Complex earth surface soil humidity high-precision intelligent inversion method based on satellite-borne GNSS-S

Abstract

The invention relates to a satellite-borne GNSS-S-based high-precision intelligent inversion method for soil humidity of a complex earth surface, which comprises the following steps of: acquiring multi-angle double-station image signals of the complex earth surface through a GNSS-S radar carried by a low-orbit satellite by taking a plurality of navigation satellites as signal sources; the data after multi-view processing is used for: (1) realizing full-link radiometric calibration from signal emission to signal reception by using a precision measurement device; (2) constructing a multi-parameter characteristic database containing surface physical characteristics and scattering characteristics; and (3) based on GNSS-S radar observation data and the feature database, establishing a deep learning intelligent inversion model to realize soil humidity quantitative inversion.

Description

Technical Field

[0001] This invention relates to a high-precision intelligent inversion method for complex surface soil moisture based on spaceborne GNSS-S, belonging to the field of radar technology design in the electronic information industry. Background Technology

[0002] Soil moisture is the primary source of water absorption for crops and a carrier for the dissolution and transport of soil nutrients, playing a crucial role in crop growth. Therefore, achieving high-precision measurement of soil moisture is of great significance for agricultural drought monitoring and crop yield estimation. Currently, microwave remote sensing is the most promising method for measuring soil moisture. Compared to optical remote sensing, microwave remote sensing has a longer wavelength, offering advantages such as all-day, all-weather measurement capabilities, and possesses a certain degree of penetration into the soil, effectively compensating for the shortcomings of optical remote sensing in soil moisture measurement. Existing microwave remote sensing technologies mainly include SAR (Synthetic Aperture Radar, active microwave remote sensing), radiometers (passive microwave remote sensing), and GNSS-R (Global Navigation Satellite System-Reflectometry).

[0003] SAR (Soil Surface Radar) is a high-resolution active microwave imaging radar used to measure soil moisture. However, due to the limited availability of surface scattering information, SAR struggles to completely eliminate the influence of factors such as vegetation cover and surface roughness, resulting in low accuracy in soil moisture measurements under complex terrain conditions. SAR also consumes a lot of power, cannot operate across the entire orbit, and cannot measure soil moisture for extended periods. Radiometers, a passive detection technology, retrieve soil moisture by receiving microwave radiation brightness and temperature from the soil itself. Similarly, due to the limited availability of surface scattering information, their accuracy in soil moisture measurements under complex terrain conditions is also low. Radiometers face challenges such as system complexity and high cost. GNSS-R, another passive detection technology, retrieves soil moisture by receiving reflected signals from specular areas. Again, due to the limited availability of surface scattering information, its accuracy in soil moisture measurements under complex terrain conditions is also low. GNSS-R can only measure soil moisture in a limited number of specular reflection areas and cannot achieve continuous measurements over a wide swath.

[0004] In summary, the core problem with existing remote sensing satellite detection methods is that, under the influence of complex surface environments (multi-layered non-homogeneous media) such as soil roughness and vegetation cover, the applicability of various microwave remote sensing soil moisture inversion methods is limited, making it difficult to meet the high-precision monitoring needs of soil moisture on large areas, in different regions, and with different characteristics. Existing microwave remote sensing methods acquire limited surface information, enabling the construction of incomplete soil moisture inversion models (similar to an "underdetermined equation"), and failing to accurately output soil moisture information for various complex surfaces. Summary of the Invention

[0005] The technical problem solved by this invention is to address the various technical deficiencies in the existing technology by proposing a high-precision intelligent inversion method for complex surface soil moisture based on spaceborne GNSS-S.

[0006] The present invention solves the above-mentioned technical problem through the following technical solution: A high-precision intelligent inversion method for complex surface soil moisture based on spaceborne GNSS-S includes: Multi-angle GNSS-S radar images of the complex terrain in the test area were acquired using GNSS-S radar. Perform full-link radiometric calibration on the GNSS-S radar image transmission link; The surface physical properties of the test area were collected and the scattering properties were measured. A multi-parameter feature database was constructed based on the obtained information. Train and deploy an intelligent inversion model for complex surface soil moisture inversion processing.

[0007] The method for acquiring multi-angle GNSS-S radar images of complex terrain using GNSS-S radar is as follows: Within a preset altitude range, a GNSS-S radar receiver is deployed on a low-Earth orbit satellite platform to synchronously receive the original echo signals of designated band signals transmitted by each navigation satellite after being scattered by the complex surface. The original echo signal is processed by bistatic synthetic aperture radar imaging to generate corresponding bistatic angle complex image data. Multi-view processing is performed on bistatic complex image data to improve the image signal-to-noise ratio to a preset range to obtain multi-angle GNSS-S radar images.

[0008] The method for performing end-to-end radiation calibration is as follows: The entire link is divided into a receiving link and a transmitting link. The receiving link is calibrated by a beacon, and the transmitting link is calibrated by a receiver. After the calibration is completed separately, the entire link is jointly calibrated by an active transponder.

[0009] The method for constructing a multi-parameter feature database is as follows: Obtain multispectral satellite images of complex terrain in the experimental area to extract normalized vegetation index, and record vegetation type and growth cycle based on ground survey records; The root mean square height of the soil was measured using a laser topographic scanner, and the relevant lengths were measured using a surface roughness meter. A drone equipped with a GNSS-S receiver was used to collect scattering data from various angles in typical scenarios; Record the azimuth angle and bistatic scattering angle of navigation satellites, and extract the backscattering coefficient and image statistical features; After all the obtained parameter information is summarized, a multi-parameter feature database is constructed.

[0010] The method for training and deploying the intelligent inversion model is as follows: Data is collected from a multi-parameter feature database, and the required data is used to standardize the true value sampling process. A pre-defined intelligent inversion model architecture is established. Data from a multi-parameter feature database is selected as the test dataset to perform performance testing on the inversion model architecture. After passing the performance test, the model is trained using the test dataset until it meets the requirements of the inversion task.

[0011] The true value sampling specification is as follows: remove surface vegetation within the specified latitude and longitude error range of the SAR image pixel center point of the required data; Soil volumetric water content was measured using a frequency domain reflectometer, with sampling at five points: east, west, south, north, and center. The sampling interval was less than the image resolution. The average of five points is taken as the true value label of soil moisture in the current test area, i.e., the true value sampling specification.

[0012] The preset algorithm required for constructing the intelligent inversion model adopts a support vector regression machine combined with a radial basis function kernel function, uses 16-dimensional feature vector data from a multi-parameter feature database as the input layer architecture, and uses soil moisture values ​​of complex surfaces in the test area as the model output to complete the construction of the intelligent inversion model.

[0013] The intelligent inversion model uses data from a multi-parameter feature database of typical surface target scenes as input datasets. Based on the latitude and longitude information at the time of data collection, it collects and measures field data that is consistent with the scattering information in time and space to obtain a surface target feature dataset as output dataset. The intelligent inversion model is trained using the input dataset and the output dataset. After training, the deep intelligent inversion model is used to provide soil moisture information.

[0014] During the collection of surface physical characteristics in the experimental area, the true value of soil moisture was detected. By selecting vegetation cover sampling points in the surface area, recording latitude, longitude, altitude, and vegetation parameters, removing the surface cover, inserting a soil moisture meter, and recording the moisture value and temperature, the moisture value was measured at five points in the four directions of front, back, left, and right, centered on the sampling point. The average value was then taken to collect experimental data.

[0015] In the radiation calibration process, a beacon is used to calibrate the radiation characteristics of the receiving link, which is then used to calibrate the receiving antenna gain. The calibration error is ≤0.3dB. By measuring the ground power flux density of BeiDou / GNSS satellites using a receiver, and constructing a BeiDou / GNSS satellite radiated power library based on deep learning, we can predict the radiated power of BeiDou / GNSS satellites under arbitrary configurations and calibrate the radiated power of navigation satellites. The calibration error is ≤0.4dB. The full-link radiation characteristics are calibrated using a precision active transponder, and the target scattering cross section in the radar equation is calibrated. The calibration error is ≤1dB, and it is used to achieve full-link variable calibration of GNSS-S radar.

[0016] The advantages of this invention compared to the prior art are: (1) The present invention provides a high-precision intelligent inversion method for soil moisture in complex surfaces based on spaceborne GNSS-S, which can break through the limitations of the dimensions of information acquisition in complex surfaces: the synchronous acquisition of multi-angle scattering signals greatly improves the surface feature characterization ability, effectively overcomes the interference of vegetation and roughness, and significantly improves the soil moisture inversion accuracy in complex environments. (2) The design of this invention takes into account both wide coverage and efficient continuous monitoring: relying on the wide-area signal coverage advantage of the navigation satellite constellation and combined with the flexible observation capability of low-orbit satellites, it realizes long-term continuous monitoring of soil moisture, and the power consumption is much lower than that of traditional active remote sensing systems. (3) This invention solves the problem of universality of inversion model through intelligent means: the scattering characteristic analysis model based on deep learning automatically adapts to different land surface types and environmental conditions, avoids errors in manual modeling, and significantly improves the accuracy and reliability of agricultural drought monitoring. Attached Figure Description

[0017] Figure 1 A system block diagram of the high-precision intelligent inversion method for complex surface soil moisture based on GNSS-S radar provided by the present invention; Figure 2 This is a diagram illustrating the operational scenario of the GNSS-S radar provided by the present invention. Figure 3 A diagram illustrating the radiation calibration working scenario provided by this invention; Figure 4 A flowchart of the radiation calibration technique provided by this invention; Figure 5 This is a schematic diagram of GNSS-S multi-station surface target scattering provided by the present invention; Figure 6 This is a schematic diagram of GNSS-S radar surface imaging provided by the present invention. Detailed Implementation

[0018] A high-precision intelligent inversion method for soil moisture on complex surfaces based on spaceborne GNSS-S uses multiple navigation satellites as signal sources. Multi-angle dual-station image signals of complex surfaces are acquired through GNSS-S radar on low-orbit satellites. The data after multi-view processing is used for: (1) full-link radiometric calibration from signal transmission to reception using precision measurement equipment; (2) constructing a multi-parameter feature database containing surface physical characteristics and scattering characteristics; and (3) establishing a deep learning intelligent inversion model based on GNSS-S radar observation data and feature database to achieve quantitative inversion of soil moisture.

[0019] The design steps of the high-precision intelligent inversion method for complex surface soil moisture are as follows: Multi-angle GNSS-S radar images of the complex terrain in the test area were acquired using GNSS-S radar. Perform full-link radiometric calibration on the GNSS-S radar image transmission link; The surface physical properties of the test area were collected and the scattering properties were measured. A multi-parameter feature database was constructed based on the obtained information. Train and deploy an intelligent inversion model for complex surface soil moisture inversion processing.

[0020] The method for acquiring multi-angle GNSS-S radar images of complex terrain using GNSS-S radar is as follows: Within a preset altitude range, a GNSS-S radar receiver is deployed on a low-Earth orbit satellite platform to synchronously receive the original echo signals of designated band signals transmitted by each navigation satellite after being scattered by the complex surface. The original echo signal is processed by bistatic synthetic aperture radar imaging to generate corresponding bistatic angle complex image data. Multi-view processing is performed on bistatic complex image data to improve the image signal-to-noise ratio to a preset range to obtain multi-angle GNSS-S radar images.

[0021] The method for performing end-to-end radiation calibration is as follows: The entire link is divided into a receiving link and a transmitting link. The receiving link is calibrated by a beacon, and the transmitting link is calibrated by a receiver. After the calibration is completed separately, the entire link is jointly calibrated by an active transponder.

[0022] The method for constructing a multi-parameter feature database is as follows: Obtain multispectral satellite images of complex terrain in the experimental area to extract normalized vegetation index, and record vegetation type and growth cycle based on ground survey records; The root mean square height of the soil was measured using a laser topographic scanner, and the relevant lengths were measured using a surface roughness meter. A drone equipped with a GNSS-S receiver was used to collect scattering data from various angles in typical scenarios; Record the azimuth angle and bistatic scattering angle of navigation satellites, and extract the backscattering coefficient and image statistical features; After all the obtained parameter information is summarized, a multi-parameter feature database is constructed.

[0023] The method for training and deploying the intelligent inversion model is as follows: Data is collected from a multi-parameter feature database, and the required data is used to standardize the true value sampling process. A pre-defined intelligent inversion model architecture is established. Data from a multi-parameter feature database is selected as the test dataset to perform performance testing on the inversion model architecture. After passing the performance test, the model is trained using the test dataset until it meets the requirements of the inversion task.

[0024] The true value sampling specification is: remove surface vegetation within the specified latitude and longitude error range of the SAR image pixel center point of the required data; Soil volumetric water content was measured using a frequency domain reflectometer, with sampling at five points: east, west, south, north, and center. The sampling interval was less than the image resolution. The average of five points is taken as the true value label of soil moisture in the current test area, i.e., the true value sampling specification.

[0025] The pre-set algorithm required for building the intelligent inversion model adopts a support vector regression machine combined with a radial basis function kernel function. It uses 16-dimensional feature vector data from a multi-parameter feature database as the input layer architecture and soil moisture values ​​of complex surfaces in the experimental area as the model output to complete the construction of the intelligent inversion model.

[0026] The intelligent inversion model uses data from a multi-parameter feature database of typical surface target scenes as input datasets. Based on the latitude and longitude information at the time of data collection, it collects and measures field data that are consistent with the scattering information in time and space to obtain a surface target feature dataset as output dataset. The intelligent inversion model is trained using the input dataset and the output dataset. After training, the deep intelligent inversion model is used to provide soil moisture information.

[0027] During the collection of surface physical characteristics in the experimental area, the true value of soil moisture was detected. By selecting vegetation cover sampling points in the surface area, recording latitude, longitude, altitude, and vegetation parameters, removing the surface cover, inserting a soil moisture meter, and recording the moisture value and temperature, the moisture value was measured at five points in the four directions of front, back, left, and right, centered on the sampling point. The average value was then taken to collect experimental data.

[0028] In the radiation calibration process, a beacon is used to calibrate the radiation characteristics of the receiving link, which is then used to calibrate the receiving antenna gain. The calibration error is ≤0.3dB. By measuring the ground power flux density of BeiDou / GNSS satellites using a receiver, and constructing a BeiDou / GNSS satellite radiated power library based on deep learning, we can predict the radiated power of BeiDou / GNSS satellites under arbitrary configurations and calibrate the radiated power of navigation satellites. The calibration error is ≤0.4dB. The full-link radiation characteristics are calibrated using a precision active transponder, and the target scattering cross section in the radar equation is calibrated. The calibration error is ≤1dB, and it is used to achieve full-link variable calibration of GNSS-S radar.

[0029] The following description, in conjunction with the accompanying drawings and preferred embodiments, provides further details: In this embodiment, the following steps are performed: Step S1: Synchronous acquisition of multi-angle GNSS-S scattering signals (1) Deploy a GNSS-S radar receiver on a low-orbit satellite platform at an altitude of 500-800km to simultaneously receive the echoes of L-band (1.2-1.6GHz) signals emitted by at least 8 navigation satellites (BeiDou / GPS / Galileo) after they are scattered by the ground surface; (2) Perform bistatic synthetic aperture radar (SAR) imaging processing on the original echo to generate multiple complex image data with different bistatic angles (covering a range of 15°-60°); (3) Perform multi-view processing to improve the image signal-to-noise ratio to over 10dB; Step S2: Full-link radiation calibration (1) Receiver link calibration: A standard gain horn antenna is used as a high-precision beacon to measure the gain of the receiving antenna. The calibration error is controlled within 0.3dB; (2) Transmit link calibration: GNSS satellite ground power flux density was measured using a ground-based broadband high-sensitivity receiver. ; A GNSS satellite radiated power library was constructed based on an LSTM deep learning model to realize the radiated power under arbitrary spatiotemporal conditions. Prediction error ≤ 0.4dB). (3) Joint calibration across the entire supply chain: Precision active transponders (transponder gain error ≤ 0.1dB) are deployed in the detection area. The target cross-section is calibrated to satisfy the radar equations, and the total system calibration error is ≤1dB. Step S3: Construction of Multi-parameter Feature Database (1) Collection of surface physical characteristics: Vegetation parameters: Normalized Difference Vegetation Index (NDVI) was extracted from multispectral satellite imagery and combined with ground survey records of vegetation type and growth cycle; Soil parameters: The root mean square height of the soil was measured using a laser topographic scanner (1.2-5.6 cm), and the relevant length was measured using a surface roughness meter (8-35 cm). (2) Actual measurement of scattering characteristics; A drone equipped with a GNSS-S receiver was used to collect multi-angle scattering data in typical scenarios such as farmland and forest. Record the azimuth angle and bistatic scattering angle of navigation satellites, and extract the backscattering coefficient (-25~-10 dB) and image statistical features; Step S4: Intelligent Inversion Model Training and Deployment (1) Truth sampling specification: Remove surface vegetation at the center point of SAR image pixels (latitude and longitude error ≤ 0.001°); Soil volumetric water content was measured using a frequency domain reflectometer (FDR), with samples taken at five points: east, west, south, north, and center (interval smaller than the image resolution). The average of five points is taken as the true soil moisture label for this pixel (unit: m³ / m³). (2) Model architecture design Input layer: 16-dimensional feature vector (including backscattering coefficient, NDVI, roughness, etc.); Algorithm: Support Vector Regression (SVR) with Radial Basis Function (RBF); Output: Soil moisture value (0.05~0.45 m³ / m³); (3) Performance verification The root mean square error (RMSE) of the test set is ≤0.03 m³ / m³, and the correlation coefficient is [missing information]. ≥0.92; Furthermore, such as Figure 1 The diagram shown illustrates a high-precision intelligent inversion method for complex surface soil moisture based on GNSS-S radar, as proposed in this embodiment. The method includes: GNSS-S Radar 10 is used to acquire multi-angle GNSS-S radar images; Multi-view processing 20 is used for multi-view processing of GNSS-S radar images; Radiation calibration 30 is used to complete end-to-end radiation calibration. The soil moisture true value of the detection area is 40, which is used to complete the soil parameter sampling at the sampling points in the test area. A multi-layered non-uniform medium surface target feature database 50 is used to construct a target feature database for complex surfaces; Deep learning intelligent inversion model 60, used to construct deep learning intelligent inversion models; The output value is 70, which is used to output the soil moisture of complex surfaces.

[0030] like Figure 1 As shown, GNSS-S radar 10 is used to acquire multi-angle GNSS-S radar images of complex terrain. Multi-view processing 20 is used to perform multi-view processing on GNSS-S radar images to obtain bistatic images with an image signal-to-noise ratio (SNR) ≥ X dB, where X ≥ 10; Radiometric calibration 30 is used to complete the full-link radiometric calibration of GNSS-S radar. The radiometric calibration accuracy should be better than 1dB to meet the requirements of soil moisture inversion. The true soil moisture value of the detection area is as follows: ① Select a sampling point with vegetation cover near the ground surface (original "landmark") and record the latitude, longitude, altitude, and vegetation parameters; ② Remove the surface cover, insert a soil moisture meter, and record the moisture value and temperature; ③ With the sampling point as the center, take one point in each of the four directions (distance from the center point x < image resolution), for a total of five points, measure the moisture value, and take the average. Multilayer Non-homogeneous Medium Surface Target Feature Database 50: This module studies and analyzes the model parameters of domestic and foreign research on the scattering mechanism of surface targets, and summarizes the parameters that can accurately describe the scattering characteristics of surface targets. For surface vegetation, vegetation index is used to describe it, and for surface soil, root mean square height, surface correlation length, and surface roughness are used. In addition, other parameters are explored based on a large amount of collected sample data, thus forming a feature database that can comprehensively characterize the scattering characteristics of surface targets. The deep learning intelligent inversion model 60 includes: 601, which determines whether the inversion model parameters have been trained; and 602, which trains the deep learning intelligent inversion model using a multi-station scattered echo signal database of typical surface target scenes as the input dataset. Based on the latitude and longitude information at the time of data collection, it collects and measures field data that is spatiotemporally consistent with the scattered information to obtain a surface target feature dataset as the output dataset. A support vector regression model is used to train the deep learning model using the established input and output datasets. 603, the deep learning intelligent inversion model 603, provides soil moisture information based on the input. Output 70: This output provides the soil moisture distribution within the detection area. like Figure 2 As shown, the GNSS-S radar uses a multi-station navigation signal source with multi-angle characteristics as an external radiation source. The radar receiver is mounted on an aircraft or satellite to receive the echo signal backscattered after passing through the complex surface. After signal processing, it can realize the detection and imaging of targets on the complex surface. Soil moisture can be further retrieved later. like Figure 3 As shown, this embodiment uses a high-precision signal receiver to calibrate the transmit link, a high-precision signal generator to calibrate the receive link, and an active transponder to calibrate the entire link. The three measuring instruments are spaced apart. Less than half the system resolution; like Figure 4 As shown, the radiation calibration process consists of the following three steps: ① Calibrate the radiation characteristics of the receiving link using a high-precision beacon, that is, calibrate the receiving antenna gain in the radar equation. ① The calibration error is ≤0.3dB; ② The landing power flux density of BeiDou / GNSS satellites is measured using a high-precision receiver, and a BeiDou / GNSS satellite radiated power library is constructed based on deep learning to achieve prediction of BeiDou / GNSS satellite radiated power under arbitrary configurations, that is, to calibrate the navigation satellite radiated power in the radar equation. ③ The calibration error is ≤0.4dB; ③ The full-link radiation characteristics are calibrated using a precision active transponder, that is, the target scattering cross-section in the radar equation is calibrated. The calibration error is ≤1dB, thereby realizing full-link variable calibration of GNSS-S radar and improving the accuracy of quantitative remote sensing inversion of land and sea surfaces. like Figure 5 As shown, the Earth's surface environment consists of a vegetation layer and a soil layer, forming a highly complex, non-homogeneous medium structure. The echo signal received by GNSS is a composite signal formed by the navigation signal propagating through the atmosphere, through direct scattering by vegetation, direct scattering by soil, and mixed scattering by soil and vegetation. Existing GNSS empirical models and physical scattering models are insufficient to accurately describe this. Figure 5 The multipath scattering process and its interaction with vegetation, soil, and geometric structures are illustrated. Therefore, this patent employs a strategy combining extensive measured data with deep learning methods to study the scattering mechanism of complex surfaces and establish a mapping relationship between measured data and surface target characteristics. like Figure 6 As shown, this embodiment successfully conducted an airborne remote sensing test using a self-developed GNSS-S new radar payload, obtained measured data of various types of surface soil, completed the development of surface imaging algorithms, and obtained the world's first GNSS-S radar imaging image.

[0031] Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make possible changes and modifications to the technical solutions of the present invention by utilizing the methods and techniques disclosed above without departing from the spirit and scope of the present invention. Therefore, any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solutions of the present invention shall fall within the protection scope of the technical solutions of the present invention.

[0032] The contents not described in detail in this specification are common knowledge to those skilled in the art.

Claims

1. A high-precision intelligent inversion method for complex surface soil moisture based on spaceborne GNSS-S, characterized in that... include: Multi-angle GNSS-S radar images of the complex terrain in the test area were acquired using GNSS-S radar. Perform full-link radiometric calibration on the GNSS-S radar image transmission link; The surface physical properties of the test area were collected and the scattering properties were measured. A multi-parameter feature database was constructed based on the obtained information. Train and deploy an intelligent inversion model for complex surface soil moisture inversion processing.

2. The high-precision intelligent inversion method for complex surface soil moisture based on spaceborne GNSS-S as described in claim 1, characterized in that: The method for acquiring multi-angle GNSS-S radar images of complex terrain using GNSS-S radar is as follows: Within a preset altitude range, a GNSS-S radar receiver is deployed on a low-Earth orbit satellite platform to synchronously receive the original echo signals of designated band signals transmitted by each navigation satellite after being scattered by the complex surface. The original echo signal is processed by bistatic synthetic aperture radar imaging to generate corresponding bistatic angle complex image data. Multi-view processing is performed on bistatic complex image data to improve the image signal-to-noise ratio to a preset range to obtain multi-angle GNSS-S radar images.

3. The high-precision intelligent inversion method for complex surface soil moisture based on spaceborne GNSS-S as described in claim 2, characterized in that: The method for performing end-to-end radiation calibration is as follows: The entire link is divided into a receiving link and a transmitting link. The receiving link is calibrated by a beacon, and the transmitting link is calibrated by a receiver. After the calibration is completed separately, the entire link is jointly calibrated by an active transponder.

4. The high-precision intelligent inversion method for complex surface soil moisture based on spaceborne GNSS-S as described in claim 3, characterized in that: The method for constructing a multi-parameter feature database is as follows: Obtain multispectral satellite images of complex terrain in the experimental area to extract normalized vegetation index, and record vegetation type and growth cycle based on ground survey records; The root mean square height of the soil was measured using a laser topographic scanner, and the relevant lengths were measured using a surface roughness meter. A drone equipped with a GNSS-S receiver was used to collect scattering data from various angles in typical scenarios; Record the azimuth angle and bistatic scattering angle of navigation satellites, and extract the backscattering coefficient and image statistical features; After all the obtained parameter information is summarized, a multi-parameter feature database is constructed.

5. The high-precision intelligent inversion method for complex surface soil moisture based on spaceborne GNSS-S according to claim 4, characterized in that: The method for training and deploying the intelligent inversion model is as follows: Data is collected from a multi-parameter feature database, and the required data is used to standardize the true value sampling. A pre-defined intelligent inversion model architecture is established. Data from a multi-parameter feature database is selected as the test dataset to perform performance testing on the inversion model architecture. After passing the performance test, the model is trained using the test dataset until it meets the requirements of the inversion task.

6. The high-precision intelligent inversion method for complex surface soil moisture based on spaceborne GNSS-S as described in claim 5, characterized in that: The true value sampling specification is as follows: remove surface vegetation within the specified latitude and longitude error range of the SAR image pixel center point of the required data; Soil volumetric water content was measured using a frequency domain reflectometer, with sampling at five points: east, west, south, north, and center. The sampling interval was less than the image resolution. The average of five points is taken as the true value label of soil moisture in the current test area, i.e., the true value sampling specification.

7. The high-precision intelligent inversion method for complex surface soil moisture based on spaceborne GNSS-S as described in claim 5, characterized in that: The preset algorithm required for constructing the intelligent inversion model adopts a support vector regression machine combined with a radial basis function kernel function, uses 16-dimensional feature vector data from a multi-parameter feature database as the input layer architecture, and uses soil moisture values ​​of complex surfaces in the test area as the model output to complete the construction of the intelligent inversion model.

8. The high-precision intelligent inversion method for complex surface soil moisture based on spaceborne GNSS-S as described in claim 5, characterized in that: The intelligent inversion model uses data from a multi-parameter feature database of typical surface target scenes as input datasets. Based on the latitude and longitude information at the time of data collection, it collects and measures field data that is consistent with the scattering information in time and space to obtain a surface target feature dataset as output dataset. The intelligent inversion model is trained using the input dataset and the output dataset. After training, the deep intelligent inversion model is used to provide soil moisture information.

9. The high-precision intelligent inversion method for complex surface soil moisture based on spaceborne GNSS-S according to claim 8, characterized in that: During the collection of surface physical characteristics in the experimental area, the true value of soil moisture was detected. By selecting vegetation cover sampling points in the surface area, recording latitude, longitude, altitude, and vegetation parameters, removing the surface cover, inserting a soil moisture meter, and recording the moisture value and temperature, the moisture value was measured at five points in the four directions of front, back, left, and right, centered on the sampling point. The average value was then taken to collect experimental data.

10. The high-precision intelligent inversion method for complex surface soil moisture based on spaceborne GNSS-S according to claim 9, characterized in that: In the radiation calibration process, a beacon is used to calibrate the radiation characteristics of the receiving link, which is then used to calibrate the receiving antenna gain. The calibration error is ≤0.3dB. By measuring the ground power flux density of BeiDou / GNSS satellites using a receiver, and constructing a BeiDou / GNSS satellite radiated power library based on deep learning, we can predict the radiated power of BeiDou / GNSS satellites under arbitrary configurations and calibrate the radiated power of navigation satellites. The calibration error is ≤0.4dB. The full-link radiation characteristics are calibrated using a precision active transponder, and the target scattering cross section in the radar equation is calibrated. The calibration error is ≤1dB, and it is used to achieve full-link variable calibration of GNSS-S radar.

Images