A method for extracting three-dimensional features of a space target component

Abstract

A kind of space target component three-dimensional feature extraction method, obtains the radar echo signal of three base stations, obtains ISAR two-dimensional image by range-doppler processing, carries out image registration to ISAR two-dimensional image and converts image to gray scale chart, calculates the best threshold of gray scale chart, converts gray scale chart into binary image, dilates image, hole filling, obtains the boundary of binary image, detects linear line segment on boundary image, obtains the beginning and end coordinates and direction of the line segment, searches the rectangular feature parameters of component in binary image, obtains radar line of sight to imaging plane projection matrix by interference to three ISAR images, searches target component in radar imaging plane projection information, obtains the real size of target component in space in combination with radar imaging plane projection matrix.This application solves the problems of missing scattering point information, target occlusion, etc., and improves the three-dimensional attitude information extraction capability of space target.

Description

Technical Field

[0001] This invention relates to a method for extracting three-dimensional features of space target components, and more particularly to a method for extracting three-dimensional features of space target components based on interferometric ISAR images. Background Technology

[0002] With increasing emphasis and investment in space utilization by various countries, the scale of space activities continues to expand. Precise attitude measurement of space targets and critical payload components can effectively estimate the operational status of space targets, enabling real-time monitoring and attitude prediction of non-cooperative space targets. Traditional target attitude measurement methods are divided into active and passive measurement. Active measurement requires the spacecraft to be equipped with a dedicated measurement system, resulting in high system complexity; while attitude measurement using photoelectric sensors is easily affected by factors such as weather and time. Currently, the acquisition of space target attitude information involves estimating the projection angle from a two-dimensional image sequence and inverting it to three-dimensional space. However, the unknown projection angle introduces uncertainty into the target inversion, resulting in significant errors and greatly limiting target identification capabilities. Summary of the Invention

[0003] The purpose of this invention is to provide a method for extracting three-dimensional features of space target components, which solves problems such as missing scattering point information and target occlusion, and improves the ability to extract three-dimensional attitude information of space targets.

[0004] To achieve the above objectives, the present invention provides a method for extracting three-dimensional features of spatial target components, comprising:

[0005] Step S1: Obtain radar echo signals from three base stations, perform range-Doppler processing, acquire ISAR two-dimensional images, perform image registration on the ISAR two-dimensional images, and convert the images into grayscale images;

[0006] Step S2: Calculate the optimal threshold of the grayscale image using the Otsu method (maximum inter-class variance method), convert the grayscale image into a binary image based on the optimal threshold, and then perform dilation and hole filling on the image.

[0007] Step S3: Use the Sobel edge detection algorithm to obtain the boundary of the binary image, use Hough transform to detect linear line segments on the boundary image, and obtain the start and end coordinates and direction of the line segment;

[0008] Step S4: Based on the Hough transform detection results, use an optimization algorithm to search for the rectangular feature parameters of the component in the binary image;

[0009] Step S5: Interferometric analysis of the three ISAR images to obtain the radar elevation and azimuth angles, and to obtain the projection matrix from the radar line of sight to the imaging plane;

[0010] Step S6: Use the particle swarm optimization algorithm to search for the projection information of the target component on the radar imaging plane, and combine it with the radar imaging plane projection matrix to obtain the true three-dimensional spatial dimensions of the target component.

[0011] Step S1 includes:

[0012] Step S1.1: Set up a radar to transmit electromagnetic wave signals and set the radar position as the origin O of the spatial coordinate system. Set up three radar base stations A, B, and C to receive the signals.

[0013] The signals received by each base station are acquired and range-Doppler signal processing is performed. Assuming A is the main channel, the ISAR two-dimensional image domain expression is as follows:

[0014]

[0015] Where λ is the wavelength of the electromagnetic wave, γ is the modulation frequency, and f r and f a These represent the range and azimuth frequencies, respectively. Let R be the position of the p-th scattering point. ref For reference distance, and These are the distances from antennas A and O at the initial time, respectively;

[0016] Taking the interference of channels A and B as an example, the differences between the three channels for the same scattering point are as follows:

[0017]

[0018] in, The positions of the same scattering point in the A and B channel images are Δf, respectively. r , Δf a Δ represents the frequency shift of the scattering point's distance and orientation across channels A and B. n Δ m R represents the positional offset of the scattering point's distance and orientation in channels A and B. A0 R B0 Let ω be the distances from antennas A and B to O at the initial moments. z The target rotational speed;

[0019] Step S1.2: Optimize the solution of rotation offset ω using the average distance minimum entropy method. z The entropy of an image is minimized when the image envelope does not move within a distance cell.

[0020] ω z =argmin(-∑g(ω) z_est )·lng(ω z_est ))

[0021] Where, ω z_est For ω z Estimated value, g(ω) z_est ) for using ω z_est The integral of the range image envelope amplitude value along the azimuth direction after envelope correction;

[0022] Using channel A as a reference, mismatch compensation is performed on channel B. The phase form of the compensation offset is as follows:

[0023] Δ comp (n,m)=exp(j2π / NΔ n ·n)exp(j2π / MΔ m ·m)

[0024] Where n and m are the discrete sampling points in the range and azimuth directions of the image, respectively, and N and M are the total number of samples. A matrix is ​​constructed in the time domain for the two compensated channels. When the image scattering points of the two channels are perfectly aligned, for The image energy is maximized when the scattering points obtained by performing a two-dimensional Fourier transform are superimposed.

[0025] The registration method for channels A and C is the same as that for channels A and B, thus obtaining the registered three-channel grayscale image.

[0026] Step S2 includes:

[0027] Step S2.1: Use a grayscale image (BMP) A For example, the Otsu's method is used to calculate BMP. A The optimal threshold TH for the image;

[0028] The image is divided into two parts: background and target, denoted as C1 (less than TH) and C2 (greater than TH). The probabilities of each pixel in the image being classified as background or target are p1 and p2, respectively. The maximum inter-class variance is:

[0029]

[0030] Where THG is the global mean. This represents the cumulative mean when the gray level is m. The m at which the maximum variance is obtained is the OTSU threshold TH.

[0031] Step S2.2: Convert the grayscale image into a binary image according to the OTSU threshold, and perform dilation and hole filling on the image to obtain binary images BI1, BI2, and BI3.

[0032] Step S3 includes:

[0033] Step S3.1: Use the Sobel edge detection algorithm to obtain the target binary boundary image after processing in step S2;

[0034] Step S3.2: Use Hough transform to obtain the linear line segments in the boundary image, including the start and end coordinates and direction of the line segments;

[0035] Convert the Cartesian coordinate system to the polar coordinate system:

[0036] ρ=xcosθ+ysinθ

[0037] In the boundary image, each pixel (x, y) in the Cartesian coordinate system on the edge corresponds to a sine curve in the polar coordinate system (ρ, θ). After performing a Hough transform on all edge points, the point (ρ0, θ0) with the most intersection in the polar coordinate system is extracted, and the longest line segment l of the edge is found. L Obtain the initial and final coordinates (x1, y1) and (x2, y2), and the direction of the line segment.

[0038] Step S4 includes:

[0039] Step S4.1: Based on the Hough transform results, use the rectangle descriptor to search for the rectangle features of the component;

[0040] The shorter side and longer side of the rectangle are perpendicular to each other. The initial coordinates of the longer side (x1, y1) are taken as the initial coordinates of the shorter side. The search continues with the search for the coordinates of the end point of the shorter side (x1, y1). i ,y i ):

[0041] minl D -l(x i ,y i )

[0042]

[0043] Among them, Fp i To search for the corresponding rectangular region of pixels when searching for the shorter side, l D Let l(x) be the length of the shorter side of the rectangle of the component. i ,y i ) represents the search length of the shorter side of the component rectangle, n i n i-1 These represent the number of pixels within the rectangular component coverage area of ​​the current search value and the previous search value, respectively.

[0044] Step S4.2: After determining the length of the short side, search for the start and end points of the long side to determine the four corner points of the target rectangular component, thereby obtaining the two-dimensional dimensions and orientation of the target component.

[0045] Step S5 includes:

[0046] Step S5.1: After performing phase correction on the three-channel radar echo signal obtained in step S1, the three-channel phase form is expressed as follows:

[0047]

[0048]

[0049]

[0050] Among them, R ΔAPO R ΔBPO R ΔCPO These are the signal path histories received by the three receiving radars, R. ref The set reference distance;

[0051] The original interference phase difference of the three channels is:

[0052]

[0053]

[0054] Theoretically, the phase difference between the three channels is:

[0055] Δφ AB =2πL1 sinα / λ

[0056] Δφ AC =2πL2sinβ / λ

[0057] Where L1 and L2 are the baseline lengths between channels AB and AC, α is the azimuth angle, and β is the elevation angle;

[0058] The interference phase is unwrapped using the dereference plane method. A reference plane is selected, tangent to the target's motion direction and perpendicular to the line of sight of antenna A. The actual interference phase φ′ is obtained by subtracting the interference phase on the reference plane corresponding to the slant range from the original interference phase and taking the principal value. AB , φ′ AC ;

[0059] Taking channels A and B as examples, the interference phase after removing the reference plane phase is:

[0060] φ A ′ B (R ΔAP0 )=φ(R ΔAP0 )-φ g (R ΔAP0 )+2kπ;

[0061] Where k is an integer, and its value satisfies -π < φ′ AB ≤π,φ(RΔAP0 (R) represents the distance difference. ΔAP0 The corresponding true phase, φ g (R ΔAP0 (R) represents the distance difference. ΔAP0 The phase on the corresponding reference plane, R ΔAP0 This is the difference between the slant range of antenna A corresponding to this pixel and the reference distance;

[0062] Solving for the X, Y, and Z positions yields:

[0063]

[0064] Simultaneously, the radar line-of-sight angle parameters (α, β) are obtained;

[0065] Step S5.2: Based on the angle obtained in step 5.1, construct the radar imaging plane projection matrix:

[0066]

[0067]

[0068]

[0069] Where, ρ r For distance dimension resolution, ρ a It is a Doppler resolution.

[0070] Step S6 includes:

[0071] Step S6.1: After obtaining the two-dimensional features of the target component, the three-dimensional projection pose parameters are obtained through a minimum optimization problem based on the component's rectangular prior, with the cost function being:

[0072]

[0073]

[0074]

[0075] in, 0°≤ψ≤180° To determine the orientation of the target linear structure, a classical particle swarm optimization algorithm is used to solve for the projection angle of the target component on the imaging plane. The optimal solution;

[0076] Step S6.2: Based on the comparison between the projection information of the generated target component in the image and the extracted two-dimensional information, determine the true size of the target component in three dimensions:

[0077]

[0078]

[0079] Among them, l L l D These are the actual size parameters of the target rectangular component, l L-RD l D-RD These are the dimensional parameters of the extracted target rectangular component projected onto the ISAR imaging plane.

[0080] The present invention has the following beneficial effects:

[0081] 1. Based on ISAR images acquired from different angles by multi-channel radar, interferometric phase information is extracted, the elevation and azimuth angles of the radar are derived, the projection matrix on the radar imaging plane is calculated, and interferometric radar is used for imaging. While ensuring resolution, it is possible to search for target projection parameters using only the interferometric image at a single moment, without the need for sequential imaging, thereby obtaining the target's three-dimensional attitude parameters.

[0082] 2. Using ISAR images for 3D attitude inversion has lower system complexity compared to traditional active attitude measurement methods, and has the advantage of being available all day and all weather compared to photoelectric measurement methods.

[0083] 3. By utilizing the multi-view information of interferometric radar, the problem of inaccurate size estimation caused by traditional two-dimensional imaging projection can be solved, and attitude estimation can be achieved through a single high-resolution three-dimensional imaging.

[0084] 4. Based on the component information extracted from the two-dimensional image, the particle swarm optimization algorithm is used to search for the target body coordinate projection angle and solve the projection matrix. This non-parametric search method is beneficial for obtaining the attitude parameters of non-cooperative targets with limited prior information. Attached Figure Description

[0085] Figure 1 This is a flowchart of a three-dimensional feature extraction method for a space target component provided by the present invention.

[0086] Figure 2 It is a geometric diagram of the InISAR radar layout and a diagram of the three-dimensional angle calculation relationship.

[0087] Figure 3a To register the first three channels of ISAR images

[0088] Figure 3b These are three-channel ISAR images after registration.

[0089] Figure 4a This is a feature extraction map of target components in an ISAR image.

[0090] Figure 4b It is the extracted target component interference phase diagram.

[0091] Figure 5a It is a three-dimensional inversion diagram of the entire target.

[0092] Figure 5b It is the extracted 3D inversion image of the target component. Detailed Implementation

[0093] The following is based on Figure 1 Figure 5 illustrates a preferred embodiment of the present invention.

[0094] Developing a method for extracting 3D features of space target components based on interferometric ISAR images can improve the accuracy of space target estimation and has practical value. Using radar sensors for attitude measurement allows for all-weather, 24 / 7 operation.

[0095] like Figure 1 As shown, the present invention provides a method for extracting three-dimensional features of space target components, comprising the following steps:

[0096] Step S1: Obtain radar echo signals from three base stations, perform range-Doppler processing, acquire ISAR two-dimensional images, perform image registration on the ISAR two-dimensional images, and convert the images into grayscale images;

[0097] Step S1.1: Set up a radar to transmit electromagnetic wave signals and set the radar position as the origin O of the spatial coordinate system. Set up three radar base stations A, B, and C to receive the signals.

[0098] The signals received by each base station are acquired and range-Doppler signal processing is performed. Assuming A is the main channel, the ISAR two-dimensional image domain expression is as follows:

[0099]

[0100] Where λ is the wavelength of the electromagnetic wave, γ is the modulation frequency, and f r and f a These represent the range and azimuth frequencies, respectively. Let R be the position of the p-th scattering point. ref For reference distance, and These are the distances from antennas A and O at the initial time, respectively;

[0101] It can be seen that, for the same scattering point, the differences between the three channels are as follows (taking the interference of channels A and B as an example):

[0102]

[0103] in, The positions of the same scattering point in the A and B channel images are Δf, respectively. r , Δfa Δ represents the frequency shift of the scattering point's distance and orientation across channels A and B. n Δ m R represents the positional offset of the scattering point's distance and orientation in channels A and B. A0 R B0 Let ω be the distances from antennas A and B to O at the initial moments. z The target rotational speed;

[0104] Step S1.2: Optimize the solution of rotation offset ω using the average distance minimum entropy method. z The entropy of an image is minimized when the image envelope does not move within a distance cell.

[0105] ω z =argmin(-∑g(ω) z_est )·lng(ω z_est ))

[0106] Where, ω z_est For ω z Estimated value, g(ω) z_est ) for using ω z_est The integral of the range image envelope amplitude value along the azimuth direction after envelope correction;

[0107] Using channel A as a reference, mismatch compensation is performed on channel B. The phase form of the compensation offset is as follows:

[0108] Δ comp (n,m)=exp(j2π / NΔ n ·n)exp(j2π / MΔ m ·m)

[0109] Where n and m are the discrete sampling points in the range and azimuth directions of the image, respectively, and N and M are the total number of samples. A matrix is ​​constructed in the time domain for the two compensated channels. When the image scattering points of the two channels are perfectly aligned, for The image energy is maximized when the scattering points obtained by performing a two-dimensional Fourier transform are superimposed.

[0110] The registration method for channels A and C is the same as that for channels A and B, thus obtaining the registered three-channel grayscale images;

[0111] Step S2: Calculate the optimal threshold of the grayscale image using the Otsu's method (OTSU), convert the grayscale image into a binary image based on the optimal threshold, and then perform dilation and hole filling on the image.

[0112] Step S2.1: Use a grayscale image (BMP) A For example, the Otsu's method is used to calculate BMP. AThe optimal threshold TH for the image;

[0113] The image is divided into two parts: background and target, denoted as C1 (less than TH) and C2 (greater than TH). The probabilities of each pixel in the image being classified as background or target are p1 and p2, respectively. The maximum inter-class variance is:

[0114]

[0115] Where THG is the global mean. This represents the cumulative mean when the gray level is m. The m at which the maximum variance is obtained is the OTSU threshold TH.

[0116] Step S2.2: Convert the grayscale image into a binary image according to the OTSU threshold, and perform dilation and hole filling on the image to obtain binary images BI1, BI2, and BI3;

[0117] Step S3: Use the Sobel edge detection algorithm to obtain the boundary of the binary image, use Hough transform to detect linear line segments on the boundary image, and obtain the start and end coordinates and direction of the line segment;

[0118] Step S3.1: Use the Sobel edge detection algorithm to obtain the target binary boundary image after processing in step S2;

[0119] Step S3.2: Use Hough transform to obtain linear line segments in the boundary image, including the start and end coordinates and direction of the line segment;

[0120] Convert the Cartesian coordinate system to the polar coordinate system:

[0121] ρ=xcosθ+ysinθ

[0122] In the boundary image, each pixel (x, y) in the Cartesian coordinate system on the edge corresponds to a sine curve in the polar coordinate system (ρ, θ). After performing a Hough transform on all edge points, the point (ρ0, θ0) with the most intersection in the polar coordinate system is extracted, and the longest line segment l of the edge is found. L Obtain the initial and final coordinates (x1, y1) and (x2, y2), and the direction of the line segment.

[0123] Step S4: Based on the Hough transform detection results, use an optimization algorithm to search for the rectangular feature parameters of the component in the binary image;

[0124] Step S4.1: Based on the Hough transform results, use the rectangle descriptor to search for the rectangle features of the component;

[0125] The shorter side and longer side of the rectangle are perpendicular to each other. The initial coordinates of the longer side (x1, y1) are taken as the initial coordinates of the shorter side. The search continues with the search for the coordinates of the end point of the shorter side (x1, y1). i,y i ):

[0126] minl D -l(x i ,y i )

[0127]

[0128] Among them, Fp i To search for the corresponding rectangular region of pixels when searching for the shorter side, l D Let l(x) be the length of the shorter side of the rectangle of the component. i ,y i ) represents the search length of the shorter side of the component rectangle, n i n i-1 These represent the number of pixels within the rectangular component coverage area of ​​the current search value and the previous search value, respectively.

[0129] Step S4.2: After determining the length of the short side, search for the start and end points of the long side (the principle is the same as in step 4.1) to determine the four corner points of the target rectangular component, thereby obtaining the two-dimensional dimensions and orientation of the target component.

[0130] Step S5: Interferometric analysis of the three ISAR images to obtain the radar elevation and azimuth angles, and to obtain the projection matrix from the radar line of sight to the imaging plane;

[0131] Step S5.1: After performing phase correction on the three-channel radar echo signal obtained in step S1, the three-channel phase form is expressed as follows:

[0132]

[0133]

[0134]

[0135] Among them, R ΔAPO R ΔBPO R ΔCPO These are the signal path histories received by the three receiving radars, R. ref The set reference distance;

[0136] The original interference phase difference of the three channels is:

[0137]

[0138]

[0139] like Figure 2 Given the geometric relationship, the theoretical phase difference between the three channels is:

[0140] Δφ AB =2πL1sinα / λ

[0141] Δφ AC =2πL2sinβ / λ

[0142] Where L1 and L2 are the baseline lengths between channels AB and AC, α is the azimuth angle, and β is the elevation angle;

[0143] As can be seen from the above formula, the path difference between the two antennas is much larger than the wavelength, that is, the true value of the phase difference between the channels is much larger than 2π, resulting in phase entanglement.

[0144] The interference phase is unwrapped using the dereference plane method. A reference plane is selected, tangent to the target's motion direction and perpendicular to the line of sight of antenna A. The actual interference phase φ′ is obtained by subtracting the interference phase on the reference plane corresponding to the slant range from the original interference phase and taking the principal value. AB , φ′ AC ;

[0145] Taking channels A and B as examples, the interference phase after removing the reference plane phase is φ. A ′ B (R ΔAP0 )=φ(R ΔAP0 )-φ g (R ΔAP0 )+2kπ;

[0146] Where k is an integer, and its value satisfies -π < φ A ′ B ≤π,φ(R ΔAP0 (R) represents the distance difference. ΔAP0 The corresponding true phase, φ g (R ΔAP0 (R) represents the distance difference. ΔAP0 The phase on the corresponding reference plane, R ΔAP0 This is the difference between the slant range of antenna A corresponding to this pixel and the reference distance;

[0147] according to Figure 2 Given the angular relationships, the X, Y, and Z positions can be calculated as follows:

[0148]

[0149] Simultaneously, the radar line-of-sight angle parameters (α, β) are obtained;

[0150] Step S5.2: Based on the angle obtained in step 5.1, construct the radar imaging plane projection matrix:

[0151]

[0152]

[0153]

[0154] Where, ρ r For distance dimension resolution, ρ a Doppler resolution;

[0155] Step S6: Use the particle swarm optimization algorithm to search for the projection information of the target component on the radar imaging plane, and combine it with the radar imaging plane projection matrix to obtain the true three-dimensional spatial dimensions of the target component.

[0156] Step S6.1: After obtaining the two-dimensional features of the target component, the three-dimensional projection pose parameters can be obtained through a minimum optimization problem based on the component's rectangular prior, with the cost function being:

[0157]

[0158]

[0159]

[0160] in, 0°≤ψ≤180° To determine the orientation of the target linear structure, a classical particle swarm optimization algorithm is used to solve for the projection angle of the target component on the imaging plane. The optimal solution;

[0161] Step S6.2: Based on the comparison between the projection information of the generated target component in the image and the extracted two-dimensional information, determine the true size of the target component in three dimensions:

[0162]

[0163]

[0164] Among them, l L l D These are the actual size parameters of the target rectangular component, l L-RD l D-RD These are the dimensional parameters of the extracted target rectangular component projected onto the ISAR imaging plane.

[0165] Figure 3a To register the first three channels of ISAR images. Figure 3b Figure 4 shows the component feature extraction information and component interferometric phase map, and Figure 5 shows the three-dimensional inversion map of the target and components. From... Figure 3a and Figure 3b It can be seen that, with channel A as the reference, the positions of the three channels are basically the same after registration. Figure 4aThe projected length of the longer side onto the imaging plane is 5.5449 m, and the angle between the target's attitude and the imaging plane is 29.1659°. Figure 5a The estimated length of the middle and long side is 6.35m, and the estimated angle between the target attitude and the imaging plane is 29.9172°. The calculated projection length of the long side is 5.5038m, with an error of 0.0411m.

[0166] This invention addresses the need for acquiring the three-dimensional attitude of space targets by proposing a method for extracting three-dimensional features of space target components based on interferometric ISAR images. By utilizing the angular information extracted from the interferometric images, the three-dimensional attitude of the target components is inverted, solving problems such as missing scattering point information and target occlusion, and improving the ability to extract three-dimensional attitude information of space targets.

[0167] The innovation and advantages of this invention are as follows:

[0168] 1. Based on ISAR images acquired from different angles by multi-channel radar, interferometric phase information is extracted, the elevation and azimuth angles of the radar are derived, the projection matrix on the radar imaging plane is calculated, and interferometric radar is used for imaging. While ensuring resolution, it is possible to search for target projection parameters using only the interferometric image at a single moment, without the need for sequential imaging, thereby obtaining the target's three-dimensional attitude parameters.

[0169] 2. Using ISAR images for 3D attitude inversion has lower system complexity compared to traditional active attitude measurement methods, and has the advantage of being available all day and all weather compared to photoelectric measurement methods.

[0170] 3. By utilizing the multi-view information of interferometric radar, the problem of inaccurate size estimation caused by traditional two-dimensional imaging projection can be solved, and attitude estimation can be achieved through a single high-resolution three-dimensional imaging.

[0171] 4. Based on the component information extracted from the two-dimensional image, the particle swarm optimization algorithm is used to search for the target body coordinate projection angle and solve the projection matrix. This non-parametric search method is beneficial for obtaining the attitude parameters of non-cooperative targets with limited prior information.

[0172] It should be noted that, in the embodiments of the present invention, the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing the embodiments and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the present invention. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0173] Although the present invention has been described in detail through the preferred embodiments above, it should be understood that the above description should not be considered as a limitation of the present invention. Various modifications and substitutions to the present invention will be apparent to those skilled in the art after reading the above description. Therefore, the scope of protection of the present invention should be defined by the appended claims.

Claims

1. A method for three-dimensional feature extraction of space object components, characterized in that, The application relates to a method for obtaining three-dimensional information of a target component, and belongs to the field of radar target recognition. Step S1: radar echo signals of three base stations are acquired, distance-Doppler processing is carried out, ISAR two-dimensional images are acquired, image registration is carried out on the ISAR two-dimensional images, and the images are converted into gray-scale images; Step S2: the optimal threshold of the gray-scale images is calculated by adopting an OTSU method, the gray-scale images are converted into binary images according to the optimal threshold, and the images are inflated and hole filling is carried out; Step S3: the boundary of the binary images is acquired by adopting a sobel edge detection algorithm, linear line segments on the boundary images are detected by adopting a Hough transformation, and the start and end coordinates and the pointing direction of the line segments are acquired; Step S4: according to the Hough transformation detection result, an optimization algorithm is adopted to search for the component rectangular feature parameters in the binary images; Step S4.1: according to the Hough transformation result, a rectangular descriptor is used to search for the component rectangular feature; The short side and the long side of the rectangle are perpendicular to each other, the initial coordinate (x1, y1) of the long side is taken as the initial coordinate of the short side, and the end coordinate (x i ,y i ) of the short side is searched. min l D -l(x i ,y i ) s.t. Wherein, Fp i is the pixel point of the corresponding rectangular region when searching the short side, l D is the length of the short side of the component rectangle, l(x i ,y i ) is the search length of the short side of the component rectangle, n i , n i-1 are the number of pixel points in the rectangular component coverage range of the current search value and the previous search value, respectively; Step S4.2, after determining the short side length, the start and end points of the long side are searched to determine the four corner points of the target rectangular component, and the two-dimensional size of the target component and the direction are obtained Step S5: three ISAR images are interfered to obtain radar elevation angles and azimuth angles, and a radar line-of-sight projection matrix to the imaging plane is obtained; Step S6: a particle swarm optimization algorithm is used to search for the target component in the radar imaging plane projection information, and the target component space three-dimensional real size is acquired in combination with the radar imaging plane projection matrix.

2. The method of claim 1, wherein, The step S1 comprises: Step S1.1: a radar is arranged to emit electromagnetic wave signals, and the radar position is set as an original point O of a space coordinate; three radar base stations A, B and C are arranged to receive the signals; The signals received by the base stations are acquired, distance-Doppler signal processing is carried out, and the A base station is taken as an example; the ISAR two-dimensional image domain expression is as follows: where λ is the wavelength of the electromagnetic wave, γ is the frequency modulation, and f r and f a denote the range and azimuth directions, respectively. bit frequency, R is the position of the pth scattering point, ref R is the reference distance, and R is the distance from the antenna A and O, respectively, at the initial time. The differences among the three channels for the same scattering point are as follows: wherein, respectively the position of the same scattering point in the A-channel and B-channel images, Δf r , Δf a the frequency offset of the scattering point distance and orientation on the A, B two channels, Δ n , Δ m the position offset of the scattering point distance and orientation on the A, B two channels, R A0 , R B0 respectively the distance of antennas A, B to O at the initial moment, ω z the target rotation speed; Step S1.2, the rotation offset ω is solved by using the average distance minimum entropy method z The entropy value of the image is minimum when the image envelope does not move within a distance cell. ω z = argmin(-∑g(ω z_est )·lng(ω z_est )) Where, ω z_est For ω z Estimated value, g(ω) z_est ) for using ω z_est The integral of the range image envelope amplitude value along the azimuth direction after envelope correction; The B channel is mismatched and compensated according to the A channel; the phase form of the compensation offset is as follows: Δ comp (n,m) = exp(j2π / NΔ n ·n) exp(j2π / MΔ m ·m) where n and m are the discrete sampling points in the range and azimuth directions, respectively, and N and M are the total number of samples. The compensated two channels are constructed into a matrix in the time domain When the two channels of image scattering points are completely aligned, the image energy is maximum. The two-dimensional Fourier transform is performed on the image scattering points to obtain an image stack. The registration methods of the A and C channels are the same as those of the A and B channels, and thus the three-channel gray-scale images after registration are obtained.

3. The method of claim 2, wherein, The step S2 comprises: Step S2.1, in gray-scale image BMP A For example, the optimal threshold TH of the BMP image is calculated by the maximum inter-class variance method. A For example, the optimal threshold TH of the BMP image is calculated by the maximum inter-class variance method. The image is divided into a background and a target, and C1 is less than TH and C2 is greater than TH; the probabilities that each pixel in the image is divided into the background and the target are p1 and p2 respectively; and the maximum inter-class variance is as follows: wherein THG is a global mean value, represents the cumulative mean value when the gray level is m, and m when the maximum variance is obtained is the OTSU threshold TH. Step S2.2: the gray-scale images are converted into binary images according to the OTSU threshold value; the images are inflated and hole filling is carried out, and the binary images BI1, BI2 and BI3 are obtained.

4. The method of claim 3, wherein, The step S3 comprises: Step S3.1: the sobel edge detection algorithm is used to acquire the target binary boundary image processed in the step S2; Step S3.2: the Hough transformation is used to acquire the linear line segments in the boundary image, including the start and end coordinates and the pointing direction of the line segments; The Cartesian coordinate system is converted into a polar coordinate system: The step S5 comprises: Each pixel point (x, y) in the Cartesian coordinate system on the edge of the boundary image is mapped to a corresponding sinusoidal curve in the polar coordinate system (p, q). After the Hough transform of all the edge points, the point (p0, q0) with the most intersections in the polar coordinate system is extracted, and the longest edge segment l of the edge is calculated L , to obtain the start and end coordinates (x1, y1) and (x2, y2), and the line segment direction 5. The method of claim 4, wherein, Step S5.1: after the three-channel radar echo signals acquired in the step S1 are phase-corrected, the three-channel phase form is as follows: The original interference phase difference of the three channels is as follows: wherein R ΔAPO , R ΔBPO , R ΔCPO are the signal path histories of the three receiving radars, respectively, and R ref is the set reference distance; The theoretical phase difference of the three channels is as follows: Wherein, L1 and L2 are the baseline lengths between the AB and AC channels, alpha is the azimuth angle, and beta is the elevation angle; Δφ AB = 2πL1sinα / λ Δφ AC = 2πL2sinβ / λ The interference phase after the reference surface phase is removed is as follows: The unwrapping of the interferometric phase is performed using the unwrapping method of the reference plane, selecting the direction tangent to the target motion direction and perpendicular to the antenna A line of sight as the reference plane, subtracting the original interferometric phase from the interferometric phase on the reference plane corresponding to the slant range, and taking the principal value to obtain the actual interferometric phase AB φ' AC ; The X, Y and Z positions are solved as follows: φ' AB (R ΔAP0 ) = φ(R ΔAP0 )- φ g (R ΔAP0 )+ 2kπ; wherein k is an integer, and its value satisfies -π < φ' < π AB ≤ π, φ(R ΔAP0 ) is the phase corresponding to the real distance difference R ΔAP0 (R g ) is the phase on the reference surface corresponding to the real distance difference R ΔAP0 (R ΔAP0 ) is the phase on the reference surface corresponding to the real distance difference R ΔAP0 is the difference between the slant distance of the pixel corresponding to the antenna A and the reference distance; ​ At the same time, the radar line-of-sight angle parameters (α, β) are obtained; Step S5.2, constructing the radar imaging plane projection matrix according to the angle obtained in step S5.1: where p r is the range dimension resolution, p a is the Doppler dimension resolution.

6. The method of claim 5, wherein: The step S6 includes: Step S6.1, after obtaining the two-dimensional feature of the target component, the three-dimensional projection pose parameters are obtained through a minimum optimization problem based on the component rectangular prior, and the cost function is: wherein 0° < ψ < 180°, The classical particle swarm optimization algorithm is used to solve the optimal solution of the projection angle of the target component in the imaging plane as the target linear structure posture direction Step S6.2, comparing the projection information of the generated target component in the image with the extracted two-dimensional extraction information, determining the real size of the target component in three dimensions: wherein, l L , l D are real size parameters of the target rectangular component, respectively, l L-RD , l D-RD are size parameters of the extracted target rectangular component projected to the ISAR imaging plane, respectively.

Images