Ground-based interferometric synthetic aperture radar atmospheric phase correction method for optimizing XGBoost model based on SSA algorithm

Abstract

The invention relates to a ground-based interferometric synthetic aperture radar atmospheric phase correction method for optimizing an XGBoost model based on an SSA algorithm, and the method comprises the steps: obtaining a plurality of interferograms based on a time sequence radar image sequence; selecting a permanent scatterer according to a coherence coefficient method and an amplitude deviation dual-threshold method to estimate an atmospheric phase screen; calculating a standard deviation through phase accumulation, and establishing a feature matrix; model parameters are optimized through a sparrow search algorithm in combination with XGBoost, so that an atmospheric phase, a noise phase and a deformation phase are classified; and fitting the atmospheric phase screen. According to the embodiments of the invention, the interference of a deformation area and noise is effectively eliminated, so that the area influenced by the atmosphere can be more accurately identified and corrected.

Description

Technical Field

[0001] The present disclosure relates to the technical field of atmospheric correction in ground-based interferometric synthetic aperture radar slope monitoring, and in particular to a ground-based interferometric synthetic aperture radar atmospheric phase correction method based on an SSA algorithm to optimize an XGBoost model. Background Art

[0002] The existing methods for atmospheric phase correction are mainly divided into three categories.

[0003] The first method is based on meteorological data correction. This method estimates the atmospheric phase by constructing an atmospheric refractive index model and combining multi-dimensional meteorological data such as temperature, humidity and air pressure. However, the correction accuracy of this method is mainly limited by the spatial distribution characteristics of meteorological observation sites and the data quality of the meteorological parameter collection system. In large-scale monitoring scenarios, for monitoring of distant target points, errors in meteorological data may lead to lower correction accuracy.

[0004] The second type of method is model-based correction, which uses artificially set stable points or permanent scatterers (PS) to construct a regression model to estimate the atmospheric phase and correct the atmospheric phase. This method does not require the acquisition of meteorological data in advance, and is simple and easy to implement. However, it is based on the assumption of a uniform atmospheric medium, and the correction accuracy is limited by the construction of the regression model. When the study area is at high altitude or in a complex environment, due to the drastic atmospheric changes, this model based on the assumption of atmospheric homogeneity is difficult to effectively correct the atmospheric phase.

[0005] The third method is correction based on grid division. This method usually divides the atmospheric phase screen (APS) into several sub-areas and realizes the correction of atmospheric phase through interpolation or parameter estimation. This method can effectively deal with nonlinear atmospheric phase under complex atmospheric conditions and provide higher correction accuracy. However, this method has a large amount of calculation and takes longer to process. Summary of the invention

[0006] The present disclosure intends to provide a ground-based interferometric synthetic aperture radar atmospheric phase correction method based on the SSA algorithm to optimize the XGBoost model, which effectively eliminates the interference of deformation areas and noise, thereby more accurately identifying areas affected by the atmosphere and making corrections.

[0007] According to one of the solutions disclosed herein, a method for atmospheric phase correction of ground-based interferometric synthetic aperture radar based on an SSA algorithm-optimized XGBoost model is provided, comprising:

[0008] Based on the time-series radar image sequence, a complex interferogram is obtained;

[0009] Select permanent scatterers according to the coherence coefficient method and the amplitude deviation double-threshold method to estimate the atmospheric phase screen;

[0010] Establish a feature matrix by phase accumulation and calculation of standard deviation;

[0011] Optimize the model parameters by combining the sparrow search algorithm with XGBoost to classify the atmospheric phase, noise phase, and deformation phase;

[0012] Fit the atmospheric phase screen.

[0013] In some embodiments, based on the temporal radar image sequence, a complex interferogram is obtained, including:

[0014] Register the adjacent two scenes of data in the temporal radar image sequence as the master image and the slave image respectively, and perform conjugate multiplication to obtain a complex interferogram.

[0015] In some embodiments, the coherence coefficient method includes:

[0016] Solve the coherence coefficient value of the central pixel by the coherence information of all pixels within the selected window.

[0017] In some embodiments, the amplitude deviation double-threshold method includes:

[0018] Select the target point;

[0019] Calculate the deviation of the amplitude of the target point and its surrounding pixels, and statistically analyze the amplitude deviation of all pixels in this area;

[0020] Obtain the amplitude deviation index as a quantization index.

[0021] In some embodiments,

[0022] Phase accumulation includes: accumulating the phases of the corresponding windows of multiple interferograms;

[0023] Calculation of standard deviation includes: statistically analyzing the interferometric phases within each window and calculating the standard deviation to measure the degree of phase dispersion.

[0024] In some embodiments, establishing a feature matrix includes:

[0025] Extract the corresponding eigenvalue as the label value for the deformation area, atmospheric area, and noise area in the permanent scatterer points, and establish a feature matrix based on phase accumulation, standard deviation, and label value.

[0026] In some embodiments, using the sparrow search algorithm includes:

[0027] It is determined that the hyperparameters to be optimized using the sparrow search algorithm are the maximum number of iterations, the depth of the tree, and the learning rate;

[0028] For each individual in the population, calculate the objective function value;

[0029] Sort and select the current optimal solution and its fitness;

[0030] In each iteration, according to the fitness ranking, divide the sparrow population into three roles: discoverers, joiners, and anti-catchers;

[0031] By continuously updating the positions of the three, calculate the newly generated hyperparameter combinations again, and perform boundary control on all updated individuals;

[0032] Calculate the new fitness of each individual after the update, re-sort the population, update the global optimal solution and the corresponding fitness until the optimal solutions of the maximum number of iterations, the depth of the tree, and the learning rate are obtained.

[0033] In some embodiments, wherein,

[0034] For discoverers, update the position of the solution by simulating the group behavior of sparrows;

[0035] For joiners, adjust according to the position of the discoverers in the population;

[0036] For anti-catchers, adjust the position according to the relationship between their fitness value and the current optimal solution.

[0037] In some embodiments, wherein, optimize the model parameters by combining the sparrow search algorithm with XGBoost, including:

[0038] Based on the obtained feature matrix and the updated hyperparameters, input into the model;

[0039] Calculate the average value as the prediction value of the first tree and calculate the loss function;

[0040] According to the obtained gradient, construct a new decision tree, and fit the residual of the current prediction value, and add the prediction value of the newly constructed decision tree to the current prediction value;

[0041] After multiple iterations, superimpose the prediction results of several decision trees to obtain the sample prediction result.

[0042] In some embodiments, wherein, fitting the atmospheric phase screen, including

[0043] Use the least squares method to fit the obtained atmospheric phase components to the atmospheric phase screen;

[0044] Based on the classification result and the estimated atmospheric phase, obtain the compensated phase.

[0045] The ground-based interferometric synthetic aperture radar atmospheric phase correction method based on the SSA algorithm to optimize the XGBoost model in various embodiments of the present disclosure is at least based on a time series of radar image sequences to obtain complex interferograms; selects permanent scatterers according to the coherence coefficient method and the amplitude deviation double-threshold method to estimate the atmospheric phase screen; establishes a feature matrix through phase accumulation and standard deviation calculation; combines the sparrow search algorithm with the XGBoost optimization model parameters to classify the atmospheric phase, noise phase, and deformation phase; fits the atmospheric phase screen, and thus classifies through a machine learning model, effectively excluding the interference of the deformation area and noise, and thus more accurately identifying the area affected by the atmosphere and performing correction. This method not only significantly improves the accuracy and reliability of phase correction, but also has a small amount of calculation and effectively improves the operation speed.

[0046] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the claimed present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In the drawings, which are not necessarily drawn to scale, like reference numerals in different views may represent like components. Like reference numerals with alphabetic suffixes or like reference numerals with different alphabetic suffixes may represent different instances of like components. The drawings generally illustrate various embodiments by way of example and not limitation, and are used in conjunction with the description and the claims to explain the disclosed embodiments.

[0048] Figure 1 The method flow chart of the ground-based interferometric synthetic aperture radar atmospheric phase correction method based on the SSA algorithm to optimize the XGBoost model in an embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0049] In order to make the objectives, technical solutions, and advantages of the embodiments of the present disclosure clearer, the technical solutions of the embodiments of the present disclosure will be clearly and completely described below with reference to the drawings of the embodiments of the present disclosure. Obviously, the described embodiments are some, but not all, of the embodiments of the present disclosure. All other embodiments obtained by those of ordinary skill in the art based on the described embodiments of the present disclosure without creative efforts shall fall within the scope of protection of the present disclosure.

[0050] Landslide disasters pose a significant threat to the ecological environment, infrastructure, and human safety. Precise monitoring and early warning are key technical requirements for geological disaster prevention and control. The GB-SAR technology has unique technical value in the field of landslide deformation monitoring due to its high-resolution measurement ability, all-day and all-weather working characteristics, and non-contact monitoring advantages. However, in practical engineering applications, the measurement accuracy of this technology is significantly affected by environmental factors, and the change of atmospheric conditions is one of the main interference sources.

[0051] During the landslide monitoring process of the GB-SAR system, affected by the dynamic changes of the atmospheric environment, especially the spatio-temporal distribution inhomogeneity of temperature, humidity, and air pressure, the radar signal propagation path undergoes non-linear distortion, resulting in a significant atmospheric disturbance effect. This disturbance is manifested as the phase delay error in the radar echo signal, directly affecting the extraction accuracy of deformation parameters and reducing the reliability of monitoring results. In the landslide monitoring scenario, due to the significant terrain undulation in the monitoring area, the spatial gradient change of atmospheric parameters is more complex, making the atmospheric disturbance effect show obvious spatio-temporal heterogeneity. Therefore, how to effectively correct the atmospheric phase has become one of the key challenges in improving the performance of the GB-InSAR technology.

[0052] Combined with the content recorded in the background technology part above, this disclosure exemplarily records the corresponding solutions in the form of embodiments to solve the defects existing in the prior art, but does not limit the scope of patent rights required for this disclosure.

[0053] As one of the solutions, as Figure 1 shows the schematic diagram of the implementation process of the ground-based interferometric synthetic aperture radar atmospheric phase correction method based on the SSA algorithm to optimize the XGBoost model in an embodiment of this disclosure. The embodiment of this disclosure provides a ground-based interferometric synthetic aperture radar atmospheric phase correction method based on the SSA algorithm to optimize the XGBoost model, including:

[0054] Based on the time-series radar image sequence, obtain the complex interferogram;

[0055] Select permanent scatterers according to the coherence coefficient method and the amplitude deviation double-threshold method to estimate the atmospheric phase screen;

[0056] Establish the feature matrix by phase accumulation and calculating the standard deviation;

[0057] Optimize the model parameters by combining the sparrow search algorithm with the XGBoost to classify the atmospheric phase, noise phase, and deformation phase;

[0058] Fit the atmospheric phase screen.

[0059] In view of the foregoing, with the continuous development of deep learning technology, the atmospheric phase correction method based on machine learning has gradually become a new hotspot in the research field. Especially when dealing with the atmospheric phase delay problem in complex spatio-temporal data, machine learning has demonstrated significant potential. The embodiments of the present disclosure aim to propose a ground-based interferometric synthetic aperture radar atmospheric phase correction method that optimizes the XGBoost model based on the SSA algorithm. By classifying through a machine learning model, the interference of deformation regions and noise is effectively excluded, thereby more accurately identifying the regions affected by the atmosphere and performing correction. This method not only significantly improves the accuracy and reliability of phase correction, but also has a small amount of calculation and effectively improves the operation speed.

[0060] In some specific embodiments, the present disclosure may be to obtain a complex interferogram based on a temporal radar image sequence, including:

[0061] By registering the adjacent two scenes of data in the temporal radar image sequence as the master image and the slave image respectively, and performing conjugate multiplication, a complex interferogram is obtained.

[0062] Exemplarily, by accurately registering the adjacent two scenes of data in the temporal radar image sequence as the master image and the slave image respectively, and performing conjugate multiplication, a complex interferogram is obtained:

[0063]

[0064] Among them, S1S2 represents the master and slave images of two radars, and * represents conjugate multiplication.

[0065] The embodiments of the present disclosure adopt a double-threshold method of coherence coefficient and amplitude deviation to select permanent scatterers (PS). PS has stable electromagnetic wave backscattering characteristics, and its radar echo signal shows significant temporal coherence and amplitude stability in a long time series, and can maintain reliable phase information in a complex environment. Therefore, PS points can be used to estimate the atmospheric phase screen.

[0066] In some specific embodiments, the present disclosure may be a coherence coefficient method, including:

[0067] The coherence coefficient value of the central pixel is solved through the coherence information of all pixels within the selected window.

[0068] Exemplarily, the coherence coefficient method is to first select a certain window, and the coherence coefficient value of the central pixel is solved through the coherence information of all pixels within this window. Its calculation formula is as follows:

[0069]

[0070] Among them, S1 and S2 represent the local pixel information of the main and auxiliary images of the radar, * represents conjugate multiplication, and m and n represent the selected window sizes.

[0071] In some specific embodiments, the present disclosure may be an amplitude deviation double-threshold method, including: selecting a target point;

[0072] By calculating the deviation of the amplitudes of the target point and its surrounding pixels, the amplitude deviations of all pixels in this area are statistically analyzed;

[0073] An amplitude deviation index is obtained as a quantization index.

[0074] Exemplarily, the amplitude deviation method is to first select a target point. By calculating the deviation of the amplitudes of the target point and its surrounding pixels, the amplitude deviations of all pixels in this area are statistically analyzed, and a quantized index can be calculated, which is called the amplitude deviation index (ADI). Usually, more than 20 radar images are used. Its calculation formula is as follows:

[0075]

[0076] Among them, σ A represents the standard deviation of the amplitude time series of the target point, and m A represents the mean value of the amplitude time series of the target point.

[0077] Then, by setting the coherence coefficient threshold γ threshold and the amplitude deviation threshold D threshold , those satisfying the conditions γ > γ threshold and D A < D threshold are used as PS points. Here, γ threshold = 0.7 and D threshold = 0.3 are used to preliminarily screen PS points to make them have deformation, atmosphere, and noise components.

[0078] In some specific embodiments, the present disclosure may be phase accumulation, including: accumulating the phases of corresponding windows of multiple interferograms; calculating the standard deviation, including: statistically analyzing the interferometric phases within each window and calculating the standard deviation to measure the degree of phase dispersion.

[0079] Exemplarily, by accurately registering the adjacent two scenes of data in the time-series radar image sequence as the main image and the slave image respectively, conjugate multiplication is performed to obtain a single complex interferogram, and the phases of corresponding windows of multiple interferograms are accumulated. Its calculation formula is as follows:

[0080]

[0081] Among them, represents the accumulated phase, Represents the phase of the i-th image.

[0082] Perform statistical analysis on the interference phases within each window, calculate its standard deviation, which is used to measure the degree of phase dispersion. The calculation formula is as follows:

[0083]

[0084] Where, σ represents the phase standard deviation, represents the average value of the phases.

[0085] In some specific implementation scenarios, the present disclosure can be to establish a feature matrix, including:

[0086] Extract corresponding eigenvalue as the label value for the deformation region, atmospheric region, and noise region in the permanent scatterer points, and establish a feature matrix based on phase accumulation, standard deviation, and label value.

[0087] Exemplarily, for the input of establishing a model, the deformation region, atmospheric region, and noise region in the PS points can be extracted to obtain corresponding eigenvalue as the label value, and then a feature matrix is established with phase accumulation, standard deviation, and label value, and is input into the XGBoost model for classification.

[0088] For the constructed feature matrix X and label value Y (deformation phase = 1, atmospheric phase = 2, noise phase = 3), at the same time, divide the training set and test set in the ratio of 8:2, and then perform normalization processing on the feature matrix to train the model:

[0089]

[0090] Where, X is an n×1-dimensional observation matrix, and Y is an n×1-dimensional label matrix.

[0091] XGBoost (eXtreme Gradient Boosting), as an efficient gradient boosting framework, shows excellent performance in classification and regression tasks, but its model performance highly depends on the optimal configuration of hyperparameters. Traditional parameter tuning methods have problems such as low computational efficiency and being prone to falling into local optima. Therefore, the Sparrow Search Algorithm (SSA) is introduced. This algorithm simulates the foraging behavior of a sparrow group and realizes efficient global search through the explorer-follower mechanism, with advantages such as fast convergence speed and low parameter sensitivity. By combining SSA with XGBoost, intelligent optimization of model parameters can be achieved, significantly improving the prediction accuracy and generalization ability of the model.

[0092] In some specific implementation scenarios, the present disclosure can be to utilize the Sparrow Search Algorithm, including:

[0093] It is determined that the hyperparameters to be optimized using the sparrow search algorithm are the maximum number of iterations, the depth of the tree, and the learning rate;

[0094] For each individual in the population, calculate the objective function value;

[0095] Sort and select the current optimal solution and its fitness;

[0096] In each iteration, according to the fitness ranking, divide the sparrow population into three roles: discoverers, joiners, and anti-predators;

[0097] By continuously updating the positions of the three, calculate the newly generated hyperparameter combinations again, and perform boundary control on all updated individuals;

[0098] Calculate the new fitness of each individual after the update, re-sort the population, update the global optimal solution and the corresponding fitness until the optimal solutions of the maximum number of iterations, the depth of the tree, and the learning rate are obtained.

[0099] Exemplarily, the sparrow population exhibits the following three main behavioral patterns during foraging: Discoverers (Producers) are responsible for exploring new food sources and have strong global search capabilities; Scroungers (Scroungers) follow the discoverers to conduct local searches near known food sources and have strong exploitation capabilities; Anti-predation Behavior enables sparrows to adjust their positions to avoid being preyed upon when perceiving danger, and this mechanism helps to jump out of local optima.

[0100] First, it is determined that the hyperparameters to be optimized using the sparrow search algorithm are the maximum number of iterations, the depth of the tree, and the learning rate. Let the number of sparrows be pop = 5, and the hyperparameters to be optimized be dim = 3. As in the search space, use the initialization function to randomly generate N solution vectors within the upper and lower bounds lb j and ub j to form the initial population matrix:

[0101]

[0102] At the same time, the boundary constraints need to be satisfied:

[0103] lb j ≤x ij ≤ub j

[0104] where lb j and ub j are the upper and lower bounds of the j-th dimension respectively. Set the upper and lower bounds of the maximum number of iterations, the depth of the tree, and the learning rate to Ib = [1, 1, 0.01] and ub = [100, 12, 0.01] respectively.

[0105] Then, for each individual \(x\) in the population i (\(i = 1,\ldots,N\)), calculate the objective function value \(f(x\) i ). Then, after sorting, select the current optimal solution \(X\) best . and its fitness \(f\) best .

[0106] In each iteration, according to the fitness sorting, the sparrow population is divided into three roles: discoverers, joiners, and anti - predators.

[0107] First are the discoverers. Update the position of the solution by simulating the group behavior of sparrows: The discoverers decide to perform local search or global search according to the warning value \(ST = 0.8\) and the random number \(R2\). If \(R2\lt ST\), the discoverers perform local search through the exponential decay formula; otherwise, perform global search through the random perturbation formula. The calculation formulas are as follows:

[0108]

[0109] where: \(x\) j is the current position of the \(j\) - th discoverer, \(R2\) is a random number, \(R2\in[0,1]\), \(ST\) is the warning value used to control the behavior of the discoverers, \(\alpha\) is a random number, \(\alpha\in[0,1]\), \(\eta\) is a random number, (standard normal distribution), and \(1\) is a vector of all \(1\)s.

[0110] For the joiners, they are adjusted according to the position \(j\) of the discoverers in the population: If the sparrows with a relatively forward position in the population (i.e., the first half of the joiners), they will approach the optimal solution \(x1\); if the sparrows with a relatively backward position in the population (i.e., the second half of the joiners), they will be adjusted according to the position of the last sparrow in the population. The calculation formulas are as follows:

[0111]

[0112] \(x1\) is the current optimal solution (the position of the discoverers), \(A\) is a random vector, \(A=[a1,a2,\ldots,a\) dim , where \(a\) i \in{- 1,1\}\), \(\eta\) is a random number, \(x\) end is the position of the last sparrow in the current population (i.e., the worst solution).

[0113] The anti - predators adjust their positions according to the relationship between their fitness values and the current optimal solution: If the fitness value \(f(x\) j ) of sparrow \(j\) is greater than the current optimal fitness value \(BestF\), then the sparrow will approach the current optimal solution \(x1\). If the fitness value \(f(x\) j) When it is equal to the current optimal fitness value BestF, it indicates that the sparrows are already in a relatively optimal position. Its calculation formula is as follows:

[0114] x1 + η·|x j - x1|, iff (x j ) > BestF

[0115]

[0116] BestF is the current optimal fitness value, K and η are random numbers, ∈[-1, 1], and ∈ is a very small value used to avoid the denominator being zero.

[0117] Then, by continuously updating the positions of the three, calculate the newly generated hyperparameter combination again, and perform boundary control on all updated individuals to ensure that each component satisfies

[0118]

[0119] Calculate the new fitness of each individual after the update And re - sort the population to update the global optimal solution X best and the corresponding fitness f best , until the optimal solutions of the maximum number of iterations, the depth of the tree, and the learning rate are obtained.

[0120] XGBoost (eXtreme Gradient Boosting) is an efficient ensemble learning framework based on decision trees, which optimizes the objective function by iteratively constructing additive models. The core of this algorithm lies in adopting the gradient boosting strategy, gradually minimizing the loss function by sequentially adding decision trees, where the construction of each new tree is dedicated to correcting the prediction error of the previous tree. XGBoost significantly improves the computational efficiency while maintaining high prediction accuracy by introducing regularization terms and parallel computing optimizations.

[0121] In some specific implementation schemes, the present disclosure can be to optimize the model parameters by combining the sparrow search algorithm with XGBoost, including:

[0122] Based on the obtained feature matrix and the updated hyperparameters, input them into the model;

[0123] Calculate the loss function by calculating the average value as the prediction value of the first tree;

[0124] According to the obtained gradient, construct a new decision tree, and fit the residual of the current prediction value, and add the prediction value of the newly constructed decision tree to the current prediction value;

[0125] After multiple iterations, superimpose the prediction results of several decision trees to obtain the sample prediction result.

[0126] For exemplary illustration, the 80% feature matrix obtained as described above and the updated hyperparameters are input into the model.

[0127] Calculate the average value as the prediction value of the first tree, and then calculate the loss function, whose calculation formula is as follows:

[0128]

[0129] Where: y i is the true label of the i-th sample, is the prediction value of the model for the i-th sample. is the loss function, which measures the difference between the prediction value and the true value y i for the i-th sample.

[0130] Then calculate the first-order gradient and the second-order gradient, whose calculation formulas are as follows:

[0131] First-order gradient:

[0132] Second-order gradient:

[0133] Where, y i is the true label of the i-th sample, is the previous prediction value for the i-th sample.

[0134] Then, based on the gradient, construct a new decision tree and fit the residual of the current prediction value. Add the prediction value of the newly constructed decision tree to the current prediction value, thereby gradually reducing the prediction error:

[0135]

[0136] Where, I is all the samples in the node, I L I R are the samples of the left and right nodes; g i , h i are the first-order and second-order derivatives of the input samples; γ and λ are regularization parameters.

[0137] The goal of constructing the tree is to optimize the objective function, which is used to measure the current classification effect. Its calculation formula is as follows:

[0138]

[0139] Where, Ω(f k ) is regularization, and its calculation formula is as follows:

[0140]

[0141] Among them, T is the number of leaf nodes of the tree, and w j is the weight of the j-th leaf node, and γ and λ are regularization parameters.

[0142] After each decision tree is established, the predicted value of each leaf node is determined as follows:

[0143]

[0144] After multiple iterations, the predicted results of several decision trees are superimposed to obtain the sample prediction result, and its calculation formula is as follows:

[0145] Among them, K is the total number of trees, and f k is the model of the K-th tree, and x i is the i-th sample, the predicted result of the sample, and f k (x i ) is the sub-predicted result of the sample.

[0146] According to the above steps, the atmospheric phase, noise phase, and deformation phase can be classified.

[0147] In some specific implementation manners, the present disclosure may be to fit an atmospheric phase screen, including fitting the obtained atmospheric phase component to the atmospheric phase screen by using the least squares method;

[0148] Based on the classification result and the estimated atmospheric phase, the compensated phase is obtained.

[0149] Exemplarily, fitting the obtained atmospheric phase component to the atmospheric phase screen by using the least squares method, and its calculation formula is as follows:

[0150] ΔΦ = Xβ + ε

[0151]

[0152] Among them, m is the number of atmospheric points after classification, r i are respectively the expanded phase and the distance from the radar of the i-th atmospheric point after classification. X is an m×3-dimensional observation matrix, and β is a 3×1-dimensional vector to be estimated. ε is an m×1-dimensional vector containing random errors, and the unknown vector β can be estimated by least squares regression:

[0153]

[0154] Then the estimated value of the atmospheric phase is:

[0155]

[0156] Subtract the estimated atmospheric phase from ΔΦ after the classification result That is the compensated phase, and the unbiased estimator of the variance error is given by the following expression:

[0157]

[0158] The ground-based interferometric synthetic aperture radar atmospheric phase correction method based on the SSA algorithm to optimize the XGBoost model in each embodiment of the present disclosure has beneficial effects compared with the prior art, which are at least reflected in:

[0159] 1. When using a GB-InSAR system for observation, if extreme weather such as strong wind, rain, snow, etc. occurs or the observation area deforms, it will affect the analysis and judgment of the observation area; the ground-based interferometric synthetic aperture radar (GB-InSAR) atmospheric phase correction method based on the sparrow search algorithm (SSA) to optimize the XGBoost model classification can identify deformation and noise, and achieve reasonable correction of the atmospheric phase.

[0160] 2. The ground-based interferometric synthetic aperture radar (GB-InSAR) atmospheric phase correction method based on the sparrow search algorithm (SSA) to optimize the XGBoost model classification combines the global optimization ability of SSA and the powerful classification performance of XGBoost, realizes high-precision and high-efficiency classification, and thus performs correction.

[0161] As one of the solutions, the embodiment of the present disclosure provides a device for the ground-based interferometric synthetic aperture radar atmospheric phase correction method based on the SSA algorithm to optimize the XGBoost model, configured with at least one processing module to implement the ground-based interferometric synthetic aperture radar atmospheric phase correction method based on the SSA algorithm to optimize the XGBoost model described in the above embodiments.

[0162] Specifically, one of the inventive concepts of the present disclosure is that, based on at least a sequence of temporal radar images in various embodiments of the present disclosure, a complex interferogram is obtained; permanent scatterers are selected according to the coherence coefficient method and the amplitude deviation double-threshold method to estimate the atmospheric phase screen; a feature matrix is established by phase accumulation and standard deviation calculation; the sparrow search algorithm is combined with the XGBoost optimization model parameters to classify the atmospheric phase, noise phase, and deformation phase; the atmospheric phase screen is fitted, and thus classification is performed through a machine learning model, effectively excluding the interference of the deformation area and noise, and thus more accurately identifying the area affected by the atmosphere and performing correction. This method not only significantly improves the accuracy and reliability of phase correction, but also has a small amount of calculation and effectively improves the operation speed.

[0163] The present disclosure also provides a computer-readable storage medium, on which computer-executable instructions are stored. When the computer-executable instructions are executed by a processor, it mainly implements the ground-based interferometric synthetic aperture radar atmospheric phase correction method for optimizing the XGBoost model based on the SSA algorithm as described above, including at least:

[0164] Based on the time-series radar image sequence, obtain a complex interferogram;

[0165] Select permanent scatterers according to the coherence coefficient method and the amplitude deviation double-threshold method to estimate the atmospheric phase screen;

[0166] Establish a feature matrix by phase accumulation and standard deviation calculation;

[0167] Optimize the model parameters by combining the sparrow search algorithm with XGBoost to classify the atmospheric phase, noise phase, and deformation phase;

[0168] Fit the atmospheric phase screen.

[0169] The above embodiments are only exemplary embodiments of the present disclosure and are not used to limit the present disclosure. The protection scope of the present disclosure is defined by the claims. Those skilled in the art can make various modifications or equivalent replacements within the essence and protection scope of the present disclosure, and such modifications or equivalent replacements should also be regarded as falling within the protection scope of the present disclosure.

Claims

1. The atmospheric phase correction method of ground-based interferometric synthetic aperture radar based on SSA algorithm to optimize the XGBoost model includes: Based on the time-series radar image sequence, a complex interferogram is obtained; Permanent scatterers are selected based on the coherence coefficient method and the amplitude deviation double threshold method to estimate the atmospheric phase screen. Establish the characteristic matrix by phase accumulation and calculation of standard deviation; The model parameters are optimized by combining the sparrow search algorithm with XGBoost to classify the atmospheric phase, noise phase, and deformation phase; Fitting the atmospheric phase screen.

2. The method according to claim 1, wherein: Based on the time-series radar image sequence, a complex interferogram is obtained, including: The complex interference pattern is obtained by registering two adjacent scenes in the time-series radar image sequence as the main image and the slave image and performing conjugate multiplication.

3. The method according to claim 2, wherein: Coherence coefficient method, including: The coherence coefficient value of the central pixel is solved by the coherence information of all pixels in the selected window.

4. The method according to claim 3, wherein: Amplitude deviation dual threshold method, including: Select the target point; By calculating the deviation of the amplitude of the target point and its surrounding pixels, the amplitude deviation of all pixels in the area is counted; The amplitude deviation index is obtained as a quantitative indicator.

5. The method according to claim 4, wherein: Phase accumulation, including: accumulating the phases of corresponding windows of multiple interference patterns; Calculating the standard deviation includes: performing statistical analysis on the interference phase in each window and calculating the standard deviation to measure the dispersion degree of the phase.

6. The method according to claim 5, establishing a feature matrix, comprising: The corresponding eigenvalues ​​of the deformation area, atmospheric area, and noise area in the permanent scatterer point are extracted as label values, and a feature matrix is ​​established based on phase accumulation, standard deviation, and label value.

7. The method according to claim 6, wherein: Utilizes the Sparrow search algorithm, including: Determine the hyperparameters that need to be optimized using the sparrow search algorithm as the maximum number of iterations, the depth of the tree, and the learning rate; For each individual in the population, calculate the value of the objective function; Sort and select the current optimal solution and its fitness; In each iteration, the sparrow population is divided into three roles: discoverers, joiners, and anti-captors according to the fitness ranking; By continuously updating the positions of the three, the newly generated hyperparameter combination is calculated again, and boundary control is performed on all updated individuals; Calculate the new fitness of each individual after the update, re-sort the population, update the global optimal solution and the corresponding fitness, until the optimal solution of the maximum number of iterations, tree depth, and learning rate is obtained.

8. The method according to claim 7, wherein: For the finder, the position of the solution is updated by simulating the group behavior of sparrows; For joiners, adjustments are made based on the finder’s position in the population; For the anti-predator, the position is adjusted according to the relationship between its fitness value and the current optimal solution.

9. The method according to claim 8, wherein: The model parameters are optimized by combining the sparrow search algorithm with XGBoost, including: Based on the obtained feature matrix and the updated hyperparameters, input the model; Calculate the loss function by calculating the average value as the prediction value of the first tree; According to the obtained gradient, a new decision tree is constructed, and the residual of the current prediction value is fitted, and the prediction value of the newly constructed decision tree is added to the current prediction value; After multiple iterations, the prediction results of several decision trees are superimposed to obtain the sample prediction results.

10. The method according to claim 9, wherein: Fitting atmospheric phase screen, including The obtained atmospheric phase components are fitted to the atmospheric phase screen using the least square method; Based on the classification results and the estimated atmospheric phase, the compensated phase is obtained.

Images