On-orbit attitude and rotation parameter inversion method for slow-rotating space target

Abstract

The application discloses a method for in-orbit attitude and rotation parameter inversion of a slow-rotating space target, comprising the following steps: acquiring an ISAR image sequence of radar echo data; modeling an ISAR projection model; constructing a training data set based on the ISAR projection model to train a pre-designed deep learning network, and using the trained network to extract a target region sequence from the ISAR image sequence; obtaining a projection region sequence corresponding to the ISAR image sequence based on the projection model, and establishing an optimization objective function in combination with the target region sequence; and solving the optimization objective function to obtain optimal target attitude parameters and rotation parameters. The method guarantees the network training quality, improves the accuracy of ISAR image target region extraction, and can jointly analyze the in-orbit attitude and rotation parameters of the target, realizes the in-orbit attitude and rotation parameter inversion of the slow-rotating space target, and has stronger feasibility and robustness in actual application.

Description

Technical Field

[0001] This invention belongs to the field of radar technology, specifically relating to a method for inverting the on-orbit attitude and rotation parameters of a slowly spinning space target. Background Technology

[0002] Inverse Synthetic Aperture Radar (ISAR) is one of the most effective methods for imaging and observing space targets. It continuously observes space targets over a long period and at a wide angle by emitting a series of broadband electromagnetic pulse signals, and performs range pulse compression and azimuth coherent accumulation on the echo signals to obtain a high-resolution two-dimensional image of the observed target. However, the two-dimensional ISAR image obtained by this method is merely a projection of the three-dimensional structure of the space target onto the radar imaging plane. Due to the limitation of the observation dimension, the two-dimensional information obtained from ISAR image interpretation cannot accurately describe the true state of the target in three-dimensional space. With the increasing frequency of aerospace activities worldwide and the growing complexity of the space environment, simple two-dimensional ISAR image interpretation is insufficient to meet the information requirements for fine situational awareness in complex space environments. Therefore, a detailed study of on-orbit attitude inversion methods for space targets based on ISAR images is necessary.

[0003] Currently, on-orbit attitude inversion methods for space targets based on ISAR images can be broadly classified into two categories. One category is attitude inversion methods based on traditional image processing feature extraction. These methods utilize traditional image processing techniques to obtain the target's ISAR image sequence features, derive the projection relationship between the target's attitude parameters and the ISAR imaging results, generate a target model projection sequence, and adjust the model's attitude parameters to match the model's projection feature sequence with the ISAR image feature sequence, thereby obtaining the optimal estimate of the target's attitude parameters. Some researchers have transformed radar observations in the Earth-Centered Inertial Coordinates (ECI) system to the centroid coordinate system and used the Radon transform to extract target edge features.

[0004] Another type is the pose estimation method based on deep learning feature extraction. This method introduces deep learning into ISAR image feature extraction, utilizing the network's powerful nonlinear mapping capabilities to establish a mapping relationship between ISAR images and image features, achieving accurate and automatic extraction of image features. Some researchers have proposed ISAR image keypoint feature extraction networks based on residual neural networks to extract key points on target structures in ISAR images. Key points are linked pairwise in a fixed order to form the projection feature vectors of the target's key structures. Other researchers have used pix2pix networks to extract typical component regions of targets in ISAR images and combined this with principal component analysis for region principal direction estimation to obtain the projection feature vectors of the extracted components. After completing the image projection sequence feature extraction, under the constraint of the projection relationship of a three-axis stable target, the corresponding three-dimensional vectors of the projection feature sequence are solved in reverse to estimate the size and orientation of the target structure and components.

[0005] However, for the first type of attitude inversion method based on traditional image processing feature extraction, the anisotropy of electromagnetic scattering in ISAR images makes feature extraction difficult, especially under low signal-to-noise ratio conditions, limiting the accuracy of the extraction and restricting its application in real-world scenarios. For the second type of attitude estimation method based on deep learning feature extraction, a large number of training samples need to be manually labeled during network training, and the accuracy of this labeling largely determines the network performance after training. Considering the sparse distribution and incomplete structure of scattering centers in ISAR images, the quality of network training is difficult to guarantee; furthermore, this method is limited to three-axis stable targets and is not applicable to fixed-axis slowly rotating targets. Summary of the Invention

[0006] To address the instability in target region extraction and the inability to analyze the on-orbit attitude of slowly rotating space targets in existing technologies, this invention provides a method for inverting the on-orbit attitude and rotation parameters of slowly rotating space targets. The technical problem to be solved by this invention is achieved through the following technical solution:

[0007] A method for inverting the on-orbit attitude and rotation parameters of a slowly rotating space target, comprising:

[0008] Step 1: Acquire ISAR image sequences from radar echo data;

[0009] Step 2: Model the ISAR projection model;

[0010] Step 3: Construct a training dataset based on the ISAR projection model to train the pre-designed deep learning network, and use the trained network to extract the target region from the ISAR image sequence to obtain the target region sequence;

[0011] Step 4: Obtain the projection region sequence corresponding to the ISAR image sequence based on the ISAR projection model, and establish an optimization objective function in conjunction with the target region sequence;

[0012] Step 5: Solve the optimization objective function to obtain the optimal target attitude parameters and rotation parameters.

[0013] The beneficial effects of this invention are:

[0014] 1. This invention first establishes an ISAR projection model, then uses the model's attitude parameters as a training set to train a deep learning network, ensuring network training quality and improving the accuracy of target region extraction from ISAR images. Next, it uses the ISAR projection model to obtain a projection region sequence, and establishes an optimization objective function in conjunction with the target region sequence obtained by the deep learning network. Finally, based on the ISAR imaging principle, an optimization algorithm is used to jointly analyze the target's on-orbit attitude and rotation parameters, simultaneously realizing the inversion of the on-orbit attitude and rotation parameters of slowly rotating space targets. This method does not limit the on-orbit attitude of space targets to be three-axis stable, solving the problems of instability in target region extraction and inability to analyze the on-orbit attitude of slowly rotating space targets in traditional attitude estimation methods, and has stronger feasibility and robustness in practical applications.

[0015] 2. This invention selects the pix2pix network as an ascending learning network to obtain the target region sequence, and utilizes the powerful nonlinear mapping capability of the pix2pix network to further enhance the accuracy of target region extraction from ISAR images.

[0016] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating a method for inverting the on-orbit attitude and rotation parameters of a slowly rotating space target according to an embodiment of the present invention.

[0018] Figure 2 This is a schematic diagram of the projection relationship of ISAR imaging of a space target provided in an embodiment of the present invention;

[0019] Figure 3 This is a spatial target model according to an embodiment of the present invention;

[0020] Figure 4 These are images showing imaging results at different signal-to-noise ratios provided in embodiments of the present invention.

[0021] Figure 5 These are partial images of the training dataset in this embodiment of the invention;

[0022] Figure 6The target extraction results using the method of this invention are compared with the target extraction results using the traditional method;

[0023] Figure 7 The initial pose estimation results are obtained using the traditional contour matching method at a signal-to-noise ratio of 0 dB.

[0024] Figure 8 The results show the target rotation vector estimation using the traditional contour matching method at a signal-to-noise ratio of 0 dB.

[0025] Figure 9 The initial attitude estimation result is obtained using the method of this invention at a signal-to-noise ratio of 0 dB.

[0026] Figure 10 The result is the target rotation vector estimation using the method of this invention at a signal-to-noise ratio of 0 dB. Detailed Implementation

[0027] The present invention will be further described in detail below with reference to specific embodiments, but the implementation of the present invention is not limited thereto.

[0028] Example 1

[0029] Please see Figure 1 , Figure 1 This is a flowchart illustrating a method for inverting the on-orbit attitude and rotation parameters of a slowly rotating space target according to an embodiment of the present invention, which includes:

[0030] Step 1: Obtain ISAR image sequences of radar echo data.

[0031] In this embodiment, step 1 specifically includes:

[0032] 11) Receive the inverse synthetic aperture radar echo of space targets and divide the echo data into several frames.

[0033] 12) Perform high-speed compensation, range compression, and translational compensation operations sequentially on each frame of echo data;

[0034] 13) Based on the data obtained in step 12), a high-resolution two-dimensional ISAR image sequence is obtained using the RD algorithm.

[0035] Specifically, firstly, the wideband echo acquired by the radar receiver over a long period of time at a large angle is divided into several frames. Then, envelope alignment is performed on each frame of echo data using the adjacent correlation method to eliminate envelope offset caused by the target's translational motion relative to the radar. Next, a self-focusing algorithm based on the minimum entropy criterion is used to compensate for the initial phase error caused by translational motion. Finally, the RD algorithm is used to obtain a high-resolution two-dimensional ISAR image sequence.

[0036] The specific implementation process of high-speed compensation, distance compression and translational compensation operations can be referred to the existing related technologies, and will not be described in detail in this embodiment.

[0037] Step 2: Model the ISAR projection model.

[0038] 21) Establish the equivalent rotation vector relationship relative to the radar line of sight during the target imaging process.

[0039] For details, please see Figure 2 , Figure 2 This is a schematic diagram of the projection relationship of ISAR imaging of a space target provided in an embodiment of the present invention, wherein the equivalent rotation vector relative to the radar line of sight during the target imaging process is... Rotation of radar line of sight and the rotation of the target itself Synthesis. ISAR performs long-term continuous observations of targets to obtain... A sequence of images composed of frames of ISAR images. For each frame of ISAR image, the midpoint of its CPI is the imaging time, and the radar line of sight at that time is called the imaging line of sight. Let the [frame name] be the [frame name]. Frame radar imaging line of sight is The radar rotation vector is Then we have the rotation vector relation:

[0040] ;

[0041] In the formula, Indicates the first step in the target imaging process The equivalent rotation vector of a frame radar image relative to the radar line of sight. This indicates that the target itself rotates and remains unchanged during observation. Indicates the first The radar rotation vector for frame radar imaging.

[0042] 22) Obtain effective rotation components based on equivalent rotation vector relationships.

[0043] Specifically, according to the Doppler generation principle, rotation perpendicular to the radar line of sight causes the scattering center to shift relative to the radar line of sight, thus causing Doppler modulation of the echo from the scattering center. Rotation along the radar line of sight does not produce Doppler modulation on the echo. Therefore, The rotation component perpendicular to the line of sight is called the first... The effective rotation components of a frame are then calculated using the following formula:

[0044] ;

[0045] In the formula, T This indicates transpose.

[0046] 23) Derive the attitude rotation matrix between the body attitude target and the initial attitude target, and the rotation matrix corresponding to the target rotation vector.

[0047] Specifically, let Indicates the first on the target The coordinate vectors of the scattering centers are called the initial attitude at the start of the ISAR imaging observation. Then, there is a rotational transformation relationship between the main target attitude and the initial attitude target.

[0048] ;

[0049] In the formula, Indicates the target's first position under the initial attitude. The coordinate vectors of the scattering centers This is the attitude rotation matrix, which is derived from the initial attitude angles. , , The values ​​are determined in the following order: roll angle around the X-axis, pitch angle around the Y-axis, and yaw angle around the Z-axis.

[0050] Then the calculation relationship is:

[0051] ;

[0052] In the formula, , , These are rotation matrices around the X-axis, Y-axis, and Z-axis, respectively.

[0053] Furthermore, considering that the target itself orbits a fixed axis The rotation of , then in the , At the time of image imaging, the target's first frame The coordinate vectors of the scattering centers are calculated as follows:

[0054] ;

[0055] In the formula, Let represent the angle through which the target rotates under the influence of the rotation vector from the initial moment to the imaging moment. Indicates the time from the initial moment to the th The time interval between the imaging moments of a frame image, also known as the first frame. At the frame imaging time, we have:

[0056] ;

[0057] Let the rotation matrix corresponding to the target rotation vector be:

[0058] ;

[0059] In the formula, It is the identity matrix. express , and

[0060] .

[0061] 24) Based on the effective rotation component Attitude rotation matrix and the rotation matrix corresponding to the target rotation vector Get the target number The formula for calculating the theoretical projection position of each scattering center on the imaging plane is:

[0062] ;

[0063] In the formula, For the goal of The projected coordinates of the scattering centers on the imaging plane The wavelength of the radar signal. , These represent the Doppler resolution unit and the range resolution unit, respectively. Indicates the first The imaging projection matrix of the frame. Indicates the first on the target The coordinate vectors of the scattering centers.

[0064] make The pulse repetition frequency of the radar signal. This represents the number of coherently accumulated echoes in a single frame of ISAR imaging. Represents the speed of light. Represents signal bandwidth, then Medium resolution unit

[0065] .

[0066] Step 3: Construct a training dataset based on the ISAR projection model to train the pre-designed deep learning network, and use the trained network to extract target regions from the ISAR image sequence to obtain the target region sequence.

[0067] 31) ISAR data simulation is performed using actual test scenarios or PO electromagnetic calculation methods to obtain training ISAR images.

[0068] Specifically, ISAR images are obtained by using ISAR images from actual measured scenarios or by simulating ISAR data using electromagnetic calculation methods, combined with the Polar Format Algorithm (PFA).

[0069] Meanwhile, while ensuring that the image orientation does not overlap, the target rotation direction and speed are randomly set, and different levels of noise are added to the ISAR image pulse compression data to serve as training ISAR images.

[0070] 32) Based on the ISAR projection model established in step 2, generate target projection region sequences for the training ISAR images as their labels to obtain the training set.

[0071] In this embodiment, it is preferable to use the target model and attitude parameters to generate the target projection region sequence as the label, that is, to use the projection model constructed in step 2 to obtain the target projection region sequence of the training ISAR image in step 31) as its label.

[0072] Alternatively, labels can be directly applied to the training ISAR images in step 31) using manual labeling to obtain the training set.

[0073] 33) Input the training set into a pre-designed deep learning network for training.

[0074] Preferably, this embodiment selects the existing pix2pix network as the deep learning network, which has a strong nonlinear mapping capability and can enhance the accuracy of target region extraction in ISAR images.

[0075] It is understood that this embodiment can also select other existing network structures as deep learning neural networks or design its own network structure.

[0076] Furthermore, the training process of the network can be implemented with reference to existing related technologies, and will not be described in detail in this embodiment.

[0077] 34) Input the two-dimensional ISAR image sequence into the trained network to extract the target region and obtain the target region sequence.

[0078] Specifically, by further processing the ISAR image sequence to be identified obtained in step 1 using the trained neural network, the corresponding target region sequence can be obtained.

[0079] Step 4: Obtain the projection region sequence corresponding to the ISAR image sequence based on the ISAR projection model, and establish an optimization objective function in conjunction with the target region sequence.

[0080] Optionally, this embodiment utilizes the characteristic that when the true 3D points of a spatial target are projected onto an ISAR image sequence, the projected sequence and the target region sequence achieve optimal matching and the maximum intersection-over-union (IoU) ratio, using the IoU ratio of the target region extraction sequence and the target region projection sequence as the optimization function. It is understood that other objective functions besides the IoU ratio can also be used to optimize the target attitude and rotation parameters.

[0081] The following section will take intersection-union ratio as an example to explain the specific implementation process of step 4 in detail.

[0082] Specifically, step 4 includes:

[0083] 41) Obtain the projection region sequence corresponding to the ISAR image sequence based on the ISAR projection model established in step 2.

[0084] First of all, let The first in the model The coordinates of each point are denoted as the model point cloud coordinate matrix. Then there is And calculate the first The model point cloud coordinate distribution at the imaging time of the frame ISAR image is calculated using the following formula:

[0085] ;

[0086] In the formula, Let be the attitude rotation matrix between the main body attitude target and the initial attitude target. This is the rotation matrix corresponding to the target rotation vector.

[0087] Then, the target model is calculated at the 1st... The set of projected coordinate points of the imaging plane of the frame ISAR image is:

[0088] ;

[0089] In the formula, Indicates the first The imaging projection matrix of the frame.

[0090] Next, according to The first point cloud model is computed. Projecting the nth point onto the nth... The pixel coordinates behind the imaging plane of the frame image.

[0091] Specifically, for For an ISAR image of a certain size, if the projected coordinates of the region exceeding the image pixels are left blank, then the first... Projecting the nth point onto the nth... The pixel coordinates behind the imaging plane of the frame image are:

[0092] ;

[0093] In the formula, Indicates rounding down; for the model Perform the same processing on each point to obtain the set of model projection pixel coordinates under the pose estimation. .

[0094] Finally, regarding the first Set of projected pixel coordinates of a frame ISAR image Binarization is performed to obtain a binarized model projection image. .

[0095] Specifically, in the Within the pixel coordinate range of the frame ISAR image, according to The coordinates of the pixels indicate the location, and the pixel value at the corresponding pixel position is set to 1. Pixel values ​​at other positions without coordinates are uniformly set to 0. After assigning pixel values, a binarized model projection image is obtained.

[0096] .

[0097] Because the target point cloud is densely distributed, after the projection image is generated, the model's projected pixel regions are automatically connected into a connected region.

[0098] 42) Calculate the intersection-union ratio (IUU) of the projected region sequence and the target region sequence. The expression is:

[0099] ;

[0100] In the formula, The first part represents the target region sequence. Frame Image and the first of the projection region sequence Frame Image The intersection and union ratio, Representing an image and images The number of pixels that intersect in the target regions. Representing an image and images The number of pixels in the union of the target regions;

[0101] 43) Using the intersection-union ratio (IUGR) as the matching degree between the target extraction region and the target projection region of the two images, the target region sequence matching degree corresponding to the pose parameter estimation value is obtained, and it is used as the optimization objective function, the expression of which is:

[0102] ;

[0103] In the formula, This represents the estimated values ​​of attitude and rotation parameters. The total number of image frames in the target region sequence or the projection region sequence.

[0104] Step 5: Solve the objective function to obtain the optimal target attitude parameters and rotation parameters.

[0105] Optionally, as one implementation, this embodiment uses the PSO (Particle Swarm Optimization) algorithm to optimize the objective function and search for the optimal attitude estimation parameters. Specifically:

[0106] (a) Initialize the PSO algorithm:

[0107] The number of particles in the PSO algorithm is set to Randomly initialize the position vectors of each particle, and denote the position vector of the first particle as... The position vectors of the particles are This vector has six parameters: the first three represent the Euler angle parameters of the target's initial attitude, and the last three represent the target's rotation vector parameters; the six-dimensional velocity parameters of the particle are randomly initialized, denoted as the... The velocity parameters of the particles are: Let vector and Record the optimal fitness of each particle and the position parameters corresponding to the optimal fitness of all particles during the iteration process, and initialize them. and Set a counter With the algorithm iteration termination count threshold ;

[0108] (b) According to the first The position parameters of each particle, and the initial attitude parameters are decomposed and combined. and target rotation parameters The attitude parameters are substituted into the target point cloud distribution calculation function to generate the model attitude sequence. ;

[0109] (c) According to the first The attitude parameters of each particle are used to calculate the projection matrix at each imaging time. The model pose sequence is then projected onto the corresponding imaging plane to generate the target projection sequence. ;

[0110] (d) Calculate the matching degree between the target extraction sequence and each group of target projection sequences, and use it as the fitness of the particle, denoted as the... The result of the group sequence calculation is ;

[0111] (e) Update based on the fitness of each particle. and ;

[0112] like Then let ;like Then let , ,in For set The index corresponding to the maximum value; otherwise. No updates will be made. And update the values ​​of each particle according to the following formula. and ;

[0113] ;

[0114] In the formula, Indicates being between Random numbers between , It is a learning factor;

[0115] (f) Repeat steps (b)-(d) until the condition is met. Then output The corresponding optimal attitude parameters are estimated and used to form the initial attitude estimation parameters. and target rotation estimation parameters .

[0116] Finally, according to and It can determine the attitude change sequence of a target slowly rotating around a fixed axis during the imaging observation period and complete attitude estimation.

[0117] This invention first establishes an ISAR projection model, and then uses the model's attitude parameters as a training set to train a deep learning network, ensuring network training quality and improving the accuracy of target region extraction from ISAR images. Next, it uses the ISAR projection model to obtain a sequence of projected regions, and establishes an optimization objective function by combining the target region sequence obtained from the deep learning network. Finally, based on the ISAR imaging principle, it employs an optimization algorithm to jointly analyze the target's on-orbit attitude and rotation parameters, realizing the inversion of the on-orbit attitude and rotation parameters of a slowly rotating space target.

[0118] The beneficial effects of the present invention will be verified and explained through simulation experiments below.

[0119] Specifically, with Figure 3 Taking the space target model shown as an example, simulation tests were conducted using the method of this invention and the traditional method, respectively. The results are shown in [reference]. Figures 4-10 As shown.

[0120] Figure 4 The images provided in this embodiment of the invention are imaging results under different signal-to-noise ratios; wherein, (a) is an image with a signal-to-noise ratio of 20dB; and (b) is an image with a signal-to-noise ratio of 0dB. Figure 5 These are partial images of the training dataset in this embodiment of the invention.

[0121] Figure 6 Figure 1 shows the target extraction results using the method of the present invention and the target extraction results using the traditional method; wherein, (a) Figure 2 shows the target contour extraction results using the traditional method at a signal-to-noise ratio of 20dB; (b) Figure 3 shows the target contour extraction results using the traditional method at a signal-to-noise ratio of 0dB; (c) Figure 4 shows the target region extraction results using the method of the present invention at a signal-to-noise ratio of 20dB; and (d) Figure 5 shows the target region extraction results using the method of the present invention at a signal-to-noise ratio of 0dB.

[0122] from Figure 6 The results show that the target region extraction results of the method of the present invention are better than those of the traditional method under both high signal-to-noise ratio and low signal-to-noise ratio conditions. Furthermore, the method of the present invention can still accurately extract the target region under low signal-to-noise ratio conditions.

[0123] Figures 7-10 This is a comparison chart showing the estimation results of attitude parameters of a fixed-axis slowly rotating spatial target using the method of this invention and the estimation results using traditional methods; wherein, Figure 7 The initial pose estimation results are obtained using the traditional contour matching method at a signal-to-noise ratio of 0 dB. Figure 8 The results show the target rotation vector estimation using the traditional contour matching method at a signal-to-noise ratio of 0 dB. Figure 9 The initial attitude estimation result is obtained using the method of this invention at a signal-to-noise ratio of 0 dB. Figure 10 The result is the target rotation vector estimation using the method of this invention at a signal-to-noise ratio of 0 dB.

[0124] contrast Figures 7-10 As can be seen, the initial attitude estimation and target rotation vector estimation results of the method of the present invention are better than those of the traditional method under low signal-to-noise ratio conditions. Furthermore, the method of the present invention can still accurately estimate the initial attitude and rotation vector of the target under low signal-to-noise ratio conditions.

[0125] The above description, in conjunction with specific preferred embodiments, provides a further detailed explanation of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, various simple deductions or substitutions can be made without departing from the concept of the present invention, and all such modifications and substitutions should be considered within the scope of protection of the present invention.

Claims

1. A method for inverting the on-orbit attitude and rotation parameters of a slowly rotating space target, characterized in that, include: Step 1: Acquire ISAR image sequences from radar echo data; Step 2: Model the ISAR projection model; Step 3: Construct a training dataset based on the ISAR projection model to train the pre-designed deep learning network, and use the trained network to extract the target region from the ISAR image sequence to obtain the target region sequence; Step 4: Obtain the projection region sequence corresponding to the ISAR image sequence based on the ISAR projection model, and establish an optimization objective function in conjunction with the target region sequence; Step 5: Solve the optimization objective function to obtain the optimal target attitude parameters and rotation parameters.

2. The method for inverting the on-orbit attitude and rotation parameters of a slowly rotating space target according to claim 1, characterized in that, Step 1 includes: 11) Receive the inverse synthetic aperture radar echo of space targets and divide the echo data into several frames; 12) Perform high-speed compensation, range compression, and translational compensation operations sequentially on each frame of echo data; 13) Based on the data obtained in step 12), a high-resolution two-dimensional ISAR image sequence is obtained using the RD algorithm.

3. The method for inverting the on-orbit attitude and rotation parameters of a slowly rotating space target according to claim 1, characterized in that, Step 2 includes: 21) Establish the equivalent rotation vector relationship relative to the radar line of sight during the target imaging process, and its expression is: ； In the formula, Indicates the first step in the target imaging process The equivalent rotation vector of a frame radar image relative to the radar line of sight. This represents the rotation vector of the target itself. Indicates the first The radar rotation vector for frame radar imaging; 22) Based on the equivalent rotation vector relationship, the effective rotation component is obtained, and its expression is: ； In the formula, Indicates the first The effective rotation component of the frame, Indicates the first Frame radar imaging line of sight, T Indicates transpose; 23) Derive the attitude rotation matrix between the body attitude target and the initial attitude target, and the rotation matrix corresponding to the target rotation vector. The expressions are as follows: ； ； In the formula, Let be the attitude rotation matrix between the main body attitude target and the initial attitude target. , , The initial attitude angles are the roll angle around the X-axis, the pitch angle around the Y-axis, and the yaw angle around the Z-axis, respectively. Let the rotation matrix be the rotation matrix corresponding to the target rotation vector. Let be the angle through which the target has rotated under the action of the rotation vector from the initial moment to the imaging moment. It is the identity matrix. express , and ； 24) Based on the effective rotation component The attitude rotation matrix and the rotation matrix corresponding to the target rotation vector. Get the target number The formula for calculating the theoretical projection position of each scattering center on the imaging plane is: ； In the formula, For the goal of The projected coordinates of the scattering centers on the imaging plane The wavelength of the radar signal. , These represent the Doppler resolution unit and the range resolution unit, respectively. Indicates the first The imaging projection matrix of the frame. Indicates the first on the target The coordinate vectors of the scattering centers.

4. The method for inverting the on-orbit attitude and rotation parameters of a slowly rotating space target according to claim 2, characterized in that, Step 3 includes: 31) ISAR data simulation is performed using actual measured scenarios or PO electromagnetic calculation methods to obtain training ISAR images; 32) Generate target projection region sequences as labels for the training ISAR images based on the ISAR projection model established in step 2, thereby obtaining the training set; 33) Input the training set into a pre-designed deep learning network for training; 34) Input the two-dimensional ISAR image sequence into the trained network to extract the target region and obtain the target region sequence.

5. The method for inverting the on-orbit attitude and rotation parameters of a slowly rotating space target according to claim 1, characterized in that, In step 3, the pre-designed deep learning network is a pix2pix network.

6. The method for inverting the on-orbit attitude and rotation parameters of a slowly rotating space target according to claim 1, characterized in that, Step 4 includes: 41) Obtain the projection region sequence corresponding to the ISAR image sequence based on the ISAR projection model established in step 2; 42) Calculate the intersection-union ratio (IUU) of the projected region sequence and the target region sequence, expressed as: ； In the formula, The first part represents the target region sequence. Frame Image and the first of the projection region sequence Frame Image The intersection and union ratio, Representing an image and images The number of pixels that intersect in the target regions. Representing an image and images The number of pixels in the union of the target regions; 43) Using the intersection-union ratio as the matching degree between the target extraction region and the target projection region of the two images, the target region sequence matching degree corresponding to the pose parameter estimation value is obtained, and it is used as the optimization objective function, the expression of which is: ； In the formula, This represents the estimated values ​​of attitude and rotation parameters. The total number of image frames in the target region sequence or the projection region sequence.

7. The method for inverting the on-orbit attitude and rotation parameters of a slowly rotating space target according to claim 1, characterized in that, Step 5 includes: The PSO algorithm is used to optimize the objective function and search for the optimal attitude estimation parameters.

8. The method for inverting the on-orbit attitude and rotation parameters of a slowly rotating space target according to claim 7, characterized in that, The objective function is optimized using the PSO algorithm, including: (a) Initialize the PSO algorithm: The number of particles in the PSO algorithm is set to Randomly initialize the position vectors of each particle, and denote the position vector of the first particle as... The position vectors of the particles are This vector has six parameters: the first three represent the Euler angle parameters of the target's initial attitude, and the last three represent the target's rotation vector parameters; the six-dimensional velocity parameters of the particle are randomly initialized, denoted as the... The velocity parameters of the particles are: Let vector and Record the optimal fitness of each particle and the position parameters corresponding to the optimal fitness of all particles during the iteration process, and initialize them. and Set a counter With the algorithm iteration termination count threshold ; (b) According to the first The position parameters of each particle, and the initial attitude parameters are decomposed and combined. and target rotation parameters The attitude parameters are substituted into the target point cloud distribution calculation function to generate the model attitude sequence. ; (c) According to the first The attitude parameters of each particle are used to calculate the projection matrix at each imaging time. The model pose sequence is then projected onto the corresponding imaging plane to generate the target projection sequence. ; (d) Calculate the matching degree between the target extraction sequence and each group of target projection sequences, and use it as the fitness of the particle, denoted as the... The result of the group sequence calculation is ; (e) Update based on the fitness of each particle. and ; like Then let ;like Then let , ,in For set The index corresponding to the maximum value; otherwise. No updates will be made. And update the values ​​of each particle according to the following formula. and ; ； In the formula, Indicates being between Random numbers between , It is a learning factor; (f) Repeat steps (b)-(d) until the condition is met. Then output The corresponding optimal attitude parameters are estimated and used to form the initial attitude estimation parameters. and target rotation estimation parameters .

Images