ISAR Satellite Attitude and Size Estimation Method Based on Component 3D Segmentation

Abstract

This invention proposes an ISAR satellite attitude and size estimation method based on component 3D segmentation to address the poor robustness of existing satellite attitude and size estimation techniques. The implementation steps are as follows: 1) Calculate the range and azimuth resolution of the ISAR image of the satellite target; 2) Perform 3D reconstruction of the satellite target; 3) Perform 3D segmentation of the satellite target components; 4) Obtain the estimated satellite attitude and size results from the ISAR image. This invention obtains the 3D point set of the satellite target body and the 3D point set of the solar panel through 3D reconstruction and component 3D segmentation, and then estimates the true attitude and size of the satellite target from these sets. The entire process does not require solving non-convex optimization functions, thus ensuring the robustness of the algorithm.

Description

Technical Field

[0001] This invention belongs to the field of radar technology and relates to an ISAR satellite attitude and size estimation method. Specifically, it is an ISAR image satellite target attitude and size estimation method based on component three-dimensional segmentation, which can be used for satellite target classification and recognition, satellite working status determination, etc. Background Technology

[0002] Satellite attitude and size can fully reflect the physical structure of satellite targets, providing rich information for satellite target classification and identification and determining satellite operational status, which is of great significance for maintaining the security of space assets and monitoring satellite dynamics. Based on the availability of prior information about the target, satellite targets can be divided into two main categories: cooperative targets and non-cooperative targets. It is hoped that satellite attitude and size estimation methods can be applied to both cooperative and non-cooperative targets, while simultaneously considering both the accuracy and robustness of the estimation.

[0003] Inverse Synthetic Aperture Radar (ISAR) has the capability for all-weather, all-day, and long-range imaging. Imaging satellite targets using ISAR can yield ISAR images containing rich geometric features of the satellite targets; using multiple ISAR images of satellite targets observed from different angles can enable estimation of the satellite's true attitude and the dimensions of its major components.

[0004] Currently, satellite attitude and size estimation methods can be mainly divided into three categories: The first category uses attitude sensors to record the changes in the satellite's three-dimensional rotation angle, and then uses filtering methods to estimate the satellite's attitude. This type of method is only applicable to spacecraft equipped with three-axis gyroscopes and attitude sensors. The second category establishes a complete template library of targets to be tested using electromagnetic simulation technology, and matches the measured data of the targets to be tested with the template library to achieve satellite target attitude determination. This type of method requires obtaining the three-dimensional model of the target to be tested in advance. The first two categories of methods are not applicable to non-cooperative targets. The third category first extracts the two-dimensional geometric features of the satellite from multiple ISAR images, and then constructs an optimization function to invert the three-dimensional structure of the satellite, thereby achieving the estimation of satellite attitude and size. For example, patent application CN112946646A, entitled "A Method for Estimating Satellite Target Attitude and Size Based on ISAR Image Interpretation," discloses a method for estimating satellite target attitude and size based on ISAR images and parameter optimization. This method first segments the ISAR image sequence to obtain images of the satellite body, then extracts the scattering points of the satellite body images and the two-dimensional orientation of the satellite body in each image, and finally constructs an optimization function to solve for the attitude and size of the satellite body. Using similar steps, this method can also obtain the attitude and size of the satellite's solar panels. This method is applicable to non-cooperative targets and has high estimation accuracy; however, the optimization function constructed in this method is non-convex, and the process of solving the optimization function often converges to a local optimum rather than a global optimum, making attitude and size estimation prone to errors and affecting the robustness of the method. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of the existing technology by proposing an ISAR satellite attitude and size estimation method based on three-dimensional component segmentation, which solves the technical problem of poor robustness in satellite attitude and size estimation in the existing technology.

[0006] To achieve the above objectives, the technical solution adopted by the present invention includes the following steps:

[0007] Step 1) Calculate the range resolution and azimuth resolution of the satellite target ISAR image:

[0008] Obtain Num ISAR images of size M×N containing satellite targets, I={I1,I2,...,I... num ,...,I Num}, and calculate I for each ISAR image num Distance resolution Δr num and azimuth resolution Δd num This yields the distance resolution set Δr = {Δr1, Δr2, ..., Δr}. num ,...,Δr Num} and the azimuth resolution set Δd={Δd1,Δd2,...,Δd num ,...,Δd Num}, where Num≥3, M≥100, N≥100, I num This represents the num-th satellite target ISAR image, where num = 1, 2, ..., Num;

[0009] Step 2) Perform three-dimensional reconstruction of the satellite target:

[0010] Step 2a) with δ num For each ISAR image, the threshold is set as follows: num Perform binarization segmentation and then segment I. num Morphological operations are performed on the binarized segmentation results to obtain the set of separated binary images Itb = {Itb1, Itb2, ..., Itb} corresponding to the ISAR image set I. num ,...,Itb Num}, where Itb num Indicates to I num The coarsely separated binary image Ita obtained by binarization segmentation num The corresponding binary image of target and noise separation, Itb num The pixel locations with median values ​​of 1 and 0 belong to satellite targets and background noise, respectively.

[0011] Step 2b) Calculate I for each ISAR image num The corresponding three-dimensional pointing vector r of the distance axis num And the three-dimensional pointing vector d of the Doppler axis num And through r num and d num Calculate I num The corresponding imaging projection matrix H num This yields the set of imaging projection matrices H = {H1, H2, ..., H} corresponding to the ISAR image set I. num ,...,H Num},in:

[0012]

[0013] in,(·) T Indicates the transpose operation;

[0014] Step 2c) The voting projection method is used, and the satellite target is reconstructed in three dimensions by separating the binary map set Itb and the imaging projection matrix set H:

[0015] Step 2c1) Initialize the reconstruction threshold as follows: There are Gp preparatory points PR = {pr1, pr2, ..., pr...} distributed at equal intervals in three-dimensional space.gp ,...,pr Gp}, where pr gp Representing three-dimensional coordinates as The gp-th 3D preparatory point, Gp≥10;

[0016] Step 2c2) Use the imaging projection matrix H num Calculate each 3D preparatory point pr gp In each binary image of target and noise separation Itb num Projection position on And projected energy E gp (num), and select the three-dimensional preparatory point set PR that satisfies The G preparatory points constitute the three-dimensional reconstruction point set Rec = {rec1, rec2, ..., rec...} g ,...,rec G},in:

[0017] E gp (num) = Itb num (x gp,num ,y gp,num )

[0018] Among them, Itb num (x gp,num ,y gp,num ) represents the image Itb num The xth gp,num line y gp,num The column's pixel values, where ∑ represents the accumulation operation, and rec... g Let G represent the g-th 3D reconstruction point, and G represent the point in PR that satisfies... The number of preparatory points, g = 1, 2, ..., G;

[0019] Step 3) Perform 3D segmentation of the satellite target components:

[0020] Step 3a) For each ISAR image I num Semantic segmentation is performed to obtain a set of binary satellite subject images Ib = {Ib1, Ib2, ..., Ib} corresponding to the ISAR image set I. num ,...,Ib Num The set of binary graphs of satellite solar panels is = {Is1, Is2, ..., Is...} num ,...,Is Num}, where Ib num Is num They represent I respectively num The corresponding binary image of the satellite body and the binary image of the satellite solar panel, Ib numPixels with a median value of 1 represent the satellite body, while pixels with a median value of 0 represent background noise or the satellite's solar panels. Is num Pixels with a median value of 1 belong to the satellite's solar panels, while pixels with a median value of 0 belong to background noise or the satellite itself.

[0021] (3b) Perform 3D segmentation of components using the 3D reconstruction point set Rec of the satellite target using Ib and Is:

[0022] Step 3b1) Initialize the binary image of each satellite subject Ib num The confidence coefficient representing the integrity of the satellite's main body shape is: Each satellite solar panel binary image Is num The confidence coefficient representing the integrity of the satellite's solar panel shape is ζ. num The segmentation threshold is ξ;

[0023] Step 3b2) Use the imaging projection matrix H num Calculate rec for each 3D reconstruction point g In each satellite main binary image Ib num Projection position on And in each satellite solar panel binary image Is num Projection position on And calculate point rec g In each satellite main binary image Ib num Projected energy Eb g (num) and Is in each satellite solar panel binary image num Projected energy Es g (num):

[0024] Eb g (num) = Ib num (xb g,num ,yb g,num )

[0025] Es g (num) = Is num (xs g,num ,ys g,num )

[0026] Among them, Ib num (xb g,num ,yb g,num ) represents image Ib num The xb g,num line yb g,num The pixel value of the column, Is num (xs g,num ,ys g,num ) represents the image Is num The xsg,num line ys g,num The pixel values ​​of the column;

[0027] Step 3b3) Calculate rec for each 3D reconstruction point g Subject confidence level Cb g and solar panel confidence level Cs g And reconstruct all points in the point set Rec that satisfy Cb using 3D reconstruction. g ≥Cs g And Cb g Points ≥ξ constitute the three-dimensional point set Bod={bod1,bod2,...,bod gb ,...,bod Gb}, use the 3D reconstructed point set Rec to find all points that satisfy Cs g ≥Cb g And Cs g Points ≥ ξ constitute the three-dimensional point set Sol = {sol1, sol2, ..., sol...} of the satellite solar panel. gs ,...,sol Gs},in:

[0028]

[0029]

[0030] Among them, bod gb This indicates that the gb-th element in Rec satisfies Cb g ≥Cs g And Cb g For any point ≥ξ, Gb represents a point in Rec that satisfies Cb. g ≥Cs g And Cb g The number of points ≥ ξ, gb = 1, 2, ..., Gb, sol gs This indicates that the gs-th element in Rec satisfies Cs. g ≥Cb g And Cs g For points ≥ξ, Gs represents the points in Rec that satisfy Cs. g ≥Cb g And Cs g The number of points ≥ ξ, gs = 1, 2, ..., Gs;

[0031] Step 4) Obtain the estimated satellite attitude and size results from the ISAR image;

[0032] Step 4a) Using principal component analysis, extract the principal axis pointing l0 of the satellite target body from the three-dimensional point set Bod of the satellite body, and extract the wingspan direction l1, plane normal vector l2 and plate width direction l3 of the satellite target solar panel from the three-dimensional point set Sol of the satellite solar panel. The principal axis pointing l0 of the body and the plane normal vector l2 of the solar panel are the attitude of the satellite target.

[0033] Step 4b) Calculate the main axis dimension Len0 of the satellite target using the main axis pointing l0, and calculate the wingspan dimension Len1 and panel width dimension Len2 of the satellite solar panel using the wingspan direction l1 and panel width direction l3 respectively. Len0, Len1, and Len2 are the dimensions of the satellite target.

[0034]

[0035]

[0036]

[0037] Here, dot(·) represents the inner product operation.

[0038] Compared with the prior art, the present invention has the following advantages:

[0039] This invention first performs 3D segmentation of satellite targets in ISAR images to obtain 3D point sets of the main body of the satellite target and 3D point sets of the solar panel. Then, the attitude and size of the satellite target are estimated using the 3D point sets of the main body of the satellite target and the 3D point sets of the solar panel. This avoids the shortcomings of existing technologies that use the 2D geometric features of the target in multiple images to estimate the attitude and size of the satellite target, which are prone to errors when solving non-convex optimization functions. This invention effectively improves the robustness of satellite attitude and size estimation while ensuring estimation accuracy. Attached Figure Description

[0040] Figure 1 This is a flowchart illustrating the implementation of the present invention;

[0041] Figure 2 This is a flowchart illustrating the implementation of the three-dimensional reconstruction of satellite targets according to the present invention;

[0042] Figure 3 This is a tag example diagram of semantic segmentation in this invention;

[0043] Figure 4 This is a diagram of the CGAN generator subnetwork structure of the present invention;

[0044] Figure 5 This is a diagram of the CGAN discriminator subnetwork structure of the present invention; Detailed Implementation

[0045] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0046] Reference Figure 1 The present invention includes the following steps:

[0047] Step 1) Calculate the range resolution and azimuth resolution of the satellite target ISAR image. The range resolution and azimuth resolution represent the vertical length and horizontal length corresponding to each pixel in the ISAR image, respectively.

[0048] Obtain Num ISAR images of size M×N containing satellite targets, I={I1,I2,...,I... num ,...,I Num}, and calculate I for each ISAR image num Distance resolution Δr num and azimuth resolution Δd num :

[0049]

[0050]

[0051] Where Num≥3, M≥100, N≥100, I num This represents the num-th satellite target ISAR image, where num = 1, 2, ..., Num, B num , λ num θ num They represent I respectively num The corresponding pulse width of the transmitted signal, the wavelength of the transmitted signal, and the target's rotation angle relative to the radar during the imaging time; thus, the range resolution set Δr = {Δr1, Δr2, ..., Δr} corresponding to the ISAR image set I is obtained. num ,...,Δr Num} and the azimuth resolution set Δd={Δd1,Δd2,...,Δd num ,...,Δd Num In this embodiment, Num = 15, M = 512, N = 512, B num =10 9 Hz, λ num =0.0375m;

[0052] Step 2) Refer to Figure 2 Perform three-dimensional reconstruction of the satellite target:

[0053] Step 2a) For each ISAR image I num Separate the target from the noise:

[0054] Step 2a1) Calculate I for each ISAR image num The segmentation threshold δ num Existing techniques for calculating the segmentation threshold include the two-dimensional maximum entropy method, the maximum inter-class variance method, or the co-occurrence matrix segmentation method. Given the advantage of the maximum inter-class variance method in its simplicity of implementation, this example uses, but is not limited to, the maximum inter-class variance method. Its calculation formula is as follows:

[0055]

[0056]

[0057]

[0058] in, This represents finding δ that maximizes the objective function. num N1 and N2 represent images I and N2 respectively. num The pixel value is not less than δ num The number of pixels and pixel values ​​are less than δ num The number of pixels, satisfying N1 + N2 = M × N, imb k ,ims h Representing image I num The pixel value is not less than δ num The pixels and pixel values ​​less than δ num The pixels, k = 1, 2, ..., N1, h = 1, 2, ..., N2;

[0059] Step 2a2) with δ num For each ISAR image, the threshold is set as follows: num Binarization segmentation yields a coarsely separated binary image Ita. num Image I num All pixel values ​​greater than δ num Set the pixel of image I to 1. num All pixel values ​​are not greater than δ num The pixel is set to 0;

[0060] Step 2a3) For each ISAR image I num Binarized segmentation result Ita num Morphological operations are performed to obtain a binary image (Itb) separating the target and noise. num :

[0061] Itb num =Close(Open(Ita) num ))

[0062] Where Open(·) represents the morphological opening operation, Close(·) represents the morphological closing operation, and Itb numPixels with median values ​​of 1 and 0 belong to satellite targets and background noise, respectively; thus, the target-noise separation binary image set Itb = {Itb1, Itb2, ..., Itb} corresponding to ISAR image set I is obtained. num ,...,Itb Num};

[0063] Step 2b) The ISAR imaging process can be viewed as a projection of a satellite target in three-dimensional space onto a two-dimensional imaging plane, which is determined by the radar ray direction and the Doppler direction. Reconstructing the target from multiple ISAR images requires knowing the imaging projection matrix of each ISAR image. The following calculation is used to determine the ISAR image... num The corresponding imaging projection matrix H num :

[0064]

[0065]

[0066]

[0067]

[0068]

[0069]

[0070]

[0071]

[0072]

[0073]

[0074]

[0075] Where, r num d num Each ISAR image represents I. num The corresponding three-dimensional pointing vectors of the distance axis and the three-dimensional pointing vectors of the Doppler axis, (·) T The expression represents the transpose operation, fix(·) represents the floor operation, and ||·||2 represents the modulo operation. This represents the partial derivative with respect to m, where Φ and λ represent the geographical longitude and latitude of the radar observation station, respectively, R0 represents the Earth's radius, and los num (:,m) represents image I num The m-th of the corresponding M radar observation rays, m = 1, 2, ..., M, βnum,m α num,m and R num,m Representing image I num The azimuth, elevation, and range corresponding to the m-th radar observation line of sight, β num,m α num,m and R num,m Provided by radar tracking information; thus obtaining the imaging projection matrix set H = {H1, H2, ..., H...} corresponding to the ISAR image set I. num ,...,H Num};

[0076] Step 2c) Use the separated binary map set Itb and the imaging projection matrix set H to perform 3D reconstruction of the satellite target; the voting projection method first projects points in 3D space onto each ISAR image, and then determines whether a point belongs to the target based on whether the projection of the point falls on the target area in the image, thus completing the 3D reconstruction; the target 3D point set obtained by the voting projection method is relatively dense and has high accuracy, therefore the voting projection method is used here to perform 3D reconstruction of the satellite target:

[0077] Step 2c1) Initialize the reconstruction threshold as follows: There are Gp preparatory points PR = {pr1, pr2, ..., pr...} distributed at equal intervals in three-dimensional space. gp ,...,pr Gp}, where pr gp Representing three-dimensional coordinates as The gp-th 3D preparatory point, Gp≥10;

[0078] Step 2c2) Use the imaging projection matrix H num Calculate each 3D preparatory point pr gp In each binary image of target and noise separation Itb num Projection position on And projected energy E gp (num), and select the three-dimensional preparatory point set PR that satisfies The G preparatory points constitute the three-dimensional reconstruction point set Rec = {rec1, rec2, ..., rec...} g ,...,rec G},in:

[0079]

[0080] E gp (num) = Itb num (x gp,num ,y gp,num )

[0081] Where ⊙ represents the Hadamard product operation; Itbnum (x gp,num ,y gp,num ) represents the image Itb num The xth gp,num line y gp,num The column's pixel values, where ∑ represents the accumulation operation, and rec... g Let G represent the g-th 3D reconstruction point, and G represent the point in PR that satisfies... The number of preparatory points, g = 1, 2, ..., G;

[0082] Step 3) Perform 3D segmentation of the satellite target components:

[0083] Step 3a) For each ISAR image I num Semantic segmentation: Existing techniques for image semantic segmentation include segmentation methods based on Conditional Generative Adversarial Network (CGAN), Fully Convolutional Network (FCN), or SegNet. This example uses, but is not limited to, the segmentation method based on Conditional Generative Adversarial Network (CGAN), and its specific implementation steps are as follows:

[0084] Step 3a1) Generate training data: Obtain Numl satellite target ISAR images as the initial training images. Itr = {Itr1, Itr2, ..., Itr} numl ,...,Itr Numl}; For each initial image Itr numl The corresponding labeled image is obtained by marking. numl , markup example Figure 3 :in, Figure 3 (a) is the initial image Itr numl , Figure 3 (b) is Itr numl Corresponding labeled image numl Each initial image Itr is colored with two different colors. numl The satellite body and solar panels in the image can be manually labeled at the pixel level to obtain Itr. numl Corresponding labeled image numl Where Numl ≥ 20, numl = 1, 2, ..., Numl; for each pair of initial images Itr numl and image label numl The training data is expanded by performing translation and mirroring operations; the expanded dataset contains Ke×Numl pairs of initial images and labeled images, where Ke≥12;

[0085] Step 3a2) Construct a CGAN segmentation network containing a generator subnetwork and a discriminator subnetwork; the generator subnetwork is as follows: Figure 4As shown: It includes an encoder consisting of 8 convolutional layers, a decoder consisting of 8 transposed convolutional layers, a prediction layer consisting of 1 convolutional layer, and 8 jumper layers; the discriminator subnetwork is as follows... Figure 5 As shown: It includes 4 convolutional layers and 1 discriminant layer; and the loss function L of the CGAN network is set. CGAN for:

[0086] L CGAN =E Itr,label [logDt(Itr i ,label i )]+E Itr [log(1-Dt(Itr i ,Ge(Itr i ))]

[0087] +λE Itr,label [||label i -Ge(Itr i )||1]

[0088] Where E[·] represents the expected value, Ge(·) represents the prediction result of the generator, Dt(·) represents the discrimination result of the discriminator, λ represents the L1 loss weight coefficient, and ||·||1 represents the L1 norm.

[0089] Step 3a3) Train the CGAN segmentation network: First, train the initial image Itr numl The input is fed into the generator subnetwork to obtain the prediction result Ge(Itr) numl Then the initial image Itr numl and the corresponding labeled image numl The discriminator subnetwork, which inputs the discriminator, obtains the discrimination result Dt(Itr). i ,label i ), and the initial image Itr numl And the predicted result Ge(Itr) numl The discriminator subnetwork, which shares the input, yields the discrimination result Dt(Itr). i ,Ge(Itr i Finally, the value L of the loss function of the Conditional Generative Adversarial Network (CGAN) is calculated. CGAN The network parameters are updated using gradient descent; after multiple iterations, the trained network is obtained when the discriminator can no longer distinguish between the predicted result and the labeled image.

[0090] Step 3a4) Use the trained network to process each ISAR image. num Marking is performed to obtain image I num Corresponding labeled image I la According to image I laDifferent pixel colors generate images I num Corresponding satellite main body binary image Ib num And the binary image of the solar panel Is num Thus, we obtain the set of binary satellite subject images Ib = {Ib1, Ib2, ..., Ib} corresponding to the ISAR image set I. num ,...,Ib Num The set of binary graphs of satellite solar panels is = {Is1, Is2, ..., Is...} num ,...,Is Num};

[0091] Step 3b) Perform 3D segmentation of components using the 3D reconstruction point set Rec of the satellite target based on Ib and Is:

[0092] Step 3b1) Initialize the binary image of each satellite subject Ib num The confidence coefficient representing the integrity of the satellite's main body shape is: Each satellite solar panel binary image Is num The confidence coefficient representing the integrity of the satellite's solar panel shape is ζ. num The segmentation threshold is ξ;

[0093] Step 3b2) Use the imaging projection matrix H num Calculate rec for each 3D reconstruction point g In each satellite main binary image Ib num Projection position on And in each satellite solar panel binary image Is num Projection position on And calculate point rec g In each satellite main binary image Ib num Projected energy Eb g (num) and Is in each satellite solar panel binary image num Projected energy Es g (num):

[0094] Eb g (num) = Ib num (xb g,num ,yb g,num )

[0095] Es g (num) = Is num (xs g,num ,ys g,num )

[0096]

[0097]

[0098] Among them, Ib num (xb g,num ,yb g,num ) represents image Ib num The xb g,num line yb g,num The pixel value of the column, Is num (xs g,num ,ys g,num ) represents the image Is num The xs g,num line ys g,num The pixel values ​​of the column;

[0099] Step 3b3) Calculate rec for each 3D reconstruction point g Subject confidence level Cb g and solar panel confidence level Cs g And reconstruct all points in the point set Rec that satisfy Cb using 3D reconstruction. g ≥Cs g And Cb g Points ≥ξ constitute the three-dimensional point set Bod={bod1,bod2,...,bod gb ,...,bod Gb}, use the 3D reconstructed point set Rec to find all points that satisfy Cs g ≥Cb g And Cs g Points ≥ ξ constitute the three-dimensional point set Sol = {sol1, sol2, ..., sol...} of the satellite solar panel. gs ,...,sol Gs},in:

[0100]

[0101]

[0102] Among them, bod gb This indicates that the gb-th element in Rec satisfies Cb g ≥Cs g And Cb g For any point ≥ξ, Gb represents a point in Rec that satisfies Cb. g ≥Cs g And Cb g The number of points ≥ ξ, gb = 1, 2, ..., Gb, sol gs This indicates that the gs-th element in Rec satisfies Cs. g ≥Cb g And Cs g For points ≥ξ, Gs represents the points in Rec that satisfy Cs. g ≥Cbg And Cs g The number of points ≥ ξ, gs = 1, 2, ..., Gs;

[0103] Step 4) Obtain the estimated satellite attitude and size results from the ISAR image:

[0104] The satellite body is mostly axisymmetric. This invention extracts the main axis of the satellite body to reflect the attitude and size of the satellite body. The satellite solar panel is approximately planar rectangular. This invention extracts the plane normal vector of the solar panel to reflect the attitude of the satellite solar panel, and estimates the wingspan and width of the solar panel to reflect the size of the satellite solar panel.

[0105] Step 4a) Using principal component analysis, extract the principal axis pointing l0 of the satellite target body from the three-dimensional point set Bod of the satellite body. At the same time, extract the wingspan direction l1, plane normal vector l2, and plate width direction l3 of the satellite target solar panel from the three-dimensional point set Sol of the satellite solar panel. The principal axis pointing l0 and the plane normal vector l2 of the solar panel are the attitude of the satellite target. The specific steps are as follows:

[0106] Step 4a1) Calculate the covariance matrix Covb of the satellite target main point set Bod, and calculate the covariance matrix Covs of the satellite target solar panel point set Sol:

[0107]

[0108]

[0109] Step 4a2) Perform eigenvalue decomposition on the covariance matrix Covb and the covariance matrix Covs:

[0110]

[0111]

[0112] Where λ1, λ2, and λ3 are the eigenvalues ​​of Covb, λ4, λ5, and λ6 are the eigenvalues ​​of Covs, and vb1, vb2, vb3, vs1, vs2, and vs3 are the eigenvectors corresponding to the eigenvalues ​​λ1, λ2, λ3, λ4, λ5, and λ6, respectively.

[0113] Step 4a3) Use the eigenvector corresponding to the largest eigenvalue of Covb as the principal axis pointing to l0 of the satellite target body; use the eigenvector corresponding to the largest eigenvalue of Covs as the wingspan direction l1 of the solar panel, and use the eigenvector corresponding to the smallest eigenvalue of Covs as the plane normal vector l2 of the solar panel. Use l1 and l2 to calculate the width direction l3 of the panel:

[0114]

[0115] Where × represents the cross product operation.

[0116] Step 4b) Calculate the main axis dimension Len0 of the satellite target using the main axis pointing l0, and calculate the wingspan dimension Len1 and panel width dimension Len2 of the satellite solar panel using the wingspan direction l1 and panel width direction l3 respectively. Len0, Len1, and Len2 are the dimensions of the satellite target.

[0117]

[0118]

[0119]

[0120] Here, dot(·) represents the inner product operation.

[0121] The technical effects of the present invention will be explained below with reference to simulation experiments.

[0122] 1. Simulation conditions and content:

[0123] The data used in this experiment consisted of 15 ISAR images, each 512×512 pixels, containing the "Tiangong-2" satellite target. The hardware platform was an Intel(R) Core(TM) i7-10750H CPU processor, Windows 10 operating system, and MATLAB R2019b simulation software platform.

[0124] Using the present invention and existing satellite target attitude and size estimation methods based on ISAR image interpretation, 100 Monte Carlo experiments were conducted on the experimental data.

[0125] 2. Simulation Result Analysis:

[0126] This example uses the angle between the estimated principal axis pointing vector or solar panel plane normal vector and the actual principal axis pointing vector or solar panel plane normal vector to measure the error of the satellite target attitude estimation result; this example uses relative size error to measure the estimation error of the satellite principal axis length, satellite solar panel wingspan, and satellite solar panel width. The relative error calculation formula is as follows:

[0127]

[0128] Wherein, |·| represents the absolute value operation; the estimation results of the present invention and the prior art are shown in Table 1.

[0129] Table 1

[0130]

[0131] As can be seen from Table 1, in 100 Monte Carlo experiments, the mean and best values ​​of the satellite attitude and size estimation results of the present invention are not much different from the mean and best values ​​of the satellite attitude and size estimation results of the existing methods. However, the worst value of the satellite attitude and size estimation results of the present invention is much better than the worst value of the satellite attitude and size estimation results of the existing methods. It can be seen that the method of the present invention can improve the robustness of the algorithm and avoid erroneous estimation while ensuring the accuracy of satellite attitude and size estimation.

[0132] The above description is merely a specific example of the present invention and does not constitute any limitation on the present invention. Obviously, those skilled in the art, after understanding the content and principles of the present invention, may make various modifications and changes in form and details without departing from the principles and structure of the present invention. However, these modifications and changes based on the ideas of the present invention are still within the scope of protection of the claims of the present invention.

Claims

1. A method for estimating satellite attitude and size in ISAR images based on component 3D segmentation, characterized in that... Includes the following steps: (1) Calculate the range resolution and azimuth resolution of the satellite target ISAR image: Get The size of the area containing the satellite target is ISAR images And calculate each ISAR image Distance resolution and azimuth resolution To obtain the range resolution set and azimuth resolution set ,in, , , , Indicates the first A satellite target ISAR image, ; (2) Perform three-dimensional reconstruction of the satellite target: (2a) with Threshold for each ISAR image Perform binarization segmentation and then... Morphological operations are performed on the binarized segmentation results to obtain an ISAR image set. The corresponding set of separated binary graphs ,in, Indicates to Coarsely separated binary image obtained by binarization segmentation The corresponding binary image of target and noise separation. The pixel locations with median values ​​of 1 and 0 belong to satellite targets and background noise, respectively. (2b) Calculate each ISAR image The corresponding three-dimensional pointing vector of the distance axis and the three-dimensional pointing vector of the Doppler axis and through and calculate Corresponding imaging projection matrix The ISAR image set was obtained. The corresponding set of imaging projection matrices ,in: ； in, Indicates the transpose operation; (2c) The voting projection method is adopted, and the binary graph set is separated. and the set of imaging projection matrices Perform 3D reconstruction of the satellite target: (2c1) Initialize the reconstruction threshold as follows: Equally spaced in three-dimensional space One preparatory point ,in, Representing three-dimensional coordinates as The Three-dimensional preparatory points, 10; (2c2) Using the imaging projection matrix Calculate each 3D preparatory point In each binary image of target and noise separation Projection position on and projected energy And select a three-dimensional preparatory point set. China satisfies of The set of three-dimensional reconstruction points consists of several preparatory points. ,in: ； in, Representing an image No. Line number Column pixel values, This indicates an accumulation operation. Indicates the first A three-dimensional reconstruction point, express China satisfies The number of reserve points, ; (3) Perform three-dimensional segmentation of satellite target components: (3a) For each ISAR image Semantic segmentation is performed to obtain an ISAR image set. The corresponding set of binary images of the satellite body Collection of binary images of satellite solar panels ,in, , They represent The corresponding binary image of the satellite body and the binary image of the satellite solar panel. Pixels with a median value of 1 represent the satellite body, while pixels with a median value of 0 represent background noise or the satellite's solar panels. Pixels with a median value of 1 belong to the satellite's solar panels, while pixels with a median value of 0 belong to background noise or the satellite itself. (3b) Utilize and 3D reconstruction point set of satellite target Perform 3D segmentation of components: (3b1) Initialize the binary image of each satellite subject The confidence coefficient representing the integrity of the satellite's main body shape is: Binary images of each satellite solar panel The confidence coefficient representing the integrity of the satellite's solar panel shape is: The segmentation threshold is ; (3b2) Using an imaging projection matrix Calculate each 3D reconstruction point In each satellite main binary image Projection position on And in each satellite solar panel binary image Projection position on and calculate the points In each satellite main binary image Projected energy And in each satellite solar panel binary image Projected energy : ； ； in, Representing an image No. Line number Column pixel values, Representing an image No. Line number The pixel values ​​of the column; (3b3) Calculate each 3D reconstruction point Subject confidence and solar panel confidence And reconstruct the point set in three dimensions All of the above satisfy and The points constitute the three-dimensional point set of the satellite body. Reconstructing point sets in 3D All of the above satisfy and The points constitute the three-dimensional point set of the satellite solar panel. ,in: ； ； in, express The Middle One satisfies and point, express China satisfies and The number of points, , express The Middle One satisfies and point, express China satisfies and The number of points, ; (4) Obtain the estimation results of satellite attitude and size from ISAR images; (4a) Using principal component analysis, from the three-dimensional point set of the satellite body Extracting the main axis direction of the satellite target body Simultaneously, from the three-dimensional point set of the satellite solar panels Extract the wingspan direction of the satellite target's solar panel. Plane normal vector and the width direction of the plate The main axis points to and the normal vector of the solar panel plane This refers to the attitude of the satellite target; (4b) Using the main axis pointing Calculate the dimensions of the main axis of the satellite target body. Using wingspan direction and the width direction of the plate Calculate the wingspan of the satellite's solar panels separately. and board width dimensions , , and That is, the size of the satellite target: ； ； ； in, This indicates the inner product operation.

2. The method for estimating satellite attitude and size in ISAR images based on component 3D segmentation according to claim 1, characterized in that, The calculation of each ISAR image in step (1) Distance resolution and azimuth resolution The calculation formulas are as follows: ； ； in, Represents the speed of light. , , They represent the first ISAR images The corresponding pulse width of the transmitted signal, the wavelength of the transmitted signal, and the angle of rotation of the target relative to the radar during the imaging time.

3. The method for estimating satellite attitude and size in ISAR images based on component 3D segmentation according to claim 1, characterized in that, The threshold mentioned in step (2a) The calculation can be performed using the two-dimensional maximum entropy method, the maximum inter-class variance method, or the co-occurrence matrix partitioning method. Binary image of target and noise separation The calculation formula is: ； in, This indicates a morphological opening operation. This indicates a morphological closing operation.

4. The method for estimating satellite attitude and size in ISAR images based on component 3D segmentation according to claim 1, characterized in that, The calculation of each ISAR image described in step (2b) The corresponding three-dimensional pointing vector of the distance axis and the three-dimensional pointing vector of the Doppler axis The calculation formulas are as follows: ； ； ； ； ； ； ； ； ； ； in, This indicates the rounding operation. This indicates a modulo operation. Indicates a request for information about The partial derivatives, , These represent the geographical longitude and geographical latitude of the radar observation station, respectively. Represents the Earth's radius. Representing an image corresponding The first radar observation ray in the indivual, , , and Representing images respectively The The azimuth, elevation, and range corresponding to each radar observation line of sight , and Provided by radar tracking information.

5. The method for estimating satellite attitude and size in ISAR images based on component 3D segmentation according to claim 4, characterized in that, The use of the imaging projection matrix described in step (2c2) Calculate each 3D preparatory point In each binary image of target and noise separation Projection position on The calculation formula is as follows: ； in, This indicates the Hadamard product operation.

6. The method for estimating satellite attitude and size in ISAR images based on component 3D segmentation according to claim 4, characterized in that, The step (3a) described for each ISAR image Semantic segmentation is performed using segmentation methods based on conditional adversarial generative networks (CGAN), fully convolutional networks (FCN), or SegNet.

7. The method for estimating satellite attitude and size in ISAR images based on component 3D segmentation according to claim 4, characterized in that, The use of an imaging projection matrix as described in step (3b4) Calculation points In each satellite main binary image Projection position on And in each satellite solar panel binary image Projection position on The steps are as follows: 。 8. The method for estimating satellite attitude and size in ISAR images based on component 3D segmentation according to claim 1, characterized in that, The principal component analysis method described in step (4a) is used to analyze the three-dimensional point set of the satellite body. Extracting the main axis direction of the satellite target body From the three-dimensional point set of satellite solar panels Extract the wingspan direction of the satellite target's solar panel. Plane normal vector and the width direction of the plate The steps are as follows: (4a1) Calculate the main point set of satellite targets covariance matrix Calculate the solar panel point set of the satellite target. covariance matrix : ； ； (4a2) For the covariance matrix Perform eigenvalue decomposition on the covariance matrix. Perform eigenvalue decomposition: ； ； in, , , for eigenvalues, , , for eigenvalues, , , , , , Eigenvalues , , , , , The corresponding feature vector; (4a3) will The eigenvector corresponding to the largest eigenvalue is used as the principal axis pointing to the satellite target. ;Will The eigenvector corresponding to the largest eigenvalue is used as the wingspan direction of the solar panel. ,Will The eigenvector corresponding to the smallest eigenvalue is used as the plane normal vector of the solar panel. ,use and Calculation board width direction : ； in, This indicates the cross product operation.

Images