SAR polarization calibration reference distributed target extraction method

Abstract

The invention discloses an SAR polarization calibration reference distributed target extraction method, and the method comprises the steps: carrying out the threshold method pixel classification through employing a polarization distortion parameter and a polarization orientation angle extracted from an uncalibrated full-polarization SAR image, and obtaining a pixel-level high-confidence volume scattering dominant target and a non-volume scattering dominant target; performing super-pixel segmentation on the uncalibrated full-polarization SAR image to obtain super-pixel units; extracting two polarization characteristics of the superpixel unit; processing the pixel-level high-confidence volume scattering dominant and non-volume scattering dominant targets by using a pixel proportion threshold method to obtain super-pixel unit high-confidence volume scattering dominant and non-volume scattering dominant targets, and combining the super-pixel unit high-confidence volume scattering dominant and non-volume scattering dominant targets with the two polarization features to obtain super-pixel unit positive and negative samples; and training the positive and negative samples of the super-pixel unit by using a random forest classifier to obtain a total image classification result, and the scattering dominant target of the super-pixel unit in the total image classification result being the reference distributed target meeting the polarization calibration requirement.

Description

Technical Field

[0001] The invention relates to a reference distributed target extraction method, device, equipment and storage medium, and belongs to the field of synthetic aperture radar polarization processing. Background Art

[0002] Polarimetric calibration plays a pivotal role in the quantitative analysis and application of polarimetric synthetic aperture radar (PolSAR) images. Distributed target polarimetric calibration methods, such as the Van Zyl algorithm, Quegan algorithm, and Ainsworth algorithm, have become increasingly important due to the reduced use of manual calibrators and the ability to cover larger calibration areas. However, Quegan-like algorithms rely on distributed targets dominated by volume scattering with reciprocity and reflection symmetry, and may encounter challenges in accurately estimating polarimetric distortion parameters when these properties are not met, resulting in reduced accuracy and stability.

[0003] In most images, there are both natural and artificial targets. Only a part of natural targets show reciprocity and reflection symmetry and can be directly used as reference targets in the Quegan algorithm. In order to solve the problem of manually selecting reference targets, it is necessary to use polarization features to extract reference targets. In existing schemes, the reference targets can be extracted by combining the selected polarization features with the threshold method, but the threshold setting requires a lot of prior information and has no specific physical meaning. It is easy to extract reference targets incorrectly and it is difficult to achieve robust automatic processing. Summary of the invention

[0004] The technical problem solved by the present invention is: to overcome the shortcomings of the prior art and provide a SAR polarization calibration reference distributed target extraction method, so as to solve the technical problems in the prior art that the reference target extraction method has high difficulty in setting the threshold, is prone to erroneous extraction of reference targets, and is difficult to achieve robust automated processing.

[0005] The technical solution of the present invention is: a SAR polarization calibration reference distributed target extraction method, comprising:

[0006] Polarization distortion parameters and polarization orientation angles extracted from uncalibrated fully polarimetric SAR images are used to perform threshold pixel classification to obtain high-confidence volume scattering-dominated targets and non-volume scattering-dominated targets at the pixel level.

[0007] Perform super-pixel segmentation on the uncalibrated fully polarized SAR image to obtain super-pixel units;

[0008] Extract two polarization features of circular polarization basis correlation coefficient and reflection symmetry coefficient from superpixel units;

[0009] The pixel ratio threshold method is used to process the pixel-level high-confidence volume scattering-dominated and non-volume scattering-dominated targets to obtain the high-confidence volume scattering-dominated and non-volume scattering-dominated targets of the superpixel unit, and combined with the two polarization features of the circular polarization basis correlation coefficient and the reflection symmetry coefficient to obtain the positive and negative samples of the superpixel unit;

[0010] A random forest classifier is used to train the positive and negative samples of superpixel units to obtain the classification results of the entire image. The superpixel unit volume scattering dominant target in the full image classification result is the reference distributed target that meets the requirements of polarization calibration.

[0011] The method uses the polarization distortion parameters and polarization orientation angles extracted from the uncalibrated fully polarized SAR image to perform threshold method pixel classification to obtain pixel-level high-confidence volume scattering-dominated targets and non-volume scattering-dominated targets, including:

[0012] The reciprocity angle is calculated using the uncalibrated fully polarized SAR image, and the reciprocity angle is compared with the reciprocity judgment threshold to eliminate the targets corresponding to the reciprocity angle exceeding the reciprocity judgment threshold.

[0013] The Quegan polarization calibration algorithm is used to process the uncalibrated fully polarimetric SAR image after elimination to obtain the crosstalk distortion parameters and channel imbalance distortion parameters.

[0014] Set the physical meaning threshold of the phase standard deviation of the channel imbalance distortion parameter and the physical meaning threshold of the crosstalk distortion parameter, and judge whether the phase standard deviation of the channel imbalance distortion parameter and the crosstalk distortion parameter of each pixel exceed the two set thresholds. If one of them exceeds the two thresholds, the classification result of the pixel is a high-confidence non-volume scattering dominated target.

[0015] The polarization orientation angle is calculated for each pixel in the uncalibrated fully polarimetric SAR image after the elimination process, and the physical significance threshold of the polarization orientation angle standard deviation is set. The classification result of the pixel corresponding to the polarization orientation angle standard deviation exceeding the physical significance threshold of the polarization orientation angle standard deviation is a pixel-level high-confidence volume scattering dominated target.

[0016] The physical significance threshold of the phase standard deviation of the channel imbalance distortion parameter is set to 140°, and the physical significance threshold of the crosstalk distortion parameter is set to -6 dB.

[0017] The physically significant threshold value of the polarization orientation angle standard deviation is set to 29°.

[0018] The calculation of the polarization orientation angle of the uncalibrated fully polarized SAR image after the elimination process is specifically:

[0019]

[0020] Where POA represents the polarization orientation angle; <·> represents the spatial average operation, * represents the conjugate operation; H represents horizontal polarization, V represents vertical polarization, M represents vertical polarization, ij represents the observed polarization scattering matrix element.

[0021] The linear iterative clustering algorithm is used to perform super-pixel segmentation on the uncalibrated fully polarimetric SAR image to obtain super-pixel units, including:

[0022] 1) Set the number of superpixels and seed points, and evenly distribute the seed points in the uncalibrated fully polarimetric SAR image;

[0023] 2) The uncalibrated fully polarimetric SAR image is processed to obtain the Pauli pseudo-color image of the uncalibrated fully polarimetric SAR image. The gradient values ​​of all pixels in the Pauli pseudo-color image are calculated in the 3×3 neighborhood of each seed point. The seed point is moved to the place with the smallest gradient in the neighborhood to obtain the reselected initial seed point, i.e., the initial cluster center.

[0024] 3) Use the initial cluster center to search for pixels in a 2S×2S neighborhood. For each searched pixel, calculate the color distance d between it and the initial cluster center in the surrounding 2S×2S range. c , spatial distance d s ; and calculate the distance metric D; the S represents the distance between two adjacent seed points;

[0025] 4) Sort the distance metrics between the pixel points and the initial cluster centers, and define the initial cluster center corresponding to the minimum distance metric D as the new cluster center of the pixel point;

[0026] 5) According to the pixels assigned to each new cluster center, calculate the average spatial position of these pixels to obtain the spatial position of the new cluster center;

[0027] 6) Iterate steps 3)-5) until the spatial position of the cluster center no longer changes, and obtain the final clustering result; define the pixel point of each cluster center as a superpixel unit.

[0028] The setting of the number of superpixels and the number of seed points includes: setting the number of superpixels=the number of seed points=the total number of pixels / (S×S).

[0029] The distance metric Among them, N c Indicates the brightness range of the Lab color model.

[0030] The extracting of two polarization features of the circular polarization basis correlation coefficient and the reflection symmetry coefficient from the superpixel unit includes:

[0031] Perform spatial averaging on superpixel units to obtain the second-order statistical covariance matrix;

[0032] The reflection symmetry coefficient is calculated using the second-order statistics covariance matrix;

[0033] Perform polarization basis transformation on the uncalibrated fully polarized SAR image, converting the linear polarization basis into the circular polarization basis, that is, converting the horizontal polarization and the vertical polarization into the left-hand circular polarization and the right-hand circular polarization, and obtaining the circular polarization basis image;

[0034] The circular polarization basis correlation coefficient is calculated using the obtained circular polarization basis image.

[0035] Reflection symmetry coefficient Q rs The calculation method is as follows:

[0036]

[0037] C ′ (θ)=U(θ)C ′ U T (θ)

[0038]

[0039] Among them, det(·) represents the determinant operation of the matrix, C' represents the covariance matrix of the second-order statistics, and C 11 is the upper left corner 2×2 matrix of C', C 22 is the lower right 1×1 matrix of C'.

[0040] The circular polarization basis correlation coefficient ρ is calculated using the obtained circular polarization basis image. RRLL ,include:

[0041]

[0042] Among them, R represents right-hand circular polarization, L represents left-hand circular polarization; i represents the imaginary part.

[0043] The pixel-level high-confidence volume scattering-dominated and non-volume scattering-dominated targets are processed by the pixel ratio threshold method to obtain high-confidence volume scattering-dominated and non-volume scattering-dominated targets of superpixel units, and the non-volume scattering-dominated targets are combined with the two polarization features of the circular polarization basis correlation coefficient and the reflection symmetry coefficient to obtain positive and negative samples of superpixel units, including:

[0044] The pixel ratio threshold method is used to calculate the ratio of pixel-level high-confidence volume scattering-dominated targets in each superpixel unit, and a ratio threshold is set. The superpixel units whose ratio of high-confidence volume scattering-dominated targets exceeds the ratio threshold are determined as volume scattering-dominated targets. The superpixel unit positive samples are obtained by combining the circular polarization basis correlation coefficient and reflection symmetry coefficient characteristics of the superpixel unit.

[0045] The pixel ratio threshold method is used to calculate the proportion of high-confidence non-volume scattering dominated targets at the pixel level in each superpixel unit. A ratio threshold is set, and the superpixel units whose proportion of high-confidence non-volume scattering dominated targets exceeds the ratio threshold are judged as non-volume scattering dominated targets. The superpixel units are fused with the circular polarization basis correlation coefficient and reflection symmetry coefficient features to obtain superpixel unit negative samples.

[0046] The ratio threshold is set to 50%.

[0047] The beneficial effects of the present invention are as follows: in the present invention, polarization distortion parameters and polarization orientation angles are extracted from uncalibrated fully polarized SAR images, and threshold pixel classification is performed to obtain pixel-level high-confidence volume scattering-dominated and non-volume scattering-dominated targets. The advantage of this method is that it reduces the dependence on complex prior information, can more intuitively distinguish targets, and improves the accuracy of classification; further, a linear iterative clustering algorithm is used to perform superpixel segmentation on the uncalibrated fully polarized SAR images to obtain superpixel units for distributed target extraction. The advantage of superpixel segmentation is that the image is divided into smaller areas with similar features, which reduces the complexity of calculation, while retaining the edge information of the target and improving the accuracy of target extraction; the circular polarization basis correlation coefficient, reflectance coefficient and other factors of the superpixel unit obtained in the previous step are calculated. The advantage of this step is that it can improve the feature expression ability of superpixels and provide richer and more accurate feature information for subsequent classification. Finally, the pixel-level high-confidence volume scattering-dominated and non-volume scattering-dominated targets are combined with the pixel ratio threshold method to obtain the superpixel unit high-confidence volume scattering-dominated and non-volume scattering-dominated targets, which are then fused with the circular polarization basis correlation coefficient and the reflection symmetry coefficient features to obtain the positive and negative samples of the superpixel unit. The random forest classifier is used to train the samples to obtain the classification results of the entire image, in which the superpixel unit high-confidence volume scattering-dominated targets are the reference distributed targets that meet the requirements of polarization calibration. This method has good generalization ability, can realize automatic classification of the entire image, and effectively extract reference distributed targets that meet the requirements of polarization calibration. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 The figure is a flow chart of the method of the present invention.

[0049] Figure 2 The Q of the built-up area target dominated by non-volume scattering (A) and the vegetation target dominated by volume scattering (B) is shown in Figure 2. rs Feature sequence; DETAILED DESCRIPTION

[0050] The embodiment of the present invention provides a SAR polarization calibration reference distributed target extraction method, referring to Figure 1 , Figure 1It is a flowchart of the first embodiment of the SAR polarization calibration reference distributed target extraction method of the present invention.

[0051] In this embodiment, the SAR polarization calibration reference distributed target extraction method includes the following steps:

[0052] Step S10: extract polarization distortion parameters and polarization orientation angles from the uncalibrated fully polarized SAR image, perform threshold pixel classification, and obtain pixel-level high-confidence volume scattering-dominated and non-volume scattering-dominated targets.

[0053] It should be noted that the distribution of ground objects is often complex and uneven, and cannot meet the reciprocity assumption of simple, uniform distributed targets. Existing polarization calibration algorithms require the reference target to meet the reciprocity assumption. In the polarization data actually obtained, many distributed targets often fail to meet the reciprocity due to propagation effects, interactions with special objects, weak backscattering cross sections, etc. Therefore, non-reciprocal scatterers are first eliminated. The projection angle θ between the target and the subspace can be used to measure:

[0054]

[0055] in, is the unit vector of the observed polarization scattering matrix:

[0056]

[0057] H=|M HH | 2 +|M HV | 2 +|M VH | 2 +|M VV | 2

[0058] Among them, H represents horizontal polarization, V represents vertical polarization, and M ij represents the observed polarization scattering matrix element.

[0059] The norm of the reciprocal projection is expressed as:

[0060]

[0061] When θ = 0, it means that the scattering matrix is ​​strictly reciprocal, and when θ = π / 2, it means that the scattering matrix does not conform to reciprocity at all. π / 4 can be regarded as the reciprocity judgment threshold, and the targets exceeding the threshold are eliminated, and the remaining targets are processed later.

[0062] It should be noted that the polarization distortion model can be approximately expressed by the following formula:

[0063] M=R·S·T+N

[0064] Where M is the observed polarization scattering matrix, R and T represent the receiving and transmitting channel distortion matrices, respectively, S is the theoretical polarization scattering matrix, and N is the system noise matrix. Ignoring the system noise, the above formula can be rewritten as:

[0065]

[0066] Among them, H represents horizontal polarization, V represents vertical polarization, S ij represents the theoretical polarization scattering matrix element, r ij represents the receive channel distortion matrix element, t ij Represents the transmit channel distortion matrix element.

[0067] After vectorizing the scattering matrix S, we get the following equation:

[0068] m=D·s

[0069] m=[M HH ,M HV ,M VH ,M VV ] T ,s=[Y HH ,S HV ,S VH ,S VV ] T

[0070] The superscript T represents the transpose operation, and D represents the redefined distortion matrix, which is composed of the distortion matrix R and T and is defined as follows:

[0071]

[0072] in, is the Kronecker product, Y is the system gain, α is the cross-polarization channel imbalance distortion, u, v, w, z are the crosstalk distortions, and k is the co-polarization channel imbalance distortion. The definitions of these parameters are as follows:

[0073]

[0074] Among them, N is the total number of pixels in the sliding window, α i is the α, μ of the i-th pixel in the sliding window α is the mean value of α in the sliding window.

[0075] In built-up areas and other man-made areas, the crosstalk amplitude |u| estimated by the Quegan algorithm is much higher than the actual system distortion, and its value is usually between -5dB and -20dB; in water areas, the cross-polarization channel imbalance phase Std(arg(α)) estimated by the Quegan algorithm is very high, fluctuating between 20° and 150°. Therefore, the two parameters |u| and Std(arg(α)) estimated by the Quegan algorithm are set with physical thresholds slightly lower than the upper limit of the above range: -6dB and 140°, and pixel classification is performed respectively. The classification results exceeding the threshold are taken as the union to obtain non-volume scattering-dominated targets such as high-confidence man-made structures. These two parameters have clear physical meanings, and the polarization distortion range of the SAR system is easy to roughly estimate, which provides prior information for threshold setting and reduces the difficulty of threshold setting.

[0076] Polarization orientation angle (POA) indicates the angular relationship between the target plane and the radar line of sight. Vegetation areas are usually dominated by volume scattering components. The Std(POA) of these areas is usually distributed between 10° and 30°, while the Std(POA) of other objects is more stable, usually within 5°. Therefore, a physical Std(POA) threshold slightly lower than the upper limit of the above range is selected: 29° for pixel classification. The classification results exceeding the threshold are high-confidence volume scattering targets such as forests, bushes, and grasslands. The estimation method of POA is as follows:

[0077]

[0078] Where <·> represents the spatial average operation, Re(·) represents the real part operation, * represents the conjugate operation, N is the total number of pixels in the sliding window, POA i is the POA of the i-th pixel in the sliding window, μ POA is the mean value of POA in the sliding window.

[0079] The above polarization characteristic parameters combined with thresholds with clear physical meanings can be used to extract high-confidence volume scattering-dominated and non-volume scattering-dominated targets. The advantage of this method is that it reduces the reliance on complex prior information, can distinguish targets more intuitively, and improves the accuracy of classification.

[0080] Step S20: Use a linear iterative clustering algorithm to perform superpixel segmentation on the uncalibrated fully polarimetric SAR image to obtain superpixel units for distributed target extraction.

[0081] In the specific implementation, we first set the number of pixels contained in each superpixel to 10×10, and calculate the number of superpixels based on the image size: total number of pixels / (10×10). We evenly distribute seed points in the image with a step size of S=10, and the number of seed points is the total number of pixels / (10×10).

[0082] Then the uncalibrated fully polarimetric SAR image is processed to obtain the Pauli pseudo-color image. The gradient values ​​of all pixels in the 3×3 neighborhood of the seed point are calculated, and the seed point is moved to the place with the smallest gradient in the neighborhood to obtain the reselected initial seed point, that is, the initial cluster center.

[0083] Use the initial cluster center obtained in the previous step to search for pixels in a 2S×2S neighborhood. For each searched pixel, calculate its color and spatial distance with the surrounding cluster center within the 2S×2S range to synthesize the distance metric.

[0084] Color distance d c The calculation method is as follows:

[0085]

[0086] Here, l, a, and b represent the brightness, the range from magenta to green, and the range from yellow to blue in the Lab color model, respectively.

[0087] Spatial distance d s The calculation method is as follows:

[0088]

[0089] Among them, x and y are the horizontal and vertical coordinates of the pixel respectively.

[0090] The distance metric synthesis method is as follows:

[0091]

[0092] Among them, N c Indicates the maximum color distance, the value range is [1,40], and is generally 10.

[0093] Sort the distance measures between the pixel point and each cluster center in the previous step, and assign the cluster center with the smallest distance as the new cluster center of the pixel point.

[0094] According to the pixels newly assigned to each cluster center, the average spatial position of these pixels is calculated to obtain the new spatial position of the cluster center.

[0095] The above steps are iterated until the spatial position of the cluster center no longer changes, and the final clustering result is obtained.

[0096] The final clustering results obtained in the above steps are processed, and the discontinuous and undersized superpixels are reallocated to neighboring superpixels to obtain the final superpixel segmentation result. The advantage of superpixel segmentation is that it divides the image into smaller areas with similar features, reducing the complexity of calculation, while retaining the edge information of the target and improving the accuracy of target extraction.

[0097] Step S30: extracting two polarization features, namely, circular polarization basis correlation coefficient and reflection symmetry coefficient, from the superpixel unit obtained in the previous step.

[0098] It should be noted that in order to extract the target dominated by volume scattering with reflection symmetry in PolSAR images, it is necessary to select appropriate polarization features, including: reflection symmetry coefficient Q rs and circular polarization basis correlation coefficient ρ RRLL .

[0099] Perform spatial averaging on the superpixel units obtained in the previous step to obtain the second-order statistical covariance matrix C ′ :

[0100]

[0101] The reflection symmetry coefficient Q is calculated using the superpixel covariance matrix. rs The calculation method is as follows:

[0102]

[0103] Among them, det(·) represents the determinant operation of the matrix, C' represents the covariance matrix, and C 11 is the upper left corner 2×2 matrix of C', C 22 is the 1×1 matrix in the lower right corner. When the reflection symmetry is satisfied and the polarization distortion is small, C' approaches the theoretical value C ′ t , so the stronger the reflection symmetry, the higher the Q rs The smaller the value of .

[0104] The polarization matrices observed from the same target in different directions may be very different. The existence of diversity in target scattering directions contains rich information, and the use of information on the diversity of target scattering directions may reveal the polarization characteristics of different targets. The reflection symmetry coefficient can be extended to the rotation domain through polarization basis transformation, and the ground cover can be classified according to its change law under different polarization bases. The POA rotation method can be used to convert the scalar feature Q rs Expanded to vector Q rs , providing a more accurate and robust target description by using the target’s changing characteristics with POA. Q at different orientation angles θ rs The calculation method is as follows:

[0105]

[0106] C ′ (θ)=U(θ)C ′ U T (θ)

[0107]

[0108] Among them, Q rs (θ) is Q under different POA rs By sampling θ at equal intervals in the range [0,90°), the scalar feature of the target can be expanded into a vector Q rs . Figure 2 The Q of built-up areas dominated by non-volume scattering and vegetation targets dominated by volume scattering are shown. rs Feature sequence, the features shown can effectively distinguish these two types of targets.

[0109] The uncalibrated fully polarized SAR image is transformed into a polarization basis from a linear polarization basis to a circular polarization basis, that is, the horizontal polarization and the vertical polarization are converted into left-hand circular polarization and right-hand circular polarization:

[0110]

[0111] Among them, R represents right-hand circular polarization, L represents left-hand circular polarization, and i represents an imaginary number.

[0112] The circular polarization base image obtained in the previous step is used to calculate the superpixel circular polarization correlation coefficient ρ RRKL Calculation:

[0113]

[0114] The polarization features obtained in the above steps are used to construct feature vectors to achieve accurate description of superpixels. The advantage of this step is that it can improve the feature expression ability of superpixels and provide richer and more accurate feature information for subsequent classification.

[0115] Step S40: Pixel-level high-confidence volume scattering-dominated and non-volume scattering-dominated targets are obtained by using the pixel ratio threshold method to obtain superpixel unit high-confidence volume scattering-dominated and non-volume scattering-dominated targets, which are fused with the circular polarization basis correlation coefficient and reflection symmetry coefficient features to obtain superpixel unit positive and negative samples, and the samples are trained using a random forest classifier to obtain full-image classification results, wherein the superpixel unit high-confidence volume scattering-dominated targets are the reference distributed targets that meet the polarization calibration requirements.

[0116] In the specific implementation, the pixel ratio threshold method is used to calculate the proportion of pixel-level high-confidence volume scattering dominant targets in each superpixel unit, and the ratio threshold is set to 50%. The superpixel unit whose ratio of high-confidence volume scattering dominant targets exceeds the ratio threshold is judged as the volume scattering dominant target, and the circular polarization basis correlation coefficient and reflection symmetry coefficient features are fused to obtain the superpixel unit positive sample; the pixel ratio threshold method is used to calculate the proportion of pixel-level high-confidence non-volume scattering dominant targets in each superpixel unit, and the ratio threshold is set to 50%. The superpixel unit whose ratio of high-confidence non-volume scattering dominant targets exceeds the ratio threshold is judged as the non-volume scattering dominant target, and the circular polarization basis correlation coefficient and reflection symmetry coefficient features are fused to obtain the superpixel unit negative sample. The random forest classifier is used to train the samples obtained in the above steps to obtain the classification result of the whole picture, in which the superpixel unit high-confidence volume scattering dominant target is the reference distributed target that meets the requirements of polarization calibration. This method has good generalization ability, can realize automatic classification of the whole picture, and effectively extract reference distributed targets that meet the requirements of polarization calibration.

[0117] The above are only preferred embodiments of the present invention, and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A SAR polarization calibration reference distributed target extraction method, characterized in that: include: Polarization distortion parameters and polarization orientation angles extracted from uncalibrated fully polarimetric SAR images are used to perform threshold pixel classification to obtain high-confidence volume scattering-dominated targets and non-volume scattering-dominated targets at the pixel level. Perform super-pixel segmentation on the uncalibrated fully polarized SAR image to obtain super-pixel units; Extract two polarization features of circular polarization basis correlation coefficient and reflection symmetry coefficient from superpixel units; The pixel ratio threshold method is used to process the pixel-level high-confidence volume scattering-dominated and non-volume scattering-dominated targets to obtain the high-confidence volume scattering-dominated and non-volume scattering-dominated targets of the superpixel unit, and combined with the two polarization features of the circular polarization basis correlation coefficient and the reflection symmetry coefficient to obtain the positive and negative samples of the superpixel unit; A random forest classifier is used to train the positive and negative samples of superpixel units to obtain the classification results of the entire image. The superpixel unit volume scattering dominant target in the full image classification result is the reference distributed target that meets the requirements of polarization calibration.

2. The SAR polarization calibration reference distributed target extraction method according to claim 1, characterized in that: The method uses the polarization distortion parameters and polarization orientation angles extracted from the uncalibrated fully polarized SAR image to perform threshold method pixel classification to obtain pixel-level high-confidence volume scattering-dominated targets and non-volume scattering-dominated targets, including: The reciprocity angle is calculated using the uncalibrated fully polarized SAR image, and the reciprocity angle is compared with the reciprocity judgment threshold to eliminate the targets corresponding to the reciprocity angle exceeding the reciprocity judgment threshold. The Quegan polarization calibration algorithm is used to process the uncalibrated fully polarimetric SAR image after elimination to obtain the crosstalk distortion parameters and channel imbalance distortion parameters. Set the physical meaning threshold of the phase standard deviation of the channel imbalance distortion parameter and the physical meaning threshold of the crosstalk distortion parameter, and judge whether the phase standard deviation of the channel imbalance distortion parameter and the crosstalk distortion parameter of each pixel exceed the two set thresholds. If one of them exceeds the two thresholds, the classification result of the pixel is a high-confidence non-volume scattering dominated target. The polarization orientation angle is calculated for each pixel in the uncalibrated fully polarimetric SAR image after the elimination process, and the physical significance threshold of the polarization orientation angle standard deviation is set. The classification result of the pixel corresponding to the polarization orientation angle standard deviation exceeding the physical significance threshold of the polarization orientation angle standard deviation is a pixel-level high-confidence volume scattering dominated target.

3. The SAR polarization calibration reference distributed target extraction method according to claim 2, characterized in that: The physical significance threshold of the phase standard deviation of the channel imbalance distortion parameter is set to 140°, and the physical significance threshold of the crosstalk distortion parameter is set to -6 dB.

4. The SAR polarization calibration reference distributed target extraction method according to claim 2, characterized in that: The physically significant threshold value of the polarization orientation angle standard deviation is set to 29°.

5. The SAR polarization calibration reference distributed target extraction method according to claim 2, characterized in that: The calculation of the polarization orientation angle of the uncalibrated fully polarized SAR image after the elimination process is specifically: Where POA represents the polarization orientation angle; <·> represents the spatial average operation, * represents the conjugate operation; H represents horizontal polarization, V represents vertical polarization, M represents vertical polarization, ij represents the observed polarization scattering matrix element.

6. The SAR polarization calibration reference distributed target extraction method according to claim 3, characterized in that: The linear iterative clustering algorithm is used to perform super-pixel segmentation on the uncalibrated fully polarimetric SAR image to obtain super-pixel units, including: 1) Set the number of superpixels and seed points, and evenly distribute the seed points in the uncalibrated fully polarimetric SAR image; 2) The uncalibrated fully polarimetric SAR image is processed to obtain the Pauli pseudo-color image of the uncalibrated fully polarimetric SAR image. The gradient values ​​of all pixels in the Pauli pseudo-color image are calculated in the 3×3 neighborhood of each seed point. The seed point is moved to the place with the smallest gradient in the neighborhood to obtain the reselected initial seed point, i.e., the initial cluster center. 3) Use the initial cluster center to search for pixels in a 2S×2S neighborhood. For each searched pixel, calculate the color distance d between it and the initial cluster center in the surrounding 2S×2S range. c , spatial distance d s ; and calculate the distance metric D; the S represents the distance between two adjacent seed points; 4) Sort the distance metrics between the pixel points and the initial cluster centers, and define the initial cluster center corresponding to the minimum distance metric D as the new cluster center of the pixel point; 5) According to the pixels assigned to each new cluster center, calculate the average spatial position of these pixels to obtain the spatial position of the new cluster center; 6) Iterate steps 3)-5) until the spatial position of the cluster center no longer changes, and obtain the final clustering result; define the pixel point of each cluster center as a superpixel unit.

7. The SAR polarization calibration reference distributed target extraction method according to claim 6, characterized in that: The setting of the number of superpixels and the number of seed points includes: setting the number of superpixels=the number of seed points=the total number of pixels / (S×S).

8. The SAR polarization calibration reference distributed target extraction method according to claim 6, characterized in that: The distance metric Among them, N c Indicates the brightness range of the Lab color model.

9. The SAR polarization calibration reference distributed target extraction method according to claim 6, characterized in that: The extracting of two polarization features of the circular polarization basis correlation coefficient and the reflection symmetry coefficient from the superpixel unit includes: Perform spatial averaging on superpixel units to obtain the second-order statistical covariance matrix; The reflection symmetry coefficient is calculated using the second-order statistics covariance matrix; Perform polarization basis transformation on the uncalibrated fully polarized SAR image, converting the linear polarization basis into the circular polarization basis, that is, converting the horizontal polarization and the vertical polarization into the left-hand circular polarization and the right-hand circular polarization, and obtaining the circular polarization basis image; The circular polarization basis correlation coefficient is calculated using the obtained circular polarization basis image.

10. The SAR polarization calibration reference distributed target extraction method according to claim 9, characterized in that: Reflection symmetry coefficient Q rs The calculation method is as follows: C ′ (θ)=U(θ)C ′ U t (i) Among them, det(·) represents the determinant operation of the matrix, C' represents the covariance matrix of the second-order statistics, and C 11 is the upper left corner 2×2 matrix of C', C 22 is the lower right 1×1 matrix of C'.

11. The SAR polarization calibration reference distributed target extraction method according to claim 9, characterized in that: The circular polarization basis correlation coefficient ρ is calculated using the obtained circular polarization basis image. RRLL ,include: Among them, R represents right-hand circular polarization, L represents left-hand circular polarization; i represents the imaginary part.

12. A SAR polarization calibration reference distributed target extraction method according to any one of claims 1 to 11, characterized in that: The pixel-level high-confidence volume scattering-dominated and non-volume scattering-dominated targets are processed by the pixel ratio threshold method to obtain high-confidence volume scattering-dominated and non-volume scattering-dominated targets of superpixel units, and the non-volume scattering-dominated targets are combined with the two polarization features of the circular polarization basis correlation coefficient and the reflection symmetry coefficient to obtain positive and negative samples of superpixel units, including: The pixel ratio threshold method is used to calculate the ratio of pixel-level high-confidence volume scattering-dominated targets in each superpixel unit, and a ratio threshold is set. The superpixel units whose ratio of high-confidence volume scattering-dominated targets exceeds the ratio threshold are determined as volume scattering-dominated targets. The superpixel unit positive samples are obtained by combining the circular polarization basis correlation coefficient and reflection symmetry coefficient characteristics of the superpixel unit. The pixel ratio threshold method is used to calculate the proportion of high-confidence non-volume scattering dominated targets at the pixel level in each superpixel unit. A ratio threshold is set, and the superpixel units whose proportion of high-confidence non-volume scattering dominated targets exceeds the ratio threshold are judged as non-volume scattering dominated targets. The superpixel units are fused with the circular polarization basis correlation coefficient and reflection symmetry coefficient features to obtain superpixel unit negative samples.

13. A SAR polarization calibration reference distributed target extraction method according to claim 12, characterized in that: The ratio threshold is set to 50%.

Images