A deep learning network training method for ionospheric error correction of insar

Abstract

The embodiment of the application discloses a kind of InSAR ionospheric error correction methods of deep learning network training.The method comprises: obtaining training sample and ionospheric phase initial estimation parameter, training sample includes unwrapping interference phase and azimuth offset, and training sample input initial error correction network;Unwrapping interference phase and azimuth offset are processed, and the probability value that the ionospheric category corresponding to training sample belongs to background ionosphere, ionospheric disturbance, ionospheric scintillation respectively and target ionospheric estimation parameter are obtained;Multi-constraint loss function is constructed, and target loss is calculated based on probability value, ionospheric phase initial estimation parameter and target ionospheric estimation parameter;The network parameter of initial error correction network is adjusted, and the error correction network of training completion is obtained.The above-mentioned mode adopts double-channel input unwrapping interference phase and the azimuth offset, and the perception ability of double-feature complementary promotion network to complex ionospheric scene is improved.

Description

Technical Field

[0001] This invention relates to the field of error correction technology for synthetic aperture radar interferometry, and more particularly to an InSAR ionospheric error correction method trained by a deep learning network. Background Technology

[0002] Interferometry synthetic aperture radar (InSAR) technology achieves precise monitoring of surface deformation by analyzing the phase difference between two or more SAR images. For multi-temporal InSAR subsidence monitoring, the stability and continuity of the results are highly dependent on the quality of the interferometric phase, and ionospheric perturbation error is one of the important external factors affecting the accuracy of long-wavelength InSAR. The ionosphere is a dispersive medium, and SAR signals will experience frequency-dependent propagation delays when crossing the ionosphere, causing phase screen changes at different spatial scales in the InSAR interferogram. This, in turn, affects the reliability of subsidence rate and deformation field estimation in time-series InSAR technology, especially for low-frequency long-wavelength L-band and P-band SAR data. Currently, several long-wavelength SAR satellites have been launched, including LuTan-1, ALOS-4, and NISAR. In addition, low-frequency SAR satellites such as Tandem-L, ROSE-L, and BIOMASS are being deployed intensively in the future. With the increasing amount of long-wavelength SAR data, ionospheric error has gradually become one of the key factors restricting high-precision deformation extraction by InSAR.

[0003] Existing InSAR ionospheric error correction methods can be categorized into two types: 1) based on external data; and 2) based on the InSAR data itself. Methods based on external data primarily utilize TEC (Transmission and Controlled Ionospheric Map) retrieved from Global Navigation Satellite System (GNSS) or Global Ionospheric Image (GIM) published by Global Data Centers to estimate the ionospheric line-of-sight delay. Methods based on the InSAR data itself mainly utilize the dispersion, polarization, geometric, or frequency domain characteristics of ionospheric errors in SAR observations for estimation and separation. Typical methods include the range-split spectrum method, azimuth migration method, and Faraday rotation method. However, existing methods generally suffer from strong dependence on external data or specific observation conditions, insufficient characterization of non-stationary nonlinear ionospheric disturbances, and difficulty in simultaneously addressing low-frequency background and high-frequency detail recovery, making it difficult to simultaneously achieve accuracy, applicability, and efficiency. Summary of the Invention

[0004] The purpose of this invention is to propose an InSAR ionospheric error correction method trained by deep learning networks, which aims to solve the problem of insufficient accuracy in InSAR ionospheric error correction in existing technologies.

[0005] To address the aforementioned technical problems, the first aspect of this application provides a method for InSAR ionospheric error correction trained using a deep learning network. The method includes: acquiring training samples and initial ionospheric phase estimation parameters, wherein the training samples include unwrapped interferometric phase and azimuth offset of a synthetic aperture radar interferogram, and the initial ionospheric phase estimation parameters serve as training pseudo-labels; constructing dual-channel input features based on the unwrapped interferometric phase and the azimuth offset, and inputting these features into an initial error correction network; processing the unwrapped interferometric phase and the azimuth offset through the initial error correction network to obtain probability values ​​indicating that the ionospheric category corresponding to the training samples belongs to the background ionosphere, ionospheric disturbance, and ionospheric scintillation, respectively, and target ionospheric estimation parameters; constructing a loss function with multiple constraints, and calculating a target loss based on the probability values, the initial ionospheric phase estimation parameters, and the target ionospheric estimation parameters using the loss function; and adjusting the network parameters of the initial error correction network according to the target loss to obtain a trained error correction network.

[0006] In one embodiment, the step of processing the unwrapped interferometric phase and the azimuth offset through the initial error correction network to obtain the probability values ​​of the ionospheric category corresponding to the training sample belonging to the background ionosphere, ionospheric disturbance, and ionospheric scintillation, respectively, and the target ionospheric estimation parameters, includes: performing multi-scale feature extraction on the unwrapped interferometric phase and the azimuth offset to obtain multi-scale spatial features; performing classification regression based on the multi-scale spatial features to obtain a first probability that the ionospheric category corresponding to the current training sample belongs to the background ionosphere, a second probability that it belongs to the ionospheric disturbance, a third probability that it belongs to the ionospheric scintillation, and the current round of ionospheric phase estimation; and determining the target ionospheric estimation parameters using a gated recurrent strategy based on the second probability, the third probability, and the current round of ionospheric phase estimation.

[0007] In one embodiment, the step of determining the target ionospheric estimation parameters based on the second probability, the third probability, and the current ionospheric phase estimation using a gated loop strategy includes: using the second probability and the third probability to perform gated modulation on the current ionospheric phase estimation to obtain the current ionospheric correction amount; Based on the current ionospheric correction and the current residual phase input, determine the next round of residual phase input; using the next round of residual phase input as the new current round residual phase input, update the first probability, second probability, third probability, and current round ionospheric phase estimate; repeatedly execute the steps from "using the second probability and third probability to perform gating modulation on the current round ionospheric phase estimate to obtain the current round ionospheric correction" to "using the next round of residual phase input as the new current round residual phase input, update the first probability, second probability, third probability, and current round ionospheric phase estimate" until the number of iterations reaches the set number; based on the current round ionospheric phase estimate output by each gating cycle, determine the target ionospheric estimation parameters.

[0008] In one embodiment, the loss function for the multiple constraints includes at least one of the following: a pseudo-label supervision loss function, a sub-band difference constraint loss function, a spatial continuity constraint loss function, and a classification supervision loss function; the step of calculating the target loss based on the probability value, the initial ionospheric phase estimation parameter, and the target ionospheric estimation parameter using the loss function includes: calculating a first loss based on the probability value and the initial ionospheric phase estimation parameter using the pseudo-label supervision loss function; calculating a second loss based on the target ionospheric estimation parameter using the sub-band difference constraint loss function; calculating a third loss based on the target ionospheric estimation parameter using the spatial continuity constraint loss function; and calculating a fourth loss based on the probability value using the classification supervision loss function.

[0009] In one embodiment, calculating the first loss based on the probability value, the initial ionospheric phase estimation parameters, and the target ionospheric estimation parameters using the pseudo-label supervision loss function includes: calculating the first loss based on the difference between the target ionospheric estimation parameters and the pseudo-label using the pseudo-label supervision loss function.

[0010] In one embodiment, calculating the second loss based on the target ionospheric estimation parameters using the subband difference constraint loss function includes: mapping the target ionospheric estimation parameters to the subband difference domain to obtain equivalent parameters; determining the interference phase difference between the high-frequency subband and the low-frequency subband corresponding to the synthetic aperture radar interferogram; and determining the second loss based on the equivalent parameters and the phase difference using the subband difference constraint loss function.

[0011] In one embodiment, calculating the third loss based on the target ionosphere estimation parameters using the spatial continuity constraint loss function includes: calculating the third loss based on the first-order spatial gradient of the target ionosphere estimation parameters using the spatial continuity constraint loss function.

[0012] To address the aforementioned technical problems, a second aspect of this application provides an InSAR ionospheric error correction method. The method includes: acquiring a synthetic aperture radar interferogram (SAR) to be processed; performing interferometric unwrapping on the SAR to obtain an unwrapped interferometric phase, and estimating the azimuth offset of the SAR to obtain an azimuth offset; constructing a dual-channel input feature based on the unwrapped interferometric phase and the azimuth offset, and inputting it into an error correction network; processing the unwrapped interferometric phase and the azimuth offset through the error correction network to obtain predicted values ​​indicating whether the ionospheric category corresponding to the SAR belongs to the background ionosphere, ionospheric disturbance, or ionospheric scintillation, respectively, and an estimated ionospheric phase value; performing error correction on the unwrapped interferometric phase based on the estimated ionospheric phase value to obtain an error-corrected interferometric phase; and determining the ionospheric category corresponding to the SAR based on the predicted values.

[0013] To address the aforementioned technical problems, a third aspect of this application provides a terminal device including a processor and a memory coupled to each other; the memory stores a computer program, and the processor executes the computer program to implement the steps of the methods provided in the first / second aspects above.

[0014] To address the aforementioned technical problems, a fourth aspect of this application provides a computer-readable storage medium storing program data, which, when executed by a processor, implements the steps of the method provided in the first / second aspect above.

[0015] The embodiments of this invention have the following beneficial effects: Unlike existing technologies, this application obtains training samples and initial ionospheric phase estimation parameters. The training samples include the unwrapped interferometric phase and azimuth offset of a synthetic aperture radar interferogram, and the initial ionospheric phase estimation parameters serve as training pseudo-labels. A dual-channel input feature is constructed based on the unwrapped interferometric phase and the azimuth offset, and input into an initial error correction network. The initial error correction network processes the unwrapped interferometric phase and the azimuth offset to obtain the probability values ​​of the ionospheric category corresponding to the training samples belonging to the background ionosphere, ionospheric disturbance, and ionospheric scintillation, respectively, as well as the target ionospheric estimation parameters. A loss function with multiple constraints is constructed, and the target loss is calculated based on the probability values, the initial ionospheric phase estimation parameters, and the target ionospheric estimation parameters. The network parameters of the initial error correction network are adjusted according to the target loss to obtain the trained error correction network. The above method uses dual-channel input of unwrapped interference phase and azimuth offset. Unwrapped interference phase provides the main observation features for ionospheric error estimation, while azimuth offset provides auxiliary discrimination features for ionospheric non-uniform structure. The dual features complement each other to improve the network's perception ability of complex ionospheric scenes, especially enhancing the recognition accuracy of ionospheric scintillation. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art 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.

[0017] in: Figure 1 This is a schematic flowchart of an embodiment of the InSAR ionospheric error correction method for deep learning network training in this application; Figure 2 This is a flowchart illustrating an embodiment of step S13 of this application; Figure 3 This is a flowchart illustrating an embodiment of step S23 of this application; Figure 4 This is a schematic flowchart of an embodiment of the InSAR ionospheric error correction method of this application; Figure 5 This is a schematic block diagram of the structure of an embodiment of the terminal device of this application; Figure 6 This is a schematic block diagram of an embodiment of a computer-readable storage medium of this application. Detailed Implementation

[0018] 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.

[0019] We propose a deep learning model for ionospheric error correction. Considering that ionospheric effects in interferograms not only manifest as smooth long-wavelength variations caused by the background ionosphere, but also as stripes, ripples, and local irregular structures in the presence of disturbances or scintillation, and may cause apparent shifts in SAR azimuth imaging, simply relying on the original unwrapped interferometric phase as input is insufficient to effectively distinguish between true deformation information and anomalous responses caused by ionospheric inhomogeneities.

[0020] Please see Figure 1 , Figure 1 This is a schematic flowchart of an embodiment of the InSAR ionospheric error correction method for deep learning network training according to this application. It should be noted that if substantially the same result is obtained, this embodiment is not necessarily identical. Figure 1 The illustrated process sequence is limited. It includes the following steps S11~S15: S11: Obtain training samples and initial ionospheric phase estimation parameters. The training samples include the unwrapped interferometric phase and azimuth offset of the synthetic aperture radar interferogram, and the initial ionospheric phase estimation parameters are used as training pseudo-labels.

[0021] The unwrapped interferometric phase is obtained by interferometric unwrapping based on the synthetic aperture radar interferogram. Specifically, a set of T interferometric pairs is constructed, and the t-th interferogram is used as the input feature to avoid phase... Periodicity leads to regression instability. A preset number of time-series InSAR data sets can be collected for a specific area. Each set contains a primary and secondary SLC image pair. After registration of each SLC image set, conjugate multiplication yields a synthetic aperture radar interferogram (SAR). Registration, interferometry, terrain removal, and phase unwrapping of the SAR are then performed to obtain the unwrapped interferometric phase. Unwrapped interference phase This is a continuous differential interferometric phase that does not include terrain phase. Furthermore, the unwrapped interferometric phase is processed before input. The data is normalized and based on the coherence coefficient. Construct a hard threshold mask, retaining only High-confidence sample regions participate in supervised training, here This refers to the Mt(x) parameter of the loss function, which will be discussed later.

[0022] The azimuth offset is estimated based on the synthetic aperture radar interferogram (SAR). Specifically, it is obtained by estimating the azimuth offset of the primary and secondary SAR images within a local window. It is used to characterize the geometric response of ionospheric inhomogeneities in the SAR azimuth direction, serving as an auxiliary feature of ionospheric disturbance. The azimuth offset can be specifically expressed as: in, This represents the azimuth offset estimation operator. , These are the primary and secondary SAR images corresponding to the synthetic aperture radar interferogram, respectively, where x is the pixel located at position x in the synthetic aperture radar interferogram.

[0023] The initial ionospheric phase estimation parameters are: the approximate ionospheric phase values ​​obtained by the traditional spectral method, which are used as pseudo-labels for supervised training to guide the network to learn the ionospheric distribution pattern.

[0024] S12: Construct dual-channel input features based on unwrapped interference phase and azimuth offset, and input them into the initial error correction network.

[0025] The initial error correction network may include a first input channel, a second input channel, an encoder, and a decoder. The decoder includes a classification branch and a regression branch. The unwrapped interferometric phase and azimuth offset are used as input features for the first and second input channels, respectively. The encoder extracts features based on the unwrapped interferometric phase and azimuth offset. The classification branch outputs the probabilities of the ionosphere belonging to three categories: background ionosphere, ionospheric disturbance, and ionospheric scintillation. The regression branch outputs the estimated parameters of the target ionosphere.

[0026] S13: The unwrapped interference phase and azimuth offset are processed by the initial error correction network to obtain the probability values ​​of the ionospheric categories corresponding to the training samples belonging to the background ionosphere, ionospheric disturbance, and ionospheric scintillation, respectively, as well as the target ionospheric estimation parameters.

[0027] Among them, the background ionosphere indicates that the background ionosphere component in the sample is mainly smooth and gradually changing, spatially continuous and with weak local anomalies; ionospheric disturbance indicates that in addition to the background component, there are obvious continuous stripes, ripples or directional anomalous structures in the sample; ionospheric scintillation indicates that there are stronger irregular body effects in the sample, manifested as local fragmentation, rapid fluctuations, discontinuous jumps or patchy anomalies.

[0028] The target ionosphere estimation parameters are obtained by stripping the ionosphere through a network, and are used to characterize the stripped ionosphere.

[0029] S14: Construct a loss function with multiple constraints, and calculate the target loss based on the probability value, the initial ionospheric phase estimation parameter, and the target ionospheric estimation parameter using the loss function.

[0030] The loss function may include at least one of the following: pseudo-label supervised loss function, subband difference constraint loss function, spatial continuity constraint loss function, and classification supervised loss function.

[0031] in: The pseudo-label supervision loss function can calculate the loss based on the difference between the estimated parameters of the target ionosphere and the pseudo-label, guiding the regression branch output to fit the real ionosphere distribution. The subband difference constraint loss function can calculate the loss based on the result of mapping the target ionospheric estimation parameters to the subband difference domain and the phase difference of the actual subband. Since the ionospheric phase satisfies the dispersion relation, the ionospheric estimation output by the network should be able to reasonably explain the phase difference between the high and low subbands. Therefore, the subband difference constraint loss is introduced to constrain the network output to conform to the ionospheric dispersion characteristics. The spatial continuity constraint loss function can be calculated based on the first-order spatial gradient of the target ionospheric estimation parameters. The ionospheric phase screen typically exhibits a low-frequency continuous field; therefore, the spatial continuity constraint loss can be used to constrain the spatial continuity of the ionospheric phase. The classification supervision loss function can calculate the loss based on the ionospheric category probability and the ionospheric state label corresponding to the pseudo-label, guiding the classification branch to accurately determine the ionospheric state.

[0032] S15: Adjust the network parameters of the initial error correction network according to the target loss to obtain the trained error correction network.

[0033] Based on the target loss, the network parameters are updated through backpropagation using the Adam optimizer. After a finite number of iterations, the training is completed, resulting in a trained error correction network.

[0034] Considering that the ionospheric effect in interferograms not only manifests as smooth long-wavelength variations caused by the background ionosphere, but also as stripes, ripples, and local irregular structures when there are disturbances or scintillation, and may cause apparent offsets in SAR azimuth imaging, it is difficult to effectively distinguish between real deformation information and abnormal responses caused by the non-uniform structure of the ionosphere by simply relying on the original unwrapped interferometric phase as input. In order to reduce the interference of real deformation on auxiliary features and enhance the model's ability to identify different spatial patterns of the background ionosphere, disturbed ionosphere, and scintillation ionosphere, the above embodiments of the present invention introduce azimuth offset as a second channel input to characterize the geometric response of the non-uniform structure of the ionosphere in the SAR azimuth, and provide auxiliary discrimination information for subsequent classification branch identification of different ionospheric states of samples.

[0035] The above method uses dual-channel input of unwrapped interferometric phase and azimuth offset. Unwrapped interferometric phase provides the main observation features for ionospheric error estimation, while azimuth offset provides auxiliary discrimination features for ionospheric non-uniform structure. The dual features complement each other to improve the network's perception ability of complex ionospheric scenes, especially enhancing the recognition accuracy of ionospheric scintillation.

[0036] In one embodiment, the initial ionospheric phase estimation parameters are obtained as follows: Spectral analysis was performed on the synthetic aperture radar interferogram to obtain low-frequency and high-frequency subbands; The initial parameters for estimating the ionospheric phase are determined based on the center frequency of the low-frequency subband, the interference phase of the low-frequency subband, the center frequency of the high-frequency subband, the interference phase of the high-frequency subband, and the center frequency of the synthetic aperture radar interferogram.

[0037] Specifically, the range spectrum of the primary and secondary SLC images corresponding to the synthetic aperture radar interferogram is bandpass filtered and segmented to obtain low-frequency and high-frequency sub-bands. For a center frequency of... InSAR interferogram, its unwrapped phase It mainly consists of nondispersive phase terms that include surface deformation, topographic errors, and convection delays. Dispersive phase term caused by ionosphere The ionospheric phase delay is inversely proportional to frequency, while non-dispersive phases such as topography / deformation do not change with frequency. Therefore, the high-low subband interference phase difference is contributed only by the ionosphere, and the initial ionospheric phase estimation parameters can be roughly represented by the high-low subband interference phase difference. Specifically, the initial ionospheric phase estimation parameters can be expressed as follows: in, This represents the initial estimate of the ionospheric phase parameters. This indicates the interference phase of the low-frequency subband. This represents the interference phase of the high-frequency subband, and x represents the pixel at position x in the synthetic aperture radar interferogram. Indicates the center frequency of the high-frequency sub-band. Indicates the center frequency of the low-frequency sub-band. This indicates the center frequency of the synthetic aperture radar interferogram.

[0038] In one embodiment, please refer to the following: Figure 2 Step S13 includes the following steps: S21: Multi-scale feature extraction is performed on the unwrapped interference phase and azimuth offset to obtain multi-scale spatial features.

[0039] The encoder downsamples the unwrapped interferometric phase and azimuth offset of the dual-channel input stepwise, and extracts multi-scale encoded features through convolution and nonlinear activation layers to obtain multi-scale spatial features. .

[0040] S22: Based on multi-scale spatial features, perform classification regression to obtain the first probability that the ionospheric category corresponding to the current training sample belongs to the background ionospheric category, the second probability that it belongs to the ionospheric disturbance category, the third probability that it belongs to the ionospheric scintillation category, and the current ionospheric phase estimation.

[0041] Pooling layers with classification branches are used to process multi-scale spatial features. Global average pooling is performed to obtain And use a fully connected layer to A linear mapping to logits yields... The output layer uses softmax to... The classification can be done using the following formula: in, Represents the weight parameters of the fully connected layer. The bias parameter represents the k-th type of ionospheric state in the fully connected layer, where k=1,2,3 correspond to the background ionosphere, ionospheric disturbance, and ionospheric scintillation, respectively. This means iterating through categories 1, 2, and 3 and summing the results. This represents the softmax operation. This represents the probability value of belonging to the k-th type of ionospheric state, satisfying... =1.

[0042] In one embodiment, the current round of ionospheric phase estimation is the network's regression branch pair decoding features. Regression analysis yields the following ionospheric phase estimate: in, To combine multi-scale spatial features in the decoder and the feature representation obtained after stepwise upsampling and recovery of skip connections. This represents a regression estimation process guided by classification. This indicates that the ionospheric phase estimate for this round is not yet available.

[0043] S23: Based on the second probability, the third probability, and the current ionospheric phase estimation, the target ionospheric estimation parameters are determined using a gated cyclic strategy.

[0044] This embodiment improves the accuracy and stability of InSAR ionospheric error correction by optimizing ionospheric phase estimation through gated iterative optimization. Without the need for external observation data and explicit physical propagation models, it achieves intelligent estimation and robust removal of InSAR ionospheric phase screens, effectively adapting to complex scenarios such as low coherence, strong noise, and complex stripes, thereby improving the accuracy and processing efficiency of wide-area deformation monitoring.

[0045] In one embodiment, please refer to Figure 3 Step S23 may include the following steps: S31: The ionospheric phase estimation of this round is gated and modulated using the second probability and the third probability to obtain the ionospheric correction amount of this round.

[0046] Specifically, when the ionospheric category belongs to ionospheric perturbation or ionospheric scintillation, it indicates that the ionospheric category of the current training sample is the foreground ionosphere. This step makes a preliminary estimate of the initial ionospheric phase based on the second and third probabilities corresponding to the foreground ionosphere. Specifically, the ionospheric correction amount in this round... Determined according to the following formula: Understandably, when the network determines that the ionospheric category of the current training sample is closer to the background ionospheric category, the foreground probability... When the value is small, the regression output is suppressed; however, when the network determines that the current training sample belongs to the ionospheric perturbation or ionospheric scintillation category... The larger the value, the more actively the network will output ionospheric corrections.

[0047] S32: Determine the residual phase input for the next round based on the current ionospheric correction and the current residual phase input.

[0048] The initial residual phase corresponding to the first gating cycle is the untangling interference phase.

[0049] The residual phase input for the next round is determined based on the difference between the residual phase input of the current round and the ionospheric correction of the current round. This is the residual phase input for the next gating cycle. It can be expressed as the following formula: in, Indicates the residual phase input for the next round. This indicates the residual phase input in this round. This indicates the amount of ionospheric correction in this round.

[0050] S33: Use the next round's residual phase input as the new current round's residual phase input to update the first probability, second probability, third probability, and current round's ionospheric phase estimate.

[0051] Specifically, the next round of residual phase input is used as the new current round of residual phase input. The initial error correction network processes the new current round of residual phase input and dual-channel input characteristics to output new first probability, second probability, third probability and current round of ionospheric phase estimate.

[0052] S34: Repeat steps S31 to S33 until the set number of iterations is reached.

[0053] S35: Determine the target ionospheric estimation parameters based on the current ionospheric phase estimation output of each gating cycle.

[0054] Specifically, the target ionospheric phase estimates are obtained by summing the historical outputs from the network training process. in, Indicates the estimated parameters of the target ionosphere. Indicates the number of gated cycles.

[0055] The gated recurrent loop, as a recursive optimization structure within the network, participates in the training process. In the forward propagation of a single sample, it doesn't perform only one ionospheric phase estimation; instead, it updates the residual phase input for the next round based on the current ionospheric phase estimate, and continues to input this data into the next round of network optimization. The classification probability in each round is not fixed but dynamically changes with the update of the residual phase during the recursive process. The classification branch, based on the current round's residual phase and its corresponding dual-channel features, re-outputs the probability that the sample belongs to one of three states: background ionosphere, ionospheric perturbation, or ionospheric scintillation. Since the significant ionospheric anomalous components in the samples have been gradually weakened in previous rounds of recursion, the residual phase in subsequent rounds is usually smoother. Therefore, the sample's class probability can gradually shift from ionospheric perturbation or ionospheric scintillation to background ionosphere. Based on this dynamic classification result, the output of the current round's regression branch is then gated to achieve adaptive estimation of residual ionospheric components at different stages.

[0056] In one embodiment, the loss function for multiple constraints includes at least one of the following: pseudo-label supervision loss function, subband difference constraint loss function, spatial continuity constraint loss function, and classification supervision loss function; The steps described above for calculating the target loss based on the probability value, the initial ionospheric phase estimation parameters, and the target ionospheric estimation parameters using a loss function include: The first loss is calculated using a pseudo-label supervised loss function based on probability values ​​and initial ionospheric phase estimation parameters. The second loss is calculated based on the target ionosphere estimation parameters using the sub-band difference constraint loss function. The third loss is calculated based on the target ionosphere estimation parameters by using a spatial continuity-constrained loss function. A fourth loss is calculated based on probability values ​​using a classification-supervised loss function.

[0057] Specifically, in order to ensure that the ionospheric phase estimation results are more reliable and that the corrected phase maintains reasonable spatial continuity, a multi-constraint joint loss function is set up, consisting of a pseudo-label supervision loss function, a sub-band difference constraint loss function, a spatial continuity constraint loss function, and a classification supervision loss function.

[0058] The loss function for multiple constraints can be expressed as: in, , , , For hyperparameters, , , , as well as These represent: loss function with multiple constraints, pseudo-label supervision loss function, subband difference constraint loss function, spatial continuity constraint loss function, and classification supervision loss function, respectively.

[0059] This embodiment combines pseudo-label supervision and physical constraints to ensure that the network output not only closely matches the actual ionospheric distribution but also conforms to the dispersion characteristics and spatial continuity, thus solving the problems of high noise and weak anti-interference ability of traditional spectral methods.

[0060] In one embodiment, a pseudo-label supervision loss function is used to calculate the first loss based on the difference between the estimated parameters of the target ionosphere and the pseudo-label, as shown in the following formula: in, This represents the summation operation of synthetic aperture radar interferograms over all time series. This represents the summation operation over k iterations of a gated loop. This represents the weight of the k-th round of the gated loop. This represents the summation of all pixels in the synthetic aperture radar interferogram. Represents pixel-level mask weights. This represents the target ionospheric estimation parameters corresponding to the t-th synthetic aperture radar interferogram. This indicates a pseudo-tag.

[0061] In one embodiment, the second loss is calculated as follows: The target ionospheric estimation parameters are mapped to the subband difference domain to obtain equivalent parameters; the interference phase difference between the high-frequency and low-frequency subbands corresponding to the synthetic aperture radar interferogram is determined; and the second loss is determined based on the equivalent parameters and phase difference using the subband difference constraint loss function.

[0062] Specifically, since the ionospheric phase satisfies a dispersion relation, the ionospheric estimate output by the network should be able to reasonably explain the phase difference between high and low subbands. Therefore, a subband difference consistency loss is introduced. The ionospheric phase approximately satisfies... The calculation of the second loss is specifically based on the following formula: in, and These represent the interference phases of the low-frequency and high-frequency sub-bands, respectively, where x is the pixel located at position x in the synthetic aperture radar interferogram. Indicates the estimated parameters of the target ionosphere. This indicates the estimated parameters of the target ionosphere. Equivalent parameters mapped to the subband difference domain.

[0063] In one embodiment, a spatial continuity constraint loss function is used to calculate a third loss based on the first-order spatial gradient of the target ionosphere estimation parameters. Specifically, the Laplace smoothing regularization loss is calculated, as shown in the following formula: in, Represents pixel-level mask weights. This represents the absolute value of the first-order spatial gradient of the target ionosphere estimation parameters.

[0064] In one embodiment, the classification supervision loss function is the cross-entropy loss function, and the fourth loss can be calculated according to the following formula: .

[0065] Please see Figure 4 , Figure 4 This is a schematic flowchart of an InSAR ionospheric error correction method according to this application, which includes the following steps: S41: Obtain the synthetic aperture radar interferogram to be processed; S42: Perform interferometric unwrapping on the synthetic aperture radar interferogram to be processed to obtain the unwrapped interferometric phase to be processed, and perform azimuth offset estimation on the synthetic aperture radar interferogram to obtain the azimuth offset amount to be processed. S43: Construct a dual-channel input feature based on the unwrapped interference phase and the azimuth offset to be processed, and then use an input error correction network; S44: The error correction network processes the unwrapped interferometric phase and the azimuth offset to be processed, and obtains the predicted values ​​of the ionospheric categories corresponding to the synthetic aperture radar interferogram to be processed, namely background ionospheric, ionospheric disturbance, and ionospheric scintillation, as well as the estimated value of the ionospheric phase. S45: Based on the ionospheric phase estimate, perform error correction on the unwrapped interferometric phase to be processed to obtain the error-corrected interferometric phase; and, based on the predicted value, determine the ionospheric category corresponding to the synthetic aperture radar interferogram to be processed.

[0066] In one embodiment, the error correction network is trained using the training method of any of the above embodiments. In practical applications, the trained error correction network is used to process the unwrapped interferometric phase and azimuth offset to obtain predicted values ​​of the ionospheric category corresponding to the synthetic aperture radar interferogram to be processed, namely, the background ionosphere, ionospheric disturbance, and ionospheric scintillation, as well as the estimated ionospheric phase value. Based on the estimated ionospheric phase value, error correction is performed on the unwrapped interferometric phase to be processed, that is, the estimated ionospheric phase value is subtracted from the unwrapped interferometric phase to be processed to obtain the correction result. The ionospheric category corresponding to the background ionosphere, ionospheric disturbance, and ionospheric scintillation with the highest predicted value is determined as the ionospheric category corresponding to the synthetic aperture radar interferogram to be processed.

[0067] Understandably, in other implementations, error correction networks obtained using other training methods can also be applied to process the unwrapped interference phase and the azimuth offset to be processed. The specific settings can be configured according to actual needs, and no specific limitations are made here.

[0068] The above method uses dual-channel input of the unwrapped interference phase to be processed and the azimuth offset to be processed. The unwrapped interference phase to be processed provides the main features of the ionospheric phase, and the azimuth offset to be processed provides the auxiliary features of the ionospheric phase. The dual features complement each other to improve the perception of complex ionospheric scenes and enhance the recognition accuracy of ionospheric scintillation.

[0069] In one embodiment, multi-scale feature extraction is performed on the unwrapped interference phase and the azimuth offset to be processed to obtain multi-scale spatial features; Based on the multi-scale spatial features, classification and regression processing is performed to obtain the first predicted value of the ionospheric category corresponding to the synthetic aperture radar interferogram to be processed, the second predicted value of the ionospheric disturbance, the third predicted value of the ionospheric scintillation, and the ionospheric phase estimate.

[0070] Please see Figure 5 , Figure 5This is a schematic block diagram of a terminal device according to an embodiment of the present application. The terminal device 900 includes a processor 910 and a memory 920 coupled to each other. The memory 920 stores a computer program. The processor 910 is used to execute the computer program to implement the InSAR ionospheric error correction method for deep learning network training described above, and any embodiment and any non-conflicting combination of the InSAR ionospheric error correction method.

[0071] For a description of each step of the processing, please refer to the description of each step of the InSAR ionospheric error correction method and the InSAR ionospheric error correction method embodiment of the above application, and will not be repeated here.

[0072] The memory 920 can be used to store program data and modules. The processor 910 executes various functional applications and data processing by running the program data and modules stored in the memory 920. The memory 920 may mainly include a program storage area and a data storage area. The program storage area may store the operating system, applications required for at least one function, etc.; the data storage area may store data created based on the use of the terminal device 900, etc. In addition, the memory 920 may include high-speed random access memory, and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other volatile solid-state storage device. Accordingly, the memory 920 may also include a memory controller to provide the processor 910 with access to the memory 920.

[0073] In the various embodiments of this application, the disclosed methods, apparatus, and devices can be implemented in other ways. For example, the embodiments of the terminal devices described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0074] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0075] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0076] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium.

[0077] See Figure 6 , Figure 6 This is a schematic block diagram of a computer-readable storage medium according to an embodiment of the present application. The computer-readable storage medium 700 stores program data 710. When the program data 710 is executed, it implements the steps of the InSAR ionospheric error correction method and various embodiments of the InSAR ionospheric error correction method as described above for deep learning network training.

[0078] For a description of each step of the processing, please refer to the description of each step of the InSAR ionospheric error correction method and the InSAR ionospheric error correction method embodiment of the above application, and will not be repeated here.

[0079] Any references to memory, storage, database, or other media used in the embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM), etc.

[0080] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0081] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features claimed herein.

Claims

1. A method for InSAR ionospheric error correction based on deep learning network training, characterized in that, The method includes: Acquire training samples and initial ionospheric phase estimation parameters. The training samples include the unwrapped interferometric phase and azimuth offset of the synthetic aperture radar interferogram. The initial ionospheric phase estimation parameters are used as training pseudo-labels. A dual-channel input feature is constructed based on the unwrapped interference phase and the azimuth offset, and then input into the initial error correction network. The initial error correction network processes the unwrapped interferometric phase and the azimuth offset to obtain the probability values ​​of the ionospheric category corresponding to the training sample belonging to the background ionosphere, ionospheric disturbance, and ionospheric scintillation, respectively, as well as the target ionospheric estimation parameters. This includes: extracting multi-scale features from the unwrapped interferometric phase and the azimuth offset to obtain multi-scale spatial features; performing classification regression based on the multi-scale spatial features to obtain the first probability that the ionospheric category corresponding to the current training sample belongs to the background ionosphere, the second probability that it belongs to the ionospheric disturbance, the third probability that it belongs to the ionospheric scintillation, and the current round of ionospheric phase estimation; and determining the target ionospheric estimation parameters using a gated recurrent strategy based on the second probability, the third probability, and the current round of ionospheric phase estimation. A loss function with multiple constraints is constructed, and the target loss is calculated based on the probability value, the initial ionospheric phase estimation parameters, and the target ionospheric estimation parameters using the loss function. Adjust the network parameters of the initial error correction network according to the target loss to obtain the trained error correction network.

2. The method according to claim 1, characterized in that, The determination of the target ionosphere estimation parameters based on the second probability, the third probability, and the current ionospheric phase estimation using a gated cyclic strategy includes: The ionospheric phase estimation of this round is gated and modulated using the second probability and the third probability to obtain the ionospheric correction amount of this round. Based on the current ionospheric correction and the current residual phase input, determine the next residual phase input; Using the next round of residual phase input as the new current round of residual phase input, update the first probability, the second probability, the third probability, and the current round of ionospheric phase estimation; The process is repeated from "gating and modulating the current ionospheric phase estimate using the second probability and the third probability to obtain the current ionospheric correction" to "updating the first probability, the second probability, the third probability and the current ionospheric phase estimate using the next round residual phase input as the new current round residual phase input" until the number of iterations reaches the set number. The target ionosphere estimation parameters are determined based on the current ionospheric phase estimate output from each round of gating loop.

3. The method according to claim 1, characterized in that, The loss function for the multiple constraints includes at least one of the following: Pseudo-label supervision loss function, sub-band difference constraint loss function, spatial continuity constraint loss function, and classification supervision loss function; The step of calculating the target loss using the loss function based on the probability value, the initial ionospheric phase estimation parameters, and the target ionospheric estimation parameters includes: The first loss is calculated using the pseudo-label supervised loss function, based on the probability value and the initial ionospheric phase estimation parameters. The second loss is calculated based on the target ionosphere estimation parameters using the sub-band difference constraint loss function. The third loss is calculated based on the target ionosphere estimation parameters using the spatial continuity constraint loss function. The fourth loss is calculated based on the probability value using the classification supervision loss function.

4. The method according to claim 3, characterized in that, The step of calculating the first loss using the pseudo-label supervised loss function, based on the probability value, the initial ionospheric phase estimation parameters, and the target ionospheric estimation parameters, includes: The first loss is calculated based on the difference between the estimated parameters of the target ionosphere and the pseudo-label using the pseudo-label supervision loss function.

5. The method according to claim 3, characterized in that, The step of calculating the second loss based on the target ionosphere estimation parameters using the sub-band difference constraint loss function includes: The target ionosphere estimation parameters are mapped to the sub-band difference domain to obtain equivalent parameters; Determine the interference phase difference between the high-frequency subband and the low-frequency subband corresponding to the synthetic aperture radar interferogram; The second loss is determined using the sub-band difference constraint loss function, based on the equivalent parameters and the phase difference.

6. The method according to claim 4, characterized in that, The calculation of the third loss based on the target ionosphere estimation parameters using the spatial continuity constraint loss function includes: The third loss is calculated using the spatial continuity constraint loss function based on the first-order spatial gradient of the target ionosphere estimation parameters.

7. An InSAR ionospheric error correction method, characterized in that, The method includes: Obtain the synthetic aperture radar interferogram to be processed; Interferometric unwrapping is performed on the synthetic aperture radar interferogram to be processed to obtain the unwrapped interferometric phase to be processed, and azimuth offset estimation is performed on the synthetic aperture radar interferogram to obtain the azimuth offset to be processed. A dual-channel input feature and an input error correction network are constructed based on the unwrapped interference phase to be processed and the azimuth offset to be processed. The error correction network processes the unwrapped interferometric phase and the azimuth offset to be processed, obtaining predicted values ​​for the ionospheric category corresponding to the synthetic aperture radar interferogram to be processed, indicating whether it belongs to the background ionosphere, ionospheric disturbance, or ionospheric scintillation, as well as an estimated ionospheric phase value. This includes: extracting multi-scale features from the unwrapped interferometric phase and the azimuth offset to obtain multi-scale spatial features; performing classification regression based on the multi-scale spatial features to obtain a first probability that the ionospheric category corresponding to the synthetic aperture radar interferogram to be processed belongs to the background ionosphere, a second probability that it belongs to ionospheric disturbance, a third probability that it belongs to ionospheric scintillation, and an estimated ionospheric phase value for the current round; and determining the estimated ionospheric phase value using a gated loop strategy based on the second probability, the third probability, and the estimated ionospheric phase value for the current round. The unwrapped interferometric phase to be processed is corrected for errors based on the ionospheric phase estimate to obtain the error-corrected interferometric phase; and the ionospheric category corresponding to the synthetic aperture radar interferogram to be processed is determined based on the predicted value.

8. A terminal device, characterized in that, The terminal device includes a processor and a memory coupled to each other; the memory stores a computer program, and the processor executes the computer program to implement the steps of the method as described in any one of claims 1-7.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores program data that, when executed by a processor, implements the steps of the method as described in any one of claims 1-7.

Images