Ionized layer TEC determination method based on satellite-ground active measurement and SAR image joint inversion

Abstract

The invention relates to an ionized layer TEC determination method and system based on satellite-ground active measurement and SAR image joint inversion. The method comprises the steps that S1, a mapping relation database of ionized layer scattering characteristics measured by a dual-frequency SAR satellite, ionized layer penetration characteristics measured by satellite-ground active measurement and ionized layer TEC measured by a GPS receiving station is established; s2, establishing and training a TEC inversion deep learning model by using a multi-source amplitude-phase parameter neural network according to the mapping relation database, and outputting a corresponding TEC value by the TEC inversion deep learning model by taking an ionized layer amplitude-phase parameter as an input; the ionosphere amplitude-phase parameters are ionosphere scattering signal amplitude-phase parameters and / or ionosphere transmission signal amplitude-phase parameters; and S3, performing on-orbit intelligent rapid inversion on the ionized layer TEC over the globe according to the TEC inversion deep learning model. According to the method, the TEC inversion model based on SAR satellite measurement data is established, the method has the advantages of high precision and high timeliness, and high-precision inversion of any space-time TEC can be realized.

Description

Technical Field

[0001] The invention relates to the field of remote sensing technology, and in particular to an ionospheric TEC determination method based on satellite-ground active measurement and SAR image joint inversion. Background Art

[0002] In satellite remote sensing, the satellite altitude is usually above 500km, while the orbit altitude of geosynchronous orbit SAR reaches 36,000km. Satellite SAR imaging usually uses active emission of electromagnetic waves and receiving echo signals for processing and imaging. Usually the signals pass through the ionosphere on the earth's surface. The background ionosphere and ionospheric irregularities will have a serious impact on the quality of satellite-borne SAR imaging.

[0003] The ionosphere affects SAR imaging in three main ways: first, the "scintillation" effect. Ionospheric scintillation causes a large change in the amplitude of electromagnetic wave transmission, affecting the accuracy and consistency of image radiation; second, the "dispersion" effect. The ionosphere causes the path of electromagnetic wave propagation to deflect and the transmission delay to be inconsistent, which seriously causes the frequency and amplitude of remote sensing radar signals to change, affecting the spatial resolution, geometric distortion, and even image performance such as peak sidelobe ratio and integrated sidelobe ratio of SAR; third, "Faraday rotation". For linear polarization remote sensing radars, the Faraday rotation effect causes changes in electromagnetic polarization, causing polarization mismatch between transmitter and receiver, affecting the amplitude of the received signal, and ultimately affecting the quality of SAR images. The ionosphere also affects non-imaging remote sensing systems such as microwave radiometers, microwave scatterometers, and radar altimeters in terms of measurement accuracy and consistency.

[0004] Therefore, the ionosphere must be accurately measured and precisely compensated. Traditional ionospheric measurement methods usually use two methods: one is to directly detect the ionosphere through independent equipment, invert the ionosphere information, and then use the obtained ionosphere information to correct and compensate the SAR image. The second is to use the SAR echo characteristics to invert the ionosphere information. However, for high-resolution, long integration time or heavy-orbit interferometric imaging, the above methods have problems such as incomplete ionospheric path measurement, incomplete information, insufficient timeliness, and even large deviations in space and time of remote sensing signals. It is difficult to meet the requirements of high-precision and high-timeliness of ionosphere measurement, and new measurement and inversion methods must be adopted. Summary of the invention

[0005] In order to solve the technical problems existing in the above-mentioned prior art, the purpose of the present invention is to provide an ionospheric TEC determination method and system based on satellite-ground active measurement and SAR image joint inversion, which has the advantages of high accuracy and high timeliness and can realize high-precision inversion of TEC in any space and time.

[0006] To achieve the above-mentioned object of the invention, the present invention provides a method for determining ionospheric TEC based on satellite-ground active measurement and SAR image joint inversion, comprising:

[0007] Step S1, establishing a mapping relationship database of ionospheric scattering characteristics measured by dual-frequency SAR satellites, ionospheric penetration characteristics measured by satellite-ground active measurement, and ionospheric TEC measured by GPS receiving stations;

[0008] Step S2, according to the mapping relationship database, using a multi-source amplitude-phase parameter neural network, establishing and training a TEC inversion deep learning model, wherein the TEC inversion deep learning model takes ionospheric amplitude-phase parameters as input and outputs corresponding TEC values; the ionospheric amplitude-phase parameters are ionospheric scattering signal amplitude-phase parameters and / or ionospheric transmission signal amplitude-phase parameters;

[0009] Step S3: Perform on-orbit intelligent rapid inversion of the ionospheric TEC over the globe according to the TEC inversion deep learning model.

[0010] According to a technical solution of the present invention, in step S1, the mapping relationship database includes Q sample data, each sample data includes the amplitude and phase parameters of the ionospheric scattering signal measured by the dual-frequency SAR satellite for the simultaneous and spatial ionosphere, the amplitude and phase parameters of the ionospheric transmission signal measured by the satellite-ground active measurement, and the ionospheric TEC value measured by the GPS receiving station;

[0011] The sample data of the mapping relationship database is represented as in, are the amplitude and phase parameters of the ionospheric scattering signal measured by the dual-frequency SAR satellite, is the amplitude and phase parameters of the ionospheric transmission signal measured by satellite-to-ground active measurement, Γ q is the ionospheric TEC value measured by the GPS receiving station; A and B represent the signal amplitudes of the ionospheric scattering signal and the ionospheric transmission signal, respectively; They respectively represent the signal phase; m represents the signal frequency, m=1 or 2, m=1 represents P band, and m=2 represents L band; n represents the polarization mode of the signal, n=1, 2, 3 or 4, n=1 is VV polarization, n=2 is VH polarization, n=3 is HV polarization, and n=4 is HH polarization; q represents the sample number, q=1,2,3,…,Q; d represents the signal propagation direction of the ionospheric transmission signal, d=1 represents satellite transmission and ground reception, and d=2 represents ground transmission and satellite reception.

[0012] According to a technical solution of the present invention, in step S1, it specifically includes:

[0013] Step S11, using the ground transceiver station and the dual-frequency SAR satellite to perform joint measurement, obtain multiple ionospheric transmission signal amplitude and phase parameter samples, and establish an ionospheric transmission signal sub-database for satellite-ground active measurement; the sample data in the ionospheric transmission signal sub-database for satellite-ground active measurement is expressed as

[0014] Step S12, using dual-frequency SAR satellite measurement, obtain multiple ionospheric scattering signal amplitude and phase parameter samples that are simultaneously and spatially identical to the ionospheric transmission signal measured by the satellite-ground active measurement in step S11, and establish a dual-frequency SAR ionospheric scattering signal sub-database; the sample data in the dual-frequency SAR ionospheric scattering signal sub-database is expressed as

[0015] Step S13, using the GPS receiving station to obtain multiple ionospheric TEC values ​​at the same time and space as the ionospheric transmission signal measured by the satellite-ground active measurement in step S11, and establish a TEC value sub-database; the samples in the TEC value sub-database are recorded as Γ q , the measurement accuracy of samples in the TEC value sub-database reaches 0.5TECU;

[0016] Step S14: establishing the mapping relationship database according to the ionospheric transmission signal sub-database of the satellite-ground active measurement, the dual-frequency SAR ionospheric scattering signal sub-database and the TEC value sub-database.

[0017] According to a technical solution of the present invention, the duration of satellite-to-ground active measurement performed by a ground transceiver station and a dual-frequency SAR satellite is τ1, τ1≤20ms;

[0018] The duration of the measurement time slot of the satellite-to-ground active measurement is τ2, τ2≤200ms;

[0019] The duration of the ionospheric scattering signal measured by the dual-frequency SAR satellite is τ3, τ3≤150ms.

[0020] According to a technical solution of the present invention, in the step S2, it specifically includes:

[0021] Step S21, constructing and utilizing the multi-source amplitude and phase parameter neural network to establish a TEC inversion deep learning model, wherein the TEC inversion deep learning model includes a feature extraction network, a feature fusion network and a TEC regression network;

[0022] Step S22, according to the mapping relationship database, obtain the amplitude and phase parameters s1 and s3 of the ionospheric scattering signal of satellite-to-ground and ground-to-satellite in the sample data, and input them into the feature extraction network to obtain the extracted feature graphs s'1=Net(s1), s'3=Net(s3);

[0023] Step S23, according to the mapping relationship database, the amplitude and phase parameters of the ionospheric scattering signal spontaneously transmitted and received by the dual-frequency SAR satellite in the sample data are obtained and input into the feature extraction network to obtain the extracted feature graph s'2=Net(s2);

[0024] Step S24, simultaneously inputting the feature graphs s'1, s'2 and s'3 extracted by the feature extraction network into the feature fusion network for feature splicing and deep feature fusion;

[0025] Step S25, performing TEC estimation on the fused deep features through the TEC regression network to obtain a TEC estimation value;

[0026] Step S26, based on the obtained TEC estimation value and the ionospheric TEC value measured by the GPS receiving station in the sample data, the inversion result is estimated and back-propagated to update the network parameters;

[0027] Step S27, repeating the above steps S22 to S26, training for several rounds, saving and outputting the TEC inversion deep learning model;

[0028] In the training phase, the learning rate is set to 0.01, the optimizer is SGD, and the training is repeated for 100 rounds. Estimate the TEC inversion effect for the loss function, and back-propagate and update the network parameters of the TEC prediction deep learning module;

[0029] Among them, Γ k is the TEC value measured by the GPS receiving station and ground equipment, Γ' k is the TEC estimate of the intelligent inversion model, k is a variable representing the batch input for a single forward training, and K represents the size of the batch value.

[0030] According to a technical solution of the present invention, the feature extraction network includes a convolutional layer conv1, a first batch of normalization layers, a first activation function, a convolutional layer conv2, a second batch of normalization layers, and a second activation function connected in series in sequence;

[0031] The convolution kernel size of the convolution layer conv1 is (1,1), the step size is 1, the padding is 0, the number of input channels is 8, and the number of output channels is set to 64;

[0032] The first activation function and the second activation function are linear rectification functions relu=max(0,x1), wherein max() represents the maximum value between 0 and x1, and x1 is an intermediate value generated in the forward process of the neural network;

[0033] The convolution kernel size of the convolution layer conv2 is (1,1), the step size is 1, the padding is 0, the number of input channels is 64, and the number of output channels is set to 128;

[0034] The calculation method of the first batch of normalized layers and the second batch of normalized layers is x1'=γ*x norm +βwhere x norm is the feature after regularization, γ and β are two learnable feature linear transformation parameters, Among them, ∈ is an adjustment factor to prevent the denominator from being zero, and its value is set to 0.0001; is the mean, k is the batch size input for a single forward training, is the variance, x k Represents the intermediate value generated during the forward process of the kth batch of neural networks in a single forward training;

[0035] In the step S24, the feature splicing method is channel connection, expressed as s"=concat(s'1,s'2,s'3), where concat is a channel connection function; the deep feature fusion is performed through a deep fusion convolution layer, expressed as s'"=conv(s"), where conv is a deep fusion convolution layer, the convolution kernel size is (1,1), the step size is 1, the padding is 0, the number of input channels is 128, and the number of output channels is set to 64;

[0036] The number of input channels of the deep fusion convolution layer is consistent with the number of channels of the feature map s" obtained after the feature splicing, where s'1 and s'3 are the feature maps of the amplitude and phase parameters of the ionospheric scattering signals of satellite-to-ground and ground-to-satellite transmission and reception, respectively. Before feature fusion, there is a z% probability that they are replaced by Gaussian noise to adapt to the input of only spontaneous and self-received signal features in the on-orbit prediction stage, where Gaussian noise is a normal distribution with a mean of 0 and a variance of 1.

[0037] According to a technical solution of the present invention, in step S3, in the on-orbit intelligent rapid inversion stage, the input of the feature fusion network is the feature map extracted after the ionospheric scattering signal measured by the dual-frequency SAR satellite passes through the feature extraction network and the randomly generated Gaussian signal, and the feature map extracted after the feature extraction network and the randomly generated Gaussian signal are feature fused and then deep feature fusion is performed through the deep fusion convolution layer, and the number of input channels of the deep fusion convolution layer is consistent with the number of channels of the feature map extracted after passing through the feature extraction network.

[0038] According to a technical solution of the present invention, in step S3, it specifically includes:

[0039] Step S31, obtaining and processing the dual-frequency SAR satellite scattered echo signal to obtain the ionospheric scattering amplitude and phase parameters of the dual-frequency SAR satellite scattered echo signal;

[0040] Step S32, performing standardized preprocessing on the ionospheric scattering amplitude and phase parameters of the dual-frequency SAR satellite scattered echo signal extracted and inputting them into the TEC inversion deep learning model;

[0041] Step S33: Perform on-orbit intelligent rapid inversion of the ionospheric scattering amplitude and phase parameters of the dual-frequency SAR satellite scattered echo signal through the TEC inversion deep learning model to obtain the ionospheric TEC value Γ' above the dual-frequency SAR satellite observation area. q .

[0042] According to one aspect of the present invention, a system for implementing the above-mentioned ionospheric TEC determination method based on satellite-ground active measurement and SAR image joint inversion is provided, comprising:

[0043] FPGA, used for receiving and acquiring scattered echo signals of dual-frequency SAR satellites, and preprocessing the scattered echo signals to obtain ionospheric scattering amplitude and phase parameters thereof;

[0044] The GPU platform is used to load and configure the TEC inversion deep learning model, and to invert the TEC value of the ionosphere above the measurement position of the dual-frequency SAR satellite through the TEC inversion deep learning model.

[0045] The present invention proposes a method and system for determining ionospheric TEC based on satellite-ground active measurement and SAR image joint inversion, which has the following advantages:

[0046] The present invention synchronously uses dual-frequency SAR satellites and ground transceiver stations to measure the electromagnetic wave scattering amplitude and phase parameters and transmission amplitude and phase parameters of the ionosphere, thereby improving the ability to describe the propagation characteristics of electromagnetic waves in a complex ionosphere. The electromagnetic wave scattering amplitude and phase parameters and transmission amplitude and phase parameters of the ionosphere are used to train and optimize a large number of measured data samples using a deep learning network model to obtain a high-precision TEC inversion model. Combined with multiple scattered signal amplitude and phase parameters measured by the dual-frequency SAR satellite, TEC inversion in any area is achieved, which significantly improves the inversion accuracy of TEC over areas such as the far sea and remote areas where stations cannot be deployed. The model has the advantages of high accuracy and high timeliness, and can achieve high-precision inversion of TEC in any space and time. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention, and for ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0048] Figure 1 A flowchart schematically shows a method for determining ionospheric TEC based on satellite-ground active measurement and SAR image joint inversion according to one embodiment of the present invention;

[0049] Figure 2 A specific flow chart of a method for determining ionospheric TEC based on satellite-ground active measurement and SAR image joint inversion provided in accordance with one embodiment of the present invention is schematically shown;

[0050] Figure 3 A schematic diagram schematically shows a time arrangement diagram of active measurement and SAR imaging according to an embodiment of the present invention;

[0051] Figure 4 A flowchart schematically showing a TEC inversion deep learning model according to an embodiment of the present invention;

[0052] Figure 5 Schematically showing a workflow diagram of a feature extraction network according to an embodiment of the present invention;

[0053] Figure 6 Schematically showing a workflow diagram of a feature fusion network according to an embodiment of the present invention;

[0054] Figure 7 The flowchart of on-orbit intelligent inversion of ionospheric TEC based on a hybrid architecture of GPU and FPGA according to another embodiment of the present invention is schematically shown. DETAILED DESCRIPTION

[0055] The description of the embodiments of this specification should be combined with the corresponding drawings, which should be considered as part of the complete specification. In the drawings, the shape or thickness of the embodiments may be enlarged and indicated for simplification or convenience. Furthermore, the parts of each structure in the drawings will be described separately. It is worth noting that the elements not shown in the drawings or not described in words are in a form known to ordinary technicians in the relevant technical field.

[0056] The description of the embodiments herein and any reference to directions and orientations are only for the convenience of description and are not to be construed as any limitation on the scope of protection of the present invention. The following description of the preferred embodiments may involve combinations of features, which may exist independently or in combination, and the present invention is not particularly limited to the preferred embodiments. The scope of the present invention is defined by the claims.

[0057] like Figure 1 As shown, the present invention provides an ionospheric TEC determination method based on satellite-ground active measurement and SAR image joint inversion, which includes three steps: database building 10, modeling 20, and inversion 30.

[0058] In the process of building database 10, the amplitude and phase parameters of ionospheric scattering signals are measured by dual-frequency SAR satellites, and the amplitude and phase parameters of ionospheric transmission signals measured by satellite-ground active measurement are obtained by joint measurement of ground transceiver stations and dual-frequency SAR satellites. The ionospheric TEC value is measured by GPS receiving stations, and a database of mapping relationships between amplitude, phase and TEC of complex propagation of electromagnetic waves in the ionosphere is established. Building database 10 includes obtaining scattering signal parameters by dual-frequency SAR satellites, obtaining transmission signal parameters by dual-frequency SAR satellites and ground transceiver stations, and obtaining ionospheric TEC values ​​by GPS receiving stations.

[0059] In the process of modeling 20, considering the complex data such as ionospheric scattering signals and transmission signals obtained by measurement, it is impossible to obtain accurate ionospheric TEC values ​​through analytical methods. A multi-source amplitude-phase parameter neural network is used, with a total of 24 amplitude-phase parameters of satellite-to-ground reception, ground-to-satellite reception, and satellite spontaneous transmission and self-reception as input, to approximate the mapping relationship between ionospheric scattering signals, transmission signals and ionospheric TEC, and provide a TEC inversion deep learning model for high-precision measurement of ionospheric TEC above the world. Modeling 20 includes multi-source feature extraction, multi-source feature fusion, and model training and storage.

[0060] In the inversion 30 process, for the TEC inversion deep learning model, the FPGA and GPU hybrid hardware platform is used to load and configure the model, and the amplitude and phase of the scattered signal of the ionosphere in any space in the world are obtained by the dual-frequency SAR satellite as the input of the model, and the TEC value of the SAR satellite observation time and space is inverted. Inversion 30 includes the input of the amplitude and phase parameters of the scattered signal of the dual-frequency SAR satellite, the configuration of the deep learning model on the hybrid platform, and the inversion of the TEC output of the ionosphere above the dual-frequency SAR satellite measurement position.

[0061] Compared with the existing ionospheric TEC measurement method, this method comprehensively considers the complex relationship between the amplitude and phase characteristics of the ionospheric scattering signal, the amplitude and phase characteristics of the penetration signal and TEC, and optimizes the model using a large amount of measured data. It has the advantages of high TEC measurement accuracy, good robustness and large measurement range.

[0062] like Figure 2 As shown, the specific process of the ionospheric TEC determination method based on satellite-ground active measurement and SAR image joint inversion of the present invention includes:

[0063] Step S1, establishing a mapping relationship database of ionospheric scattering characteristics measured by dual-frequency SAR satellites, ionospheric penetration characteristics measured by satellite-ground active measurement, and ionospheric TEC measured by GPS receiving stations;

[0064] In the step S1, the mapping relationship database includes Q sample data, generally Q≥20000, each sample data includes 8 parameters of ionospheric scattering signal amplitude and phase parameters measured by dual-frequency SAR satellites for the simultaneous and spatial ionosphere, 24 parameters of ionospheric transmission signal amplitude and phase parameters measured by satellite-ground active measurement, and ionospheric TEC value measured by GPS receiving station;

[0065] The sample data of the mapping relationship database is represented as in, are the amplitude and phase parameters of the ionospheric scattering signal measured by the dual-frequency SAR satellite, is the amplitude and phase parameters of the ionospheric transmission signal measured by satellite-to-ground active measurement, Γ q is the ionospheric TEC value measured by the GPS receiving station; A and B represent the signal amplitudes of the ionospheric scattering signal and the ionospheric transmission signal, respectively; They respectively represent the signal phase; m represents the signal frequency, m=1 or 2, m=1 represents P band, and m=2 represents L band; n represents the polarization mode of the signal, n=1, 2, 3 or 4, n=1 is VV polarization, n=2 is VH polarization, n=3 is HV polarization, and n=4 is HH polarization; q represents the sample number, q=1,2,3,…,Q; d represents the signal propagation direction of the ionospheric transmission signal, d=1 represents satellite transmission and ground reception, and d=2 represents ground transmission and satellite reception.

[0066] In the process of constructing each sample, the TEC is kept approximately unchanged, and the dual-frequency SAR satellite and the satellite-ground joint detection repeatedly measure the same area of ​​the ionosphere H times, and the H measurement results are averaged and recorded as a set of sample data.

[0067] In the step S1, it specifically includes:

[0068] Step S11, using the ground transceiver station and the dual-frequency SAR satellite to perform joint measurement, obtain multiple ionospheric transmission signal amplitude and phase parameter samples, and establish an ionospheric transmission signal sub-database for satellite-ground active measurement; the sample data in the ionospheric transmission signal sub-database for satellite-ground active measurement is expressed as

[0069] In the process of constructing the qth sample, TEC remains approximately unchanged. The dual-frequency SAR satellite and the ground station repeatedly measure the same area of ​​the ionosphere H times, and average the H measurement results as a set of sample data to improve the signal-to-noise ratio, which is recorded as in

[0070] During this process, the orbital altitude of the dual-frequency SAR satellite is about 36,000 km. The dual frequencies include P-band and L-band. The center frequency of the P-band is 435 MHz, the center frequency of the L-band is 1.26 GHz, and the polarizations include VV, VH, HV, and HH.

[0071] Step S12, using dual-frequency SAR satellite measurement, obtain multiple ionospheric scattering signal amplitude and phase parameter samples that are simultaneously and spatially identical to the ionospheric transmission signal measured by the satellite-ground active measurement in step S11, and establish a dual-frequency SAR ionospheric scattering signal sub-database; the sample data in the dual-frequency SAR ionospheric scattering signal sub-database is expressed as

[0072] In the process of constructing the qth sample, TEC remains approximately unchanged. The dual-frequency SAR satellite repeatedly measures the same area of ​​the ionosphere H times, and averages the H measurement results as a set of sample data to improve the signal-to-noise ratio, which is recorded as in

[0073] During this process, the orbital altitude of the dual-frequency SAR satellite is about 36,000 km. The dual frequencies include P-band and L-band. The center frequency of the P-band is 435 MHz, the center frequency of the L-band is 1.26 GHz, and the polarizations include VV, VH, HV, and HH.

[0074] like Figure 3 As shown, Figure 3 The time arrangement of satellite-to-ground active measurement 1021 and SAR imaging measurement 1011. In step S21 and step S22, the duration of satellite-to-ground active measurement 1021 performed by the ground transceiver station and the dual-frequency SAR satellite is τ1, τ1≤20ms; the duration of the measurement time slot 1022 of the satellite-to-ground active measurement is τ2, τ2≤200ms; the duration of the SAR imaging measurement 1011 performed by the dual-frequency SAR satellite to measure the ionospheric scattered signal is τ3, τ3≤150ms.

[0075] Step S13, using the GPS receiving station to obtain multiple ionospheric TEC values ​​at the same time and space as the ionospheric transmission signal measured by the satellite-ground active measurement in step S11, and establish a TEC value sub-database; the samples in the TEC value sub-database are recorded as Γ q The measurement accuracy of samples in the TEC value sub-database reaches 0.5TECU.

[0076] Step S14: establishing the mapping relationship database according to the ionospheric transmission signal sub-database of the satellite-ground active measurement, the dual-frequency SAR ionospheric scattering signal sub-database and the TEC value sub-database.

[0077] Step S2: According to the mapping relationship database, a multi-source amplitude-phase parameter neural network is used to establish and train a TEC inversion deep learning model, wherein the TEC inversion deep learning model takes ionospheric amplitude-phase parameters as input and outputs corresponding TEC values; the ionospheric amplitude-phase parameters are ionospheric scattering signal amplitude-phase parameters and / or ionospheric transmission signal amplitude-phase parameters.

[0078] like Figure 4 As shown in FIG. 1 , the modeling process is mainly based on the establishment of a TEC inversion deep learning model based on a mapping relationship database. The amplitude and phase parameters of satellite-to-ground, ground-to-satellite, and self-to-self signals can be passed through the feature extraction network 201 to obtain the extracted feature graph s', and then connected and fused through the feature fusion network 202 to obtain the fusion feature s", and finally the TEC estimation is completed through the TEC regression network 203 to obtain the TEC estimation result t', and back propagation is performed to update the network parameters to realize the training of the TEC inversion deep learning model.

[0079] In the step S2, it specifically includes:

[0080] Step S21, constructing and utilizing the multi-source amplitude and phase parameter neural network to establish a TEC inversion deep learning model, wherein the TEC inversion deep learning model includes a feature extraction network 201, a feature fusion network 202 and a TEC regression network 203;

[0081] Step S22, according to the mapping relationship database, the amplitude and phase parameters s1 and s3 of the ionospheric scattering signals of satellite-to-ground and ground-to-satellite in the sample data are obtained, and input into the feature extraction network 201 to obtain the extracted feature graphs s'1=Net(s1), s'3=Net(s3);

[0082] Step S23, according to the mapping relationship database, the amplitude and phase parameters s2 of the ionospheric scattering signal spontaneously transmitted and received by the dual-frequency SAR satellite in the sample data are obtained and input into the feature extraction network 201 to obtain the extracted feature graph s'2=Net(s2);

[0083] In step S22 and step S23, the feature extraction network 201 includes a convolutional layer conv12011, a first batch of normalization layers 2012, a first activation function 2013, a convolutional layer conv22014, a second batch of normalization layers 2015, and a second activation function 2016 which are sequentially connected in series;

[0084] The convolution kernel size of the convolution layer conv1 is (1,1), the step size is 1, the padding is 0, the number of input channels is 8, and the number of output channels is set to 64;

[0085] The first activation function and the second activation function are linear rectification functions relu=max(0,x1), wherein max() represents the maximum value between 0 and x1, and x1 is an intermediate value generated in the forward process of the neural network;

[0086] The convolution kernel size of the convolution layer conv2 is (1,1), the step size is 1, the padding is 0, the number of input channels is 64, and the number of output channels is set to 128;

[0087] The calculation method of the first batch of normalized layers 2012 and the second batch of normalized layers 2015 is x1'=γ*x norm +βwhere x norm is the feature after regularization, γ and β are two learnable feature linear transformation parameters, Among them, ∈ is an adjustment factor to prevent the denominator from being zero, and its value is set to 0.0001; is the mean, k is a variable, representing the batch input for a single forward training. is the variance, K represents the size of the batch value input for a single forward training, x k Represents the intermediate value generated during the forward process of the kth batch of a single forward training neural network.

[0088] Step S24, simultaneously input the feature graphs s'1, s'2, s'3 extracted by the feature extraction network 201 into the feature fusion network 202 for feature concatenation and deep feature fusion to output fused features.

[0089] In step S24, the feature concatenation method is channel connection, which is expressed as s"=concat(s'1,s'2,s'3), where concat is the channel connection function; deep feature fusion is performed through the deep fusion convolution layer, which is expressed as s'"=conv(s"), where conv is a deep fusion convolution layer, the convolution kernel size is (1,1), the step size is 1, the padding is 0, the number of input channels is 128, and the number of output channels is set to 64;

[0090] The number of input channels of the deep fusion convolution layer 2024 is consistent with the number of channels of the feature map s" obtained after the feature splicing, where s'1 is the feature extraction result 2021 of satellite transmission and ground reception, and s'3 is the feature extraction result 2022 of ground transmission and satellite reception, which has a z% probability of being replaced by Gaussian noise before feature fusion to adapt to the input of only spontaneous transmission and self-reception signal features in the on-orbit prediction stage, where the noise is a normal distribution with a mean of 0 and a variance of 1.

[0091] During the on-orbit prediction phase, the input of the feature fusion network is the feature map s'2 extracted from the spontaneously transmitted and received ionospheric signal after passing through the feature extraction network, and the randomly generated Gaussian signals g'1, g'3, and feature fusion is performed, s" = concat(g'1, s'2, g'3), and then deep feature fusion is performed through the deep fusion convolution layer 2024, s'" = conv(s"), to generate the feature map 2025s'" after feature fusion. The number of input channels of the convolution layer is consistent with the number of channels of s", and the network parameters remain unchanged.

[0092] Step S25, performing TEC estimation on the fused deep features through the TEC regression network 203 to obtain a TEC estimation value;

[0093] The TEC regression network 203 consists of two fully connected layers.

[0094] Step S26, based on the obtained TEC estimation value and the ionospheric TEC value measured by the GPS receiving station in the sample data, the inversion result is estimated and back-propagated to update the network parameters;

[0095] Step S27, repeating the above steps S22 to S26, training for several rounds, saving and outputting the TEC inversion deep learning model;

[0096] In the training phase, the learning rate is set to 0.01, the optimizer is SGD, and the training is repeated for 100 rounds. estimating the TEC inversion effect for the loss function, and back-propagating and updating the network parameters of the TEC inversion deep learning model;

[0097] Among them, Γ k is the TEC value measured by the GPS receiving station and ground equipment, Γ' k is the tec estimate of the intelligent inversion model, and k is the batch size input for one forward training.

[0098] Step S3: Perform on-orbit intelligent rapid inversion of the ionospheric TEC over the globe according to the TEC inversion deep learning model.

[0099] The present invention also provides a system for implementing an ionospheric TEC determination method based on satellite-ground active measurement and SAR image joint inversion, comprising:

[0100] FPGA, used for receiving and acquiring scattered echo signals of dual-frequency SAR satellites, and preprocessing the scattered echo signals to obtain ionospheric scattering amplitude and phase parameters thereof;

[0101] The GPU platform is used to load and configure the TEC inversion deep learning model, and to invert the TEC value of the ionosphere above the measurement position of the dual-frequency SAR satellite through the TEC inversion deep learning model.

[0102] In the step S3, it specifically includes:

[0103] Step S31, using FPGA to obtain and process the scattered echo signal measured in real time by the dual-frequency SAR satellite on orbit, and obtain the ionospheric scattering amplitude and phase parameters of the scattered echo signal of the dual-frequency SAR satellite, which are recorded as in C represents the signal amplitude of the dual-frequency SAR satellite scattered echo signal, represents the signal phase, m represents the frequency, m=1,2, m=1 represents the P band, m=2 represents the L band; n represents the polarization mode, n=1,2,3,4, n=1 represents VV polarization, n=2 represents VH polarization, n=3 represents HV polarization, and n=4 represents HH polarization. H represents the number of repeated measurements when taking a set of samples, q represents the sample number, q=1,2,3,…,Q.

[0104] Step S32, performing standardized preprocessing on the ionospheric scattering amplitude and phase parameters of the dual-frequency SAR satellite scattered echo signal extracted and inputting them into the TEC inversion deep learning model;

[0105] According to the input requirements of the intelligent inversion model, the amplitude and phase parameters of the scattered echoes in two frequency bands and four polarizations are standardized and preprocessed. The standardized eight amplitude and phase parameters of the scattered echoes are converted into Transfer to GPU at a rate greater than or equal to 10MB / s.

[0106] Step S33: Perform on-orbit intelligent rapid inversion of the ionospheric scattering amplitude and phase parameters of the dual-frequency SAR satellite scattered echo signal through the TEC inversion deep learning model to obtain the ionospheric TEC value Γ' above the dual-frequency SAR satellite observation area. q .

[0107] The ground tracking and control station uploads the trained and optimized ionospheric TEC intelligent inversion model to the on-orbit processing module on the satellite through the satellite-ground tracking and control link. The upload rate is greater than or equal to 128kb / s. Under the control of FPGA, the ionospheric TEC intelligent inversion model is configured in the GPU.

[0108] Using the GPU-configured ionospheric TEC intelligent inversion model, the eight received scattered echo amplitude and phase parameters The ionospheric TEC is inverted in real time on-orbit intelligently to obtain the ionospheric TEC value Γ' above the dual-frequency SAR satellite observation area. q .

[0109] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or terminal device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or terminal device. In the absence of further restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, article or terminal device including the elements.

[0110] Finally, it should be noted that the above is a preferred embodiment of the present invention. It should be pointed out that although the preferred embodiment of the present invention has been described, for those skilled in the art, once the basic creative concept of the present invention is known, several improvements and modifications can be made without departing from the principles of the present invention, and these improvements and modifications should also be regarded as the protection scope of the present invention. Therefore, the attached claims are intended to be interpreted as including the preferred embodiment and all changes and modifications that fall within the scope of the embodiments of the present invention.

Claims

1. A method for determining ionospheric TEC based on joint inversion of satellite-ground active measurement and SAR image, characterized in that: include: Step S1, establishing a mapping relationship database of ionospheric scattering characteristics measured by dual-frequency SAR satellites, ionospheric penetration characteristics measured by satellite-ground active measurement, and ionospheric TEC measured by GPS receiving stations; Step S2, according to the mapping relationship database, using a multi-source amplitude-phase parameter neural network, establishing and training a TEC inversion deep learning model, wherein the TEC inversion deep learning model takes ionospheric amplitude-phase parameters as input and outputs corresponding TEC values; the ionospheric amplitude-phase parameters are ionospheric scattering signal amplitude-phase parameters and / or ionospheric transmission signal amplitude-phase parameters; Step S3: Perform on-orbit intelligent rapid inversion of the ionospheric TEC over the globe according to the TEC inversion deep learning model.

2. The ionospheric TEC determination method based on satellite-ground active measurement and SAR image joint inversion according to claim 1 is characterized in that: In step S1, the mapping relationship database includes Q sample data, each sample data includes the amplitude and phase parameters of the ionospheric scattering signal measured by the dual-frequency SAR satellite for the simultaneous and spatial ionosphere, the amplitude and phase parameters of the ionospheric transmission signal measured by the satellite-ground active measurement, and the ionospheric TEC value measured by the GPS receiving station; The sample data of the mapping relationship database is represented as in, are the amplitude and phase parameters of the ionospheric scattering signal measured by the dual-frequency SAR satellite, is the amplitude and phase parameters of the ionospheric transmission signal measured by satellite-to-ground active measurement, Γ q is the ionospheric TEC value measured by the GPS receiving station; A and B represent the signal amplitudes of the ionospheric scattering signal and the ionospheric transmission signal, respectively; They respectively represent the signal phase; m represents the signal frequency, m=1 or 2, m=1 represents P band, and m=2 represents L band; n represents the polarization mode of the signal, n=1, 2, 3 or 4, n=1 is VV polarization, n=2 is VH polarization, n=3 is HV polarization, and n=4 is HH polarization; q represents the sample number, q=1,2,3,…,Q; d represents the signal propagation direction of the ionospheric transmission signal, d=1 represents satellite transmission and ground reception, and d=2 represents ground transmission and satellite reception.

3. The ionospheric TEC determination method based on satellite-ground active measurement and SAR image joint inversion according to claim 2 is characterized in that: In the step S1, it specifically includes: Step S11, using the ground transceiver station and the dual-frequency SAR satellite to perform joint measurement, obtain multiple ionospheric transmission signal amplitude and phase parameter samples, and establish an ionospheric transmission signal sub-database for satellite-ground active measurement; the sample data in the ionospheric transmission signal sub-database for satellite-ground active measurement is expressed as Step S12, using dual-frequency SAR satellite measurement, obtain multiple ionospheric scattering signal amplitude and phase parameter samples that are simultaneously and spatially identical to the ionospheric transmission signal measured by the satellite-ground active measurement in step S11, and establish a dual-frequency SAR ionospheric scattering signal sub-database; the sample data in the dual-frequency SAR ionospheric scattering signal sub-database is expressed as Step S13, using the GPS receiving station to obtain multiple ionospheric TEC values ​​at the same time and space as the ionospheric transmission signal measured by the satellite-ground active measurement in step S11, and establish a TEC value sub-database; the samples in the TEC value sub-database are recorded as Γ q , the measurement accuracy of samples in the TEC value sub-database reaches 0.5TECU; Step S14: establishing the mapping relationship database according to the ionospheric transmission signal sub-database of the satellite-ground active measurement, the dual-frequency SAR ionospheric scattering signal sub-database and the TEC value sub-database.

4. The method for determining ionospheric TEC based on satellite-ground active measurement and SAR image joint inversion according to claim 3, characterized in that: The duration of satellite-to-ground active measurement using a ground transceiver station and a dual-frequency SAR satellite is τ1, τ1≤20ms; The duration of the measurement time slot of the satellite-to-ground active measurement is τ2, τ2≤200ms; The duration of the ionospheric scattering signal measured by the dual-frequency SAR satellite is τ3, τ3≤150ms.

5. The method for determining ionospheric TEC based on satellite-ground active measurement and SAR image joint inversion according to claim 3, characterized in that: In the step S2, it specifically includes: Step S21, constructing and utilizing the multi-source amplitude and phase parameter neural network to establish a TEC inversion deep learning model, wherein the TEC inversion deep learning model includes a feature extraction network, a feature fusion network and a TEC regression network; Step S22, according to the mapping relationship database, obtain the amplitude and phase parameters s1 and s3 of the ionospheric scattering signal of satellite-to-ground and ground-to-satellite in the sample data, and input them into the feature extraction network to obtain the extracted feature graphs s'1=Net(s1), s'3=Net(s3); Step S23, according to the mapping relationship database, the amplitude and phase parameters of the ionospheric scattering signal spontaneously transmitted and received by the dual-frequency SAR satellite in the sample data are obtained and input into the feature extraction network to obtain the extracted feature graph s'2=Net(s2); Step S24, simultaneously inputting the feature graphs s'1, s'2 and s'3 extracted by the feature extraction network into the feature fusion network for feature splicing and deep feature fusion; Step S25, performing TEC estimation on the fused deep features through the TEC regression network to obtain a TEC estimation value; Step S26, based on the obtained TEC estimation value and the ionospheric TEC value measured by the GPS receiving station in the sample data, the inversion result is estimated and back-propagated to update the network parameters; Step S27, repeating the above steps S22 to S26, training for several rounds, saving and outputting the TEC inversion deep learning model; In the training phase, the learning rate is set to 0.01, the optimizer is SGD, and the training is repeated for 100 rounds. Estimate the TEC inversion effect for the loss function, and back-propagate and update the network parameters of the TEC prediction deep learning module; Among them, Γ k is the TEC value measured by the GPS receiving station and ground equipment, Γ' k is the TEC estimate of the intelligent inversion model, k is a variable representing the batch input for a single forward training, and K represents the batch value size input for a single forward training.

6. The method for determining ionospheric TEC based on satellite-ground active measurement and SAR image joint inversion according to claim 5, characterized in that: The feature extraction network includes a convolutional layer conv1, a first batch of normalization layers, a first activation function, a convolutional layer conv2, a second batch of normalization layers, and a second activation function connected in series in sequence; The convolution kernel size of the convolution layer conv1 is (1,1), the step size is 1, the padding is 0, the number of input channels is 8, and the number of output channels is set to 64; The first activation function and the second activation function are linear rectification functions relu=max(0,x1), wherein max() represents the maximum value between 0 and x1, and x1 is an intermediate value generated in the forward process of the neural network; The convolution kernel size of the convolution layer conv2 is (1,1), the step size is 1, the padding is 0, the number of input channels is 64, and the number of output channels is set to 128; The calculation method of the first batch of normalized layers and the second batch of normalized layers is x1'=γ*x norm +βwhere x norm is the feature after regularization, γ and β are two learnable feature linear transformation parameters, Among them, ∈ is an adjustment factor to prevent the denominator from being zero, and its value is set to 0.0001; is the mean, k is the batch size input for a single forward training, is the variance, x k Represents the intermediate value generated during the forward process of the kth batch of neural networks in a single forward training; In the step S24, the feature splicing method is channel connection, expressed as s"=concat(s'1,s'2,s'3), where concat is a channel connection function; the deep feature fusion is performed through a deep fusion convolution layer, expressed as s"'=conv(s"), where conv is a deep fusion convolution layer, the convolution kernel size is (1,1), the step size is 1, the padding is 0, the number of input channels is 128, and the number of output channels is set to 64; The number of input channels of the deep fusion convolution layer is consistent with the number of channels of the feature map s" obtained after the feature splicing, where s'1 and s'3 are the feature maps of the amplitude and phase parameters of the ionospheric scattering signals of satellite-to-ground and ground-to-satellite transmission and reception, respectively. Before feature fusion, there is a z% probability that they are replaced by Gaussian noise to adapt to the input of only spontaneous and self-received signal features in the on-orbit prediction stage, where Gaussian noise is a normal distribution with a mean of 0 and a variance of 1.

7. The method for determining ionospheric TEC based on satellite-ground active measurement and SAR image joint inversion according to claim 5, characterized in that: In step S3, in the on-orbit intelligent rapid inversion stage, the input of the feature fusion network is the feature map extracted from the ionospheric scattering signal measured by the dual-frequency SAR satellite after passing through the feature extraction network and the randomly generated Gaussian signal, and the feature map extracted after the feature extraction network and the randomly generated Gaussian signal are feature fused and then deep feature fusion is performed through the deep fusion convolution layer, and the number of input channels of the deep fusion convolution layer is consistent with the number of channels of the feature map extracted after passing through the feature extraction network.

8. The method for determining ionospheric TEC based on satellite-ground active measurement and SAR image joint inversion according to claim 7, characterized in that: In the step S3, it specifically includes: Step S31, obtaining and processing the dual-frequency SAR satellite scattered echo signal to obtain the ionospheric scattering amplitude and phase parameters of the dual-frequency SAR satellite scattered echo signal; Step S32, performing standardized preprocessing on the ionospheric scattering amplitude and phase parameters of the dual-frequency SAR satellite scattered echo signal extracted and inputting them into the TEC inversion deep learning model; Step S33: Perform on-orbit intelligent rapid inversion of the ionospheric scattering amplitude and phase parameters of the dual-frequency SAR satellite scattered echo signal through the TEC inversion deep learning model to obtain the ionospheric TEC value Γ' above the dual-frequency SAR satellite observation area. q .

9. A system for implementing the ionospheric TEC determination method based on satellite-ground active measurement and SAR image joint inversion as claimed in any one of claims 1 to 8, characterized in that: include: FPGA, used for receiving and acquiring scattered echo signals of dual-frequency SAR satellites, and preprocessing the scattered echo signals to obtain ionospheric scattering amplitude and phase parameters thereof; The GPU platform is used to load and configure the TEC inversion deep learning model, and to invert the TEC value of the ionosphere above the measurement position of the dual-frequency SAR satellite through the TEC inversion deep learning model.

Images