CN118379449A - Space target data set construction method and system - Google Patents

Abstract

The invention discloses a method and a system for constructing a space target data set, and relates to the technical field of data set construction. The technical key points of the invention include: establishing a plurality of satellite space geometric models by utilizing three-dimensional modeling software; establishing a space target on-orbit scene, wherein the scene comprises the following steps: a scene containing the earth background, a scene containing different sun illumination angles and effects, and a scene introducing deep space background noise; setting related parameters to enable a plurality of satellite space geometric models to move in the space target in-orbit scene, and further obtaining a plurality of groups of pictures corresponding to a plurality of satellites and different poses; and processing the plurality of groups of pictures to obtain a space target data set. The invention greatly improves the scale and diversity of the data set, remarkably improves the data precision and the authenticity, remarkably reduces the data acquisition cost and time, and greatly reduces the threshold of research and development.

Description

A method and system for constructing a space target data set

Technical Field

The present invention relates to the technical field of data set construction, and in particular to a method and system for constructing a space target data set.

Background technique

As countries around the world pay more and more attention to space technology, the importance of space situational awareness is increasing. Attacking, protecting and on-orbit servicing important space targets, mainly satellites, has become an important development direction of space technology in countries around the world, and vision-based space target behavior perception is a key link. Space target behavior perception involves the estimation of the pose of space targets, the identification of key components and the identification of space targets, which is crucial for the navigation, positioning and on-orbit servicing of spacecraft. This technology can help better understand and control the dynamic behavior of spacecraft and provide support for mission planning and decision-making. With the rapid development of artificial intelligence and machine learning technologies, space target pose estimation and target recognition technologies have also made significant progress. Deep learning models have been successfully applied to identify target morphology from complex backgrounds and estimate their pose parameters. These models can learn the characteristics of targets from a large amount of training data, including their shape, size, texture, etc., and estimate the type, position and pose of the target through these characteristics.

However, the study of space target behavior perception based on deep learning requires a large amount of space target data. Domestic and foreign scholars generally adopt two ideas: real pictures and simulated pictures. Among them, real pictures have high satellite details and often have the background of the earth, but their number is small, which makes it difficult to meet the requirements of deep learning for a large number of data sets. Therefore, some scholars simulate satellites running in the earth's orbit based on software such as STK, and capture pictures to establish satellite data sets. However, due to the modeling quality of satellites and the earth, the constructed satellite pictures are quite different from real pictures. The deep learning algorithm trained with this kind of simulated pictures is difficult to apply to real satellites in orbit. Therefore, it is necessary to study the construction method of a relatively realistic and large number of space target data sets to overcome the shortcomings of a small number of real pictures and low authenticity of simulated pictures.

The limitations of existing space object pose estimation datasets in terms of scale, diversity, and authenticity have largely hindered the development of space vehicle navigation and the application of deep learning algorithms in space object recognition, tracking, and interactive simulation. In particular, existing datasets are often small in scale, single in scene, or lack real-world complexity, which makes it difficult to meet the growing technical needs. In addition, obtaining large-scale, high-quality, and diverse space object pose data usually requires complex experimental design, which further limits the development of research and application.

Summary of the invention

In view of the above problems, the present invention proposes a method and system for constructing a space target dataset.

According to one aspect of the present invention, a method for constructing a space target dataset is proposed, the method comprising:

Step 1: Use 3D modeling software to build multiple satellite space geometry models;

Step 2: Establishing an on-orbit scene of a space target, wherein the scene includes: a scene with an earth background, a scene with different sunlight angles and effects, and a scene with deep space background noise introduced;

Step 3: setting relevant parameters so that multiple satellite space geometric models move in the space target on-orbit scene, thereby obtaining multiple sets of pictures corresponding to multiple satellites and different postures; the parameters include rotation angular velocity and moment of inertia;

Step 4: Process the multiple groups of images corresponding to the multiple satellites and different postures to obtain a space target data set.

Furthermore, the scene containing the earth background is established as follows: create an earth model, adjust the size ratio of the earth model and the satellite space geometric model; add a texture map to the material properties of the earth model, and adjust the transparency and saturation to achieve a real space earth background, wherein the texture map types include color maps, height maps, and cloud maps; set relevant parameters of the lighting and noise of the shadow part of the earth, wherein the parameters of the lighting of the shadow part of the earth include light type, color value, intensity, position, and angle, and the parameters of the noise include noise texture and transparency; set the rotation angular velocity and rotation axis position to control the rotation of the earth model, thereby realizing the earth rotation scene.

Furthermore, the scene containing different sunlight angles and effects is established as follows: establish a solar model and adjust the proportion of the solar sphere; configure the self-luminous material color of the solar sphere; set the directional light source at the center of the solar sphere and the color of the directional light source; adjust the intensity of the directional light source; set the angle range between the directional light source and the surface of the earth model.

Furthermore, the scene introducing deep space background noise is established as follows: creating a deep space background and setting the sky box material to a deep space galaxy; using an improved fusion noise algorithm to generate random and visually continuous deep space background noise, and then generating a noise map.

Further, the use of the improved fusion noise algorithm to generate random and visually continuous noise includes: selecting a noise source, the noise source includes using improved Perlin noise to generate a basic distribution of stars, using Simplex noise to increase nebula details and fog effects, and using value noise to add star highlights and sparkle effects; defining a weight function, the weight function includes a weight W _star (x, y) corresponding to the basic star field, a weight W _nebula (x, y) corresponding to the nebula details, and a weight W _glow (x, y) corresponding to the highlight and sparkle; superimposing the noise and the weight function to generate deep space background noise as follows:

Wherein, Noise _perlin (x, y) represents Perlin noise distribution; Noise _simplex (x, y) represents Simplex noise distribution; Noise _value (x, y) represents value noise distribution.

Furthermore, the weight W _star (x, y) of the corresponding basic star field is set according to the following formula:

W _star (x,y)＝1-Smootstep(Noise _perlin (x,y))

Where Smootstep represents a smooth step function, which is used to smooth the transition of noise values;

The weight W _nebula (x, y) corresponding to the nebula detail is set according to the following formula:

W _nebula (x,y)＝Noise _simplex (x,y) ^α

In the formula, α>1 represents an exponential increase in local contrast;

The weight W _glow (x, y) corresponding to highlight and shine is set according to the following formula:

W _glow (x,y)＝Noise _value (x,y)>s

Where s∈(0,1) represents a random threshold.

Further, the setting of relevant parameters so that the multiple satellite space geometric models move in the space target on-orbit scene includes: controlling the change of the position and rotational angular velocity of the satellite in the space target on-orbit scene, wherein the rotational angular velocity has components on the three axes of the satellite body coordinate system, and the angular velocities of the three axes satisfy the Euler formula; using a camera to simulate shooting images to obtain multiple groups of pictures corresponding to multiple satellites and different postures; wherein a random disturbance term is introduced to dynamically adjust the rotational angular velocity according to the following formula:

ω _i (t)=a _i +b _i ·t+c _i ·sin(d _i ·θ _i (t))

Where i represents one channel among the yaw angle, pitch angle and roll angle; a _i , b _i , c _i and d _i represent the initial velocity, acceleration, attitude influence coefficient and periodic modulation factor of attitude influence respectively; θ _i (t) represents the Euler angle at time t and satisfies the following formula:

θ _i (t+Δt)=θ _i (t)+(ω _i (t)+∈ _i (t))Δt

Where Δt represents the time interval; ∈ _i (t) represents the random disturbance term and satisfies the uniform distribution U: ∈ _i (t)～U(-δ _i (θ _i (t),t),δ _i (θ _i (t),t)), δ _i represents the dynamic adjustment function related to the disturbance intensity, current posture and time t.

Further, the processing of multiple groups of pictures corresponding to multiple satellites and different postures to obtain a space target data set includes: obtaining the initial position and posture of the satellite and the camera; using the satellite body coordinate system and the camera body coordinate system as reference points, calculating the rotation matrix and the translation vector, including: obtaining the pitch angle, yaw angle and roll angle of the satellite relative to its own body coordinate system, constructing a rotation matrix according to these angle values as follows: R=R _yaw (θ _yaw )·R _pitch (θ _pitch )·R _roll (θ _roll ), wherein R _yaw (θ _yaw ) represents a rotation about the Y axis, R _pitch (θ _pitch ) represents a rotation about the X axis, and R _roll (θ _roll ) represents a rotation about the Z axis, and generating posture data according to the rotation matrix; calculating the translation vector according to the coordinates of the center point of the satellite body coordinate system and the coordinates of the center point of the camera body coordinate system, and generating position data according to the translation vector.

According to another aspect of the present invention, a space target dataset construction system is provided, the system comprising:

A model building module configured to build a plurality of satellite space geometric models using three-dimensional modeling software;

A scene establishment module, configured to establish a space target on-orbit scene, the scene including: a scene containing the earth background, a scene containing different sunlight angles and effects, and a scene introducing deep space background noise;

An image acquisition module is configured to set relevant parameters so that multiple satellite space geometric models move in the space target on-orbit scene, thereby obtaining multiple sets of pictures corresponding to multiple satellites and different postures; the parameters include rotation angular velocity and moment of inertia;

The data set construction module is configured to process the multiple groups of pictures corresponding to the multiple satellites and in different positions to obtain a space target data set.

Furthermore, the scene including the earth background in the scene establishment module is established as follows: creating an earth model, adjusting the size ratio of the earth model and the satellite space geometric model; adding a texture map to the material properties of the earth model, and adjusting the transparency and saturation to achieve the real space earth background, wherein the texture map types include color maps, height maps, and cloud maps; setting the relevant parameters of the illumination and noise of the shadow part of the earth, wherein the parameters of the illumination of the shadow part of the earth include light type, color value, intensity, position, angle, and the parameters of the noise include noise texture and transparency; setting the rotation angular velocity and the rotation axis position to control the rotation of the earth model, thereby realizing the earth rotation scene;

The scene containing different sunlight angles and effects is established as follows: establish a sun model and adjust the proportion of the sun sphere; configure the self-luminous material color of the sun sphere; set the directional light source at the center of the sun sphere and the color of the directional light source; adjust the intensity of the directional light source; set the angle range between the directional light source and the surface of the earth model;

The scene introducing deep space background noise is established as follows: creating a deep space background and setting the sky box material to a deep space galaxy; using an improved fusion noise algorithm to generate random and visually continuous deep space background noise, and then generating a noise map.

The beneficial technical effects of the present invention are:

The present invention proposes a method and system for constructing a space target dataset, which has significant advantages over the existing technologies in terms of performance, accuracy, efficiency and cost. Specifically, first, the scale and diversity of the dataset are greatly improved: the existing technologies usually rely on limited physical experiments or field measurements to obtain data, resulting in small datasets and single scenes. The present invention can generate large-scale datasets containing tens of thousands to millions of data points, covering a variety of scenes from low-Earth orbit to deep space. This scale expansion and scene diversity are crucial for training and verifying high-performance space target tracking algorithms; second, the data accuracy and authenticity are significantly improved: existing datasets often lack the complexity of real environments, which affects the generalization ability of deep learning networks for real space three-dimensional targets. The present invention accurately simulates the physical environment in space. Characteristics (such as lighting changes, motion dynamics, etc.) can be used to generate highly realistic data sets. The error between simulated data and actual observed data can be controlled within 3%, which significantly improves the application value of the data and the generalization ability of the algorithm, and helps to improve the ability of the training network to recognize and estimate the position and posture of spatial targets in real scenes; third, it significantly reduces the cost and time of data acquisition: the traditional method of obtaining high-quality spatial target data is costly and time-consuming. The present invention can automatically generate a complete data set within a few hours, while the traditional method may take months or even years. At the same time, the cost is only less than 5% of the traditional data acquisition method, which greatly reduces the threshold for research and development.

BRIEF DESCRIPTION OF THE DRAWINGS

The present invention can be better understood by referring to the description given below in conjunction with the accompanying drawings, which together with the following detailed description are included in this specification and form a part of this specification, and are used to further illustrate the preferred embodiments of the present invention and explain the principles and advantages of the present invention.

FIG1 is a flow chart of a method for constructing a space target pose dataset according to an embodiment of the present invention.

FIG. 2 is an example diagram of satellite models established in an embodiment of the present invention; wherein (a) corresponds to the DAWN satellite model; (b) corresponds to the STARLINK satellite model; and (c) corresponds to the James Webb Space Telescope satellite model.

FIG. 3 is a schematic diagram of earth background simulation in an embodiment of the present invention.

FIG. 4 is a schematic diagram of the solar illumination angle in an embodiment of the present invention.

FIG. 5 is a deep space background image with starlight noise according to an embodiment of the present invention.

FIG6 is a schematic diagram of a certain posture of a DAWN satellite in orbit according to an embodiment of the present invention.

FIG. 7 is a schematic diagram of another posture of the DAWN satellite in orbit according to an embodiment of the present invention.

Detailed ways

In order to enable those skilled in the art to better understand the scheme of the present invention, exemplary implementations or embodiments of the present invention will be described below in conjunction with the accompanying drawings. Obviously, the described implementations or embodiments are only implementations or embodiments of a part of the present invention, not all of them. Based on the implementations or embodiments of the present invention, all other implementations or embodiments obtained by ordinary technicians in the field without creative work should fall within the scope of protection of the present invention.

The purpose of this invention is to provide a method that can mass-produce high-quality, diverse and realistic space target pose data sets, which can provide strong data support for fields such as space vehicle navigation, autonomous control, deep learning training and space science research, thereby promoting technological progress and application expansion in these fields.

An embodiment of the present invention provides a method for constructing a space target pose dataset, as shown in FIG1 , the method comprising:

Step 1: Use 3D modeling software to build multiple satellite space geometry models;

Step 2: Establishing an on-orbit scene of a space target, wherein the scene includes: a scene with an earth background, a scene with different sunlight angles and effects, and a scene with deep space background noise introduced;

Step 3: setting relevant parameters so that multiple satellite space geometric models move in the space target on-orbit scene, thereby obtaining multiple sets of pictures corresponding to multiple satellites and different postures; the parameters include rotation angular velocity and moment of inertia;

Step 4: Process the multiple groups of images corresponding to the multiple satellites and different postures to obtain a space target data set.

The method starts from step 1. In step 1, a plurality of satellite space geometric models are established by using three-dimensional modeling software.

According to an embodiment of the present invention, based on three-dimensional modeling software, a detailed satellite geometric model is established with reference to the satellite structure diagram, dimension diagram and schematic diagram, and satellite components such as the satellite body, antenna, solar panels, sensitive elements (such as high-precision lenses, lidar), antenna, etc. are established according to size requirements, and the components are assembled according to the satellite structure diagram, and each component is fixed on the satellite body.

After the model structure is established, the model texture and material need to be set, and different materials need to be assigned to different components to achieve a more realistic effect. Collect pictures and information of satellite models from NASA's official website, Satellite Tool Kit (STK) or other reliable sources. Use reference pictures to determine the material type and color scheme of the satellite. Based on the collected reference pictures, use image editing software (such as Adobe Photoshop or GIMP) to create or modify texture maps. These maps will include surface details of various parts of the satellite, such as metal texture, logos, signs of wear and tear, etc. At the same time, create seamless textures to avoid obvious seams on the model. Specify the map as a diffuse, specular or normal map in the material settings of the 3D modeling software.

Refine the model visual effect, set the surface metalness, highlight, gloss, and roughness of the metal material parts in the target to 0.5. Since the surface texture of the solar panels is relatively rich, the surface texture of the solar panels is refined, and the original overall and smooth solar panels are refined into panels with rich textures that are more similar to the real model.

After the model is built, it is exported as a file in FBX format. The built model is shown in Figures 2(a), 2(b), and 2(c).

Then, step 2 is executed. In step 2, an on-orbit scene of a space target is established. The scene includes: a scene containing an earth background, a scene containing different sunlight angles and effects, and a scene introducing deep space background noise.

According to an embodiment of the present invention, the scene including the earth background is established as follows: create an earth model, adjust the size ratio of the earth model and the satellite space geometric model; add a texture map to the material properties of the earth model, and adjust the transparency and saturation to achieve a real space earth background, wherein the texture map types include color maps, height maps, and cloud maps; set relevant parameters of the lighting and noise of the shadow part of the earth, wherein the parameters of the lighting of the shadow part of the earth include light type, color value, intensity, position, and angle, and the parameters of the noise include noise texture and transparency; set the rotation angular velocity and the rotation axis position to control the rotation of the earth model, thereby realizing the earth rotation scene.

Specifically, establishing the earth background requires considering many factors, such as the transparency and saturation of the surface atmosphere, the angular velocity of the earth's rotation, and the synthesis of light noise on the dark side of the earth.

First, create a model of the Earth. For example, create a sphere in Unity to represent the Earth. You can create a basic sphere model by selecting Sphere in the 3D Object module from Unity's GameObject menu.

Next, import the satellite space geometry model file, and adjust the model's performance in the scene by adjusting the model's position, rotation, and scale in the Inspector view of Unity4. At the same time, set the three axes (X, Y, and Z) of the target body coordinate system, and adjust the size ratio of the earth model and the space target model. You need to adjust the sphere's Scale property to match the relative size of the earth, according to the actual ratio (the relative size ratio of the earth and the space target).

Afterwards, in order to restore the earth background in real space, the earth surface is textured by mapping (including the light and dark sides of the earth surface corresponding to the day and night scenes). Using high-quality texture maps, including surface maps, cloud maps, night light maps, etc., add these maps to the corresponding texture slots in the material properties of the sphere. Taking the earth as an example, first create a Material in Unity to add textures to the earth model. Then import high-definition earth surface maps, which usually include color maps, height maps, cloud maps, etc. Then apply the map to the previously created material, apply this material to the sphere model, and adjust its transparency and saturation to achieve the real space earth background. Make sure the map is correctly mapped to the sphere to achieve a realistic earth appearance. Cloud layer mapping can be divided into: creating clouds, adding a slightly larger Sphere object outside the sphere model to serve as the cloud layer, the material of this sphere should be translucent, and apply cloud layer mapping; setting cloud layer dynamics, adding a rotation script for the cloud layer, but the rotation speed can be slightly different to simulate the movement of the cloud layer relative to the ground. The mapping parameters are shown in Table 1 below.

Table 1 Map type parameter table

Then, design the back lighting noise. First, adjust the position and angle of the directional light (sunlight) to ensure that it can correctly illuminate the earth model to form a light-dark boundary. Then add noise. On the back side of the earth (the shadow part), you can add appropriate noise textures to the material, or use Shader to create more complex lighting noise effects to increase visual details and realism. The noise parameters are shown in Table 2 below.

Table 2 Earth background illumination setting parameters

Then, set the rotation angular velocity and the rotation axis position to control the rotation of the earth model, and then realize the earth rotation scene to achieve real-time simulation of the scene where the space target is located. For example, first add a script in Unity and create a new C# script file. By writing the rotation angular velocity and rotation axis position code in the script, the rotation of the sphere can be controlled, so as to control the rotation of the earth. Figure 3 is a schematic diagram of the earth background simulation.

According to an embodiment of the present invention, the scene containing different sunlight angles and effects is established as follows: establish a solar model and adjust the proportion of the solar sphere; configure the self-luminous material color of the solar sphere; set the directional light source at the center of the solar sphere and the color of the directional light source; adjust the intensity of the directional light source; set the angle range between the directional light source and the surface of the earth model.

Specifically, first, a sun sphere is created. For example, a Sphere object is added to the Unity scene as the sun. Considering that the sun is mainly used as a light source in this scene, its proportion adjustment is kept the same as that of the earth. Secondly, considering the sun material, the present invention uses a simple self-luminous material, whose main function is to act as a light source.

Secondly, configure the directional light source and its properties. For example, select a directional light source, and in Unity's light source options, select Directional Light; adjust the light source position and move the directional light source to the center of the sun sphere. Simulate the effect of the sun as a luminous body to ensure that the light can be evenly irradiated from the sun's position; set the radius of the directional light source. In Unity, the directional light source simulates the coverage of sunlight by adjusting the angle and range of the light source. The directional light source mainly simulates parallel light from a long distance and sets the relationship between its direction and the earth. In addition, the present invention also adjusts the intensity of the directional light property, increases the intensity of the directional light source to simulate the brightness of sunlight, enhances the brightness of the sun sphere in Unity, and highlights its role as the main light source; then the color setting, adjusts the color of the light source to reflect the sunlight effect at different times. The present invention uses warmer tones (yellow to orange) to simulate the light at sunrise and sunset, and uses whiter light to simulate sunlight at noon, making the effect more realistic.

Again, adjust the relative position of the sun and the earth. The present invention sets the sun sphere away from the earth and adjusts the position of the sun sphere so that it is away from the earth model in the scene. In Unity, a sufficient visual distance is maintained between the sun and the earth to enhance the light and shadow effects. Secondly, simulate the sun angle. By rotating the sun sphere or directly adjusting the direction of the directional light source, the angle between the sun's rays and the earth can be simulated, thereby affecting the lighting effect and the day and night changes on the earth, thereby achieving a true simulation of the space scene. The design of the sunlight parameters is shown in Table 3 below. The illumination angle is shown in Figure 4.

Table 3 Sunlight Parameter Design Table

According to an embodiment of the present invention, the scene for introducing deep space background noise is established as follows: a deep space background is created, and the sky box material is set to a deep space galaxy; an improved fusion noise algorithm is used to generate random and visually continuous deep space background noise, and then a noise map is generated.

Specifically, a method for introducing deep space background noise in a three-dimensional space simulation environment is provided, aiming to improve the simulation of the real deep space background environment, and to provide a more realistic space scene for generating space target data sets by simulating the background noise effect in the space environment. The Unity game development engine is used to generate background noise through a specific algorithm and integrate it into the space scene to simulate different noise levels and characteristics from outer space to deep space.

First, create a deep space background. Build a background model and create a spherical skybox in the simulation software Unity as the basis for simulating the deep space background. The skybox provides an all-round visual background for observing space targets to enhance the realism and immersion of the simulated environment. Set the skybox material, select the material of the deep space galaxy, and apply it to the inner wall of the skybox. The material should contain high-resolution starry sky images, reflecting elements such as starlight, galaxies, and nebulae in the deep space. Next, use the improved fusion noise algorithm to generate background noise. The algorithm can generate random but visually continuous noise patterns, which is suitable for simulating the optical noise naturally present in the deep space background. Finally, generate a noise map and use the noise algorithm to generate a noise map. This map simulates the optical noise effect caused by distant galaxies and nebulae. The deep space background with starlight background noise is shown in Figure 5.

The improved fusion noise algorithm uses the superposition noise method to improve the traditional noise distribution. The specific steps are as follows:

1) Select the noise source. The selected noise sources mainly come from the following noises: Basic star field, using improved Perlin noise to generate the basic distribution of stars. Nebula details, using Simplex noise or OpenSimplex noise to increase the details and fog effects of the nebula. Highlights and shine, using value noise to add the highlights and shine effects of stars to increase the dynamic range of the scene.

2) Define a weight function, which includes a weight W _star (x, y) corresponding to the basic star field, a weight W _nebula (x, y) corresponding to the nebula details, and a weight W _glow (x, y) corresponding to the highlight and flare.

The star density weight W _star (x, y) is used to control the density variation of the basic star field. It is set based on a gradient or random distribution to simulate the uneven distribution of stars in space. The settings are as follows:

W _star (x,y)＝1.0-Smoothstep(Noise _perlin (x,y))

Where: Smoothstep is a smooth step function used to smooth the transition of noise values; Noise _perlin (x, y) represents the Perlin noise distribution.

The nebula detail weight W _nebula (x,y) is used to control the visibility and complexity of the nebula. It depends on the local intensity of another noise pattern and is set as follows:

W _nebula (x,y)＝Noise _simplex (x,y) ^α

Where: α>1 is an exponent used to increase the local contrast in order to better highlight the nebula area.

The highlight and shine weight W _glow (x,y) is a random threshold to increase the brightness of the star at a specific position. The settings are as follows:

W _glow (x,y)＝Noise _value (x,y)>s

Where: s∈(0,1) represents the random threshold range.

3) Superimposing noise and weight, the final deep space background noise is:

Where: represents the Simplex noise distribution; Noise _value (x, y) represents the value noise distribution.

By adjusting the influence weight parameters between different noises, the deep space background noise suitable for the on-orbit scene of space targets can be obtained.

Then execute step three, in which relevant parameters are set so that multiple satellite space geometric models move in the on-orbit scene of the space target, thereby obtaining multiple groups of pictures corresponding to multiple satellites and different postures; the parameters include angular velocity of rotation and moment of inertia.

According to an embodiment of the present invention, the Transform component in Unity is used to control the position and rotation of the target in the scene, and the changes of these parameters are set by scripts to make the movement of the target reach the expected state. For example, if the target is set to rotate around the X axis at an angular velocity of 2° per second, the transform.eulerAngles.x parameter can be set to change by 2° per second by creating a script. In order to include target images with different poses in the pose estimation data set, a rotation angular velocity is set for the target, and the angular velocity has components on the three axes of the body, so that images in different poses can be obtained as much as possible.

To take into account the target motion law in the real space scene, the angular velocity of the three axes satisfies the Euler formula, where I _x , I _y , and I _z are the moments of inertia around the object's X, Y, and Z axes. Unity does not provide an option to set the moment of inertia directly in the editor, but you can set the moment of inertia indirectly through the script. Access the inertiaTensor and inertiaTensorRotation properties of the Rigidbody through the script, set the moment of inertia of the three axes to 100, 100, and 100. The expression of the Euler formula is as follows:

Since the satellite is also subject to external disturbances during its movement (such as solar radiation pressure, the earth's magnetic field, air resistance, etc.), these disturbances are not only random, they may also change with the satellite's attitude and orbital position. Therefore, a random disturbance term is introduced to reflect this dependence, and the angular velocity is dynamically adjusted according to the following formula:

ω _i (t)=a _i +b _i ·t+c _i ·sin(d _i ·θ _i (t))

Among them, i∈{ψ,θ,α}, respectively represents yaw, pitch, and roll; a _i , b _i , c _i and d _i respectively represent the initial velocity, acceleration, attitude influence coefficient and periodic modulation factor of attitude influence; θ _i (t) is the Euler angle at time t; ∈ _i (t) represents the random disturbance term, which simulates the influence of external environmental disturbance on the satellite's spin angular velocity. The Euler angle satisfies the following formula:

θ _i (t+Δt)=θ _i (t)+(ω _i (t)+∈ _i (t))Δt

Where Δt represents the time interval; ∈ _i satisfies the uniform distribution: ∈ _i (t)～U(-δ _i (θ _i (t),t),δ _i (θ _i (t),t)); δ _i represents the dynamic adjustment function related to the disturbance intensity, current attitude and time t. The specific value of δ depends on the disturbance intensity, current attitude and time t that you want to simulate. For example, for smaller satellites or satellites in lower orbits, a smaller δ value may be required to reflect the relatively small impact of environmental disturbances.

In Unity, add a camera through GameObject>Camera to shoot the target and obtain the target image. In the Hierarchy panel of the Unity editor, select the camera object to be modified. In the Inspector panel, find the Camera component of the camera object. Set the camera to perspective projection mode in the Projection module, adjust the size of the camera's field of view in the Field of View module, and set the near clipping plane and far clipping plane in Clipping Planes, which respectively represent the closest and farthest distances that the camera can display. Objects beyond this range will not be rendered. Adjust the position and size of the camera's viewport in the Viewport Rect module, and set the resolution of the image to be obtained. If the image resolution is 1024×1024, this means that the aspect ratio is 1:1. In the Unity camera component, set the "Aspect" property to 1. The position and direction of the camera can also be adjusted through the Transform component. In the process of generating the dataset image, in addition to the situation where the camera is facing the target, it is also necessary to consider the situation where the target deviates from the center of the picture in actual application. Therefore, the camera is set to rotate around its Z axis in a small range. This rotation can be set through a C# script. In the script, control the camera's transform.Rotate(xAngle,yAngle,zAngle) parameters, set zAngle to 2, and set xAngle and yAngle parameters to 0, so that the target can be located at different positions in the image. And set the target to the module corresponding to the spatial target in the Inspector panel so that the camera always shoots the target. The pictures taken in different postures are shown in Figures 6 and 7.

Then, step 4 is executed. In step 4, the multiple groups of pictures corresponding to the multiple satellites and different postures are processed to obtain a space target data set.

According to an embodiment of the present invention, a highly realistic and accurate space target pose data set is created through the Unity game development engine, including but not limited to space navigation, target tracking, and machine learning training. By simulating the on-orbit motion and operation process of the space target, a data set containing detailed position and pose information can be generated. The data set includes pose label data for the space target and the shooting camera, as well as the position label data of the target and the shooting camera.

First, get the initial position and attitude of the satellite and camera; for example, create or import the required 3D model of the space target in the Unity environment. Set the initial position and attitude of the space target and camera to ensure that they are in the appropriate initial state. Place the camera in the scene and simulate the position and orbital altitude of the observation point as needed. Adjust the position and orientation of the camera so that they can capture all angles of the space target.

Then, generate a posture data set. First, according to the above coordinate system definition, define a body coordinate system for the space target and the camera respectively. These coordinate systems are used as reference points to calculate the rotation matrix and translation vector. The first is the rotation matrix calculation: use the transformation component (Transform) in Unity to obtain the pitch angle (Pitch), yaw angle (Yaw) and roll angle (Roll) of the space target relative to its coordinate system. Based on these angle values, construct the rotation matrix R. The rotation matrix R can be calculated by the following formula:

R＝R _yaw (θ _yaw )·R _pitch (θ _pitch )·R _roll (θ _roll )

The rotation matrices correspond to rotations around the Y axis, X axis, and Z axis, respectively.

They are respectively rotation around the Z axis (Roll):

Rotation around the X axis (Pitch):

Rotation around the Y axis (Yaw):

Then, generate the position data set. First, calculate the translation vector matrix. According to the coordinates of the center point of the target body coordinate system and the center point of the camera coordinate system, calculate the translation vector T, which represents the spatial displacement from the target object coordinate system to the camera coordinate system. Then generate the position data: apply the translation vector T to update the position of the target object in the camera's view. In Unity, adjust the Transform.position property of the target object to reflect the new position relative to the camera.

Another embodiment of the present invention provides a space target dataset construction system, the system comprising:

A model building module configured to build a plurality of satellite space geometric models using three-dimensional modeling software;

A scene establishment module, configured to establish a space target on-orbit scene, the scene including: a scene containing the earth background, a scene containing different sunlight angles and effects, and a scene introducing deep space background noise;

An image acquisition module is configured to set relevant parameters so that multiple satellite space geometric models move in the space target on-orbit scene, thereby obtaining multiple sets of pictures corresponding to multiple satellites and different postures; the parameters include rotation angular velocity and moment of inertia;

The data set construction module is configured to process the multiple groups of pictures corresponding to the multiple satellites and in different positions to obtain a space target data set.

In this embodiment, optionally, the scene including the earth background in the scene establishment module is established as follows: create an earth model, adjust the size ratio of the earth model and the satellite space geometric model; add a texture map to the material properties of the earth model, and adjust the transparency and saturation to achieve the real space earth background, wherein the texture map types include color maps, height maps, and cloud maps; set the relevant parameters of the illumination and noise of the shadow part of the earth, wherein the parameters of the illumination of the shadow part of the earth include light type, color value, intensity, position, angle, and the parameters of the noise include noise texture and transparency; set the rotation angular velocity and the rotation axis position to control the rotation of the earth model, thereby realizing the earth rotation scene;

The scene containing different sunlight angles and effects is established as follows: establish a sun model and adjust the proportion of the sun sphere; configure the self-luminous material color of the sun sphere; set the directional light source at the center of the sun sphere and the color of the directional light source; adjust the intensity of the directional light source; set the angle range between the directional light source and the surface of the earth model;

The scene introducing deep space background noise is established as follows: creating a deep space background and setting the sky box material to a deep space galaxy; using an improved fusion noise algorithm to generate random and visually continuous deep space background noise, and then generating a noise map.

The functions of the space target dataset construction system described in this embodiment can be described by the aforementioned space target dataset construction method. Therefore, for the parts not described in detail in this embodiment, please refer to the above method embodiments and will not be repeated here.

Although the present invention has been described according to a limited number of embodiments, it will be apparent to those skilled in the art, with the benefit of the above description, that other embodiments are contemplated within the scope of the invention thus described. The disclosure of the present invention is intended to be illustrative rather than restrictive of the scope of the invention, which is defined by the appended claims.

Claims (10)

1. A method for constructing a space target dataset, comprising: Step 1: Use 3D modeling software to build multiple satellite space geometry models; Step 2: Establishing an on-orbit scene of a space target, wherein the scene includes: a scene with an earth background, a scene with different sunlight angles and effects, and a scene with deep space background noise introduced; Step 3: setting relevant parameters so that multiple satellite space geometric models move in the space target on-orbit scene, thereby obtaining multiple sets of pictures corresponding to multiple satellites and different postures; the parameters include rotation angular velocity and moment of inertia; Step 4: Process the multiple groups of images corresponding to the multiple satellites and different postures to obtain a space target data set. 2. According to a method for constructing a space target data set according to claim 1, it is characterized in that the scene containing the earth background is established as follows: create an earth model, adjust the size ratio of the earth model and the satellite space geometric model; add a texture map to the material properties of the earth model, and adjust the transparency and saturation to achieve a real space earth background, wherein the texture map types include color maps, height maps, and cloud maps; set relevant parameters of the illumination and noise of the shadow part of the earth, wherein the parameters of the illumination of the shadow part of the earth include light type, color value, intensity, position, and angle, and the parameters of the noise include noise texture and transparency; set the rotation angular velocity and the rotation axis position to control the rotation of the earth model, thereby realizing the earth rotation scene. 3. According to a method for constructing a space target data set according to claim 1, it is characterized in that the scene containing different sunlight angles and effects is established as follows: establish a solar model and adjust the proportion of the solar sphere; configure the self-luminous material color of the solar sphere; set the directional light source at the center of the solar sphere and the color of the directional light source; adjust the intensity of the directional light source; set the angle range between the directional light source and the surface of the earth model. 4. According to claim 1, a method for constructing a space target dataset is characterized in that the scene introducing deep space background noise is established as follows: creating a deep space background and setting the sky box material to a deep space galaxy; using an improved fusion noise algorithm to generate random and visually continuous deep space background noise, and then generating a noise map. 5. According to claim 4, a method for constructing a space target dataset is characterized in that the use of an improved fusion noise algorithm to generate random and visually continuous noise includes: selecting a noise source, the noise source includes using improved Perlin noise to generate a basic distribution of stars, using Simplex noise to increase nebula details and fog effects, and using value noise to add star highlights and sparkle effects; defining a weight function, the weight function includes a weight W _star (x, y) corresponding to the basic star field, a weight W _nebula (x, y) corresponding to the nebula details, and a weight W _glow (x, y) corresponding to the highlight and sparkle; superimposing the noise and the weight function to generate deep space background noise as follows: Wherein, Noise _perlin (x, y) represents Perlin noise distribution; Noise _simplex (x, y) represents Simplex noise distribution; Noise _value (x, y) represents value noise distribution. 6. A method for constructing a space target dataset according to claim 5, characterized in that the weight W _star (x, y) corresponding to the basic star field is set according to the following formula: W _star (x,y)＝1-Smootstep(Noise _perlin (x,y)) Where Smootstep represents a smooth step function, which is used to smooth the transition of noise values; The weight W _nebula (x, y) corresponding to the nebula detail is set according to the following formula: W _nebula (x,y)＝Noise _simplex (x,y) ^α In the formula, α>1 represents an exponential increase in local contrast; The weight W _glow (x, y) corresponding to highlight and shine is set according to the following formula: W _glow (x,y)＝Noise _value (x,y)>s Where s∈(0,1) represents a random threshold. 7. A method for constructing a space target data set according to claim 1, characterized in that the setting of relevant parameters so that multiple satellite space geometric models move in the space target on-orbit scene includes: controlling the change of the position and rotational angular velocity of the satellite in the space target on-orbit scene, wherein the rotational angular velocity has components on the three axes of the satellite body coordinate system, and the angular velocities of the three axes satisfy the Euler formula; using a camera to simulate shooting images to obtain multiple groups of pictures corresponding to multiple satellites and different postures; wherein a random disturbance term is introduced to dynamically adjust the rotational angular velocity according to the following formula: ω _i (t)=a _i +b _i ·t+c _i ·sin(d _i ·θ _i (t)) Where i represents one channel among the yaw angle, pitch angle and roll angle; a _i , b _i , c _i and d _i represent the initial velocity, acceleration, attitude influence coefficient and periodic modulation factor of attitude influence respectively; θ _i (t) represents the Euler angle at time t and satisfies the following formula: θ _i (t+Δt)=θ _i (t)+(ω _i (t)+∈ _i (t))Δt Where Δt represents the time interval; ∈ _i (t) represents the random disturbance term and satisfies the uniform distribution U: ∈ _i (t)～U(-δ _i (θ _i (t),t),δ _i (θ _i (t),t)), δ _i represents the dynamic adjustment function related to the disturbance intensity, current posture and time t. 8. A method for constructing a space target data set according to claim 1, characterized in that the processing of multiple groups of pictures corresponding to multiple satellites and different postures to obtain the space target data set includes: obtaining the initial position and posture of the satellite and the camera; using the satellite body coordinate system and the camera body coordinate system as reference points, calculating the rotation matrix and the translation vector, including: obtaining the pitch angle, yaw angle and roll angle of the satellite relative to its own body coordinate system, constructing a rotation matrix according to these angle values as follows: R=R _yaw (θ _yaw )·R _pitch (θ _pitch )·R _roll (θ _roll ), wherein R _yaw (θ _yaw ) represents rotation around the Y axis, R _pitch (θ _pitch ) represents rotation around the X axis, and R _roll (θ _roll ) represents rotation around the Z axis, and generating posture data according to the rotation matrix; calculating the translation vector according to the coordinates of the center point of the satellite body coordinate system and the coordinates of the center point of the camera body coordinate system, and generating position data according to the translation vector. 9. A space target dataset construction system, comprising: A model building module configured to build a plurality of satellite space geometric models using three-dimensional modeling software; A scene establishment module, configured to establish a space target on-orbit scene, the scene including: a scene containing the earth background, a scene containing different sunlight angles and effects, and a scene introducing deep space background noise; An image acquisition module is configured to set relevant parameters so that multiple satellite space geometric models move in the space target on-orbit scene, thereby obtaining multiple sets of pictures corresponding to multiple satellites and different postures; the parameters include rotation angular velocity and moment of inertia; The data set construction module is configured to process the multiple groups of pictures corresponding to the multiple satellites and in different positions to obtain a space target data set. 10. A space target data set construction system according to claim 9, characterized in that the scene containing the earth background in the scene establishment module is established as follows: create an earth model, adjust the size ratio of the earth model and the satellite space geometric model; add a texture map to the material properties of the earth model, and adjust the transparency and saturation to achieve the real space earth background, wherein the texture map types include color maps, height maps, and cloud maps; set the relevant parameters of the illumination and noise of the shadow part of the earth, wherein the parameters of the illumination of the shadow part of the earth include light type, color value, intensity, position, angle, and the parameters of the noise include noise texture and transparency; set the rotation angular velocity and the rotation axis position to control the rotation of the earth model, thereby realizing the earth rotation scene; The scene containing different sunlight angles and effects is established as follows: establish a sun model and adjust the proportion of the sun sphere; configure the self-luminous material color of the sun sphere; set the directional light source at the center of the sun sphere and the color of the directional light source; adjust the intensity of the directional light source; set the angle range between the directional light source and the surface of the earth model; The scene introducing deep space background noise is established as follows: creating a deep space background and setting the sky box material to a deep space galaxy; using an improved fusion noise algorithm to generate random and visually continuous deep space background noise, and then generating a noise map.