Space debris typical inertia feature extraction method based on X-ray simulation

Abstract

The application discloses a space debris typical inertia feature extraction method based on X-ray simulation, solves the problem that traditional visible light imaging and radar detection means cannot accurately acquire the internal structure of space debris, thereby leading to inaccurate inertia estimation, and the method is based on an image dataset under an X-ray source, combines deep learning and a rigid body dynamics model to perform trajectory estimation, and when dynamic analysis is performed on the space debris, the motion trajectory and speed of the space debris can be accurately estimated.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of artificial intelligence and machine learning, and particularly relates to a space debris typical inertia feature extraction method based on X-ray simulation. BACKGROUND

[0002] With the increase of space activities, the threat of space debris to spacecraft is becoming increasingly serious. Traditional monitoring methods (such as visible light imaging and radar detection) can only obtain surface information of the debris, and it is difficult to analyze the internal structure and mass distribution of the debris, resulting in insufficient dynamic prediction accuracy. Existing simulation-based modeling methods mostly rely on visible light images, which cannot penetrate objects, limiting the accurate estimation of the disintegration characteristics and inertia parameters of the debris.

[0003] X-ray imaging technology can obtain internal density distribution information of an object due to its penetration advantage, providing a new idea for dynamic analysis of the debris. However, existing X-ray simulation methods are mostly concentrated in the medical or industrial fields, and have not been effectively applied to dynamic prediction of space debris. SUMMARY

[0004] The purpose of the present application is to solve the problem that traditional visible light imaging and radar detection methods cannot accurately obtain the internal structure of space debris, resulting in inaccurate inertia estimation, and to provide a space debris typical inertia feature extraction method based on X-ray simulation.

[0005] In order to achieve the above purpose, the present application adopts the following technical scheme:

[0006] A space debris typical inertia feature extraction method based on X-ray simulation, characterized in that it comprises the following steps:

[0007] Step 1: Simulate space debris models of different types, materials, shapes and sizes randomly generated under X-ray source, and obtain single-class space debris images; superimpose the gray levels of the single-class space debris images to obtain multi-class space debris images;

[0008] Step 2: Rotate and project the single-class space debris images and the multi-class space debris images respectively to obtain projection images, and respectively perform back projection on the projection images to obtain single-class space debris back projection images and multi-class space debris back projection images;

[0009] Step 3: Perform original path reverse superposition on the single-class space debris back projection images and the multi-class space debris back projection images respectively to generate single-class preliminary X-ray images and multi-class preliminary X-ray images;

[0010] Step 4, randomly combining single-category preliminary X-ray images and multi-category preliminary X-ray images to generate a combined image, labeling the position and category of the single-category preliminary X-ray image and the multi-category preliminary X-ray image in each combined image, and obtaining an image dataset;

[0011] Step 5, constructing a YOLOv11 network and training the YOLOv11 network through the image dataset to obtain an optimized YOLOv11 network;

[0012] Step 6, inputting a to-be-detected space debris image into the optimized YOLOv11 network to obtain the position of the target detection frame at different times;

[0013] Step 7, determining the trajectory, speed and size of the to-be-detected space debris according to the position of the target detection frame at different times.

[0014] Further, step 1 is specifically:

[0015] Step 1.1, using simulation technology and CT imaging simulation equipment to generate space debris models of multiple single types, materials, shapes and sizes;

[0016] Step 1.2, adding an X-ray source to the space debris model obtained in step 1.1, adjusting the control parameters of the X-ray source to simulate different imaging conditions, and using random noise adding technology to simulate noise interference in actual imaging, thereby obtaining multiple single-category space debris images;

[0017] Step 1.3, superimposing the gray levels of at least two randomly selected single-category space debris images to obtain multiple multi-category space debris images.

[0018] Further, step 2 is specifically: respectively rotating the single-category space debris images and the multi-category space debris images by 180° on a plane, collecting a projection image every 1° of rotation, and respectively performing back projection on the projection images to obtain 180 single-category space debris back projection images corresponding to each single-category space debris image and 180 multi-category space debris back projection images corresponding to each multi-category space debris image.

[0019] Further, the YOLOv11 network adopts a CSPDarknet53 backbone network, which includes: five first convolutional layers Conv, four C3K2 modules, an SPPF module and a C2PSA module.

[0020] The first first convolutional layer Conv, the second first convolutional layer Conv, the first C3K2 module, the third first convolutional layer Conv, the second C3K2 module, the fourth first convolutional layer Conv, the third C3K2 module, the fifth first convolutional layer Conv, the fourth C3K2 module, the SPPF module and the C2PSA module are sequentially arranged in the order of input and output.

[0021] Further, in the five first convolutional layers Conv, k=3, s=2, and p=1.

[0022] Wherein, k is the size of the control convolution kernel;

[0023] s is the control of the down-sampling rate and the output resolution;

[0024] p is the size of balancing input and output;

[0025] The calculation formula is:

[0026] ;

[0027] Wherein, is the input size of the first convolutional layer Conv, is the output size of the first convolutional layer Conv.

[0028] Further, step 7 is specifically:

[0029] Step 7.1, calculate the centroid position of the target detection frame at different times according to the position of the target detection frame at different times, and draw the trajectory of the centroid position of the target detection frame, that is, the trajectory of the spatial debris position;

[0030] Step 7.2, calculate the centroid speed of the target detection frame according to the change of the centroid position of the target detection frame of adjacent frames, that is, the speed of the spatial debris to be detected;

[0031] Step 7.3, threshold segmentation and connected domain analysis are performed on the position of the target detection frame to determine the size of the spatial debris to be detected.

[0032] Further, in step 7.1, the formula for calculating the centroid position of the target detection frame at different times is:

[0033]

[0034]

[0035] Wherein, ( x center , y center ) represents the centroid position of the target detection frame, ( xmin , y min ) and ( x max , y max ) are the left upper corner coordinates and the right lower corner coordinates of the target detection frame, respectively.

[0036] Further, in step 7.2, the formula for calculating the centroid speed of the target detection frame is:

[0037] ;

[0038] wherein, V center represents the centroid speed of the target detection frame, ( x t , y t ) and ( x t-1 , y t-1 ) represent the centroid coordinates of the target detection frame in the current frame and the previous frame, respectively, and Δt represents the frame interval time.

[0039] The beneficial effects of the present application are:

[0040] (1) The present application provides a space debris typical inertia feature extraction method based on X-ray simulation, which combines X-ray penetration characteristics and simulation technology to construct a high-simulation image dataset of various types, shapes, materials and sizes, and the training samples have diversity, which can improve the YOLOv11 network optimization effect, and further improve the space debris inertia feature extraction precision.

[0041] (2) The present application provides a space debris typical inertia feature extraction method based on X-ray simulation, which is based on the image dataset under the X-ray source, combines deep learning and rigid body dynamics model for trajectory estimation, and can accurately estimate the motion trajectory and speed of the space debris when dynamically analyzing the space debris.

[0042] (3) The present application provides a space debris typical inertia feature extraction method based on X-ray simulation, which analyzes the size information change of the X-ray image, combines the scale change of the space debris in different frames of images, and accurately calculates the size of the space debris and its change trend. BRIEF DESCRIPTION OF DRAWINGS

[0043] Figure 1 A simulation flowchart in a space debris typical inertia feature extraction method based on X-ray simulation provided by the present application;

[0044] Figure 2A kind of space debris category graph in the space debris typical inertia feature extraction method based on X-ray simulation embodiment provided by the application;

[0045] Figure 3 Single-class space debris simulation graph in the space debris typical inertia feature extraction method based on X-ray simulation embodiment provided by the application;

[0046] Figure 4 Multi-class space debris simulation graph in the space debris typical inertia feature extraction method based on X-ray simulation embodiment provided by the application;

[0047] Figure 5 Preliminary X-ray image of 14 categories in the space debris typical inertia feature extraction method based on X-ray simulation embodiment provided by the application;

[0048] Figure 6 Combined image in the space debris typical inertia feature extraction method based on X-ray simulation embodiment provided by the application;

[0049] Figure 7 Partial annotation effect diagram in the space debris typical inertia feature extraction method based on X-ray simulation embodiment provided by the application;

[0050] Figure 8 Different space debris X-ray simulation sequence image in the space debris typical inertia feature extraction method based on X-ray simulation embodiment provided by the application;(a) is a cuboid sequence diagram, (b) is a cylindrical sequence diagram, (c) is a prism sequence diagram, (d) is a multi-class space debris superposition sequence diagram;

[0051] Figure 9 For Figure 8 Cuboid motion trajectory diagram in (a); Figure 9 In (a), (a) represents the cuboid running trajectory diagram of the 1st X-ray image sequence, (b) represents the cuboid running trajectory diagram of the 2nd X-ray image sequence;

[0052] Figure 10 Cylindrical motion trajectory diagram in the space debris typical inertia feature extraction method based on X-ray simulation embodiment provided by the application;

[0053] Wherein, (a) represents the cylindrical running trajectory diagram of the 1st X-ray image sequence, (b) represents the cylindrical running trajectory diagram of the 2nd X-ray image sequence;

[0054] Figure 11 Prism motion trajectory diagram in the space debris typical inertia feature extraction method based on X-ray simulation embodiment provided by the application;

[0055] Wherein, wherein, (a) indicates the prism running track diagram of the No. 1 X-ray image sequence, (b) indicates the prism running track diagram of the No. 2 X-ray image sequence;

[0056] Figure 12 A multi-target motion track diagram in a space debris typical inertia feature extraction method embodiment based on X-ray simulation is provided for the application;

[0057] Wherein, (a) indicates the running track diagram of the superposition of the cuboid, cylinder and prism of the No. 1 X-ray image sequence; (b) indicates the running track diagram of the superposition of the cuboid, cylinder and prism of the No. 2 X-ray image sequence;

[0058] Figures 9-12 In the formula, the abscissa indicates Figure 8 In the formula, the abscissa indicates Figure 8 In the formula, the abscissa indicates DETAILED DESCRIPTION

[0059] The technical solutions of the application will be described clearly and completely below in combination with embodiments. Obviously, the described embodiments are only a part of the embodiments of the application, rather than all the embodiments. Based on the embodiments of the application, all other embodiments obtained by those skilled in the art without creative labor fall within the protection scope of the application.

[0060] The embodiment provides a space debris typical inertia feature extraction method based on X-ray simulation, as shown in the formula: Figure 1 The method comprises the following steps:

[0061] Step 1: Since the space debris includes discarded spacecraft (including failed satellites, spacecraft shells), rocket bodies, debris, discarded components (including circuit boards, sunshields, rocket bodies, antennas, debris, electric control boxes, lenses, nuts, spacecraft shells, failed satellites, screws, solar sail 1, solar sail 2 and wheels), as shown in the formula: Figure 2 The formula includes both single-class space debris images with single shapes and multi-class space debris images with complex shapes; on this basis, a single-class space debris image is first constructed:

[0062] A plurality of single-type, material, shape and size space debris models are generated by using simulation technology and CT imaging simulation equipment, X-ray sources are added to the space debris models, the control parameters of the X-ray sources are adjusted, different imaging conditions are simulated, random noise adding technology is used to simulate noise interference in actual imaging, a single-class space debris image is obtained, and the simulation image is as shown in the formula: Figure 3

[0063] ​The following types, materials, shapes and sizes of single category space debris are simulated with specific data in this embodiment as follows:

[0064] I. The space debris in the image is a rectangle:

[0065] (1) Random size generation:

[0066] Width and height :

[0067]

[0068]

[0069] wherein, rand is a random number generation function for generating a random number between 0 and 1, max_ width and min_width represent the maximum and minimum values of the allowed width, respectively, max_height and min_height represent the maximum and minimum values of the allowed height, respectively.

[0070] Center coordinates ( ):

[0071]

[0072]

[0073] wherein, img_size represents the width or height of the image

[0074] (2) Vertex rotation transformation:

[0075] Assuming that the centroid of the rectangle is at ( ) and the coordinates of the vertex when not rotated are ( ), the rotation angle is . The vertex coordinates after rotation can be calculated by the following formula:

[0076]

[0077] .

[0078] II. The space debris in the image is a circle

[0079] (1) Random generation: outer circle radius ∈[ , ] and inner circle radius ∈[5, ] to form a ring structure, 、 represents the minimum and maximum outer circle radius, respectively.

[0080] (2) Polar coordinate transformation: through the polar coordinate formula 、 to generate a ring profile. 、 、 represents the horizontal coordinate, vertical coordinate and arc length, respectively.

[0081] III. The space fragments in the image are regular hexagons

[0082] (1) The first vertex coordinate (x1, y1) calculation: assuming the centroid of the regular hexagon is at (0, 0) and its circumscribed circle radius is R, the vertex coordinates of the hexagon can be calculated by the following formula: i

[0083]

[0084]

[0085] (2) Rotation and offset: randomly generate a rotation angle θ ∈ [0°, 360°], and the vertex coordinates are represented as:

[0086]

[0087]

[0088] where, and represent the offset of the regular hexagon in the horizontal and vertical directions, respectively. represents the circumscribed circle radius of the regular hexagon, i.e. the radius of the regular hexagon r , represents the image size.

[0089] IV. The space fragments in the image are triangles:

[0090] For an equilateral triangle, assuming its side length is s, the bottom edge is located on the horizontal line (0, 0), and the vertex coordinates can be represented as:

[0091] The lower left corner coordinate is:

[0092]

[0093] The lower right corner coordinate is:

[0094] ​​​​

[0095] Vertex coordinates are:

[0096] .

[0097] The single-class space debris image generated in the above manner is shown in Figure 3 , where (a)-(l) represent circuit board, light shield, rocket body, antenna, debris, electric control box, lens, nut, spacecraft shell, failed satellite, screw, solar sail 1 type, respectively.

[0098] On the basis of the single-class space debris image, a multi-class space debris image is reconstructed:

[0099] At least two single-class space debris images are randomly selected for row grayscale superposition, different materials (such as metal shell and plastic inner core) are combined into a multi-layer structure, and a multi-class space debris image is obtained, and the simulation effect of the multi-class space debris image is shown in Figure 4 .

[0100] Step 2, respectively, rotate the single-class space debris image and the multi-class space debris image by 180° in the plane, collect a projection image every 1° of rotation, and respectively perform back projection on the projection images to obtain 180 single-class space debris back projection images corresponding to each single-class space debris image and 180 multi-class space debris back projection images corresponding to each multi-class space debris image.

[0101] Step 3, respectively, perform original path reverse superposition on the single-class space debris back projection images and the multi-class space debris back projection images to generate single-class preliminary X-ray images and multi-class preliminary X-ray images.

[0102] In this embodiment, 7000 single-class preliminary X-ray images and multi-class preliminary X-ray images of 512x512 pixels are generated through steps 2 and 3, covering 14 categories (antenna, circuit board, light shield, lens, debris, electric control box, nut, spacecraft shell, rocket body, failed satellite, screw, solar sail 1 type, solar sail 2 type, and wheel), 500 for each category, and the preliminary X-ray images of each category are shown in Figure 5 (a)-(n).

[0103] Step 4, randomly combine the single-class preliminary X-ray images and the multi-class preliminary X-ray images to generate a combined image, as shown in Figure 6 , where (a) and (b) are randomly combined images, the position and category of the single-class preliminary X-ray images and the multi-class preliminary X-ray images in each combined image are labeled, and part of the labeling effect is shown in Figure 7 , to obtain an image dataset. Figure 7The numbers at the bottom of each figure from left to right represent the category number, the normalized x-axis coordinate of the center point of the bounding box, the normalized y-axis coordinate of the center point of the bounding box, the normalized width of the bounding box, and the normalized height of the bounding box. Figure 7 Taking the upper left corner of the figure as an example, the category number is 2, the normalized x-axis coordinate of the center point of the bounding box is 0.349222, the normalized y-axis coordinate of the center point of the bounding box is 0.675097, the normalized width of the bounding box is 0.177043, and the normalized height of the bounding box is 0.175097.

[0104] Step 5, constructing a YOLOv11 network, adopting a CSPDarknet53 backbone network, extracting feature maps of different scales of images through a large number of convolution operations, and then learning different levels of information of the image on feature maps of different scales through a hierarchical structure, the CSPDarknet53 backbone network comprising five first convolution layers Conv, four C3K2 modules, an SPPF module and a C2PSA module;

[0105] The first first convolution layer Conv, the second first convolution layer Conv, the first C3K2 module, the third first convolution layer Conv, the second C3K2 module, the fourth first convolution layer Conv, the third C3K2 module, the fifth first convolution layer Conv, the fourth C3K2 module, the SPPF module and the C2PSA module are sequentially arranged in the order of input and output, and are used to extract local features of different scales through five times of down-sampling to generate feature maps of different scales, and to perform multi-scale feature fusion on the feature maps of different scales to obtain fused features; the feature maps of different scales can capture spatial information of different scales and have good detection capability for spatial fragments of different sizes.

[0106] The first convolution layer Conv is used as the input of the CSPDarknet53 backbone network to receive an external preliminary X-ray image, and in the five first convolution layers Conv, k=3, s=2, and p=1.

[0107] k: controls the size of the convolution kernel;

[0108] s: controls the down-sampling rate and output resolution;

[0109] p: balances the size of the input and output;

[0110] The calculation formula is:

[0111] ;

[0112] Wherein, is the input size of the first convolution layer Conv, is the output size of the first convolution layer Conv.

[0113] Substitute data:

[0114]

[0115] That is, the output size of the first convolutional layer Conv is: 256x256x64.

[0116] In this embodiment, five first convolutional layers Conv are used to extract features of different scales in sequence; the C3K2 module is a specific implementation of the C3 (Cross Stage Partial) structure, where K2 represents that two Bottleneck (residual units) are contained inside, and the C3 structure itself is derived from the design of CSPNet (Cross Stage Partial Network), which divides the input features into two parts, one of which is directly reserved, and the other is spliced with the former after passing through the Bottleneck, effectively reducing redundant calculations and improving gradient flow capacity. In the C3K2 module, only two Bottleneck are used, which further reduces the amount of calculation and the size of the model under the premise of ensuring the feature extraction capability, is suitable for deployment in resource-constrained environments, and is helpful to improve the model's representation ability for multi-scale targets. Figure 1

[0117] The SPPF module is a spatial pyramid feature fusion technology that uses multi-scale pooling, and its main purpose is to enhance the network's perception ability at different scales through multi-scale feature fusion.

[0118] The C2PSA module is a CSPDarknet53 backbone network used to introduce a spatial attention mechanism to enhance the features of key regions and improve detection accuracy.

[0119] ​In this embodiment, in the first and second first convolutional layers, k = 3, s = 2, the input size is: 512x512x3, and the output size is: 256x256x64; the first C3K2 module expands the channel, the input size is 128x128x64, and the output channel number is 128, and since there is no down sampling, the output size is 128x128x128; in the third first convolutional layer, k = 3, s = 2, the input size is: 128x128x128, and the output size is: 64x64x128; the second C3K2 module expands the channel, the input size is: 64x64x128, and the output channel number is: 256, and since there is no down sampling, the output size is: 64x64x256; in the fourth first convolutional layer, k = 3, s = 2, the input size is: 64x64x256, and the output size is: 32x32x256; the third C3K2 module expands the channel, the input size is: 32x32x256, and the output channel number is: 512, and since there is no down sampling, the output size is: 32x32x512; in the fifth first convolutional layer, k = 3, s = 2, the input size is: 32x32x512, and the output size is: 16x16x512; the fourth C3K2 module expands the channel, the input size is: 16x16x512, and the output channel number is: 1024, and since there is no down sampling, the output size is: 16x16x1024; the SPPF module. The input size of the SPPF module is: 16x16x1024, the output channel number is increased to 2048, and the output size is 16x16x2048. The input size of the C2PSA module is: 16x16x2048, and the output size is: 16x16x2048.

[0120] In this embodiment, the SPPF module is obtained by multiple max pooling (k = 5, s = 1), and the output size of the SPPF module is 16x16x1024, wherein k represents the size of the convolution kernel, and s represents the step length. The expression is as follows:

[0121]

[0122] wherein, is max pooling, is a splicing operation, and F is an input tensor or feature map.

[0123] The YOLOv11 network is trained by using an image data set, and an optimized YOLOv11 network is obtained.

[0124] Step 6, inputting the space debris image to be detected into the optimized YOLOv11 network to obtain a target detection frame position.

[0125] Step 7. According to the position of the labeled target detection frame, the trajectory, speed and size of the space debris to be detected are determined; specifically:

[0126] Step 7.1. Calculate the centroid position of the target detection frame at different times according to the position of the target detection frame at different times, and draw the trajectory of the centroid position of the target detection frame, i.e. the trajectory of the space debris position.

[0127] The formula for calculating the centroid position of the target detection frame at different times is:

[0128]

[0129]

[0130] wherein, x center , y center the centroid position of the target detection frame, ( x min , y min ) and ( x max , y max ) are the coordinates of the upper left corner and the lower right corner of the target detection frame, respectively.

[0131] Step 7.2. Calculate the centroid speed of the target detection frame, i.e. the speed of the space debris to be detected, according to the change of the centroid position of the target detection frame of adjacent frames.

[0132] The formula for calculating the centroid speed of the target detection frame is:

[0133] ;

[0134] wherein, V center the centroid speed of the target detection frame, ( x t , y t ) and ( x t-1 , y t-1 ) are the centroid coordinates of the target detection frame in the current frame and the previous frame, respectively, and Δt is the frame interval time.

[0135] Step 7.3. Threshold segmentation and connected component analysis are performed on the position of the target detection frame to determine the size of the space debris to be detected.

[0136] In this embodiment, Ubuntu operating system is adopted, NVIDIA GeForce RTX 4090 GPU is carried, and the detection targets include cuboids, cylinders and prisms in typical space debris shapes. The images obtained by each simulation are arranged according to time to obtain X-ray image sequences, and 100 groups of X-ray image sequences are obtained as a test set, wherein each sequence is composed of 100 frames of images.

[0137] Figure 8 Figures are X-ray simulation sequence images of different space debris, wherein (a) is a cuboid sequence image, (b) is a cylinder sequence image, (c) is a prism sequence image, and (d) is a multi-class space debris superimposed sequence image. Two groups of sequences in the test set are randomly sampled to further illustrate the effect of the space debris typical inertia feature extraction method based on X-ray simulation in this embodiment. The first row represents the No. 1 X-ray image sequence of each type of data, and the second row represents the No. 2 X-ray image sequence. The typical inertia features of single-class space debris images are calculated as follows:

[0138] I. Space debris is a cuboid:

[0139] (1) Trajectory calculation:

[0140] Based on the detection results of the No. 1 X-ray image sequence and the No. 2 X-ray image sequence of the cuboid, the position information of the centroid of the cuboid is obtained by calculating the center coordinates of the detection frame, and the orbital velocity value of each two frames is obtained according to the velocity calculation formula. Table 1 is a table of instantaneous velocity and average velocity changes of the cuboid. In Table 1, the instantaneous velocity and the average velocity both refer to pixels / frame. It can be known that the simulation data is in non-uniform linear motion, so the average velocity of the complete pixel is calculated. Figure 9 Figures show the calculated running trajectories of the cuboid, wherein (a) represents the running trajectory diagram of the cuboid in the No. 1 X-ray image sequence, and (b) represents the running trajectory diagram of the cuboid in the No. 2 X-ray image sequence.

[0141] Table 1 Table of instantaneous velocity and average velocity changes of the cuboid

[0142]

[0143] (2) Scale calculation

[0144] Based on the detection results of the No. 1 X-ray image sequence and the No. 2 X-ray image sequence of the cuboid, the pixel number of the target in the image is obtained by a segmentation algorithm, the scale change of the space target is obtained by calculating the pixel number difference of adjacent frames, and Table 2 is a table of scale change and interval change of the two sequences of the cuboid. In Table 2, the step change of the interval is 10. It can be seen that the scale of the cuboid in the No. 1 X-ray image sequence is generally increased (+2019), and the scale of the cuboid in the No. 2 X-ray image sequence is generally decreased (-1227).

[0145] Table 2 Two sequence scale changes and interval changes of cuboid

[0146]

[0147] II. Space debris is a cylinder

[0148] (1) Trajectory calculation

[0149] Based on the detection results of the No. 1 X-ray image sequence and the No. 2 X-ray image sequence of the cylinder, the orbital velocity value of every two frames is obtained, and Table 3 is a table of instantaneous velocity and average velocity changes of the cylinder. In Table 3, the instantaneous velocity and the average velocity both refer to pixels / frame. It can be known from Table 3 that the simulation data is in non-uniform linear motion, so the average velocity of the complete pixel is obtained. Figure 10 The calculated running trajectory of the cylinder is shown in FIG. 6, wherein (a) represents a cylinder running trajectory diagram of the No. 1 X-ray image sequence, and (b) represents a cylinder running trajectory diagram of the No. 2 X-ray image sequence.

[0150] Table 3 Table of instantaneous velocity and average velocity changes of the cylinder

[0151]

[0152] (2) Scale calculation

[0153] Based on the detection results of the No. 1 X-ray image sequence and the No. 2 X-ray image sequence of the cylinder, the scale changes of the space target are obtained, Table 4 is a table of two sequence scale changes and interval changes of the cylinder, and the step change of the interval in Table 4 is 10. It can be seen that the scale of the cylinder of the No. 1 X-ray image sequence is reduced as a whole (-4125), and the scale of the cylinder of the No. 2 X-ray image sequence is reduced as a whole (-855).

[0154] Table 4 Table of two sequence scale changes and interval changes of the cylinder

[0155]

[0156] III. Space debris is a prism

[0157] (1) Trajectory calculation

[0158] Based on the detection results of the No. 1 X-ray image sequence and the No. 2 X-ray image sequence of the prism, the orbital velocity value of every two frames is obtained, and Table 5 is a table of instantaneous velocity and average velocity changes of the prism. In Table 5, the instantaneous velocity and the average velocity both refer to pixels / frame, and it can be known from Table 5 that the simulation data is in non-uniform linear motion, so the average velocity of the complete pixel is obtained. Figure 11The calculated prism running trajectory is shown in FIG. 6, wherein (a) represents a prism running trajectory diagram of the No. 1 X-ray image sequence, and (b) represents a prism running trajectory diagram of the No. 2 X-ray image sequence.

[0159] Table 5: Prism instantaneous speed and average speed change table

[0160]

[0161] (2) Scale calculation

[0162] Based on the detection results of the No. 1 X-ray image sequence and the No. 2 X-ray image sequence of the prism, the scale change of the space target is obtained, and Table 6 is a two-sequence scale change and interval change table of the prism. In Table 6, the step change of the interval is 10. It can be seen that the scale of the No. 1 X-ray image sequence of the prism increases as a whole (3183), and the scale of the No. 2 X-ray image sequence of the prism increases as a whole (393).

[0163] Table 6: Two-sequence scale change and interval change table of the prism

[0164]

[0165] The typical inertia characteristic calculation of the multi-class space debris image is as follows:

[0166] (1) Trajectory calculation

[0167] The running trajectory of the space target obtained by superimposing the cuboid, the cylinder and the prism is shown in FIG. 7. Figure 12

[0168] (2) Scale calculation

[0169] Based on the detection results of the No. 1 X-ray image sequence and the No. 2 X-ray image sequence of the cuboid, the cylinder and the prism, the scale change of the space target is obtained, and Table 7 is a two-sequence scale change and interval change table of the space target obtained by superimposing the cuboid, the cylinder and the prism. It can be seen that the scale of the space target of the No. 1 X-ray image sequence decreases as a whole (-5181), and the scale of the space target of the No. 2 X-ray image sequence decreases as a whole (-3963).

[0170] Table 7: Two-sequence scale change and interval change table of the space target obtained by superimposing

[0171]

[0172] According to the data in Tables 1-7, the motion trajectory, instantaneous speed and average speed of the target are obtained for both the single-class space debris and the multi-class space debris, and the motion trajectory obtained in the embodiment conforms to the simulation trajectory parameters through the sequence data generation algorithm verification.​

[0173] The above merely illustrates the specific embodiments of the present application, and the effects of the specific embodiments and the comparative examples are compared, but the protection scope of the present application is not limited thereto, any change or replacement within the technical scope disclosed in the present application should be covered in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method for extracting typical inertia features of space debris based on X-ray simulation, characterized in that, Includes the following steps: Step 1: Simulate spatial debris models of different types, materials, shapes and sizes randomly generated under X-ray source and obtain single-category spatial debris images; then, superimpose the single-category spatial debris images in grayscale to obtain multi-category spatial debris images. Step 2: Rotate and project the single-class spatial debris image and the multi-class spatial debris image respectively to obtain the projected image. Then, back-project the projected image to obtain the single-class spatial debris back-projected image and the multi-class spatial debris back-projected image. Step 3: Perform reverse overlay of the single-category spatial debris backprojection image and the multi-category spatial debris backprojection image along the original path to generate a single-category preliminary X-ray image and a multi-category preliminary X-ray image; Step 4: Randomly combine single-class preliminary X-ray images and multi-class preliminary X-ray images to generate composite images. Label the location and category of the single-class preliminary X-ray images and multi-class preliminary X-ray images in each composite image to obtain the image dataset. Step 5: Construct a YOLOv11 network and train it using an image dataset to obtain an optimized YOLOv11 network. The YOLOv11 network adopts the CSPDarknet53 backbone network, including: five first convolutional layers (Conv), four C3K2 modules, an SPPF module, and a C2PSA module; The first convolutional layer Conv, the second convolutional layer Conv, the first C3K2 module, the third convolutional layer Conv, the second C3K2 module, the fourth convolutional layer Conv, the third C3K2 module, the fifth convolutional layer Conv, the fourth C3K2 module, the SPPF module, and the C2PSA module are set in the order of input and output. Step 6: Input the spatial debris image to be detected into the optimized YOLOv11 network to obtain the position of the target detection box at different times; Step 7: Based on the position of the target detection box at different times, determine the trajectory, velocity, and size of the spatial debris to be detected, specifically as follows: Step 7.1: Calculate the centroid position of the target detection box at different times based on its position at different times, and draw the centroid position trajectory of the target detection box, which is the spatial debris position trajectory. Step 7.2: Calculate the centroid velocity of the target detection box based on the change in the centroid position of the target detection box in adjacent frames, which is the velocity of the spatial fragment to be detected; Step 7.3: Perform threshold segmentation and connected component analysis on the target detection box position to determine the size of the spatial fragment to be detected.

2. The method for extracting typical inertia features of space debris based on X-ray simulation according to claim 1, characterized in that, Step 1 is as follows: Step 1.1: Using simulation technology and CT imaging simulation equipment, generate various spatial fragment models of single type, material, shape, and size; Step 1.2: Add an X-ray source to the space debris model obtained in Step 1.1, adjust the control parameters of the X-ray source, simulate different imaging conditions, and at the same time use random noise addition technology to simulate noise interference in actual imaging to obtain a single-class space debris image. Step 1.3: Overlay the grayscale values ​​of at least two randomly selected single-class spatial fragment images to obtain multiple multi-class spatial fragment images.

3. The method for extracting typical inertia features of space debris based on X-ray simulation according to claim 1, characterized in that, Step 2 specifically involves rotating the single-class spatial fragment image and the multi-class spatial fragment image 180° in a plane, acquiring a projection image every 1° of rotation, and then back-projecting the projection images to obtain 180 single-class spatial fragment back-projection images corresponding to each single-class spatial fragment image and 180 multi-class spatial fragment back-projection images corresponding to each multi-class spatial fragment image.

4. The method for extracting typical inertia features of space debris based on X-ray simulation according to claim 1, characterized in that: In the five first convolutional layers Conv, k = 3, s = 2, and p = 1; Where k is the size of the convolution kernel; s controls the downsampling rate and output resolution; p represents the size of the balanced input and output; The calculation formula is: Among them, W in W is the input size of the first convolutional layer Conv. out This is the output size of the first convolutional layer, Conv.

5. The method for extracting typical inertia features of space debris based on X-ray simulation according to claim 1, characterized in that: In step 7.1, the formula for calculating the centroid position of the target detection box at different times is: Among them, (x center y center (x) represents the centroid position of the target detection bounding box. min y min ) and (x max y max ) represent the coordinates of the top left and bottom right corners of the target detection box, respectively.

6. The method for extracting typical inertia features of space debris based on X-ray simulation according to claim 4, characterized in that: In step 7.2, the formula for calculating the centroid velocity of the target detection box is: Among them, V center The centroid velocity of the target detection bounding box is represented by (x). t y t ) and (x t-1 y t-1 ) represent the centroid coordinates of the target detection box in the current frame and the previous frame, respectively, and Δt represents the frame interval time.

Images