) and the hybrid coordinate ocean model (HYCOM). These models rely on accurate forcing fields (i.e. wind and tides) and open conditions (i.e. water temperature and salinity).

[0003] Synthetic Aperture Radar (SAR) has strong atmospheric (wind and rain) detection capabilities and ocean dynamics (i.e., waves and currents) with good resolution. SAR-derived wind is crucial for SAR wave inversion algorithms, which are divided into two categories, namely, theory-based methods and empirical models or deep learning. The theory-based methods are developed based on the mapping mechanism of synthetic aperture radar. In other words, the wave spectrum can be inverted from the synthetic aperture radar intensity spectrum using three modulated model transfer functions (MTFs). This principle is the key to existing algorithms, namely, the Max-Planck Institute algorithm (MPI), the semi-parametric retrieval algorithm (SPRA), the partitioned recalibration and movement algorithm (PARSA), the parameterized first-guess spectrum method (PFSM), and the full polarimetric technique. In order to avoid the calculation of complex MTFs, the wave parameters are directly related to the synthetic aperture radar imaging variables through empirical models or deep learning.

[0004] Compared with wind and waves, the magnitude of ocean currents is relatively small, and the speed of ocean currents on the global ocean is about 2m / s. Therefore, the inversion of current velocity is a difficult point for SAR. However, there is currently a lack of methods for synthetic aperture radar current velocity inversion. Summary of the invention

[0005] In view of the defects in the prior art, the purpose of the present invention is to provide a synthetic aperture radar sea surface current velocity inversion method and device based on machine learning. The present invention provides an effective method for SAR sea surface current velocity inversion.

[0006] To solve the above problems, the technical solution of the present invention is:

[0007] A synthetic aperture radar sea surface current velocity inversion method based on machine learning includes the following steps:

[0008] Obtain original synthetic aperture radar images, corresponding OCN data, hybrid coordinate ocean model data and NDBC buoy data;

[0009] The PFSM algorithm is used to calculate the wave spectrum of synthetic aperture radar images, and the relationship between flow velocity and synthetic aperture radar image parameters is explored;

[0010] Construct a synthetic aperture radar sea surface current velocity inversion algorithm based on machine learning;

[0011] Verify the accuracy of the inversion algorithm.

[0012] Preferably, the step of obtaining the original synthetic aperture radar image, the corresponding OCN data, the hybrid coordinate ocean model data and the NDBC buoy data specifically includes: collecting multiple Sentinel-1 images in the western boundary current region, i.e., the western Pacific and the western Atlantic, and the ocean data corresponding to a single Sentinel-1 image is stored as a Level-2 product, including SAR-derived wind speed, SAR-derived wave parameters and radar echo frequency shift of the beam center, and obtaining the corresponding HY-2 effective wave height data and NDBC buoy data.

[0013] Preferably, the step of using the PFSM algorithm to calculate the wave spectrum of the synthetic aperture radar image and exploring the relationship between the flow velocity and the synthetic aperture radar image parameters specifically includes: the PFSM algorithm calculates the threshold k of the wave number k s , separate wind wave and swell retrieval:

[0014]

[0015] Among them, U 10 is the wind speed at 10 meters, V is the satellite speed, g is the acceleration due to gravity (=9.8 m / s2), R is the slant distance, is the angle of the wave propagation direction relative to the radar observation direction, and θ is the incident angle; due to the nonlinear effect of the velocity beam, the cost function J is used to invert the wind-sea spectrum P k , the cost function J has the following form:

[0016]

[0017] Where μ is the weight coefficient, A is set to 0.001 to ensure the convergence of the solution, and Q k is the synthetic aperture radar intensity spectrum, and is the SAR spectrum and tilt modulation function Fluid Dynamics Modulation Function and velocity aggregation function The mapping after multiplication of the MTF.

[0018] Preferably, the step of constructing a synthetic aperture radar sea surface current velocity inversion algorithm based on machine learning specifically includes: the input of the model includes the azimuth cutoff wavelength difference Δλ between the theory and the SAR measurement c , sea surface wind speed U 10 , Stokes drift U calculated from SAR wave spectrum s , radar incident angle and Doppler centroid anomaly f DC ; The output is flow velocity; Two-thirds of the dataset is used to train the SAR sea surface flow velocity inversion algorithm based on machine learning, and the rest is used to verify the accuracy of the algorithm.

[0019] Preferably, in the step of constructing a synthetic aperture radar sea surface current velocity inversion algorithm based on machine learning, the inversion algorithm adopts an XGBoost model.

[0020] Preferably, the step of verifying the accuracy of the inversion algorithm specifically includes: verifying the inversion result through NDBC buoy data, matching the two types of data in time and space, and screening by using time less than 1 hour and spatial distance less than 2 kilometers as conditions to evaluate the accuracy of the practical algorithm for applying SAR sea surface current inversion.

[0021] Furthermore, the present invention also provides a synthetic aperture radar sea surface current velocity inversion device based on machine learning, the device comprising a processor and a memory for storing executable instructions of the processor, the processor being configured to execute the synthetic aperture radar sea surface current velocity inversion method based on machine learning as described above by executing the executable instructions.

[0022] Compared with the prior art, the present invention provides a practical algorithm for synthetic aperture radar sea surface current velocity inversion based on machine learning, and verifies the applicability of SAR sea surface current velocity inversion based on machine learning. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Other features, objects and advantages of the present invention will become more apparent from the detailed description of non-limiting embodiments made with reference to the following drawings:

[0024] Figure 1 This is a flowchart of the method for inverting sea surface current velocity using synthetic aperture radar based on machine learning of the present invention;

[0025] Figure 2a This is an example diagram of OCN data products; Figure 2b Get DCA data graphs for OCN products;

[0026] Figure 3This is an example diagram of flow rate data of HYCOM data product;

[0027] Figure 4a This is a sub-image of the Sentinel-1 data product; Figure 4b Calculate the corresponding one-dimensional wave spectrum obtained for the image;

[0028] Figure 5a and Figure 5b is the relationship between the azimuthal cutoff wavelength difference and the sea surface wind speed and Stokes drift; Figure 5c and Figure 5d is the relationship between Doppler centroid anomaly and sea surface wind speed and Stokes drift;

[0029] Figure 6a and Figure 6b It is the relationship diagram between flow velocity, azimuthal cutoff wavelength difference and Doppler centroid anomaly;

[0030] Figure 7 Diagram of the training process for a machine learning model;

[0031] Figure 8 This is a comparison chart of SAR sea surface current velocity inversion results and NDBC data product velocity data;

[0032] Figure 9a is the algorithm inversion diagram; Figure 9b This is an example of HYCOM data for a single image. DETAILED DESCRIPTION

[0033] The present invention is described in detail below in conjunction with specific embodiments. The following embodiments will help those skilled in the art to further understand the present invention, but are not intended to limit the present invention in any form. It should be noted that, for those of ordinary skill in the art, several changes and improvements can also be made without departing from the concept of the present invention. These all belong to the protection scope of the present invention.

[0034] Specifically, the present invention provides a synthetic aperture radar sea surface current velocity inversion method based on machine learning, such as Figure 1 As shown, the method comprises the following steps:

[0035] S1: Acquire the original synthetic aperture radar image, the corresponding OCN data, the hybrid coordinate ocean model data and the NDBC buoy data;

[0036] Specifically, a total of more than 4,500 Sentinel-1 images have been collected in the western boundary current region, namely the western Pacific and western Atlantic. These VV polarization images were acquired in interferometric wide (IW) mode during 2020-2023. Currently, the ocean (OCN) data corresponding to a single Sentinel-1 image are stored as Level-2 products, which include synthetic aperture radar (SAR)-derived wind speed, SAR-derived wave parameters, and radar echo frequency shift at the beam center. And obtain 0.08° spatial resolution Hybrid Coordinate Ocean Model (HYCOM) data every 3 hours for spatiotemporal matching, as well as National Data Buoy Center (NDBC) buoy data. Figure 2a This is an example of an OCN data product. Figure 2b Get DCA data graphs for OCN products. Figure 3 This is an example diagram of flow rate data of HYCOM data product.

[0037] S2: Use the PFSM algorithm to calculate the wave spectrum of synthetic aperture radar images and explore the relationship between flow velocity and synthetic aperture radar image parameters;

[0038] Specifically, the collected Sentinel-1 remote sensing images are used to invert the wave spectrum through the parameterized first guess spectrum method (PFSM) theoretical algorithm. Figure 4a This is a sub-image of the Sentinel-1 data product; Figure 4b The corresponding one-dimensional ocean wave spectrum is calculated for this image. The azimuth cut-off wavelength is simulated by synthetic aperture radar spectrum. The dependence of the ocean surface dynamics (i.e. wind, waves and currents) on the difference between the azimuth cut-off wavelength measured by SAR and the simulated value is studied. Figure 5a and Figure 5b is the relationship between the azimuth cut-off wavelength difference and the sea surface wind speed and Stokes drift, Figure 5c and Figure 5d This is a graph showing the relationship between the Doppler centroid anomaly and the sea surface wind speed and Stokes drift. Figure 6a and Figure 6b The figure shows the relationship between the velocity and the azimuth cutoff wavelength difference and Doppler centroid anomaly. Based on these relationships, the feasibility of a synthetic aperture radar sea surface velocity inversion algorithm based on machine learning is determined.

[0039] The PFSM algorithm is used to calculate the wave spectrum of the SAR image. The advantage of the PFSM algorithm is that it calculates the threshold k of the wave number k. s , separate wind wave and swell retrieval:

[0040]

[0041] Among them, U 10is the wind speed at 10 meters, V is the satellite speed, g is the acceleration due to gravity (=9.8 m / s2), R is the slant distance, is the angle of the wave propagation direction relative to the radar observation direction, and θ is the incident angle. Due to the nonlinear effect of the velocity beam, the cost function J is used to invert the wind wave spectrum P k , the cost function J has the following form:

[0042]

[0043] Where μ is the weight coefficient, A is set to 0.001 to ensure the convergence of the solution, and Q k is the synthetic aperture radar intensity spectrum, and is the SAR spectrum and tilt modulation function Fluid Dynamics Modulation Function and velocity aggregation function The mapping after multiplication of the MTF.

[0044] S3: Construct a synthetic aperture radar sea surface current velocity inversion algorithm based on machine learning;

[0045] Specifically, compared with empirical models, the advantage of machine learning is that it can fit the relationship between weakly correlated parameters. The extreme gradient boosting (XGBoost) model is widely used in various regression analysis tasks due to its excellent performance and flexibility. The model improves the predictive ability of the model by integrating multiple weak learners. The core algorithm uses parallel processing and cache optimization techniques during execution, which greatly improves the training efficiency. The model is iteratively optimized through the gradient boosting algorithm to continuously improve the overall predictive ability of different data sets. XGBoost can not only control the complexity of the model, but also improve the generalization ability of the model by adding L1 and L2 regularization terms in the loss function to prevent overfitting and data multicollinearity. In addition, the prediction accuracy of the model can be further improved by optimizing the model parameters. For example, the learning_rate parameter controls the step size of each enhancement. By reducing the step size, the model can learn in a more detailed way, so that it gradually approaches the optimal solution in multiple iterations. The max_depth parameter determines the complexity of a single tree. Properly increasing the depth can obtain more complex patterns and relationships in the training data set.

[0046] In the present invention, the input of the model includes the azimuth cutoff wavelength difference Δλ between the theory and the SAR measurement c , sea surface wind speed U 10 , Stokes drift U calculated from SAR wave spectrum s , radar incident angle and Doppler centroid anomaly f DC The output is the flow rate. Figure 7Figure 1. The training process of the machine learning model. Two-thirds of the dataset was used to develop a machine learning-based SAR sea surface current velocity inversion algorithm, and the rest was used to verify the accuracy of the algorithm. XGBoost was trained using a dataset from 4,000 images to build a relationship between azimuthal current velocity and synthetic aperture radar measurement variables.

[0047] S4: Verify the accuracy of the inversion algorithm.

[0048] Specifically, the sea surface velocity inversion results of this method are verified by the National Data Buoy Center (NRBC) data products. The inversion results are verified by NDBC buoy data, and the two data are matched in time and space. The time is less than 1 hour and the spatial distance is less than 2 kilometers. The conditions are used to screen and evaluate the accuracy of the practical algorithm for applying SAR sea surface velocity inversion. Figure 8 To compare the results, the accuracy of the practical algorithm is verified. Figure 9a is the algorithm inversion diagram, Figure 9b This is an example of HYCOM data for a single image. Figure 9a and Figure 9b It can be clearly seen that the velocity results inverted from SAR images are consistent with the velocity data distribution of HYCOM data products.

[0049] The above describes the specific embodiments of the present invention. It should be understood that the present invention is not limited to the above specific embodiments, and those skilled in the art can make various changes or modifications within the scope of the claims, which does not affect the essence of the present invention. In the absence of conflict, the embodiments of the present application and the features in the embodiments can be combined with each other arbitrarily.

Claims

1. A synthetic aperture radar sea surface current velocity inversion method based on machine learning, characterized in that: The method comprises the following steps: Obtain original synthetic aperture radar images, corresponding OCN data, hybrid coordinate ocean model data and NDBC buoy data; The PFSM algorithm is used to calculate the wave spectrum of synthetic aperture radar images, and the relationship between flow velocity and synthetic aperture radar image parameters is explored; Construct a synthetic aperture radar sea surface current velocity inversion algorithm based on machine learning; Verify the accuracy of the inversion algorithm.

2. The method for inverting sea surface current velocity using synthetic aperture radar based on machine learning according to claim 1, characterized in that: The step of obtaining the original synthetic aperture radar image, the corresponding OCN data, the hybrid coordinate ocean model data and the NDBC buoy data specifically includes: collecting multiple Sentinel-1 images in the western boundary current area, namely the western Pacific and the western Atlantic, and storing the ocean data corresponding to a single Sentinel-1 image as a Level-2 product, including SAR-derived wind speed, SAR-derived wave parameters and radar echo frequency shift at the beam center, and obtaining the corresponding HY-2 effective wave height data and NDBC buoy data.

3. The method for inverting sea surface current velocity using synthetic aperture radar based on machine learning according to claim 1, characterized in that: The steps of using the PFSM algorithm to calculate the wave spectrum of the synthetic aperture radar image and exploring the relationship between the flow velocity and the synthetic aperture radar image parameters specifically include: the PFSM algorithm calculates the threshold k of the wave number k s , separate wind wave and swell retrieval: Among them, U 10 is the wind speed at 10 meters, V is the satellite speed, g is the acceleration due to gravity (=9.8 m / s2), R is the slant distance, is the angle of the wave propagation direction relative to the radar observation direction, and θ is the incident angle; due to the nonlinear effect of the velocity beam, the cost function J is used to invert the wind-sea spectrum P k , the cost function J has the following form: Where μ is the weight coefficient, A is set to 0.001 to ensure the convergence of the solution, and Q k is the synthetic aperture radar intensity spectrum, and is the SAR spectrum and tilt modulation function Fluid Dynamics Modulation Function and velocity aggregation function The mapping after multiplication of the MTF.

4. The method for sea surface current inversion based on synthetic aperture radar and machine learning according to claim 1, characterized in that: The steps of constructing a synthetic aperture radar sea surface current velocity inversion algorithm based on machine learning specifically include: the input of the model includes the azimuth cutoff wavelength difference Δλ between the theory and the SAR measurement c , sea surface wind speed U 10 , Stokes drift U calculated from SAR wave spectrum s , radar incident angle and Doppler centroid anomaly f DC ; The output is flow velocity; Two-thirds of the dataset is used to train the SAR sea surface flow velocity inversion algorithm based on machine learning, and the rest is used to verify the accuracy of the algorithm.

5. The method for sea surface current inversion based on synthetic aperture radar using machine learning according to claim 4 is characterized in that: In the step of constructing a synthetic aperture radar sea surface current velocity inversion algorithm based on machine learning, the inversion algorithm adopts an XGBoost model.

6. The method for sea surface current inversion based on synthetic aperture radar and machine learning according to claim 1, characterized in that: The step of verifying the accuracy of the inversion algorithm specifically includes: verifying the inversion result through NDBC buoy data, matching the two data in time and space, and screening by using time less than 1 hour and spatial distance less than 2 kilometers as conditions to evaluate the accuracy of the practical algorithm for applying SAR sea surface current inversion.

7. A synthetic aperture radar sea surface current velocity inversion device based on machine learning, characterized in that: The device includes a processor and a memory for storing executable instructions of the processor, and the processor is configured to perform the machine learning-based synthetic aperture radar sea surface current velocity inversion method as described in any one of claims 1 to 6 by executing the executable instructions.

Images