Soil moisture inversion method and system based on GNSS-R and radiometer fusion

Abstract

The application discloses a soil moisture inversion method and system based on GNSS-R and radiometer fusion, and comprises the following steps: obtaining direct signal target polarization reflection signals acquired by GNSS receivers in a target soil area, and calculating target polarization reflectivity; obtaining brightness temperature data, soil roughness and comprehensive vegetation attenuation coefficients acquired by a microwave radiometer in the target area; according to an algorithm, one or more kinds of polarization mode reflectivity can be selected, and the brightness temperature data, the soil roughness and the comprehensive vegetation attenuation coefficients are input into an LSTM model to invert soil moisture in the target soil area. The brightness temperature data, the soil roughness and the comprehensive vegetation attenuation coefficients acquired by the microwave radiometer are used to weaken the influence of external environment such as ground roughness, and the LSTM model is combined to invert soil moisture, so that the soil moisture inversion method has higher precision and stronger noise resistance due to the strong noise resistance of the LSTM itself.

Description

Technical Field

[0001] This invention relates to the field of soil moisture inversion, and in particular to a soil moisture inversion method and system based on LSTM-based GNSS-R and radiometer fusion. Background Technology

[0002] GNSS navigation satellite reflection signal detection technology is a novel remote sensing technology that lies between passive and active remote sensing. It can be viewed as a non-cooperative artificial radiation source, a separate transmit and receive L-band radar system with multiple transmitters and a single receiver. Its receiver structure is simple, and its cost, size, and weight are relatively small, allowing for flexible mounting on ground-based, airborne, and spaceborne platforms. Currently, GNSS-R technology applications are mainly concentrated in marine and land detection fields. However, in soil moisture retrieval, there are problems such as significant signal interference, high noise levels, and low data retrieval accuracy due to the influence of surface vegetation. Therefore, this paper proposes a soil moisture retrieval method and system based on LSTM-fused GNSS-R and radiometer to reduce noise impact and improve retrieval accuracy. Summary of the Invention

[0003] The purpose of this invention is to provide a soil moisture inversion method and system that integrates GNSS-R and radiometer. It utilizes brightness temperature data obtained from microwave radiometer, as well as soil roughness and comprehensive vegetation attenuation coefficient to reduce the influence of external environment such as surface roughness, and combines it with LSTM model for soil moisture inversion. By taking advantage of the strong noise resistance of LSTM itself, the soil moisture inversion method has higher accuracy and stronger noise resistance.

[0004] To achieve the above objectives, the present invention provides the following solution:

[0005] A method for soil moisture inversion integrating GNSS-R and radiometer, the method comprising:

[0006] Acquire the direct signal and target polarization reflection signal obtained by the GNSS receiver within the target soil area; the target polarization reflection signal is at least one of the following: left-hand circular polarization reflection signal, right-hand circular polarization reflection signal, vertical polarization reflection signal, and horizontal polarization reflection signal;

[0007] The target polarization reflectivity is calculated based on the direct signal and the target polarization reflection signal.

[0008] Obtain brightness temperature data, soil roughness, and comprehensive vegetation attenuation coefficient from the radiometer within the target area;

[0009] The brightness temperature data, soil roughness, comprehensive vegetation attenuation coefficient, and target polarization reflectance are input into the LSTM model to retrieve the soil moisture in the target soil area.

[0010] Optionally, the expression for the target polarization reflectivity includes:

[0011]

[0012] In the formula, Γ pq The actual reflectivity is represented by Γ, where p represents the polarization of the satellite transmitting antenna (right-hand circular polarization for GNSS-R), and q represents the polarization of the receiving antenna for the reflected signal. The four polarizations are: left-hand circular polarization, right-hand circular polarization, vertical polarization, and horizontal polarization. The reflectivity for each of these four polarizations is: Γ. rl , Γ rr , Γ rv , Γ rh ;P r This represents the correlated power of the reflected signal, including reflected signals with four polarization modes; P d G represents the correlated power of the direct signal; d For the direct antenna gain, G r Let θ be the gain of the reflecting antenna, and θ be the incident angle of the antenna. It is the azimuth angle.

[0013] Optionally, the formulas for calculating the soil roughness and the comprehensive vegetation attenuation coefficient are as follows:

[0014] TR = τ NAD +H R / 2

[0015] Where TR represents soil roughness and the overall vegetation attenuation coefficient; τ NAD Indicates optical thickness; H R Indicates soil roughness.

[0016] Optionally, before inputting the brightness temperature data, the soil roughness and the comprehensive vegetation attenuation coefficient, and the reflectance of the selected polarization mode into the LSTM model, the method further includes:

[0017] Construct a hyperparameter set for the LSTM model and set the value range of each hyperparameter in the hyperparameter set; the hyperparameters include the number of nodes, maximum number of iterations, and learning rate of each LSTM layer of the LSTM model;

[0018] The hyperparameter set and training data are input into the Bayesian optimizer, which outputs the optimized hyperparameter set.

[0019] The hyperparameters in the optimized hyperparameter set are used as the hyperparameters of the LSTM model to obtain the assigned LSTM model.

[0020] The validation data is input into the assigned LSTM model, and the root mean square error of the model is calculated; the training data and the validation data are all historical brightness temperature data, historical soil roughness and comprehensive vegetation attenuation coefficient, and historical target polarization reflectance.

[0021] When the root mean square error is greater than a preset value and the number of parameter iterations is less than the preset number of iterations, return to the step "input the hyperparameter set and historical training data into the Bayesian optimizer and output the optimized hyperparameter set";

[0022] When the root mean square error is less than a preset value or the number of parameter iterations reaches a preset number of iterations, the hyperparameter set corresponding to the currently assigned LSTM model is the optimal hyperparameter set.

[0023] The present invention also provides a soil moisture retrieval system integrating GNSS-R and radiometer, the system comprising:

[0024] The first data acquisition module is used to acquire the direct signal and the target polarization reflection signal acquired by the GNSS receiver in the target soil area; the target polarization reflection signal is at least one of the following: left-hand circular polarization reflection signal, right-hand circular polarization reflection signal, vertical polarization reflection signal, and horizontal polarization reflection signal;

[0025] The reflectivity calculation module is used to calculate the target polarization reflectivity based on the direct signal and the target polarization reflection signal.

[0026] The second data acquisition module is used to acquire brightness temperature data, soil roughness, and comprehensive vegetation attenuation coefficient obtained by microwave radiometers within the target area.

[0027] The inversion module is used to input the brightness temperature data, the soil roughness and the comprehensive vegetation attenuation coefficient, and the target polarization reflectance into the target LSTM model to invert the soil moisture in the target soil area.

[0028] Optionally, the expression for the polarization reflectivity of each target in the first data acquisition module includes:

[0029]

[0030] In the formula, Γ pq The actual reflectivity is represented by Γ, where p represents the polarization of the satellite transmitting antenna (right-hand circular polarization for GNSS-R), and q represents the polarization of the receiving antenna for the reflected signal. The four polarizations are: left-hand circular polarization, right-hand circular polarization, vertical polarization, and horizontal polarization. The reflectivity for each of these four polarizations is: Γ. rl , Γ rr , Γ rv , Γ rh Pr This represents the correlated power of the reflected signal, including reflected signals with four polarization modes; P d G represents the correlated power of the direct signal; d For the direct antenna gain, G r Let θ be the gain of the reflecting antenna, and θ be the incident angle of the antenna. It is the azimuth angle.

[0031] This invention also provides a soil moisture inversion device that integrates GNSS-R and radiometer, comprising:

[0032] Processor; and

[0033] Memory, used to store computer-readable program instructions;

[0034] The soil moisture inversion method of GNSS-R and radiometer fusion as described in any one of claims 1-4 is executed when the computer-readable program instructions are run by the processor.

[0035] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method according to any one of claims 1-4.

[0036] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0037] This invention provides a soil moisture retrieval method and system that integrates GNSS-R and radiometer data. Brightness temperature data, soil roughness, comprehensive vegetation attenuation coefficient, and reflectance with a selected polarization mode are input into an LSTM model to retrieve soil moisture within a target soil area. The introduction of brightness temperature data from the radiometer, along with soil roughness and the comprehensive vegetation attenuation coefficient, mitigates the influence of external environmental factors such as surface roughness. Furthermore, due to the black-box nature of neural networks, LSTM itself possesses strong noise resistance. Therefore, the LSTM-based GNSS-R and radiometer fusion soil moisture retrieval method offers advantages over traditional methods, including higher accuracy and stronger noise resistance. Attached Figure Description

[0038] 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 introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0039] Figure 1 Here is a flowchart of a soil moisture inversion method that integrates GNSS-R and radiometer, as provided in Embodiment 1 of the present invention;

[0040] Figure 2 The flowchart of the Bayesian optimizer provided in Embodiment 1 of the present invention is shown. Detailed Implementation

[0041] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0042] The purpose of this invention is to provide a soil moisture inversion method and system that integrates GNSS-R and radiometer. By introducing brightness temperature data output by the radiometer, as well as soil roughness and comprehensive vegetation attenuation coefficient, the influence of external environment such as surface roughness is weakened. Furthermore, due to the black-box characteristics of neural networks, LSTM itself has strong noise resistance. Therefore, the soil moisture inversion method based on LSTM that integrates GNSS-R and radiometer has advantages such as high accuracy and strong noise resistance compared with traditional methods.

[0043] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0044] Example 1

[0045] like Figure 1 As shown in the figure, this embodiment provides a soil moisture inversion method that integrates GNSS-R and radiometer, the method comprising:

[0046] S1: Acquire the direct signal and target polarization reflection signal obtained by the GNSS receiver within the target soil area; the target polarization reflection signal is at least one of the following: left-hand circular polarization reflection signal, right-hand circular polarization reflection signal, vertical polarization reflection signal, and horizontal polarization reflection signal.

[0047] In this embodiment, dual-channel or single-channel reception of reflected signals with different polarizations can be selected. For example, when sampling dual channels, the GNSS receiver has two input channels. In order to receive signals from one direct antenna and two reflective antennas with different circular polarizations, data acquisition is divided into two stages: the first stage and the second stage, which are sequential in acquisition time. In the first stage (1-LD), channel 1 is responsible for receiving the reflected left-hand circularly polarized (LHCP) signal, and channel 2 is responsible for receiving the direct right-hand circularly polarized (RHCP) signal. Then, in the second stage (2-DR), channel 1 is responsible for receiving the direct right-hand circularly polarized (RHCP) signal, and channel 2 is responsible for receiving the reflected right-hand circularly polarized (RHCP) signal. In this way, the dual-channel receiver can simultaneously receive signals from one direct antenna and two reflective antennas, and the channel differences can be eliminated by calibrating the direct signals received by different channels.

[0048] S2: Calculate the target polarization reflectivity based on the direct signal and the target polarization reflection signal.

[0049] The expression for the target polarization reflectivity includes:

[0050]

[0051] In the formula, Γ pq The actual reflectivity is represented by Γ, where p represents the polarization of the satellite transmitting antenna (right-hand circular polarization for GNSS-R), and q represents the polarization of the receiving antenna for the reflected signal. The four polarizations are: left-hand circular polarization, right-hand circular polarization, vertical polarization, and horizontal polarization. The reflectivity for each of these four polarizations is: Γ. rl , Γ rr , Γ rv , Γ rh P r This represents the correlated power of the reflected signal, including reflected signals with four polarization modes; P d G represents the correlated power of the direct signal; d For the direct antenna gain, G r For the gain of the reflective antenna, these are all related to the incident angle θ and azimuth angle of the signal entering the antenna. related.

[0052] Specifically, when acquiring dual-channel data, the left-hand circularly polarized reflectivity and right-hand circularly polarized reflectivity are calculated based on the direct signal, the left-hand circularly polarized signal, and the right-hand circularly polarized signal. The relevant calculation formulas are as follows:

[0053] K = mPD2 / mPD1

[0054] R rl =(mPR1 / mPD1) / K

[0055] R rr = (mPR2 / mPD2)*K

[0056] Among them, R rl R represents the left-hand circularly polarized reflectivity. rr denoted by , K represents the channel difference, which is the ratio of the correlated power of the direct signal received through the two channels. mPD1 and mPD2 represent the average peak value of the correlated power of the direct signal received through channel 1 and channel 2, respectively, while mPR1 and mPR2 represent the average peak value of the correlated power waveform of the left-hand circularly polarized and right-hand circularly polarized reflected signals.

[0057] S3: Obtain brightness temperature data, soil roughness, and comprehensive vegetation attenuation coefficient from the microwave radiometer within the target area.

[0058] Brightness temperature, soil roughness, and overall vegetation attenuation coefficient were obtained from the radiometer.

[0059] Soil roughness and overall vegetation attenuation coefficient TR are:

[0060] TR = τ NAD +H R / 2

[0061] Where, τ NAD H represents optical thickness. R This indicates surface roughness.

[0062] S4: Input the brightness temperature data, soil roughness, comprehensive vegetation attenuation coefficient, and target polarization reflectance into the LSTM model to retrieve the soil moisture in the target soil area.

[0063] When the target polarization reflectance is selected as left-handed circular polarization reflectance, only left-handed circular polarization reflectance can be selected and input into the LSTM model along with the brightness temperature data, the soil roughness, and the comprehensive vegetation attenuation coefficient to invert the soil moisture in the target soil area.

[0064] For LSTM models, you can choose the LSTM model. The LSTM model is an improved recurrent neural network model. In order to address the gradient vanishing or exploding problem caused by error propagation in traditional recurrent neural networks, the LSTM neural network model introduces input gate, forget gate and output gate to control the transmission of information, effectively avoiding the occurrence of gradient vanishing and exploding problems, and realizing the effective processing of long-term sequence information.

[0065] The LSTM network consists of five layers: an input layer, an LSTM layer, a Dropout layer, a fully connected layer, and an output layer. The input data are Rrl (left-handed reflectivity), TB (luminance temperature), and TR (soil roughness and combined vegetation attenuation coefficient), and the output data is SMC (soil moisture).

[0066] Before inputting the brightness temperature data, soil roughness, comprehensive vegetation attenuation coefficient, and left-handed circularly polarized reflectance into the LSTM model, the method further includes: adjusting the neural network parameters using a Bayesian algorithm.

[0067] The parameters that need to be optimized in this invention include the number of nodes in each LSTM layer, the maximum number of iterations, and the learning rate. The specific steps are as follows:

[0068] 1. Set the Bayesian optimization parameters, including the initial estimated point and the maximum number of iterations;

[0069] 2. Set the search range for the parameters to be optimized, including the range of the number of nodes in each LSTM layer, the maximum number of iterations, and the learning rate range, and create a set of parameters to be optimized;

[0070] 3. Use the training and validation data as input to create the objective function for the Bayesian optimizer.

[0071] 4. The objective function trains the LSTM network, and the optimizer returns the validation RMSE on the validation set after each iteration;

[0072] 5. Using RMSE as the evaluation metric, iterate until the maximum number of iterations is reached. The parameter combination with the minimum RMSE is the optimal hyperparameter combination for the neural network.

[0073] Using Bayesian algorithms to adjust neural network parameters specifically includes:

[0074] Construct a hyperparameter set for the LSTM model and set the value range of each hyperparameter in the hyperparameter set; the hyperparameters include the number of nodes, maximum number of iterations, and learning rate of each LSTM layer of the LSTM model.

[0075] The hyperparameter set and training data are input into the Bayesian optimizer, which outputs the optimized hyperparameter set.

[0076] The hyperparameters in the optimized hyperparameter set are used as the hyperparameters of the LSTM model to obtain the assigned LSTM model.

[0077] The validation data is input into the assigned LSTM model, and the root mean square error of the model is calculated. The training data and the validation data are historical brightness temperature data, historical soil roughness and comprehensive vegetation attenuation coefficient, and historical target polarization reflectance (left-handed circular polarization reflectance or historical right-handed circular polarization reflectance or historical vertical polarization reflectance or historical horizontal polarization reflectance).

[0078] When the root mean square error is greater than a preset value and the number of parameter iterations is less than the preset number of iterations, return to the step "input the hyperparameter set and historical training data into the Bayesian optimizer and output the optimized hyperparameter set";

[0079] When the root mean square error is less than a preset value or the number of parameter iterations reaches a preset number of iterations, the hyperparameter set corresponding to the currently assigned LSTM model is the optimal hyperparameter set.

[0080] The following is the theoretical optimization principle of the Bayesian optimizer:

[0081] The Bayesian optimization objective is defined as:

[0082]

[0083] In the formula, A and B are both random events; P(A) is the probability of event A occurring, which is called the prior probability; P(B|A) is the conditional probability of event B occurring given that event A has occurred; P(B) is the probability of event B occurring; P(A|B) is the conditional probability of event A occurring given that event B has occurred, which is called the posterior probability.

[0084] The core of the Bayesian optimization algorithm is to employ a probabilistic surrogate model and a sampling function, finding the optimal solution through an execution method similar to Bayes' theorem. The surrogate model fits the surrogate function based on the sampled data, significantly reducing computational cost. The algorithm uses a Gaussian process as the probabilistic surrogate model and estimates the mean μ(x) and variance σ of the posterior probability distribution. 2 (x). Modeling the objective function:

[0085] f(x)≈N(μ(x),σ 2 (x))

[0086] In the formula, N represents a Gaussian distribution.

[0087] The probabilistic surrogate model abstracts the data distribution mechanism P(B|A) from the observed prior probability P(A). Then, based on the updated data P(B), it obtains the posterior distribution P(A|B) containing more information, and estimates the distribution range using the above formula. Subsequently, the acquisition function balances "exploration" and "development." Exploration seeks solutions in areas with fewer data points, corresponding to areas of high uncertainty, while development seeks solutions in areas with higher expected values.

[0088] During the solution process, the upper confidence limit U is used. UCB Let μ(x) and σ represent the mean μ(x) and variance σ. 2 The weighted sum of (x) is calculated using the following formula:

[0089] U UCB ≈μ(x)+kσ 2 (x)

[0090] In the formula, k is a parameter used to balance exploration and development.

[0091] The location of the next evaluation point is determined by comparing the maximum value corresponding to the confidence upper limit. By repeating this process iteratively, the optimal hyperparameters can be found. Figure 2 The optimization flowchart of the Bayesian optimizer is shown.

[0092] In this embodiment, the brightness temperature data output by the radiometer, as well as soil roughness and comprehensive vegetation attenuation coefficient, are introduced to weaken the influence of external environmental factors such as surface roughness. Furthermore, due to the black-box nature of neural networks, LSTM itself has strong noise resistance. Therefore, the soil moisture inversion method based on LSTM and GNSS-R fusion with radiometer has advantages such as high accuracy and strong noise resistance compared with traditional methods. Moreover, since the Bayesian method is introduced to adjust the network parameters, the soil moisture inversion is automated and requires no manual intervention.

[0093] Example 2

[0094] This embodiment provides a soil moisture inversion system that integrates GNSS-R and radiometer, the system comprising:

[0095] The first data acquisition module M1 is used to acquire the direct signal and the target polarization reflection signal acquired by the GNSS receiver in the target soil area; the target polarization reflection signal is at least one of the following: left-hand circular polarization reflection signal, right-hand circular polarization reflection signal, vertical polarization reflection signal and horizontal polarization reflection signal.

[0096] The reflectivity calculation module M2 is used to calculate the target polarization reflectivity based on the direct signal and the target polarization reflection signal.

[0097] The second data acquisition module M3 is used to acquire brightness temperature data, soil roughness, and comprehensive vegetation attenuation coefficient obtained by microwave radiometers within the target area.

[0098] The inversion module M4 is used to input the target polarization reflectance, along with the brightness temperature data, the soil roughness, and the comprehensive vegetation attenuation coefficient, into the LSTM model to invert the soil moisture in the target soil area.

[0099] Example 3

[0100] This embodiment provides a soil moisture inversion device that integrates GNSS-R and radiometer, including:

[0101] Processor; and

[0102] Memory, used to store computer-readable program instructions;

[0103] The computer-readable program instructions are executed by the processor to perform the soil moisture inversion method of GNSS-R and radiometer fusion as described in Example 1.

[0104] Example 4

[0105] This embodiment provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in Embodiment 1.

[0106] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to the method section.

[0107] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for soil moisture inversion integrating GNSS-R and radiometer, characterized in that, The method includes: Acquire the direct signal and target polarization reflection signal obtained by the GNSS receiver within the target soil area; the target polarization reflection signal is at least one of the following: left-hand circular polarization reflection signal, right-hand circular polarization reflection signal, vertical polarization reflection signal, and horizontal polarization reflection signal; The target polarization reflectivity is calculated based on the direct signal and the target polarization reflection signal. Obtain brightness temperature data, soil roughness, and comprehensive vegetation attenuation coefficient from a microwave radiometer within the target soil area; The brightness temperature data, soil roughness, comprehensive vegetation attenuation coefficient, and target polarization reflectance are input into the LSTM model to retrieve the soil moisture in the target soil area. Before inputting the brightness temperature data, soil roughness, comprehensive vegetation attenuation coefficient, and target polarization reflectance into the LSTM model, the following steps are also included: Construct a hyperparameter set for the LSTM model and set the value range of each hyperparameter in the hyperparameter set; the hyperparameters include the number of nodes, maximum number of iterations, and learning rate of each LSTM layer of the LSTM model; The hyperparameter set and training data are input into the Bayesian optimizer, which outputs the optimized hyperparameter set. The hyperparameters in the optimized hyperparameter set are used as the hyperparameters of the LSTM model to obtain the assigned LSTM model. The validation data is input into the assigned LSTM model, and the root mean square error of the model is calculated; the training data and the validation data are all historical brightness temperature data, historical soil roughness and comprehensive vegetation attenuation coefficient, and historical target polarization reflectance. When the root mean square error is greater than a preset value and the number of parameter iterations is less than the preset number of iterations, return to the step "input the hyperparameter set and historical training data into the Bayesian optimizer and output the optimized hyperparameter set"; When the root mean square error is less than a preset value or the number of parameter iterations reaches a preset number of iterations, the hyperparameter set corresponding to the currently assigned LSTM model is the optimal hyperparameter set.

2. The method according to claim 1, characterized in that, The expression for the target polarization reflectivity includes: In the formula, This represents the actual reflectivity, where p represents the polarization of the satellite transmitting antenna, which for GNSS-R is right-hand circular polarization, and q represents the polarization of the receiving antenna for the reflected signal. The four polarizations are: left-hand circular polarization, right-hand circular polarization, vertical polarization, and horizontal polarization, and the corresponding reflectivity is: , , , ; This represents the power of the reflected signal, including reflected signals with four polarization modes; Indicates the power of the direct signal; For direct antenna gain, For the gain of the reflective antenna, The angle of incidence of the antenna. It is the azimuth angle.

3. The method according to claim 1, characterized in that, The formulas for calculating soil roughness and comprehensive vegetation attenuation coefficient are as follows: in, TR This represents soil roughness and the overall vegetation attenuation coefficient. Indicates optical thickness; Indicates soil roughness.

4. A soil moisture inversion system based on the method described in any one of claims 1 to 3, comprising GNSS-R and radiometer fusion, characterized in that, The system includes: The first data acquisition module is used to acquire the direct signal and the target polarization reflection signal acquired by the GNSS receiver in the target soil area; the target polarization reflection signal is at least one of the following: left-hand circular polarization reflection signal, right-hand circular polarization reflection signal, vertical polarization reflection signal, and horizontal polarization reflection signal; The reflectivity calculation module is used to calculate the target polarization reflectivity based on the direct signal and the target polarization reflection signal. The second data acquisition module is used to acquire brightness temperature data, soil roughness, and comprehensive vegetation attenuation coefficient obtained by microwave radiometers in the target soil area. The inversion module is used to input the brightness temperature data, the soil roughness and comprehensive vegetation attenuation coefficient, and the target polarization reflectance into the target LSTM model to invert the soil moisture in the target soil area; Before inputting the brightness temperature data, soil roughness, comprehensive vegetation attenuation coefficient, and target polarization reflectance into the LSTM model, the following steps are also included: Construct a hyperparameter set for the LSTM model and set the value range of each hyperparameter in the hyperparameter set; the hyperparameters include the number of nodes, maximum number of iterations, and learning rate of each LSTM layer of the LSTM model; The hyperparameter set and training data are input into the Bayesian optimizer, which outputs the optimized hyperparameter set. The hyperparameters in the optimized hyperparameter set are used as the hyperparameters of the LSTM model to obtain the assigned LSTM model. The validation data is input into the assigned LSTM model, and the root mean square error of the model is calculated; the training data and the validation data are all historical brightness temperature data, historical soil roughness and comprehensive vegetation attenuation coefficient, and historical target polarization reflectance. When the root mean square error is greater than a preset value and the number of parameter iterations is less than the preset number of iterations, return to the step "input the hyperparameter set and historical training data into the Bayesian optimizer and output the optimized hyperparameter set"; When the root mean square error is less than a preset value or the number of parameter iterations reaches a preset number of iterations, the hyperparameter set corresponding to the currently assigned LSTM model is the optimal hyperparameter set.

5. The system according to claim 4, characterized in that, The expressions for the target polarization reflectivity in the first data acquisition module include: In the formula, This represents the actual reflectivity, where p represents the polarization of the satellite transmitting antenna, which for GNSS-R is right-hand circular polarization, and q represents the polarization of the receiving antenna for the reflected signal. The four polarizations are: left-hand circular polarization, right-hand circular polarization, vertical polarization, and horizontal polarization, and the corresponding reflectivity is: , , , ; This represents the power of the reflected signal, including reflected signals with four polarization modes; Indicates the power of the direct signal; For direct antenna gain, For the gain of the reflective antenna, The angle of incidence of the antenna. It is the azimuth angle.

6. A soil moisture inversion device integrating GNSS-R and radiometer, characterized in that, include: processor; as well as Memory, used to store computer-readable program instructions; The soil moisture inversion method of GNSS-R and radiometer fusion as described in any one of claims 1-3 is executed when the computer-readable program instructions are run by the processor.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by the processor, the program implements the soil moisture inversion method of GNSS-R and radiometer fusion as described in any one of claims 1-3.

Images