Precise polarization calibration method for reciprocal coherence target of airborne full-polarization SAR (synthetic aperture radar)

Abstract

According to the airborne full-polarization SAR reciprocal coherence target polarization fine calibration method, related factors influencing polarization calibration precision are analyzed, a way and a method for improving polarization calibration quality are mastered macroscopically, and the polarization calibration quality is improved so as to facilitate application breadth and depth of polarization data. Influences of factors such as channel crosstalk, unbalance degree phase amplitude, system noise, multi-view processing and the like on airborne polarization calibration quality are researched, a new method for screening reciprocity and reflection symmetry targets is provided, and it is demonstrated that a reciprocity included angle and a coherence factor are used for judging the rationality of uncalibrated data reciprocity and coherence characteristics. And then using a polarization coherent target decomposition model to extract a symmetric dominant target from the screened reciprocal coherent target for polarization calibration parameter calculation. Through a polarization calibration experiment, the feasibility of the new method is verified, and the calculation precision is higher. Through experimental analysis of airborne complete polarization data of different land types and loads, the obtained new method is applicable and has more advantages in precision.

Description

Technical Field

[0001] The present application relates to an airborne SAR polarization calibration method, and in particular to an airborne full-polarization SAR reciprocal coherent target polarization precision calibration method, belonging to the technical field of synthetic aperture radar calibration. Background Art

[0002] With the innovation of radar and the development of technology, SAR systems have evolved from airborne to satellite-borne, and from single polarization to full polarization. The special working mode of the fully polarized SAR system can record the polarization characteristics of the echo signal of the observed target, and this characteristic is highly related to its physical characteristics such as shape, structure, texture, and material. Therefore, the use of fully polarized data can achieve quantitative inversion of the physical characteristics of the target, making it more potential in applications such as target detection, identification, and target anti-interference. This makes the fully polarized SAR system, which already has the advantages of all-day, all-weather, and active imaging, a key means of earth observation developed by various countries.

[0003] Polarimetric SAR systems are affected by factors such as channel crosstalk, unbalanced phase amplitude, system noise, and complex ionospheric transmission environment. The echo scattering characteristics obtained by the system deviate from the actual scattering characteristics of the target. To use polarimetric information for quantitative inversion of the physical characteristics of the target, the data must be polarimetrically calibrated. The accuracy of polarimetric calibration determines the inversion accuracy of the physical characteristics of the target, so how to improve the quality of polarimetric calibration is one of the key issues in the breadth and depth of polarimetric data application.

[0004] The key to polarization calibration is to obtain the polarization distortion matrix. According to the differences in the selected calibration bodies and prior knowledge, the existing technology uses a variety of calibration bodies such as metal balls, dihedral angles, trihedral angles, and active calibrators to propose a variety of point target calibration algorithms. However, due to the limitations of the calibrator, the calibration result is only valid for a limited range of the calibration area. In order to obtain the calibration result of the entire image, a large number of calibrators need to be deployed. In addition, since the accuracy of the preset calibrator product cannot be strictly controlled, and the background noise of the SAR image is complex, the calibration result has a certain error, and the calibration accuracy is difficult to guarantee.

[0005] The problems that need to be solved in the prior art airborne fully polarimetric SAR polarization calibration and the key technical difficulties of this application include:

[0006] (1) Polarimetric SAR systems are affected by factors such as channel crosstalk, unbalanced phase amplitude, system noise, and complex ionospheric transmission environment. The acquired echo scattering characteristics have certain deviations from the actual scattering characteristics of the target. To use polarimetric information for quantitative inversion of the target's physical characteristics, the data must be polarimetrically calibrated. The accuracy of polarimetric calibration determines the inversion accuracy of the target's physical characteristics. Therefore, how to improve the quality of polarimetric calibration is one of the key issues in the breadth and depth of polarimetric data application. The quality of polarimetric calibration in existing technologies is poor, and there is a lack of analysis of the principle and method of polarimetric calibration, and a lack of analysis of the relevant factors that affect the accuracy of polarimetric calibration. It is impossible to grasp the ways and methods to improve the quality of polarimetric calibration from a macro perspective. The quality of polarimetric calibration of airborne fully polarimetric SAR targets is poor and inefficient.

[0007] (2) The prior art uses a variety of calibration bodies such as metal balls, dihedral angles, trihedral angles, and active calibrators to propose a variety of point target calibration algorithms. However, due to the limitation of the calibrator, the calibration result is only valid for a limited range of the calibration area. In order to obtain the calibration result of the entire image, a large number of calibrators need to be deployed. In addition, since the accuracy of the preset calibrator product cannot be strictly controlled, and the background noise of the SAR image is complex, the calibration result has a certain error, and the calibration accuracy is difficult to guarantee. The prior art is completely based on the calibration algorithm of scattered targets. The polarization characteristics of various ground objects are not deeply understood. It is not easy to select ground objects that can be used for polarization calibration. The symmetry processing of the polarization SAR system is too rough. In addition, due to the complexity of the scattered target characteristics, it is impossible to obtain a good calibration accuracy. There is no research on the helicity and orientation angle of the polarization data, and it cannot be guaranteed that the polarization calibration process does not destroy the polarization information of the polarization data. The classic polarization calibration methods based on distributed targets in the existing technology, such as the Quegan algorithm, the Ainsworth algorithm, and their related improved algorithms, cannot get rid of the dependence on the assumption of scattering ease or symmetry of ground objects. If the calibration reference samples involved in the solution cannot meet the assumptions, errors will inevitably be introduced in the solution process, thereby affecting the calibration accuracy.

[0008] (3) The existing technology lacks research on the impact of factors such as channel crosstalk, imbalance phase amplitude, system noise, and multi-view processing on the quality of airborne polarization calibration, and lacks methods for simulating the impact of the above factors in the data and using simulated data for polarization calibration. The existing technology not only lacks excellent algorithm support for improving the quality of polarization calibration, but also lacks methods for improving the screening accuracy of calibration reference samples that meet the algorithm requirements. It is impossible to establish a method for screening reciprocity and reflection symmetry targets based on polarization coherent target decomposition, and it is impossible to determine the rationality of the reciprocity and coherence characteristics of uncalibrated data. It is impossible to use the polarization coherent target decomposition model to extract symmetry-dominated targets for polarization calibration parameter solution. The feasibility of the method is poor, the solution accuracy is low, and the applicability of the method is narrow. Summary of the invention

[0009] This application analyzes the relevant factors that affect the accuracy of polarization calibration, grasps the ways and methods to improve the quality of polarization calibration from a macro perspective, and improves the quality of polarization calibration so as to expand the breadth and depth of polarization data applications. The influence of factors such as channel crosstalk, imbalance phase amplitude, system noise, and multi-view processing on the quality of airborne polarization calibration is studied, and it is found that the accuracy of polarization calibration parameter calculation decreases with the increase of channel crosstalk; the calculation stability improves with the increase of signal-to-noise ratio and the increase of the number of views. Based on the principle of polarization coherent target decomposition, a new method for screening reciprocal and reflection symmetric targets is proposed, and the rationality of using reciprocal angle and coherence factor to determine the reciprocity and coherence characteristics of uncalibrated data is demonstrated. Then, the polarization coherent target decomposition model is used to extract the symmetry-dominated targets from the screened reciprocal coherent targets for polarization calibration parameter calculation. The feasibility of the new method was verified through polarization calibration experiments on real data; by comparing the calibration results of the reference samples screened by the conventional correlation coefficient method, it was found that the new method has higher solution accuracy; through experimental analysis of airborne full polarization data of different terrain types and different loads, it was found that the new method is not only applicable, but also has better accuracy.

[0010] In order to achieve the above technical effects, the technical solutions adopted in this application are as follows:

[0011] The precise polarization calibration method for reciprocal coherent targets of airborne fully polarimetric SAR analyzes the impact of channel crosstalk, unbalanced phase amplitude, system noise, and multi-look processing factors on the quality of airborne polarization calibration, and uses simulation data for polarization calibration; based on polarization coherent target decomposition, a new method for screening reciprocal and reflection symmetric targets is established, and the polarization coherent target decomposition model is used to extract symmetry-dominated targets from the screened reciprocal coherent targets for polarization calibration parameter calculation;

[0012] 1) Scattered target polarization calibration accuracy evaluation model: Establish an evaluation model for the effects of channel crosstalk, unbalanced phase amplitude, and system noise on polarization calibration quality, which affect polarization data quality. Calculate that polarization calibration accuracy improves with the reduction of channel crosstalk, and the stability and accuracy of parameter solution improve with the improvement of signal-to-noise ratio. Multi-view processing reduces the impact of data noise on calibration quality, and parameter solution accuracy improves with the increase of the number of views. Establish a reference model for improving polarization calibration quality, improving polarization system design, and clarifying polarization data quality requirements.

[0013] 2) Optimization of calibration reference samples based on polarization coherent target decomposition: Based on polarization coherent target decomposition, a method for screening reciprocity and reflection symmetry calibration reference samples is established, and a target scattering feature identification method based on coherent target decomposition is adopted, including reciprocity constraint and identification, coherence constraint and identification, and symmetry constraint and identification. By calculating the factors that cause the weakening of target reciprocity, a method for determining reciprocity by using the reciprocity angle is adopted, and based on the polarization distortion model, a method for determining reciprocity by using the reciprocity angle for uncalibrated data is established; then, under the premise of satisfying reciprocity, the coherence factor is calculated and the critical value is set to further screen out the strong coherence targets; finally, based on the polarization coherence decomposition model, the selected reciprocity coherence targets are decomposed, and the targets with dominant reflection symmetry are selected as calibration reference samples for polarization calibration parameter solution. Through multi-layer screening of target features, the deviation influence of uncalibrated data caused by polarization distortion is reduced, and the accuracy of calibration reference sample screening is improved.

[0014] Preferably, the influence model of crosstalk and imbalance between channels on polarization calibration accuracy: ideal data is simulated by reflection symmetry model, and the influence of crosstalk and imbalance between channels on calibration accuracy is calculated. The polarization covariance matrix is ​​as shown in Formula 1. A trihedral corner reflector model is added to the image for accuracy assessment. The signal strength is 40db, and the signal form is as shown in Formula 2. The data meets the reciprocity and reflection symmetry assumptions.

[0015]

[0016] The effects of crosstalk and cross-channel imbalance on the calibration solution are verified respectively. Only one interference variable is added to the data, and the crosstalk and cross-channel imbalance are added respectively. Then, polarization calibration is performed on the two types of generated simulation data.

[0017] Preferably, the influence model of system noise on calibration accuracy is: using Gaussian white noise simulation with an intensity of -20db, setting the data to different levels of signal-to-noise ratio, setting the ground object signal to different intensity levels, setting the interval to -20 to 10db, and obtaining simulation data with a signal-to-noise ratio range of 0 to 30db in the form of the ground signal as shown in Formula 1; arranging a trihedral corner reflector in the image for accuracy assessment, the signal intensity is 40db, the signal form is as shown in Formula 2, adding a crosstalk signal less than -12db, adding crosstalk amounts of -15db, -20db, and -30db to the generated ideal data respectively, generating three simulation data, and then performing polarization calibration simulation.

[0018] Preferably, a system noise impact model is established:

[0019] Step 1: Generate ideal data that meets symmetry based on the ideal model of the ground object and trihedral angle, with a size of 500*1600;

[0020] Step 2: Set an energy level for every 100 pixels of ideal data along the distance, with a total of 16 energy levels. The difference between each energy level is 2db, and the energy level range is -20 to 10db. Set a trihedral corner reflector in each energy level, and its signal strength is 40db.

[0021] Step 3: Add -20db Gaussian white noise to the ideal graded data with ground objects and trihedral corner reflector signals, and obtain 16 energy levels with a signal-to-noise ratio of 0 to 30db, with a difference of 2db between energy levels;

[0022] Step 4: Add -15db, -20db, and -30db crosstalk errors to the data to generate three data with different crosstalk amounts, where the amplitudes of the four crosstalk parameters are equal, i.e., v=w=z=u, the crosstalk phases are all set to 0 degrees, the channel imbalance is set to the ideal value, the amplitude is 1, and the phase is 0 degrees;

[0023] Step 5: Solve the calibration parameters for the three simulation data respectively, and calibrate the data using the calculated calibration parameters;

[0024] Step 6: Analyze the relationship between parameter solution accuracy and signal-to-noise ratio. By recalibrating the calibrated data, the residual amount of each parameter is obtained. The crosstalk calibration accuracy is evaluated by the average value of the four crosstalk parameters, and the imbalance amplitude accuracy is evaluated by the difference between the solution result and the preset value 1.

[0025] Preferably, the impact model of multi-view processing on calibration accuracy is:

[0026] Step 1: Generate ideal forest simulation data based on the Arii vegetation model, Bragg scattering model, and dihedral scattering model. The size is 2000*2000 (row*column) data. Set the forest data scattering intensity to -10db. Set a trihedral reflector in the data, and its signal strength is 40db.

[0027] Step 2: Add -22db Gaussian white noise to the data, and set the data signal-to-noise ratio to 12db;

[0028] Step 3: Add the above data to the crosstalk amount of -15db, -20db, and -30db to generate three simulation data with different crosstalk, where the amplitudes of the four crosstalk parameters are equal, that is, v=w=z=u, the crosstalk phases are all set to 0 degrees, and the channel imbalance is set to the ideal value, the amplitude is 1, and the phase is 0 degrees;

[0029] Step 4: Perform multi-view processing on the above three data respectively, and set the number of views to 2, 5, 10, 20, 50, and 100 respectively to obtain the multi-view processed data;

[0030] Step 5: Calculate the calibration parameters for the above three sets of multi-view data, and calibrate the data using the calculated parameters;

[0031] Step 6: Analyze the relationship between parameter solution accuracy and multi-view processing. By recalibrating the calibrated data, the residual amount of each parameter is obtained. The crosstalk calibration accuracy is evaluated by the average value of the four crosstalk parameters, and the imbalance amplitude accuracy is evaluated by the difference between the solution result and the preset value 1.

[0032] Preferably, the crosstalk parameter estimation accuracy analysis: when the crosstalk amount is -20dB and -30dB, when the number of views is 5, the calibration accuracy of the crosstalk parameter and the unbalanced phase amplitude tends to be stable; when the crosstalk amount is -15dB, when the number of views is 5, the crosstalk and unbalanced phase solution accuracy tends to be stable, and when the number of views is greater than 10, the change of the unbalanced amplitude accuracy tends to be stable;

[0033] Suppose the polarization response characteristics of the trihedral reflector are analyzed: with the increase of the number of views, the calibration effect gradually improves. When the crosstalk is -20dB and -30dB, when the number of views is greater than 5, the polarization response of the trihedral is calibrated to the best. When the crosstalk is -15dB, when the number of views is greater than 50, the polarization response of the trihedral is calibrated to the best.

[0034] Preferably, the calibration reference sample based on polarization coherent target decomposition preferably: represents any scattering matrix as a sum of mutually independent basic target scattering matrices:

[0035]

[0036] Where S is the scattering matrix of a coherent target, S k represents the scattering matrix of a basic target, αk is the weight of the basic target scattering matrix S in the matrix S;

[0037] When the incident wave and the scattered wave are completely polarized waves, the scattering matrix S can express the scattering characteristics of a specific target and be associated with its relevant physical characteristics. Polarization coherent target decomposition is used to accurately extract calibration reference samples that satisfy reciprocity and reflection symmetry from uncalibrated data. Based on the reciprocity of the calibration data, target coherence, and the symmetry characteristics of the coherent decomposition model, further modeling is performed.

[0038] Preferably, reciprocity constraint and identification: using the reciprocity angle as the reciprocity determination factor, by calculating the target scattering matrix in its reciprocity subspace W rec Projection, find the original matrix and the reciprocal subspace W rec The angle θ between rec , the solution of the reciprocal angle is:

[0039]

[0040] where θ rec is the reciprocity angle, P rec W rec The projection operator in space, is the normalized scattering matrix, represents the reciprocal projection of the scattering matrix, ||█|| represents the two-norm symbol, and the vector normalization is as follows:

[0041]

[0042] Where (■,■) represents the inner product symbol; the scattering matrix unit vector is obtained by normalizing the vector

[0043]

[0044] Express power in W:

[0045] W=|S HH | 2 +|S HV | 2 +|S VH | 2 +|S VV | 2 Formula 8

[0046] Among them, S xy are the four elements of the scattering matrix S, then the reciprocal projection norm is simplified to:

[0047]

[0048] Reciprocity angle θ rec Characterizes the degree to which the scattering matrix of the target object approaches reciprocity. rec = 0, indicating that the target object scattering matrix is ​​completely located in W rec subspace, the two matrices belong to the same class, and the scattering matrix is ​​strictly reciprocal; when θ rec =π / 2, indicating that the scattering matrix of the target object is orthogonal to the reciprocity scattering matrix, and is completely non-reciprocal. π / 4 is used as the critical value for reciprocity determination. Target objects with a value less than this critical value are considered to satisfy reciprocity and are reciprocity scatterers.

[0049] When the acquired observation data has deviations, the reciprocity angle needs to be derived in combination with the polarization calibration model. After ignoring the image additive noise, the observed data and the ideal scattering matrix after calibration have the following relationship:

[0050]

[0051] Among them, O is the observation matrix, O xyare the four elements of the observation matrix, ignoring the influence of crosstalk, simplifying the above relationship group and expressing it in matrix form as follows:

[0052]

[0053] This gives the power W of the observation matrix O , the formula is as follows:

[0054] W O =|O HH | 2 +|O HV | 2 +|O VH | 2 +|O VV | 2 =|kαS HH | 2 +|α -1 S HV | 2 +|αS VH | 2 +|k -1 α -1 S VV | 2 Formula 12

[0055] The reciprocal projection norm of the observation matrix is ​​obtained as follows:

[0056]

[0057] The reciprocity angle of the observation matrix is ​​expressed as:

[0058]

[0059] When selecting calibration reference samples based on the reciprocity characteristics, it is only necessary to select samples with the smallest reciprocity angle possible based on the calculated statistical distribution law of the angle.

[0060] Preferably, coherence constraint and identification: the coherence of the target is used to screen the samples, and the influence of the scattering characteristic error of the incoherent target on the polarization decomposition of the coherent target is reduced. The target coherence is characterized by the factor P. The scattering matrix of the scattered target is the result of the mutual influence of multiple adjacent scattering units. P is statistically calculated by the ensemble average method, and the expression is as follows:

[0061]

[0062] Where <█> represents the spatial set average, α is the trihedral scattering component S in the false color synthesis decomposition a ε is the maximum symmetric partial coefficient of the dihedral scattering component S in the false color synthesis decomposition;

[0063] When the coherence factor P is close to 1, the target has strong coherence and is judged as a coherence-dominated target; when the coherence factor P is close to 0, the target has weak coherence and is judged as an incoherent target; the coherence factor P value is calculated using a sliding window, and the window size is determined in combination with the pixel size. The point target is set to 2 to 5 meters, and the window is set between 2m×2m and 5m×5m;

[0064] The calculation of the coherence factor P is based on the false color synthesis decomposition method. α and ε are obtained by solving the problem, and the coherence factor P is used to determine the relevance of the target.

[0065] Preferably, symmetry constraints and identification: targets that satisfy reflection symmetry must satisfy reciprocity, which is the premise of reflection symmetry. Polarization coherent target decomposition model is used to determine target symmetry, and diagonalization method is used to perform characteristic decomposition on target scattering matrix S:

[0066] S=R(ψ)T(τ m )·S d ·T(τ m )R(-ψ) Formula 16

[0067] In the formula, τ m represents the helicity of the target, T(τ m ) and S d They are respectively expressed as follows:

[0068]

[0069] By separating the rotation matrix Obtain the rotation invariant parameters, which are the phase Helix angle τ m , scattering type parameter α s , other parameters include maximum amplitude m, polarization azimuth parameters As follows:

[0070]

[0071] in, SV represents the scattering vector;

[0072]

[0073] Substituting the eigenvalues ​​λ1, λ2, λ3 and the corresponding eigenvectors u1, u2, u3 of the parameterized coherence matrix T3 into Equation 19, we can obtain the decomposition parameters of the corresponding eigenvalues. The parameter component with the largest eigenvalue is the principal component, reflecting the dominant characteristics of the scatterer. The rotationally invariant parameters reflect the scattering law and geometric characteristics of the ground object target. m The value is the helicity angle, when τ m= 0, indicating that the scatterer is completely symmetrical, including dihedral angles and trihedral angles. m =π / 4, indicating that the scatterer is completely asymmetric, including typical left and right helices. m The spiral components that do not conform to the reflection symmetry are eliminated, most of the asymmetric targets are accurately eliminated, and the error influence of non-reflection symmetry targets on parameter solution is reduced.

[0074] Compared with the prior art, the innovations and advantages of this application are:

[0075] (1) This application analyzes the relevant factors that affect the accuracy of polarization calibration, grasps the ways and methods to improve the quality of polarization calibration from a macro perspective, and improves the quality of polarization calibration so as to expand the breadth and depth of polarization data applications. The influence of factors such as channel crosstalk, imbalance phase amplitude, system noise, and multi-view processing on the quality of airborne polarization calibration is studied, and it is found that the accuracy of polarization calibration parameter calculation decreases with the increase of channel crosstalk; the calculation stability improves with the increase of signal-to-noise ratio and the increase of the number of views. Improving the quality of polarization calibration requires not only the support of excellent algorithms, but also the improvement of the screening accuracy of calibration reference samples that meet the requirements of the algorithm. To this end, this application proposes a new method for screening reciprocity and reflection symmetry targets based on the principle of polarization coherent target decomposition, and demonstrates the rationality of using reciprocity angle and coherence factor to determine the reciprocity and coherence characteristics of uncalibrated data. Then, the polarization coherent target decomposition model is used to extract symmetry-dominated targets from the screened reciprocity coherence targets for polarization calibration parameter calculation. The feasibility of the new method was verified through polarization calibration experiments on real data; by comparing the calibration results of the reference samples screened by the conventional correlation coefficient method, it was found that the new method has higher solution accuracy; through experimental analysis of airborne full polarization data of different terrain types and different loads, it was found that the new method is not only applicable, but also has better accuracy.

[0076] (2) This application first analyzes the system indicators and multi-look processing that affect the quality of full-polarization SAR data, and then establishes an impact model for the three factors of channel crosstalk, channel imbalance, and system noise and multi-look processing, and conducts calibration experiments. The accuracy of polarization calibration parameter calculation decreases with the increase of crosstalk; the stability and accuracy of parameter calculation improve with the enhancement of signal-to-noise ratio and the increase of multi-look average processing number; when establishing a polarization SAR system, the quality of polarization data is improved by reducing system noise and increasing crosstalk between channels. In addition, when polarization calibration is performed, appropriate multi-look average processing of polarization data helps to reduce the impact of data noise and improve the quality of polarization calibration. It is helpful to use full polarization data to achieve quantitative inversion of target physical characteristics, making it more potential in applications such as target detection, recognition, and target anti-interference, so that the full polarization SAR system, which already has the advantages of all-day, all-weather, and active imaging, will become a key means of earth observation.

[0077] (3) Aiming at improving the quality of polarization calibration reference sample screening, this application innovatively introduces the polarization coherent target decomposition theory to improve the sample screening method. First, the rationality of using the reciprocity angle to perform reciprocity screening on uncalibrated data is explained, and based on this, the coherence determination method is introduced. By optimizing the reciprocity and coherence of the uncalibrated data, the error influence of the incoherent component on the polarization coherent target decomposition is reduced. On this basis, the reflection symmetry of the sample is further optimized with the help of the polarization coherent target decomposition method, and the samples that finally meet the reciprocity and reflection symmetry are used for polarization calibration parameter solution. The feasibility of the improved method is verified through real data experiments; the calibration results of the samples screened by the conventional correlation coefficient method are compared through calibration parameter residuals, false color composite images, polarization characteristic response images, and polarization covariance matrix C 23 Comparative analysis such as the imaginary part information diagram verified the advanced nature of the improved method; calibration experiments on different land types and different load data verified the reliability and universality of the improved method. BRIEF DESCRIPTION OF THE DRAWINGS

[0078] Figure 1 It is a two-part flow chart based on the polarization decomposition method and the correlation coefficient feature screening method in the experiment.

[0079] Figure 2 It is the reciprocity angle distribution diagram and its distribution histogram before data calibration.

[0080] Figure 3 It is the reciprocity angle distribution diagram and its distribution histogram after data calibration.

[0081] Figure 4 It is the helicity distribution diagram and its distribution histogram before data calibration.

[0082] Figure 5 It is a statistical diagram of the dominance of symmetry within different critical values.

[0083] Figure 6 It is the correlation coefficient graph and its distribution histogram before data calibration.

[0084] Figure 7 C before and after calibration 23 Infographic of imaginary part of term.

[0085] Figure 8 This is a schematic diagram of Wishart_H_A_alpha classification comparison. DETAILED DESCRIPTION

[0086] The following further describes the technical solution of the airborne full-polarization SAR reciprocal coherent target polarization precise calibration method provided by the present application in conjunction with the accompanying drawings, so that those skilled in the art can better understand the present application and implement it.

[0087] With the innovation of radar theory and the development of technology, SAR systems have evolved from airborne to satellite-borne, and from single polarization to full polarization. The special working mode of the fully polarized SAR system can record the polarization characteristics of the target echo signal, and this characteristic is highly related to its shape, structure, texture, material and other physical characteristics. Therefore, the use of fully polarized data can achieve quantitative inversion of the target's physical characteristics, making it more potential in applications such as target detection, identification and target anti-interference. This makes the fully polarized SAR system, which already has the advantages of all-day, all-weather and active imaging, a key means of earth observation developed by various countries.

[0088] Polarimetric SAR systems are affected by factors such as channel crosstalk, unbalanced phase amplitude, system noise, and complex ionospheric transmission environment. The echo scattering characteristics obtained by the system deviate from the actual scattering characteristics of the target. To use polarimetric information for quantitative inversion of the physical characteristics of the target, the data must be polarimetrically calibrated. The accuracy of polarimetric calibration determines the inversion accuracy of the physical characteristics of the target, so how to improve the quality of polarimetric calibration is one of the key issues in the breadth and depth of polarimetric data application.

[0089] This application first briefly summarizes the principle and method of polarization calibration, and comprehensively sorts out and summarizes the existing polarization calibration schemes, then analyzes the relevant factors affecting the accuracy of polarization calibration, and grasps the ways and methods to improve the quality of polarization calibration from a macro perspective;

[0090] 1. Scattered target polarization calibration accuracy assessment model: Establish an assessment model for the effects of channel crosstalk, unbalanced phase amplitude, and system noise on polarization calibration quality that affect polarization data quality. Calculate that polarization calibration accuracy improves with the reduction of channel crosstalk, and the stability and accuracy of parameter solution improve with the improvement of signal-to-noise ratio. Multi-view processing reduces the impact of data noise on calibration quality, and parameter solution accuracy improves with the increase of the number of views. Establish a reference model to improve polarization calibration quality, improve polarization system design, and clarify polarization data quality requirements.

[0091] 2. Optimization of calibration reference samples based on polarization coherent target decomposition: Based on polarization coherent target decomposition, a method for screening reciprocity and reflection symmetry calibration reference samples is established. The target scattering feature identification method based on coherent target decomposition is adopted, including reciprocity constraint and identification, coherence constraint and identification, symmetry constraint and identification. By calculating the factors that weaken the target reciprocity, a method for determining reciprocity by using the reciprocity angle is used, and the reciprocity is determined by using the reciprocity angle for uncalibrated data according to the polarization distortion model; then, under the premise of satisfying reciprocity, the coherence factor is calculated and the critical value is set to further screen the strong coherence target; finally, based on the polarization coherence decomposition model, the selected reciprocity coherence target is decomposed, and the target with dominant reflection symmetry is selected as the calibration reference sample for polarization calibration parameter solution. Through multi-layer screening of target features, the deviation effect of uncalibrated data caused by polarization distortion is reduced, and the accuracy of calibration reference sample screening is improved. The experiment uses the Quegan algorithm to solve the calibration parameters, and the polarization coherence decomposition uses the SDH, Cameron, and TSVM models respectively. The experiment compares the calibration results with the conventional correlation coefficient method, and conducts an in-depth analysis of the calibration effect from five aspects: calibration accuracy, polarization response diagram, residual information of the imaginary part of the cross-polarization term of polarization covariance, synthetic false color synthesis diagram, and Wishart-HA-alpha classification results. The calibration effects of the three polarization decomposition models for screening calibration reference samples for this experimental data are better than the correlation coefficient method, among which the Cameron model has the best screening effect. The experiment proves that this method is effective. Finally, the reliability and universality of this method are verified through data experiments of different land types and different loads.

[0092] 1. Scattered Target Polarization Calibration Accuracy Evaluation Model

[0093] The accuracy of polarization calibration is not only related to the selected calibration algorithm, but also to the quality of the data and the multi-view processing of the data. Factors affecting data quality include crosstalk between channels, channel imbalance, system noise, and multi-view processing.

[0094] (I) Model of the impact of inter-channel crosstalk and imbalance on polarization calibration accuracy

[0095] The ideal data is simulated using a reflection symmetry model to calculate the impact of crosstalk and imbalance between channels on calibration accuracy. The polarization covariance matrix is ​​shown in Formula 1. A trihedral corner reflector model is added to the image for accuracy assessment. The signal strength is 40db, and the signal form is shown in Formula 2. The data meets the reciprocity and reflection symmetry assumptions.

[0096]

[0097] The effects of crosstalk and cross-channel imbalance on the calibration solution are verified respectively. Only one interference variable is added to the data, and the crosstalk and cross-channel imbalance are added respectively. Then, polarization calibration is performed on the two types of generated simulation data.

[0098] The crosstalk calculation is not affected by the added cross-channel imbalance, nor is the cross-channel imbalance calculation affected by its own value. The calculation accuracy of the calibration parameters is affected by the amount of channel crosstalk in the data. The larger the crosstalk, the lower the calculation accuracy.

[0099] 2. Model of the impact of system noise on calibration accuracy

[0100] A Gaussian white noise simulation with an intensity of -20db is used, and the data is set to different levels of signal-to-noise ratio. The ground object signal is set to different intensity levels, and the interval is set to -20 to 10db. The ground signal form is as shown in Formula 1 to obtain simulation data with a signal-to-noise ratio range of 0 to 30db; a trihedral corner reflector is arranged in the image for accuracy assessment, and the signal intensity is 40db. The signal form is as shown in Formula 2, and a crosstalk signal less than -12db is added. The crosstalk amounts of -15db, -20db, and -30db are added to the generated ideal data respectively to generate three simulation data, and then polarization calibration simulation is performed.

[0101] 1. Establish a system noise impact model

[0102] Step 1: Generate ideal data that meets symmetry based on the ideal model of the ground object and trihedral angle, with a size of 500*1600 (rows*columns);

[0103] Step 2: Set an energy level for every 100 pixels of ideal data along the distance direction (row direction), with a total of 16 energy levels. The difference between each energy level is 2db, and the energy level range is -20 to 10db. Set a trihedral corner reflector in each energy level, and its signal strength is 40db.

[0104] Step 3: Add -20db Gaussian white noise to the ideal graded data with ground objects and trihedral corner reflector signals, and obtain 16 energy levels with a signal-to-noise ratio of 0 to 30db, with a difference of 2db between energy levels;

[0105] Step 4: Add -15db, -20db, and -30db crosstalk errors to the data to generate three data with different crosstalk amounts, where the amplitudes of the four crosstalk parameters are equal, i.e., v=w=z=u, the crosstalk phases are all set to 0 degrees, the channel imbalance is set to the ideal value, the amplitude is 1, and the phase is 0 degrees;

[0106] Step 5: Solve the calibration parameters for the three simulation data respectively, and calibrate the data using the calculated calibration parameters;

[0107] Step 6: Analyze the relationship between parameter solution accuracy and signal-to-noise ratio. By recalibrating the calibrated data, the residual amount of each parameter is obtained. The crosstalk calibration accuracy is evaluated by the average value of the four crosstalk parameters, and the imbalance amplitude accuracy is evaluated by the difference between the solution result and the preset value 1.

[0108] 2. Experimental analysis:

[0109] (1) Crosstalk valuation analysis

[0110] The data with preset crosstalk amounts of -15dB, -20dB, and -30dB were calibrated respectively to obtain the crosstalk parameter estimates of the data under different signal-to-noise ratios. With the increase of the data signal-to-noise ratio, the solution of the crosstalk parameters gradually tends to be stable. Under low signal-to-noise ratio conditions, the smaller the crosstalk in the data, the greater the impact of the signal-to-noise ratio on the solution stability. Under high signal-to-noise ratio conditions, the solution accuracy is related to the amount of crosstalk contained in the data. The smaller the crosstalk amount, the higher the parameter solution accuracy.

[0111] (2) Analysis of imbalance phase amplitude estimation

[0112] As the data signal-to-noise ratio increases, the channel imbalance solution gradually becomes stable; under the same signal-to-noise ratio conditions, the size of the solution error is related to the crosstalk. The greater the crosstalk, the greater the solution error. When the signal-to-noise ratio is greater than 12dB, the channel imbalance solution for each crosstalk condition tends to be stable, the accuracy tends to be stable, the residual is less than 0.023, the accuracy is better than -32dB; the phase angle is better than 0.03 degrees.

[0113] 3. Model of the impact of multi-view processing on calibration accuracy

[0114] Step 1: Generate ideal forest simulation data based on the Arii vegetation model, Bragg scattering model, and dihedral scattering model. The size is 2000*2000 (row*column) data. Set the forest data scattering intensity to -10db. Set a trihedral reflector in the data, and its signal strength is 40db.

[0115] Step 2: Add -22db Gaussian white noise to the data, and set the data signal-to-noise ratio to 12db;

[0116] Step 3: Add the above data to the crosstalk amount of -15db, -20db, and -30db to generate three simulation data with different crosstalk, where the amplitudes of the four crosstalk parameters are equal, that is, v=w=z=u, the crosstalk phases are all set to 0 degrees, and the channel imbalance is set to the ideal value, the amplitude is 1, and the phase is 0 degrees;

[0117] Step 4: Perform multi-view processing on the above three data respectively, and set the number of views to 2, 5, 10, 20, 50, and 100 respectively to obtain the multi-view processed data;

[0118] Step 5: Calculate the calibration parameters for the above three sets of multi-view data, and calibrate the data using the calculated parameters;

[0119] Step 6: Analyze the relationship between parameter solution accuracy and multi-view processing. By recalibrating the calibrated data, the residual amount of each parameter is obtained. The crosstalk calibration accuracy is evaluated by the average value of the four crosstalk parameters, and the imbalance amplitude accuracy is evaluated by the difference between the solution result and the preset value 1.

[0120] 1. Analysis of crosstalk parameter estimation accuracy

[0121] When the crosstalk amount is -20dB and -30dB, when the look number is 5, the calibration accuracy of the crosstalk parameters and unbalanced phase amplitude tends to be stable; when the crosstalk amount is -15dB, when the look number is 5, the crosstalk and unbalanced phase solution accuracy tends to be stable, and when the look number is greater than 10, the change of unbalanced amplitude accuracy tends to be stable.

[0122] 2. Analysis of polarization response characteristics of preset trihedral reflector

[0123] As the number of views increases, the calibration effect gradually improves. When the crosstalk is -20dB and -30dB, when the number of views is greater than 5, the polarization response of the trihedral angle is calibrated to the best. When the crosstalk is -15dB, when the number of views is greater than 50, the polarization response of the trihedral angle is calibrated to the best.

[0124] Polarization calibration is performed using the Quegan algorithm based on scattered targets. Appropriate multi-look processing is beneficial to reducing the impact of noise. The calculation of crosstalk parameters and imbalance phase amplitude gradually stabilizes with the increase of the number of views. When the crosstalk is less than -20dB, the impact of multi-look processing on the calibration accuracy is obvious. When the number of views is greater than 5, the calibration accuracy can be stable.

[0125] 2. Calibration reference sample selection based on polarization coherent target decomposition

[0126] An arbitrary scattering matrix can be expressed as a sum of independent elementary target scattering matrices:

[0127]

[0128] Where S is the scattering matrix of a coherent target, S k represents the scattering matrix of a basic target, α k is the weight of the basic target scattering matrix S in the matrix S;

[0129] When the incident wave and the scattered wave are completely polarized waves, the scattering matrix S can express the scattering characteristics of a specific target and be associated with its relevant physical characteristics. Polarization coherent target decomposition is used to accurately extract calibration reference samples that satisfy reciprocity and reflection symmetry from uncalibrated data. Based on the reciprocity of the calibration data, target coherence, and the symmetry characteristics of the coherent decomposition model, further modeling is performed.

[0130] 1. Target scattering feature recognition based on coherent target decomposition

[0131] 1. Reciprocity Constraints and Identification

[0132] The reciprocity angle is used as the reciprocity determination factor, and the target scattering matrix is ​​calculated in its reciprocity subspace W. rec Projection, find the original matrix and the reciprocal subspace W rec The angle θ between rec , the solution of the reciprocal angle is:

[0133]

[0134] where θ rec is the reciprocity angle, P rec W rec The projection operator in space, is the normalized scattering matrix, represents the reciprocal projection of the scattering matrix, ||█|| represents the two-norm symbol, and the vector normalization is as follows:

[0135]

[0136] Where (■,■) represents the inner product symbol; the scattering matrix unit vector is obtained by normalizing the vector

[0137]

[0138] Express power in W:

[0139] W=|S HH | 2 +|S HV | 2 +|S VH | 2 +|S VV | 2 Formula 8

[0140] Among them, S xy are the four elements of the scattering matrix S, then the reciprocal projection norm is simplified to:

[0141]

[0142] Reciprocity angle θ rec Characterizes the degree to which the scattering matrix of the target object approaches reciprocity. rec = 0, indicating that the target object scattering matrix is ​​completely located in W rec subspace, the two matrices belong to the same class, and the scattering matrix is ​​strictly reciprocal; when θ rec =π / 2, indicating that the scattering matrix of the target object is orthogonal to the reciprocity scattering matrix, and is completely non-reciprocal. π / 4 is used as the critical value for reciprocity determination. Target objects with a value less than this critical value are considered to satisfy reciprocity and are reciprocity scatterers.

[0143] When the acquired observation data has deviations, the reciprocity angle needs to be derived in combination with the polarization calibration model. After ignoring the image additive noise, the observed data and the ideal scattering matrix after calibration have the following relationship:

[0144]

[0145] Among them, O is the observation matrix, O xy are the four elements of the observation matrix, ignoring the influence of crosstalk, simplifying the above relationship group and expressing it in matrix form as follows:

[0146]

[0147] This gives the power W of the observation matrix O , the formula is as follows:

[0148] WO=|O HH | 2 +|O HV | 2 +|O VH | 2 +|O VV | 2=|kαS HH | 2 +|α -1 S HV | 2 +|αS VH | 2 +|k -1 α -1 S VV | 2 Formula 12

[0149] The reciprocal projection norm of the observation matrix is ​​obtained as follows:

[0150]

[0151] The reciprocity angle of the observation matrix is ​​expressed as:

[0152]

[0153] When selecting calibration reference samples based on the reciprocity characteristics, it is only necessary to select samples with the smallest reciprocity angle possible based on the calculated statistical distribution law of the angle.

[0154] 2. Coherence Constraints and Identification

[0155] The coherence of the target is used to screen the samples and reduce the influence of the incoherent target scattering characteristic error on the polarization decomposition of the coherent target. The target coherence is characterized by the factor P. The scattering matrix of the scattered target is the result of the mutual influence of multiple adjacent scattering units. P is statistically calculated using the ensemble average method, and the expression is as follows:

[0156]

[0157] Where <█> represents the spatial set average, α is the trihedral scattering component S in the false color synthesis decomposition a ε is the maximum symmetric partial coefficient of the dihedral scattering component S in the false color synthesis decomposition;

[0158] When the coherence factor P is close to 1, the target has strong coherence and is judged as a coherence-dominated target; when the coherence factor P is close to 0, the target has weak coherence and is judged as an incoherent target; the coherence factor P value is calculated using a sliding window, and the window size is determined in combination with the pixel size. The point target is set to 2 to 5 meters, and the window is set between 2m×2m and 5m×5m;

[0159] The calculation of the coherence factor P is based on the false color synthesis decomposition method. α and ε are obtained by solving the problem, and the coherence factor P is used to determine the relevance of the target.

[0160] 3. Symmetry constraints and identification

[0161] The target that satisfies reflection symmetry must satisfy reciprocity. Reciprocity is the premise of reflection symmetry. The polarization coherent target decomposition model is used to determine the target symmetry, and the diagonalization method is used to perform characteristic decomposition on the target scattering matrix S:

[0162] S=R(ψ)T(τ m )·S d ·T(τ m )R(-ψ) Formula 16

[0163] In the formula, τ m represents the helicity of the target, T(τ m ) and S d They are respectively expressed as follows:

[0164]

[0165] By separating the rotation matrix Obtain the rotation invariant parameters, which are the phase Helix angle τ m , scattering type parameter α s , other parameters include maximum amplitude m, polarization azimuth parameters As follows:

[0166]

[0167] in, SV represents the scattering vector;

[0168]

[0169] Substituting the eigenvalues ​​λ1, λ2, λ3 and the corresponding eigenvectors u1, u2, u3 of the parameterized coherence matrix T3 into Equation 19, we can obtain the decomposition parameters of the corresponding eigenvalues. The parameter component with the largest eigenvalue is the principal component, reflecting the dominant characteristics of the scatterer. The rotationally invariant parameters reflect the scattering law and geometric characteristics of the ground object target. m The value is the helicity angle, when τ m = 0, indicating that the scatterer is completely symmetrical, including dihedral angles and trihedral angles. m =π / 4, indicating that the scatterer is completely asymmetric, including typical left and right helices. m The spiral components that do not conform to the reflection symmetry are eliminated, most of the asymmetric targets are accurately eliminated, and the error influence of non-reflection symmetry targets on parameter solution is reduced.

[0170] (II) Correlation coefficient feature screening

[0171] For a scatterer that satisfies reflection symmetry, the assimilation and cross-polarization channel scattering components satisfy the following equations:

[0172] HV S VH * >=0,(S HH S VH * >=0,(S VV S VH * >=0, VV SVH * >=0 Formula 20

[0173] The correlation coefficients of the co-polarization and cross-polarization channels of different combinations of the same target are the same. The correlation coefficient of any combination can be calculated. The correlation coefficient is represented by ρ, and the formula is as follows:

[0174]

[0175] ​​In the formula, <█> represents the spatial set average, |█| represents the modulus, when the correlation coefficient ρ is 0, there is no correlation, when the correlation coefficient ρ value gradually increases, the correlation gradually increases, when ρ reaches 1, there is complete correlation, the correlation coefficient p value is calculated using a sliding window, and the scattering matrix of the scattering units in the window is estimated by the set average. The larger the window, the smaller the estimate, and the smaller the window, the larger the estimate.

[0176] 3. Experimental methods and procedures

[0177] (I) Experimental data

[0178] The experimental data used is an airborne X-band fully polarized SAR system, and the antenna adopts a dual-transmit and dual-receive working mode. The experiment uses 2 scene data, and 7 trihedral corner reflectors are deployed in the experimental area. The terrain in the experimental area is flat and the impact of terrain undulation is small. There are rich types of landforms in the area, including large buildings, villages, roads, farmlands, bare land, water bodies, etc.

[0179] (II) Experimental process

[0180] The overall process of this application experiment is as follows Figure 1 ,In order to compare the effects of different calibration reference sample ,screening methods on the calibration results, the experiment is divided into two parts: ,based on polarization decomposition method and correlation coefficient feature ,screening method. The specific steps are as follows:

[0181] 1. Calibration reference sample screening method based on polarization decomposition

[0182] (1) Calculate the reciprocal angle θ of each pixel in the uncalibrated data rec , determine the critical value of reciprocity judgment according to the distribution of the reciprocity angle histogram, and select the targets with reciprocity angle less than the critical value as the samples with dominant reciprocity. The critical value is set to 15 degrees;

[0183] (2) Based on the theory of false color synthesis polarization decomposition, a sliding window is used to calculate the spatial set average and the coherence factor P of the sample data. The coherence judgment critical value is determined according to the histogram distribution of the coherence factor. The targets with a coherence factor greater than the critical value are screened as samples with dominant coherence. A 7×7 sliding window is used to calculate the coherence factor, and the critical value is set to 0.8.

[0184] (3) Based on the polarization decomposition of coherent targets, different decomposition models are used to screen symmetry-dominated targets as calibration reference samples. The impact of symmetry judgment of different models on the calibration results is compared, and the data is polarized. On the premise of satisfying reciprocity coherence, symmetry-dominated targets are selected as calibration reference samples. In the decomposition results, targets with a helicity less than 10 degrees are selected as symmetry-dominated targets.

[0185] (4) After the above three steps, the calibration parameters of the selected reference samples are solved, and the polarization calibration parameters are used to perform polarization calibration on the data;

[0186] (5) Using the calibration algorithm as a verification algorithm, the calibration data is parameterized again and the residuals of each calibration parameter are analyzed.

[0187] 2. Calibration reference sample screening method based on conventional methods

[0188] (1) The correlation coefficient method is used to select targets within the appropriate visual value as calibration reference samples. The correlation coefficient is calculated using a 5×5 sliding window, and the critical value is set to 0.2. Targets with a value smaller than the critical value are judged as irrelevant targets.

[0189] (2) The results screened by the correlation coefficient method are used as calibration reference samples to solve the calibration parameters, and the calculated calibration parameters are used to perform polarization calibration on the data;

[0190] (3) Using the calibration algorithm as the verification algorithm, the calibration data is parameterized again and the residual amount of each calibration parameter is analyzed.

[0191] 4. Calculation and Analysis of Target Scattering Characteristics

[0192] 1. Reciprocity analysis

[0193] The calculation of the reciprocity angle refers to the pixel-by-pixel calculation. Through the reciprocity angle determination, the targets with dominant reciprocity are screened as samples for subsequent screening and processing. Figure 2 This is the reciprocity angle distribution diagram of the uncalibrated data in area 1. It can be seen from the figure that there are fewer high values ​​of the reciprocity angle, and the angle values ​​are concentrated in the low value area, but the boundary of the area is relatively obvious.

[0194] Figure 3This is a statistical diagram of samples along the distance direction within different critical values ​​of the reciprocity angle. When 45 degrees (0.785 radians) is selected as the critical value, the samples along the distance direction exceed 2800, and the proportion of pixels satisfying reciprocity on the image reaches more than 96.92%. At this time, there is no discrimination for reciprocity judgment. Combined with the experimental data that the reciprocity angle is concentrated in the low-value area, the visual value is further reduced on the premise of ensuring that there are sufficient samples in each distance direction, and a target closer to the ideal reciprocity is selected. When 15 degrees (0.262 radians) is selected as the critical value, the minimum number of samples in the 2000-2300 column of the distance direction is greater than 1400 pixels, and the proportion of pixels satisfying reciprocity on the image is 61.68%. When 10 degrees (0.174 radians) is selected as the critical value, the minimum number of samples in the 2000-2300 column of the distance direction is close to 800 pixels, and the number of samples is small, and the proportion of pixels satisfying reciprocity on the image is 40.72%. After comprehensive analysis, the experimental data of this application uses a reciprocity angle of 15 degrees (0.262) as the critical value for determining reciprocity of the data.

[0195] Comparing the distribution diagram of the reciprocity angle before and after calibration, it can be seen that the reciprocity of the data changes before and after calibration. The peak value of the reciprocity angle before calibration is about 0.2, about 2.8×106 pixels, and the peak value after calibration is about 0.05, about 7.7×106 pixels. The coherence factor before calibration is concentrated between 0-0.3, while the coherence factor after calibration is concentrated between 0-0.1. This phenomenon occurs because the calibration algorithm based on the reciprocity assumption, the established model determines that all targets participating in the solution satisfy the reciprocity angle of 0, so the calibration parameters obtained will further enhance the reciprocity of samples with a reciprocity angle less than 15 degrees. Since the reciprocity angle of the experimental data is concentrated in the low value area, although some targets greater than the critical value do not participate in the calibration parameter solution, due to the similar reciprocity, the calibration parameters also play a corrective role on these targets, so that their reciprocity angle is further reduced and the reciprocity is enhanced. Therefore, the overall reciprocity of the data after calibration is significantly improved. If 45 degrees is used as the critical value, 97.33% of the pixels on the image after calibration meet the reciprocity requirement, and more than 96.92% of the pixels on the image before calibration meet the reciprocity requirement. If it is assumed that the image after calibration is error-free, the error rate before calibration is less than 0.5%, indicating that 15 degrees is selected as the critical value, which maintains the original reciprocity characteristics of the image and the critical value is selected appropriately.

[0196] 2. Coherence Analysis

[0197] Because the coherence factor P value uses a sliding window to calculate the ensemble average of the scattering matrix, the window size will directly affect the calculation result of the coherence factor. The size of the sliding window is determined in combination with the pixel size. The point target is between 2m and 5m, which is used as a reference for setting the window size, except for the case where the individual point targets in the observation area are larger. Then, the calculated coherence factor diagram is further analyzed. The resolution of the experimental image is 0.8 meters in azimuth and 0.75 meters in distance. The experiment uses 2×2, 7×7, 10×10, and 20×20 windows to calculate the coherence factor and draws the coherence factor diagram calculated by different windows. The 2×2 window is used for calculation, and the coherence factor can be seen from the distribution histogram. The calculation is generally too high, and there is no obvious boundary between the objects on the coherence factor map, indicating that the window selection is too small; using a 7×7 window calculation, the boundary between the high-value area and the low-value area on the coherence factor map is obvious, the coherence of different objects is distinguished, and the histogram shows that the peak value of the coherence factor is around 0.85; using a 10×10 window calculation, the boundary between the high-value area and the low-value area on the coherence factor map is more obvious, the upper-middle value range is slightly dominant, and the coherence of different objects is clearly distinguished. The histogram shows that the peak value of the coherence factor is around 0.8; using a 200 window calculation, the upper-middle value range of the coherence factor is absolutely dominant, and the coherence of the targets in this area is not clearly distinguished. The histogram shows that the peak value of the coherence factor is around 0.75. Combined with the coherence factor map and distribution histogram calculated with different windows, 0.8 is selected as the critical value of the coherence factor in this experimental data.

[0198] Comparing the coherence diagrams of the data before and after calibration, it can be found that the coherence distribution of the data before and after calibration is similar, but there are also slight changes. The coherence factor values ​​before calibration are concentrated between 0.7 and 0.9, with a value of about 2.2×10 6 pixels, and after calibration, the value range is more evenly distributed, and the coherence factor is concentrated between 0.6 and 0.9, about 1.8×10 6 The contours and monotonicity of the low-value area and the high-value area before and after calibration are similar, and the coherence of the image remains relatively stable.

[0199] 3. Symmetry analysis

[0200] The decomposition results obtained three eigenvalues ​​and one principal component corresponding to the decomposition parameters, and the target helicity parameter τ corresponding to the dominant principal component was analyzed. m , a scatterer with a symmetrical structure τ m The value is close to 0, and the asymmetric scatterer τ m The value is close to 45 degrees. Figure 4 Principal component helicity τ m They are concentrated in the low-value area, of which there are about 1.2×105 pixels close to 0, accounting for 14.14% of the total samples.

[0201] Figure 5 is the helicity τm The statistical diagram of the number of points screened when the viewing value is set to 10 degrees and 5 degrees. When the critical value is set to 10 degrees, the number of points dominated by symmetry is about 7.06×10 6 , accounting for 83.37% of the total samples. There are too many samples. When the critical value is set to 5 degrees, the symmetry advantage is about 5.12×10 6 , accounting for 60.42% of the sample, more than half of the sample. For this experimental data, the helicity τ m A threshold of 5 degrees was used as the selected visual value for the symmetry target.

[0202] Helicity τ in the principal component after data calibration m There are about 1.25×10 pixels close to 0 6 , accounting for 14.76% of the total number of samples, which is equivalent to 14.14% before calibration. After calibration, the helicity τ m There are approximately 4.78×10 pixels with values ​​less than 5 degrees. 6 , accounting for 56.52% of the total pixels, compared with 60.42% before calibration. Overall, the target symmetry does not change much before and after calibration, and the critical value is reasonably selected.

[0203] 4. Correlation Analysis

[0204] HH and HV are used to calculate the correlation coefficient. Three windows of different sizes, 2×2, 5×5, and 10×10, are used to calculate the correlation coefficient of the sample data. When a 2×2 window is used, the correlation coefficient value calculated is too large because the window is too small, and there are very few irrelevant samples, which cannot provide sufficient samples for calibration calculation. When a 10×10 window is used, Figure 6 As shown in the figure, because the window is too large, the calculated correlation values ​​are generally small and the correlation discrimination is not high.

[0205] Using a 5×5 window for calculation, the correlation coefficient follows a normal distribution, and the texture of the image can be clearly distinguished from the correlation coefficient graph. This application uses 0.2 as the critical value, and selects pixels with a correlation coefficient less than 0.2 as irrelevant reference samples to participate in the solution. The number of points participating in the calculation along the distance direction decreases in the 600 to 800 rows in the distance direction, and the minimum number of samples is about 200.

[0206] 5. Polarization Calibration Experiment Analysis

[0207] The correlation coefficient method and coherent decomposition method are used to screen the calibration reference samples, the Quegan algorithm is used to solve the calibration parameters, and finally the Ainsworth algorithm and Quegan algorithm are used to solve the parameter residuals of the calibration results. Among them, the coherent decomposition uses three methods, SDH, Comeron, and TSVM, for symmetry screening. The solution results and residuals of each calibration parameter are obtained. By analyzing the calibration parameter solution results, it can be found that the crosstalk parameters u, v, w, and z are very small. The correlation coefficient method is generally -31dB; the coherent decomposition method is generally -30dB.

[0208] The results of residual calculation using the Ainsworth algorithm for the calibration data in the above table show that: the crosstalk residual using the correlation coefficient method is -61 to -62dB, and the channel imbalance residual is -41 to -49dB; the crosstalk residual using the coherent decomposition method is -65 to -68dB, and the channel imbalance residual is -56 to -87dB. Among them, the Comeron method has the best effect, with a crosstalk residual of -66 to -68dB and a channel imbalance residual of -68 to -87dB.

[0209] The imbalance calibration effect of the TSVM decomposition screening method is poor, ranging from -56 to -59 dB, indicating that there is a large error influence when only using helicity for reflection symmetry screening and the screening ability is limited.

[0210] The experimental results show that the calibration parameter residual calculated by the calibration reference sample screened by the coherent decomposition method is the smallest, among which the Comeron method has the best effect and the best calibration result. The correlation coefficient method is the worst among the four methods. From the residual solution results of the Quegan algorithm, similar conclusions can be drawn. However, from the residual magnitude, the residuals of the calibration results of the four methods are already very small.

[0211] 6. Polarization covariance matrix C 23 Imaginary residual analysis

[0212] According to the expression of the polarization covariance matrix C of the scattering matrix, for a scatterer that satisfies reciprocity, its scattering matrix satisfies: S 12 =S 21 , for a scatterer that satisfies reciprocity, its polarization covariance matrix C is 23 The imaginary part is zero, so the calibration effect can be judged by the amount of information in the C23 imaginary image.

[0213] Figure 7 Displayed is C 23 The information diagram of the imaginary part of the term, the brightness of the image represents the strength of the information. Figure (4-25a) shows the data before calibration. 23There is a lot of information in the imaginary part of the term. Figure (4-25b) shows that after the data was calibrated by the correlation coefficient method to select the calibration reference sample, a lot of information has been corrected, but there is still residual information with irregular distribution. Figure (4-25c) shows that after the data was calibrated by the polarization decomposition method to select the calibration reference sample, a lot of information has also been corrected. The distribution of the remaining information is related to the polarization characteristics of the ground object. For flat roads, buildings, and some bare land with strong reciprocity characteristics, the information correction is relatively thorough, while for farmland, bare land, and irregular artificial buildings with rough surfaces that cannot meet reciprocity, some information cannot be corrected. Therefore, the method of selecting calibration reference samples by the polarization decomposition method is closely related to the characteristics of the ground object and conforms to the actual characteristics of the ground object.

[0214] 7. Wishart_H_A_alpha classification

[0215] Figure 8 Regions 1 and 2 are both artificial buildings that are nearly perpendicular to the distance. A large number of dihedral angles are formed between the building and the ground, and between the front facade of the building and the structure of the top of the building that are perpendicular to the incident wave. Under normal circumstances, when performing Wishart_HAalpha classification, these strong dihedral angles should appear as low-glare multiple scattering, during which there will be some independent dipole components, which appear as low-olefin dipole scattering. The experiment found that the calibration results obtained by the calibration reference samples screened by the polarization decomposition method are more in line with the actual situation of the above-mentioned images, while the calibration results obtained by the calibration reference samples screened by the correlation coefficient method are misclassified.

[0216] The feasibility of the improved method was verified through experiments with real data. The advanced nature of the improved method was verified by comparing the calibration results of samples screened by the conventional correlation coefficient method through calibration parameter residuals, false color composite images, polarization characteristic response images, and imaginary part information images of the polarization covariance matrix C23. The reliability and universality of the improved method were verified through calibration experiments on different land types and different load data.

Claims

1. A method for precise polarization calibration of reciprocal coherent targets for airborne fully polarimetric SAR, characterized in that: Analyze the impact of channel crosstalk, imbalance phase amplitude, system noise, and multi-look processing factors on the quality of airborne polarization calibration, and use simulation data for polarization calibration; establish a new method for screening reciprocity and reflection symmetry targets based on polarization coherent target decomposition, and use the polarization coherent target decomposition model to extract symmetry-dominated targets from the screened reciprocity coherent targets for polarization calibration parameter calculation; 1) Scattered target polarization calibration accuracy evaluation model: Establish an evaluation model for the effects of channel crosstalk, unbalanced phase amplitude, and system noise on polarization calibration quality, which affect polarization data quality. Calculate that polarization calibration accuracy improves with the reduction of channel crosstalk, and the stability and accuracy of parameter solution improve with the improvement of signal-to-noise ratio. Multi-view processing reduces the impact of data noise on calibration quality, and parameter solution accuracy improves with the increase of the number of views. Establish a reference model for improving polarization calibration quality, improving polarization system design, and clarifying polarization data quality requirements. 2) Optimization of calibration reference samples based on polarization coherent target decomposition: Based on polarization coherent target decomposition, a method for screening reciprocity and reflection symmetry calibration reference samples is established, and a target scattering feature identification method based on coherent target decomposition is adopted, including reciprocity constraint and identification, coherence constraint and identification, and symmetry constraint and identification. By calculating the factors that cause the weakening of target reciprocity, a method for determining reciprocity by using the reciprocity angle is adopted, and based on the polarization distortion model, a method for determining reciprocity by using the reciprocity angle for uncalibrated data is established; then, under the premise of satisfying reciprocity, the coherence factor is calculated and the critical value is set to further screen out the strong coherence targets; finally, based on the polarization coherence decomposition model, the selected reciprocity coherence targets are decomposed, and the targets with dominant reflection symmetry are selected as calibration reference samples for polarization calibration parameter solution. Through multi-layer screening of target features, the deviation influence of uncalibrated data caused by polarization distortion is reduced, and the accuracy of calibration reference sample screening is improved.

2. The method for precise polarization calibration of reciprocal coherent targets of airborne fully polarimetric SAR according to claim 1, characterized in that: Model of the impact of inter-channel crosstalk and imbalance on polarization calibration accuracy: The ideal data is simulated using the reflection symmetry model to calculate the impact of inter-channel crosstalk and imbalance on calibration accuracy. The polarization covariance matrix is ​​shown in Formula 1. A trihedral corner reflector model is added to the image for accuracy assessment. The signal strength is 40db, and the signal form is shown in Formula 2. The data meets the reciprocity and reflection symmetry assumptions. The effects of crosstalk and cross-channel imbalance on the calibration solution are verified respectively. Only one interference variable is added to the data, and the crosstalk and cross-channel imbalance are added respectively. Then, polarization calibration is performed on the two types of generated simulation data.

3. The method for precise polarization calibration of reciprocal coherent targets of airborne fully polarimetric SAR according to claim 1, characterized in that: Model of the impact of system noise on calibration accuracy: Use Gaussian white noise with an intensity of -20db for simulation, set the data to different levels of signal-to-noise ratio, set the ground object signal to different intensity levels, set the interval to -20 to 10db, and the ground signal form is as shown in Formula 1 to obtain simulation data with a signal-to-noise ratio range of 0 to 30db; arrange trihedral corner reflectors in the image for accuracy assessment, the signal intensity is 40db, the signal form is as shown in Formula 2, add a crosstalk signal less than -12db, add -15db, -20db, and -30db crosstalk to the generated ideal data respectively, generate three simulation data, and then perform polarization calibration simulation.

4. The method for precise polarization calibration of reciprocal coherent targets of airborne fully polarimetric SAR according to claim 3, characterized in that: Model the effects of system noise: Step 1: Generate ideal data that meets symmetry based on the ideal model of the ground object and trihedral angle, with a size of 500*1600; Step 2: Set an energy level for every 100 pixels of ideal data along the distance, with a total of 16 energy levels. The difference between each energy level is 2db, and the energy level range is -20 to 10db. Set a trihedral corner reflector in each energy level, and its signal strength is 40db. Step 3: Add -20db Gaussian white noise to the ideal graded data with ground objects and trihedral corner reflector signals, and obtain 16 energy levels with a signal-to-noise ratio of 0 to 30db, with a difference of 2db between energy levels; Step 4: Add -15db, -20db, and -30db crosstalk errors to the data to generate three data with different crosstalk amounts, where the amplitudes of the four crosstalk parameters are equal, i.e., v=w=z=u, the crosstalk phases are all set to 0 degrees, the channel imbalance is set to the ideal value, the amplitude is 1, and the phase is 0 degrees; Step 5: Solve the calibration parameters for the three simulation data respectively, and calibrate the data using the calculated calibration parameters; Step 6: Analyze the relationship between parameter solution accuracy and signal-to-noise ratio. By recalibrating the calibrated data, the residual amount of each parameter is obtained. The crosstalk calibration accuracy is evaluated by the average value of the four crosstalk parameters, and the imbalance amplitude accuracy is evaluated by the difference between the solution result and the preset value 1.

5. The method for precise polarization calibration of reciprocal coherent targets of airborne fully polarimetric SAR according to claim 1, characterized in that: Model of the impact of multi-view processing on calibration accuracy: Step 1: Generate ideal forest simulation data based on the Arii vegetation model, Bragg scattering model, and dihedral scattering model. The size is 2000*2000 (row*column) data. Set the forest data scattering intensity to -10db. Set a trihedral reflector in the data, and its signal strength is 40db. Step 2: Add -22db Gaussian white noise to the data, and set the data signal-to-noise ratio to 12db; Step 3: Add the above data to the crosstalk amount of -15db, -20db, and -30db to generate three simulation data with different crosstalk, where the amplitudes of the four crosstalk parameters are equal, that is, v=w=z=u, the crosstalk phases are all set to 0 degrees, and the channel imbalance is set to the ideal value, the amplitude is 1, and the phase is 0 degrees; Step 4: Perform multi-view processing on the above three data respectively, and set the number of views to 2, 5, 10, 20, 50, and 100 respectively to obtain the multi-view processed data; Step 5: Calculate the calibration parameters for the above three sets of multi-view data, and calibrate the data using the calculated parameters; Step 6: Analyze the relationship between parameter solution accuracy and multi-view processing. By recalibrating the calibrated data, the residual amount of each parameter is obtained. The crosstalk calibration accuracy is evaluated by the average value of the four crosstalk parameters, and the imbalance amplitude accuracy is evaluated by the difference between the solution result and the preset value 1.

6. The method for precise polarization calibration of reciprocal coherent targets of airborne fully polarized SAR according to claim 1, characterized in that: Analysis of crosstalk parameter estimation accuracy: When the crosstalk amount is -20dB and -30dB, when the number of views is 5, the calibration accuracy of the crosstalk parameters and unbalanced phase amplitude tends to be stable; when the crosstalk amount is -15dB, when the number of views is 5, the crosstalk and unbalanced phase solution accuracy tends to be stable, and when the number of views is greater than 10, the change of unbalanced amplitude accuracy tends to be stable; Suppose the polarization response characteristics of the trihedral reflector are analyzed: with the increase of the number of views, the calibration effect gradually improves. When the crosstalk is -20dB and -30dB, when the number of views is greater than 5, the polarization response of the trihedral is calibrated to the best. When the crosstalk is -15dB, when the number of views is greater than 50, the polarization response of the trihedral is calibrated to the best.

7. The method for precise polarization calibration of reciprocal coherent targets of airborne fully polarized SAR according to claim 1, characterized in that: Optimization of calibration reference samples based on polarization coherent target decomposition: Any scattering matrix is ​​expressed as the sum of independent basic target scattering matrices: Where S is the scattering matrix of a coherent target, S k represents the scattering matrix of a basic target, α k is the weight of the basic target scattering matrix S in the matrix S; When the incident wave and the scattered wave are completely polarized waves, the scattering matrix S can express the scattering characteristics of a specific target and be associated with its relevant physical characteristics. Polarization coherent target decomposition is used to accurately extract calibration reference samples that satisfy reciprocity and reflection symmetry from uncalibrated data. Based on the reciprocity of the calibration data, target coherence, and the symmetry characteristics of the coherent decomposition model, further modeling is performed.

8. The method for precise polarization calibration of reciprocal coherent targets of airborne fully polarized SAR according to claim 1, characterized in that: Reciprocity constraint and identification: The reciprocity angle is used as the reciprocity determination factor, and the target scattering matrix is ​​calculated in its reciprocity subspace W. rec Projection, find the original matrix and the reciprocal subspace W rec The angle θ between rec , the solution of the reciprocal angle is: where θ rec is the reciprocity angle, P rec W rec The projection operator in space, is the normalized scattering matrix, represents the reciprocal projection of the scattering matrix, ||█|| represents the two-norm symbol, and the vector normalization is as follows: Where (■,■) represents the inner product symbol; the scattering matrix unit vector is obtained by normalizing the vector Express power in W: W=|S HH | 2 +|S HV | 2 +|S VH | 2 +|S VV | 2 Formula 8 Among them, S xy are the four elements of the scattering matrix S, then the reciprocal projection norm is simplified to: Reciprocity angle θ rec Characterizes the degree to which the scattering matrix of the target object approaches reciprocity. rec = 0, indicating that the target object scattering matrix is ​​completely located in W rec subspace, the two matrices belong to the same category, and the scattering matrix is ​​strictly reciprocal; when θ rec =π / 2, indicating that the scattering matrix of the target object is orthogonal to the reciprocity scattering matrix, and is completely non-reciprocal. π / 4 is used as the critical value for reciprocity determination. Target objects with a value smaller than this critical value are considered to satisfy reciprocity and are reciprocity scatterers. When the acquired observation data has deviations, the reciprocity angle needs to be derived in combination with the polarization calibration model. After ignoring the image additive noise, the observed data and the ideal scattering matrix after calibration have the following relationship: Among them, O is the observation matrix, O xy are the four elements of the observation matrix, ignoring the influence of crosstalk, simplifying the above relationship group and expressing it in matrix form as follows: This gives the power W of the observation matrix O , the formula is as follows: W O =|O HH | 2 +|O HV | 2 +|O VH | 2 +|O VV | 2 =|kαS HH | 2 +|a -1 S HV | 2 +|αS VH | 2 +|k -1 a -1 S VV | 2 formula 12 The reciprocal projection norm of the observation matrix is ​​obtained as follows: The reciprocity angle of the observation matrix is ​​expressed as: When selecting calibration reference samples based on the reciprocity characteristics, it is only necessary to select samples with the smallest reciprocity angle possible based on the calculated statistical distribution law of the angle.

9. The method for precise polarization calibration of reciprocal coherent targets of airborne fully polarized SAR according to claim 1, characterized in that: Coherence constraint and identification: The coherence of the target is used to screen samples to reduce the influence of the scattering characteristic error of the incoherent target on the polarization decomposition of the coherent target. The target coherence is characterized by the factor P. The scattering matrix of the scattered target is the result of the mutual influence of multiple adjacent scattering units. P is statistically calculated using the ensemble average method, and the expression is as follows: Where <█> represents the spatial set average, α is the trihedral scattering component S in the false color synthesis decomposition a ε is the maximum symmetric partial coefficient of the dihedral scattering component S in the false color synthesis decomposition; When the coherence factor P is close to 1, the target has strong coherence and is judged as a coherence-dominated target; When the coherence factor P is close to 0, the target coherence is weak and it is judged as an incoherent target. The coherence factor P value is calculated using a sliding window, and the window size is determined in combination with the pixel size. The point target is set to 2 to 5 meters, and the window is set between 2m×2m and 5m×5m. The calculation of the coherence factor P is based on the false color synthesis decomposition method. α and ε are obtained by solving the problem, and the coherence factor P is used to determine the relevance of the target.

10. The method for precise polarization calibration of reciprocal coherent targets of airborne fully polarimetric SAR according to claim 1, characterized in that: Symmetry constraints and identification: Targets that satisfy reflection symmetry must satisfy reciprocity. Reciprocity is the premise of reflection symmetry. Polarization coherent target decomposition model is used to determine target symmetry. Diagonalization method is used to perform characteristic decomposition on target scattering matrix S: S = R(ψ)T(τ m )·S d ·T(τ m )R(-ψ) Equation 16 In the formula, τ m represents the helicity of the target, T(τ m ) and S d They are respectively expressed as follows: By separating the rotation matrix Obtain the rotation invariant parameters, which are the phase Helix angle τ m , scattering type parameter α s , other parameters include maximum amplitude m, polarization azimuth parameters As follows: in, SV represents the scattering vector; Substituting the eigenvalues ​​λ1, λ2, λ3 and the corresponding eigenvectors u1, u2, u3 of the parameterized coherence matrix T3 into Equation 19, we can obtain the decomposition parameters of the corresponding eigenvalues. The parameter component with the largest eigenvalue is the principal component, reflecting the dominant characteristics of the scatterer. The rotationally invariant parameters reflect the scattering law and geometric characteristics of the ground object target. m The value is the helicity angle, when τ m = 0, indicating that the scatterer is completely symmetrical, including dihedral angles and trihedral angles. m =π / 4, indicating that the scatterer is completely asymmetric, including typical left and right helices. m The spiral components that do not conform to the reflection symmetry are eliminated, most of the asymmetric targets are accurately eliminated, and the error influence of non-reflection symmetry targets on parameter solution is reduced.

Images