GNSS-R and GNSS-S radar combined sea surface high wind speed inversion method and satellite-borne receiving system

Abstract

The invention relates to a GNSS-R and GNSS-S radar combined sea surface high wind speed inversion method and a satellite-borne receiving system, and the method comprises the following steps: S1, obtaining a reflection signal and a scattering signal in a time space under a high wind speed condition, and carrying out the full-link sampling of the reflection signal and the scattering signal; s2, performing delay Doppler mapping processing on the reflected signal to generate a sea surface delay Doppler map; performing double-station SAR imaging processing on the scattered signals to generate Q GNSS-S radar images with different double-station angles; s3, extracting statistical feature vectors of the sea surface delay Doppler map and the Q GNSS-S radar images with different bistatic angles, and forming a joint feature X; and S4, inputting the joint feature X into a pre-trained deep neural network, and outputting a sea surface wind speed result. According to the method, the characteristics of GNSS reflection signals and scattering signals can be fully utilized, and the precision of sea surface high-wind-speed inversion is effectively improved.

Description

Technical Field

[0001] This invention relates to the field of marine microwave remote sensing and radar signal processing technology, and in particular to a method for inverting high wind speeds over the sea surface using GNSS-R and GNSS-S radars, and a spaceborne receiving system. Background Technology

[0002] Sea surface wind fields are a core driving force for ocean circulation and the regulation of heat and water vapor exchange between the sea and the atmosphere. Accurate acquisition of these wind fields is crucial for ocean numerical weather prediction, typhoon track and intensity forecasting, shipping safety, and global climate monitoring. The World Meteorological Organization lists sea surface wind speed at 10 m as a fundamental global ocean climate variable. However, existing space-based observation methods generally experience performance degradation under high sea states (>20 m / s): Ku and C-band scatterometers near 30 m / s experience rapid amplification of inversion errors due to the radar backscattering coefficient σ0 saturating with increasing wind speed; radiometers are strongly affected by rainfall and cloud liquid water, often resulting in data gaps in areas such as typhoon eyewalls; while synthetic aperture radar (SAR) offers high resolution, it also faces saturation of the scattering coefficient σ0 and 180° wind direction ambiguity under single-station geometry, and high-power transmitters are difficult to operate as small satellite constellations. Therefore, accurate, continuous, and global observation of sea surface wind fields in medium to high sea states (20-50 m / s) remains a significant challenge in the field of ocean remote sensing.

[0003] In recent years, the development of GNSS technology has provided unprecedented space signal resources for GNSS "opportunity source" bistatic / multistatic radars. GNSS-R (reflected signal) and GNSS-S (scattered signal) technologies do not require active transmitters; only receivers are needed to achieve "satellite-to-satellite transmission-to-reception" measurements, offering advantages such as low power consumption, low cost, and easy networking. Furthermore, utilizing the geometric diversity of bistatic radars, high-wind-speed measurements at sea surface can be achieved through the complementary "reflection-scattering" mechanism of GNSS. GNSS-R primarily focuses on the near-nadir (0°-45° reflection angle), with σ0 decreasing monotonically with increasing wind speed; GNSS-S shows σ0 increasing monotonically with increasing wind speed within the 15°-75° scattering angle range. The combination of these two technologies can generate two sensitivity curves—one decreasing and one increasing—at the same sea surface element and time, providing a theoretical basis for high-wind-speed sea surface wind field measurements.

[0004] However, current methods still rely solely on GNSS-R information to retrieve sea surface wind speeds, failing to fully utilize the coupled information from both mechanisms. Furthermore, the scarcity of high-wind-speed samples leads to significant extrapolation errors in the models. Therefore, a precise sea surface high-wind-speed retrieval technology combining GNSS-R and GNSS-S radars is urgently needed to improve the accuracy of high-wind-speed retrieval. Summary of the Invention

[0005] To address the technical problems existing in the prior art, the present invention aims to provide a GNSS-R and GNSS-S radar combined high sea surface wind speed inversion method, achieving sea surface wind speed inversion accuracy better than 10% for extreme wind speeds.

[0006] To achieve the above-mentioned objective, this invention provides a method for inverting high wind speeds over the sea surface using a combination of GNSS-R and GNSS-S radars, comprising the following steps:

[0007] Step S1: Obtain the reflection and scattering signals in the same space under high wind speed conditions, and perform full-link sampling on the reflection and scattering signals;

[0008] Step S2: Perform delayed Doppler mapping processing on the reflected signal to generate a delayed Doppler image of the sea surface; perform bistatic SAR imaging processing on the scattered signal to generate Q-frame GNSS-S radar images with different bistatic angles;

[0009] Step S3: Extract the statistical feature vectors of the delayed Doppler image of the sea surface and the Q-frame GNSS-S radar images at different bistatic angles, and form a joint feature X;

[0010] Step S4: Input the joint feature X into the pre-trained deep neural network and output the sea surface wind speed result.

[0011] According to one technical solution of the present invention, in step S1, the reflected signal and the scattered signal are acquired by a high-gain multibeam GNSS echo receiving antenna installed on a satellite platform;

[0012] The high-gain multi-beam GNSS echo receiving antenna has both a GNSS-R wide-beam receiving channel and a GNSS-S narrow-beam receiving channel; the gain of the high-gain multi-beam GNSS echo receiving antenna satisfies the following: the GNSS-R signal-to-noise ratio (SNRR) received at an orbital altitude of 500~1000km is ≥3dB, and the GNSS-S signal-to-noise ratio (SNRS) is ≥5dB.

[0013] In step S1, the reflected signal and the scattered signal are sampled across the entire chain using a digital-to-analog converter. The digital-to-analog converter operates in the frequency range of 1.2 to 1.6 GHz and its quantization noise power is no higher than 0.1 dB.

[0014] According to one technical solution of the present invention, in step S2, the range of reflection angles corresponding to the generated sea surface time-delay Doppler image is 0°≤θ. r ≤45°; Bistatic scattering angle θ corresponding to Q-span GNSS-S radar image q The coverage area satisfies 15°≤θ q ≤75°.

[0015] According to one technical solution of the present invention, step S3 specifically includes:

[0016] Step S31: Extract the statistical feature vector of the sea surface time-delay Doppler image:

[0017]

[0018] In the formula, σ τ σ represents the standard deviation calculated after probability normalization of the sea surface time-delay Doppler image along the time-delay dimension τ. f P represents the standard deviation calculated after probability normalization of the time-delayed Doppler image of the sea surface along the Doppler dimension f; max The maximum correlation power value is represented in the delayed Doppler image of the sea surface; LES represents the slope of the rising edge of the delayed waveform; SNR represents the signal-to-noise ratio of the reflected signal;

[0019] Step S32: Extract the statistical feature vectors of Q GNSS-S radar images:

[0020]

[0021] In the formula, μ σ0,q σ represents the sample mean of the backscattering coefficients of the effective sea surface pixels in the q-th GNSS-S radar image; σ0,q α represents the sample standard deviation of the backscattering coefficients in the q-th GNSS-S radar image; inc,q R represents the average incident angle corresponding to the q-th GNSS-S radar image; Γ / σ0,q This represents the ratio of the reflection coefficient Γ to the scattering coefficient σ0;

[0022] Step S33: Construct joint feature X:

[0023] .

[0024] According to one technical solution of the present invention, in step S4, the deep neural network is a convolutional neural network, including an input layer, a convolutional layer, a pooling layer, a fully connected layer and an output layer;

[0025] The convolutional layer consists of three layers with 16, 64, and 128 input channels and 64, 128, and 256 output channels, respectively.

[0026] The pooling layer uses the max pooling method, with a pooling window size of 2×2 and a step size of 1.

[0027] The number of nodes in the fully connected layer are 512 and 100, respectively;

[0028] The number of nodes in the output layer is set to 81, and the output sea surface wind speed result U 10∈[20,50]m / s.

[0029] According to one technical solution of the present invention, in step S4, the deep neural network uses the NDBC buoy truth value U during the training phase. truth As labels, buoy samples in both the validation and test sets were collected independently to ensure that the time and space difference with satellite observations was ≤15 minutes and the distance difference was ≤5 km.

[0030] During training, 70% of the data in the sample set is used as the training set, and 30% is used as the test set; the loss function adopted is the mean absolute error loss function, expressed as:

[0031]

[0032] In the formula, This represents the wind speed result for the i-th sample output by the deep neural network. Let N be the true value of the neutral wind measured at 10 meters for the i-th sample, and N be the number of training samples.

[0033] According to one aspect of the present invention, a spaceborne receiving system is provided for implementing the above-described GNSS-R and GNSS-S radar joint high-wind-speed inversion method for the sea surface, comprising:

[0034] A GNSS direct signal receiving antenna, used to receive direct signals from multiple GNSS satellites;

[0035] A high-gain multibeam GNSS echo receiving antenna is used to receive GNSS echo signals from the sea surface. GNSS echo signals include reflected signals and scattered signals.

[0036] The host computer processes the reflected and scattered signals to generate corresponding time-delayed Doppler images and GNSS-S radar images of the sea surface. It also extracts joint features X through a deep neural network and outputs the sea surface wind speed results.

[0037] The data download and on-orbit calibration module is used to download the obtained sea surface wind speed results, joint feature X, and intermediate mass markers in real time, and receive feedback from the ground calibration field to ensure that the long-term radiometric calibration error is ≤0.3dB.

[0038] According to one technical solution of the present invention, the host includes:

[0039] Direct signal receiver, used to process direct signals from GNSS satellites, and output GNSS satellite ephemeris, navigation messages and direct signal copies;

[0040] The direct signal copy generation module is used to generate reference signals required for the processing of scattered and reflected signals;

[0041] A high-sensitivity reflected signal receiver is used to amplify, down-convert, and bandpass filter the received reflected signal to generate a reflected intermediate frequency signal.

[0042] A high-sensitivity scattered signal receiver is used to amplify, down-convert, and bandpass filter the received scattered signal to generate a scattered intermediate frequency signal.

[0043] A digital-to-analog converter is used for high-speed sampling and quantization of reflected and scattered intermediate frequency signals;

[0044] The radar main control unit is used to give commands to the direct signal receiver and the direct signal copy generation module, and to store and transmit the reflected intermediate frequency signal and the scattered intermediate frequency signal after being quantized by the digital-to-analog converter.

[0045] The DBF processing unit is used to synthesize high-gain antenna beams pointing to different sea areas;

[0046] The central processing unit is used to generate time-delayed Doppler images and GNSS-S radar images of the sea surface based on reference signals, reflected signals, and scattered signals, as well as to train deep neural networks and extract features.

[0047] According to one technical solution of the present invention, the central processing unit is an FPGA+GPU heterogeneous processing unit, including a DDMA generation module, a SAR image generation module, a feature extraction module, and a neural network training module.

[0048] Compared with the prior art, the present invention has the following beneficial effects:

[0049] This invention provides a method for accurate inversion of high sea surface wind speeds using a combination of GNSS-R and GNSS-S radars, along with a spaceborne receiving system. Through high-quantization bit reception, multi-angle scattering imaging, and deep network nonlinear fitting, it achieves operational inversion of sea surface wind fields with an accuracy better than 10% under high wind speeds of 20-50 m / s, filling a gap in global extreme sea state monitoring. This invention can significantly improve the accuracy of sea surface wind speed monitoring under extreme sea conditions such as typhoons and explosive cyclones.

[0050] In this invention, the physical properties of GNSS-R reflected signals decreasing with wind speed and GNSS-S scattered signals increasing with wind speed are utilized to achieve a resolution within 20-50ms. - ¹Even in high wind speeds, effective sea surface echoes can still be acquired. Combined with 16-bit high quantization and multi-angle imaging, the richness of information and reliability of inversion under extreme wind conditions are guaranteed.

[0051] In this invention, a lightweight deep network and on-orbit processing hardware work together to complete signal processing and wind speed output in real time. The model is updated on-orbit by transmitting incremental data back from the ground calibration field, which supports real-time monitoring of extreme sea conditions around the world and rapid early warning of typhoons and explosive cyclones, significantly improving the timeliness of observation. Attached Figure Description

[0052] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly described below. Obviously, the drawings described below are merely some embodiments of the present invention, and those skilled in the art can obtain other drawings based on these drawings without creative effort.

[0053] Figure 1 A flowchart illustrating the GNSS-R and GNSS-S radar joint high wind speed inversion method provided in an embodiment of the present invention is shown.

[0054] Figure 2 This illustration depicts a scenario where spaceborne GNSS-R and GNSS-S radars jointly detect sea surface wind speed according to an embodiment of the present invention.

[0055] Figure 3 This illustration shows the composition of a spaceborne receiving system according to an embodiment of the present invention.

[0056] Figure 4 This schematic diagram illustrates the process of accurately inverting high wind speeds over the sea surface using a combination of spaceborne GNSS-R and GNSS-S radars according to an embodiment of the present invention.

[0057] Figure 5 This diagram illustrates the construction of a joint high-wind-speed inversion network for sea surface using spaceborne GNSS-R and GNSS-S radars according to an embodiment of the present invention. Detailed Implementation

[0058] The description of the embodiments in this specification should be taken in conjunction with the accompanying drawings, which should form part of the complete specification. In the drawings, the shape or thickness of the embodiments may be exaggerated and may be indicated in a simplified or convenient manner. Furthermore, parts of the various structures in the drawings will be described separately; it is worth noting that elements not shown in the figures or not described in words are in a form known to those skilled in the art.

[0059] The descriptions of the embodiments herein, including any references to directions and orientations, are for ease of description only and should not be construed as limiting the scope of the invention. The following description of preferred embodiments involves combinations of features, which may exist independently or in combination; the invention is not particularly limited to the preferred embodiments. The scope of the invention is defined by the claims.

[0060] like Figure 1 and Figure 3 As shown, this invention provides a method for joint GNSS-R and GNSS-S radar high-wind speed inversion over the sea surface and a spaceborne receiving system. The spaceborne receiving system is used to implement the method for joint GNSS-R and GNSS-S radar high-wind speed inversion over the sea surface. The method for joint GNSS-R and GNSS-S radar high-wind speed inversion over the sea surface provided in this invention includes the following steps:

[0061] Step S1: Obtain the reflection and scattering signals in the same space under high wind speed conditions, and perform full-link sampling on the reflection and scattering signals;

[0062] Step S2: Perform delayed Doppler mapping processing on the reflected signal to generate a delayed Doppler image of the sea surface; perform bistatic SAR imaging processing on the scattered signal to generate Q-frame GNSS-S radar images with different bistatic angles;

[0063] Step S3: Extract the statistical feature vectors of the delayed Doppler image of the sea surface and the Q-frame GNSS-S radar images at different bistatic angles, and form a joint feature X;

[0064] Step S4: Input the joint feature X into the pre-trained deep neural network and output the sea surface wind speed result.

[0065] The spaceborne receiving system includes a GNSS direct signal receiving antenna 10, a high-gain multibeam GNSS echo receiving antenna 20, a main unit 30, and a data downlink and on-orbit calibration module 40. The GNSS direct signal receiving antenna 10 receives direct signals from multiple GNSS satellites. The high-gain multibeam GNSS echo receiving antenna 20 receives GNSS echo signals from the sea surface, which include reflected and scattered signals. The main unit 30 processes the reflected and scattered signals in the GNSS echoes, generating corresponding sea surface time-delay Doppler (DDMA) maps and GNSS-S radar images. It then extracts joint features X using a deep neural network and outputs the sea surface wind speed result U. 10 Sea surface wind speed result U 10 The wind speed is at a height of 10 meters. Data is downloaded and used by the on-orbit calibration module 40 for sea surface wind speed results.10 The joint characteristic X and intermediate quality indicators are transmitted in real time and feedback from the ground calibration field is received, so that the long-term radiometric calibration error is ≤0.3dB.

[0066] In this embodiment of the invention, preferably, the high-gain multi-beam GNSS echo receiving antenna 20 has a GNSS-R wide beam channel 201 and a GNSS-S narrow beam channel 202. The gain needs to ensure that the GNSS-R signal-to-noise ratio (SNRR) received at an orbital altitude of 500~1000km is ≥3dB and the GNSS-S signal-to-noise ratio (SNRS) is ≥5dB, thereby ensuring the accuracy of the inversion results.

[0067] In a preferred embodiment of the present invention, the host 30 includes a direct signal receiver 301, a direct signal copy generation module, a high-sensitivity reflected signal receiver 302, a high-sensitivity scattered signal receiver 303, a digital-to-analog converter 304, a radar main control unit 305, a DBF processing unit 306, and a central processing unit 307.

[0068] The direct signal receiver 301 processes direct signals from GNSS satellites, outputting GNSS satellite ephemeris, navigation messages, and a copy of the direct signal. The direct signal copy generation module generates reference signals required for processing scattered and reflected signals. The high-sensitivity reflected signal receiver 302 amplifies, down-converts, and bandpass-filters the GNSS reflected signal to generate a reflected intermediate frequency (IF) signal. The high-sensitivity scattered signal receiver 303 amplifies, down-converts, and bandpass-filters the GNSS scattered signal to generate a scattered IF signal. The digital-to-analog converter 304 uses a 16-bit ADC array for high-speed sampling and quantization of the GNSS reflected and scattered IF signals. The radar main control unit 305 provides command control to the direct signal receiver 301 and the direct signal copy generation module, and controls the storage and transmission of the GNSS reflected and scattered signals quantized by the 16-bit ADC array. The DBF processing unit 306 synthesizes high-gain antenna beams pointing towards different sea areas. The central processing unit 307 adopts an FPGA+GPU heterogeneous processing unit to generate sea surface time-delay Doppler images and GNSS-S radar images based on reference signals, reflected signals and scattered signals, as well as to train deep neural networks and extract features. The central processing unit includes a DDMA generation module, a SAR image generation module, a feature extraction module and a neural network training module.

[0069] like Figures 2 to 4 As shown, the present invention provides a method for inverting high wind speeds over the sea surface using a combination of GNSS-R and GNSS-S radars, comprising the following steps:

[0070] Step S1: Obtain the reflection and scattering signals in the same space under high wind speed conditions, and perform full-link sampling on the reflection and scattering signals;

[0071] In step S1, a high-gain multi-beam GNSS echo receiving antenna 20 with a diameter of 5m² to 15m² is installed on a satellite platform at an orbital altitude of 500km to 1000km. The high-gain multi-beam GNSS echo receiving antenna 20 simultaneously possesses a GNSS-R wide-beam receiving channel 201 and a GNSS-S narrow-beam receiving channel 202, synchronously receiving reflected and scattered signals from the same location under high wind speed conditions (20~50m / s). A digital-to-analog converter 304 with a quantization bit depth of 16 bits or higher is used to perform end-to-end sampling of the reflected and scattered signals. The operating frequency range is 1.2~1.6GHz, and the quantization noise power meets the following requirements:

[0072]

[0073] in, Indicates the quantization noise power. Indicates the received signal power. This indicates the target echo signal-to-noise ratio, thus ensuring that the high wind speed 20m / s ≤ U 10 The minute fluctuations ΔΓ≤0.15dB and Δσ0≤0.2dB of the sea surface reflection coefficient Γ and scattering coefficient σ0 with wind speed under conditions of ≤50m / s were accurately captured.

[0074] Step S2: Perform delayed Doppler mapping processing on the reflected signal to generate a delayed Doppler image of the sea surface; perform bistatic SAR imaging processing on the scattered signal to generate Q-frame GNSS-S radar images with different bistatic angles;

[0075] The reflected signal is processed by Delay-Doppler Mapping (DDM) to generate a Delayed-Doppler Map (DDMA) of the sea surface, wherein the reflection angle corresponding to the DDMA is 0°≤θ. r ≤45°; Perform bistatic SAR imaging processing on the scattered signal to generate Q-amplitude images with different bistatic angles θ. q GNSS-S radar images of (q=1,2,…,Q; Q≥3), and the bistatic scattering angle θ corresponding to the Q GNSS-S radar images. q The coverage area satisfies 15°≤θ q ≤75°.

[0076] Step S3: Extract the statistical feature vectors of the delayed Doppler image of the sea surface and the Q-frame GNSS-S radar images at different bistatic angles, and form a joint feature X.

[0077] Specifically, step S3 includes:

[0078] Step S31: Extract the statistical feature vector of DDMA:

[0079]

[0080] In the formula, σ τ This represents the standard deviation calculated after probability normalization of DDMA along the delay dimension τ, reflecting the dispersion of reflection path lengths caused by sea surface height variations; σ f P represents the standard deviation of DDMA calculated after probability normalization along the Doppler dimension f, reflecting the Doppler broadening caused by wave trajectory velocity at high wind speeds; max The maximum correlated power value in DDMA is represented, which is closely related to sea surface roughness (wind field intensity); LES represents the slope of the rising edge of the time delay waveform, which becomes slower as the wind speed increases; SNR represents the signal-to-noise ratio of the reflected signal.

[0081] Step S32: Extract the statistical feature vectors of Q GNSS-S images:

[0082]

[0083] In the formula, μ σ0,q σ represents the mean of the backscattering coefficients of the effective sea surface pixels in the q-th image, used to characterize the average scattering intensity at this bistatic angle; σ0,q α represents the sample standard deviation of the backscattering coefficient in the q-th image, measuring the spatial variability of sea surface scattering at high wind speeds (the value increases when breaking waves and specular points coexist); inc,q R represents the average incident angle corresponding to the q-th image; Γ / σ0,q This represents the ratio of the reflection coefficient Γ to the scattering coefficient σ0. This ratio decreases monotonically as the wind speed increases, and is used to improve the sensitivity of neural networks at high wind speeds.

[0084] Step S33: Construct joint feature X:

[0085]

[0086] The joint feature X is input into a pre-trained deep neural network, which outputs a high wind speed U at 10m above sea level. 10 The network loss function L is constructed as follows:

[0087]

[0088] In the formula, α and β are learnable weights, and ∇U pred It is a spatial gradient regularization term; and the U truth The data is derived from actual measurements taken by the US NDBC buoy, and the measured wind speed has been converted to a 10m neutral wind level based on the buoy height.

[0089] In step S3, the joint feature X is further normalized by adaptive weighting, and is represented as follows:

[0090]

[0091] In the formula, CRB(U 10 |X i Given feature X i 10-meter high wind speed U 10 The lower realm of Clamy-Lo, w i As feature weights, to improve the neural network's ability to detect high wind speeds U at 10m above sea level. 10 Sensitivity ≥35m / s.

[0092] like Figure 5 As shown, in this invention, the deep neural network is constructed as a convolutional neural network (CNN), consisting of an input layer, convolutional layers, pooling layers, fully connected layers, and an output layer. The three convolutional layers have 16, 64, and 128 input channels, and 64, 128, and 256 output channels, respectively. The pooling layer uses max pooling with a 2×2 pooling window and a stride of 1. The fully connected layers have 512 and 100 nodes, respectively. The output layer has 81 nodes and outputs a wind speed U at a height of 10 meters. 10 ∈[20,50]m / s. 70% of the data in the sample library is used as the training set, and 30% is used as the test set;

[0093] The loss function used is the Mean Absolute Error Loss (MAE), which has the following form:

[0094]

[0095] In the formula, The sea surface wind speed at 10m high for the i-th sample output by the CNN. N represents the true value of neutral wind at a distance of 10 meters, as measured by the buoy, and N is the number of training samples.

[0096] The training process uses the true value U of the NDBC buoy. truth As labels, buoy samples in both the validation and test sets are collected independently to ensure that the time and space difference with satellite observation is ≤15min and the distance difference is ≤5km.

[0097] This invention belongs to the field of marine microwave remote sensing and radar signal processing. For accurate inversion of high wind speeds at sea surface, a high-gain multibeam GNSS echo receiving antenna is first installed on a satellite platform at an orbital altitude of 500-1000 km to simultaneously receive sea surface reflection and scattering signals under high wind speed (20-50 m / s) conditions. Quantization is performed using a 16-bit or higher ADC to ensure that the reflection coefficient Γ and scattering coefficient σ0, with slight variations in wind speed (ΔΓ≤0.15dB and Δσ0≤0.2dB), are accurately captured. Then, through DDM processing and bistatic SAR imaging processing, simultaneous spatial-temporal GNSS-R time-delayed Doppler images (DDMA) and multiple bistatic GNSS-S radar images at different angles are obtained. Finally, the spatial-Doppler diffusion statistical features of the DDMA and the scattering statistical features of the multi-angle GNSS-S images are extracted to construct a joint feature set, which is then input into a deep neural network based on MAE loss optimization to achieve high wind speed inversion with a relative error ≤10%.

[0098] By combining the aforementioned GNSS-R and GNSS-S radars with a high-speed sea surface inversion and a spaceborne receiving system, the system achieves integrated "reflection-scattering" reception, high-quantization bit sampling, multi-angle imaging, and nonlinear fitting of a deep network. It also achieves sea surface wind speed inversion accuracy better than 10% for extreme wind speeds and has on-orbit real-time computing capabilities. This effectively enhances the potential of GNSS remote sensing technology in the field of early warning of large-scale marine disasters such as hurricanes, tsunamis, and typhoons worldwide.

[0099] Finally, it should be noted that the above description represents a preferred embodiment of the present invention. It should be pointed out that although preferred embodiments have been described, those skilled in the art, once they understand the basic inventive concept of the present invention, can make various improvements and modifications without departing from the principles described herein. These improvements and modifications should also be considered within the scope of protection of the present invention. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the embodiments of the present invention.

Claims

1. A method for retrieving sea surface high wind speed from GNSS-R and GNSS-S radar combination, characterized in that, The method comprises the following steps: Step S1, acquiring the reflection signal and the scattering signal in space-time under high wind speed conditions, and performing full-link sampling on the reflection signal and the scattering signal; Step S2, performing delay Doppler mapping processing on the reflection signal to generate a sea surface delay Doppler map, and performing bistatic SAR imaging processing on the scattering signal to generate a Q-amplitude GNSS-S radar image with different bistatic angles; Step S3, extracting statistical feature vectors of the sea surface delay Doppler map and the Q-amplitude GNSS-S radar image with different bistatic angles, and forming a joint feature X; Step S4, inputting the joint feature X into a pre-trained deep neural network to output a sea surface wind speed result.

2. The GNSS-R and GNSS-S radar united sea surface high wind speed retrieval method according to claim 1, characterized in that, In the step S1, the reflection signal and the scattering signal are acquired by a high-gain multi-beam GNSS echo receiving antenna installed on a satellite platform; The high-gain multi-beam GNSS echo receiving antenna simultaneously has a GNSS-R wide-beam receiving channel and a GNSS-S narrow-beam receiving channel; The gain of the high-gain multi-beam GNSS echo receiving antenna satisfies: the received GNSS-R signal-to-noise ratio SNRR≥3dB and the received GNSS-S signal-to-noise ratio SNRS≥5dB at an orbital height of 500-1000km; In the step S1, the reflection signal and the scattering signal are fully sampled by a digital-to-analog converter, the working frequency range of the digital-to-analog converter is 1.2-1.6GHz, and the quantization noise power is not higher than 0.1dB.

3. The GNSS-R and GNSS-S radar united sea surface high wind speed retrieval method according to claim 2, characterized in that, In the step S2, the generated sea surface delay Doppler map corresponds to a reflection angle range of 0°≤θ r ≤45°; the Q amplitude GNSS-S radar image corresponds to a bistatic scattering angle θ q covering a range of 15°≤θ q ≤75°.

4. The GNSS-R and GNSS-S radar united sea surface high wind speed retrieval method according to claim 3, characterized in that, In the step S3, specifically comprising: Step S31, extracting the statistical feature vector of the sea surface delay Doppler map: where σ τ represents the standard deviation calculated after the probability normalization of the sea surface delay-Doppler map along the delay dimension τ; σ f represents the standard deviation calculated after the probability normalization of the sea surface delay-Doppler map along the Doppler dimension f; P max represents the maximum correlation power value in the sea surface delay-Doppler map; LES represents the slope of the rising edge of the delay waveform; SNR represents the signal-to-noise ratio of the reflected signal; Step S32, extracting the statistical feature vector of the Q-amplitude GNSS-S radar image: where μ σ0,q represents the mean value of the backscatter coefficient samples of the valid sea surface pixels in the qth GNSS-S radar image; σ σ0,q represents the standard deviation of the backscatter coefficient samples in the qth GNSS-S radar image; α inc,q represents the corresponding mean incidence angle in the qth GNSS-S radar image; R Γ / σ0,q represents the ratio of the reflectivity coefficient Γ to the backscatter coefficient σ0. Step S33, constructing the joint feature X: 。 5. The GNSS-R and GNSS-S radar united sea surface high wind speed retrieval method according to claim 4, characterized in that, In the step S4, the deep neural network is a convolutional neural network comprising an input layer, a convolutional layer, a pooling layer, a fully connected layer and an output layer; The convolutional layer has three layers, the input channels of which are 16, 64 and 128 respectively, and the output channels of which are 64, 128 and 256 respectively; The pooling layer adopts a maximum pooling method, the pooling window size is 2x2, and the step is 1; The number of nodes of the fully connected layer is 512 and 100 respectively; The number of nodes of the output layer is set to 81, and the output sea surface wind speed result U 10 ∈ [20, 50] m / s.

6. The GNSS-R and GNSS-S radar united sea surface high wind speed retrieval method according to claim 5, characterized in that, In the step S4, the deep neural network uses the NDBC buoy true value U in the training stage truth As labels, the validation set and the test set buoy samples are independently taken, ensuring that the spatial and temporal difference with satellite observation is ≤15 min, and the distance difference is ≤5 km. During the training process, 70% of the sample set is used as a data training set, and 30% is used as a test set; the loss function adopts a mean absolute error loss function, which is expressed as: wherein is the wind speed result of the i-th sample output by the deep neural network, is the 10-meter neutral wind true value measured by the buoy corresponding to the i-th sample, and N is the number of training samples.

7. A space-borne receiving system for implementing the method for the retrieval of sea surface high wind speed from GNSS-R and GNSS-S radar combination according to any one of claims 1 to 6, characterized in that, It comprises: A GNSS direct signal receiving antenna for receiving direct signals of multiple GNSS satellites; A high-gain multi-beam GNSS echo receiving antenna for receiving GNSS echo signals from the sea surface, the GNSS echo signals comprising reflection signals and scattering signals; A host computer for processing the reflection signals and the scattering signals to generate corresponding sea surface delay Doppler maps and GNSS-S radar images, extracting features of the sea surface delay Doppler maps and the GNSS-S radar images through a deep neural network to obtain a joint feature X, and outputting a sea surface wind speed result; Data download and on-orbit calibration module for real-time downloading of obtained sea surface wind speed results, combined features X and intermediate quality marks, and receiving ground calibration field feedback to make long-term radiation calibration error ≤0.3dB.

8. The space-borne receiving system of claim 7, characterized in that The host computer comprises: A direct signal receiver for processing direct signals from GNSS satellites, outputting GNSS satellite ephemeris, navigation messages and direct signal copies; A direct signal copy generation module for generating reference signals required for processing scattered signals and reflected signals; A high-sensitivity reflected signal receiver for amplifying, down-converting and band-pass filtering the received reflected signals to generate reflected intermediate frequency signals; A high-sensitivity scattered signal receiver for amplifying, down-converting and band-pass filtering the received scattered signals to generate scattered intermediate frequency signals; A digital-to-analog converter for high-speed sampling and quantization of the reflected intermediate frequency signals and the scattered intermediate frequency signals; A radar master control unit for instructing control of the direct signal receiver and the direct signal copy generation module, and storage and transmission control of the quantized reflected intermediate frequency signals and scattered intermediate frequency signals of the digital-to-analog converter; A DBF processing unit for synthesizing high-gain antenna beams pointing to different sea areas; A central processing unit for generating sea surface delay Doppler maps and GNSS-S radar images according to reference signals, reflected signals and scattered signals, and training and feature extraction of deep neural networks.

9. The space-borne receiving system of claim 8, characterized in that, The central processing unit is an FPGA+GPU heterogeneous processing unit, comprising a DDMA generation module, a SAR image generation module, a feature extraction module and a neural network training module.

Images