A Spatial Target Recognition Sample Generation System and Method Based on Imaging Simulation

Abstract

This invention discloses a space target recognition sample generation system and method based on imaging simulation. The system includes: a basic data layer for receiving and storing three-dimensional models of space targets, typical space target spectral data, orbital attitude data, and simulation configuration parameters; a data processing layer for preprocessing texture images required for target characteristic association, managing simulation data, and generating scenario data files; an algorithm and model layer for calculating imaging geometric relationships and radiative transfer, performing impact analysis on the spacecraft platform, and mapping and labeling feature components; a service scheduling layer for implementing imaging simulation labeling tasks by calling the algorithm and model layer based on load balancing according to simulation calculation requests; and an application layer for loading the model and data from the basic data layer according to the scenario data file, associating target characteristics, constructing a space target scene, and generating and labeling simulation images by combining ray tracing.

Description

Technical Field

[0001] This invention belongs to the field of space target imaging simulation technology, and in particular relates to a space target recognition sample generation system and method based on imaging simulation. Background Technology

[0002] With the development and application of modern high technology, especially aerospace technology, space has become an important venue for human scientific exploration, space activities, and even future economic, social, and military activities. The most important elements in space are the various spacecraft operating in orbit, providing services closely related to economic and social life, such as navigation, communication, surveying, and environmental monitoring. Currently, humans have sent over 26,000 objects into space, and approximately tens of thousands remain in orbit, of which only 6% are still operational spacecraft; the rest are space debris.

[0003] The development of human space activities has spurred the rapid advancement of technologies related to space target surveillance and space operations. Space target surveillance tracks, monitors, identifies, and catalogs various spacecraft and debris in space, obtaining crucial target characteristics such as size, shape, and orbital parameters. This allows for the determination of target threat levels, alerting for space collisions, and providing safety warnings. Space operations perform tasks such as space assembly, maintenance, resupply, and recapture of spacecraft, altering their on-orbit status, repairing damaged spacecraft to extend their lifespan, recovering defunct spacecraft, and proactively removing various types of space debris.

[0004] Space target surveillance and space operations have created an urgent need for space target identification. Space target identification typically focuses on the specific properties of targets, such as geometry and spectral characteristics, leading to identification methods based on geometric features, spectral properties, radiation characteristics, attitude characteristics, and orbital characteristics. However, these single-feature-based methods have limitations. Space-based space target identification, as an emerging field, is seeing the increasing application of many popular target identification approaches, such as genetic algorithms, neural networks, pattern recognition, and deep learning. This is beneficial for further developing my country's integrated space-ground space target surveillance system and space situational awareness capabilities. However, the primary problem with applying neural networks and deep learning to space-based space target identification is the severe shortage of training sample image data.

[0005] Currently, space target recognition sample data is mainly generated through manual annotation based on existing space-based observation images. However, space-based observation images are limited by observation conditions and the number of observations, making it difficult to obtain space target images under various conditions to meet different application needs. Furthermore, the space environment is complex, and laboratory physical simulations require significant manpower, material resources, and financial investment, resulting in low experimental efficiency and difficulty in verifying equivalence. In contrast, space target imaging simulation can quickly and flexibly construct various space scenarios for different space targets, obtaining simulated images under different conditions such as time, location, and space environment, and annotating the feature components to be identified. This provides a large-sample simulated image source for various image processing algorithms such as space target recognition, demonstrating enormous application potential. Summary of the Invention

[0006] The purpose of this invention is to overcome the shortcomings of the prior art and to propose a spatial target recognition sample generation system and method based on imaging simulation.

[0007] To achieve the above objectives, this invention proposes a spatial target recognition sample generation system based on imaging simulation, the system comprising:

[0008] The basic data layer is used to receive and store 3D models of space targets, spectral data of typical space targets, orbital attitude data, and simulation configuration parameters;

[0009] The data processing layer is used to preprocess the texture images required for the correlation of target characteristics, manage simulation data based on the imaging mission planning file and satellite payload test data, and generate scenario data files based on the imaging mission planning file.

[0010] The algorithm and model layer is used to calculate imaging geometry, calculate radiative transfer, analyze the impact of aircraft platforms, and map and label feature components.

[0011] The service scheduling layer, based on simulation calculation requests and load balancing, implements imaging simulation annotation tasks by calling the algorithm and model layers; and

[0012] The application layer is used to load the model and data from the basic data layer based on the given data file, associate target characteristics, construct a spatial target scene, and combine ray tracing to generate and annotate specific simulation images by calling the service scheduling layer, thereby obtaining spatial target recognition samples.

[0013] As an improvement to the above system, the basic data layer includes a three-dimensional model of the space target, typical space target spectral data, space target orbital attitude data, spacecraft orbital attitude data, and simulation configuration parameters; the space target orbital attitude data and spacecraft orbital attitude data include historical orbital attitude data, real-time received orbital attitude data, and simulated orbital attitude data.

[0014] As an improvement to the above system, the data processing layer includes:

[0015] The texture image preprocessing module is used to collect the corresponding spectral reflectance and directional reflectance based on the spectral data of typical spatial targets and store them in the spectral database; it is also used to classify the texture images of the three-dimensional models of spatial targets using automatic and semi-automatic image classification algorithms and save the classification results.

[0016] The simulation data management module is used to generate simulation data based on satellite payload test data and imaging mission planning documents, and to perform statistical analysis to generate simulation reports; and

[0017] The mission scenario data processing module is used to set the scenario time and mission type based on the imaging mission planning file and generate the scenario file.

[0018] As an improvement to the above system, the algorithm and model layer includes:

[0019] The target pose calculation module is used to calculate the position and attitude of the space target, as well as the spatial position and attitude against the background of the starry sky, including the sun, moon and planets, based on the space target's orbital attitude data.

[0020] The aircraft attitude calculation module is used to calculate the spatial position and attitude of sensors on the aircraft platform based on the aircraft's orbital attitude data.

[0021] The imaging geometry module is used to transform scene coordinates into image coordinates through coordinate transformation between the geocentric equatorial coordinate system, the aircraft platform body coordinate system, and the camera image space coordinate system.

[0022] The radiative transfer module is used to calculate the radiance reaching the entrance pupil of the camera's optical system and form a radiance image.

[0023] The sensor module is used to simulate the sensor imaging process and generate simulated sensor images;

[0024] The platform impact module models the effects of linear vibration, high-frequency jitter, and low-frequency sinusoidal disturbances on imaging, obtaining simulated images output by the camera; and

[0025] The feature component annotation module is used to annotate simulation images and generate sample images based on the automatic feature component annotation method of full-chain simulation.

[0026] As an improvement to the above system, the sensor module processing procedure specifically includes:

[0027] The simulated camera is set to the parameters of an on-orbit real camera, including aperture, focal length and shutter speed. Using the corresponding camera spectral response curve, the characteristics of the optical camera are simulated to generate image data in various formats, including full-spectrum images and grayscale images with 8 to 16-bit grayscale levels.

[0028] Simulates image defects caused by inherent problems in camera optics or electronic systems, including overexposure, defocus, motion blur, and smear effects, generating images with non-uniform defects.

[0029] As an improvement to the above system, the processing procedure of the feature component annotation module specifically includes:

[0030] Obtain the feature component information that needs to be annotated;

[0031] Based on the feature component information, different colors are assigned to the feature components to be annotated in the original spatial target 3D model to obtain the target annotated 3D model and color mapping table;

[0032] Simulation calculations are performed on the 3D model of the target annotation to obtain the target annotation simulation image;

[0033] Extract different color regions from the target-annotated simulation image to form region description information;

[0034] Find the color mapping table to create a template for the feature component region;

[0035] The feature component region template is mapped onto the imaging simulation image to achieve the annotation of the corresponding feature component;

[0036] The parameter configuration information during the image generation process is retrieved, including the orbital attitude and sensor parameters of the aircraft and target related to the image, and these are combined with the labeled simulation image to form a sample image description file according to the sample data description format.

[0037] As an improvement to the above system, the service scheduling layer includes:

[0038] The computing service interaction module is used for computing service management, computing service control, computing process management and orchestration, and submitting simulation computing requests to the computing service scheduling module; it is also used to submit query requests for computing service call records, computing service exception logs, computing process call records and computing process exception logs to the computing service scheduling module; and it is also used to save and query computing results.

[0039] The computing service scheduling module is used to send simulation computing requests to computing service computing node modules with sufficient computing resources based on the simulation computing request status and the scheduling feedback information of the computing service computing node modules, and to realize the call based on load balancing; it is also used to determine whether existing historical computing results can be used; and it is also used to send scheduling status information, computing service call records, computing service exception logs, computing process call records, and computing process exception logs to the computing service interaction module.

[0040] The computing service computing node module is used to complete the imaging calculation of the simulated image pixels according to the scheduling information of the computing service scheduling module; it is also used to send node status information to the computing service interaction module.

[0041] As an improvement to the above system, the processing procedure of the computing service computing node module specifically includes:

[0042] Upon receiving a computing task from the computing service scheduling module, the module parses the task to obtain the computing process, initial parameters, progress, and intermediate results.

[0043] The computation process is broken down into sub-computation tasks, corresponding computation services are executed, and computation results are saved. If the entire computation process is completed, the computation service scheduling module is notified that the computation task has been completed. If there are still subsequent sub-computation tasks, the sub-computation tasks are fed back to the computation service scheduling module according to the resource usage and status.

[0044] As an improvement to the above system, the application layer includes:

[0045] The target feature association module is used to perform texture mapping based on the spatial target 3D model and typical spatial target spectral data of the basic data layer. It specifies the material type in the typical spatial target spectral data for each texture image classification result, generates the corresponding material texture mapping file, obtains target feature data, and realizes the association between the geometric information and spectral information of the spatial target 3D model in the scene.

[0046] The spatial scene organization module is used to set and deploy the initial situation according to the experimental task intent, obtain the corresponding target and aircraft position and attitude data from the algorithm and model layers, and generate a simulation scene by combining target characteristic data; and

[0047] The ray tracing module is used to calculate the transmission process of light between the light source, spatial target, and sensor to obtain the correspondence between camera pixels and the imaging scene.

[0048] As an improvement to the above system, the processing procedure of the spatial scene organization module specifically includes:

[0049] The initial situation is set and deployed according to the intention of the simulation experiment, forming a scenario file for the imaging simulation task. This file contains the three-dimensional model of the spatial target, interface relationships, and event association information required for the simulation experiment, forming a simulation scenario script file.

[0050] By parsing the simulation scenario script file, the computational model of the algorithm and model layer is called to generate the simulation scenario of the scenario task; at the same time, KD tree is used to organize and manage the entire imaging space scene including space targets, satellites, the sun, and space background.

[0051] On the other hand, this invention proposes a spatial target recognition sample generation method based on imaging simulation, implemented based on the above-mentioned system, the method comprising:

[0052] Based on the three-dimensional model of the space target and the spectral data of typical targets, multi-characteristic modeling of the target is achieved through the target characteristic association module;

[0053] Based on the multi-characteristic modeling results of the target, a simulation scene is generated using the spatial scene organization module;

[0054] The target pose calculation module and the aircraft pose calculation module are used to obtain the position and attitude data of the space target and the aircraft, respectively.

[0055] Run the ray tracing module to obtain the correspondence between camera pixels and the image scene;

[0056] The simulation calculation request is submitted to the computing service scheduling module through the computing service interaction module;

[0057] The computing service scheduling module is based on load balancing and generates computing service nodes according to the simulation computing request. Each computing service node can process the imaging calculation of one pixel of the simulation image.

[0058] The imaging geometry module enables the conversion from scene coordinates to image coordinates.

[0059] Run the radiative transfer module to calculate the radiance image at the camera's entrance pupil;

[0060] The simulated image output by the camera is obtained through the sensor module and the platform influence module;

[0061] Based on the pixel information of the feature components recorded by the feature component annotation module during the simulation process, the feature components in the simulation image are classified, labeled, and annotated.

[0062] Compared with the prior art, the advantages of the present invention are:

[0063] 1. This invention proposes a method for generating space target recognition samples, which can quickly simulate and generate a set of on-orbit high-fidelity feature component labeled images of different space targets under different conditions, providing sample data for the training, testing and verification of intelligent algorithms for space targets, and solving the problem of insufficient sample data when space-based observation images are used for intelligent algorithm applications;

[0064] 2. The method of this invention can generate simulated images with a similarity of more than 90% to actual images of the aircraft. It can be used to carry out the optimization design and demonstration evaluation of space imaging systems for final application effects, reduce the development cost of space imaging systems, and shorten the development cycle.

[0065] 3. This invention proposes an automatic feature component annotation method based on full-chain simulation, which can effectively improve the accuracy of image annotation results and the fineness of the annotation contour, while solving the problems of low efficiency and low quality of manual annotation. Attached Figure Description

[0066] Figure 1 This is a hierarchical relationship diagram of a spatial target recognition sample generation system based on imaging simulation.

[0067] Figure 2 This is a diagram showing the module composition of a space target recognition sample generation system based on imaging simulation.

[0068] Figure 3 This is a flowchart of a spatial target recognition sample generation system based on imaging simulation.

[0069] Figure 4 This is a flowchart of the automatic annotation method for feature components based on full-chain simulation. Detailed Implementation

[0070] The system of this invention is mainly aimed at the urgent need for space target recognition samples in space target monitoring and space operations. Based on the inter-satellite relative pose, target structural parameters, optical camera parameters and other parameters of real missions, it constructs space scenes under different conditions, simulates and generates high frame rate and high realism target on-orbit images, and annotates the image feature components. The annotated images are classified and managed as a test sample set, which is used to test and verify the performance of the space target intelligent recognition algorithm loaded on the spacecraft before launch and during on-orbit testing.

[0071] The main functions of this invention include:

[0072] 1. Rapidly simulate and generate a high-fidelity on-orbit feature component labeled image sample set of different space targets under different conditions, providing sample data for the training, testing and verification of intelligent algorithms for space targets, and solving the problem that the research on intelligent recognition algorithms cannot be carried out due to insufficient space-based observation images;

[0073] 2. Simulated images with a similarity greater than 90% to actual images of the aircraft can be used to conduct optimized design and evaluation of space imaging systems for final application effects, thereby reducing the development cost of space imaging systems and shortening the development cycle.

[0074] 3. The automatic annotation method for feature components based on full-chain simulation can effectively improve the accuracy of image annotation results and the fineness of the annotation contour, while solving the problems of low efficiency and low quality of manual annotation.

[0075] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and embodiments.

[0076] Example 1

[0077] like Figure 1 As shown, Embodiment 1 of the present invention proposes a spatial target recognition sample generation system based on imaging simulation. It consists of a basic data layer, a data processing layer, an algorithm and model layer, a service scheduling layer, and an application layer.

[0078] The basic data layer is used to receive and store three-dimensional models of space targets, spectral data of typical space targets, orbital attitude data, and simulation configuration parameters. The basic data layer includes three-dimensional models of space targets, spectral data of typical space targets, orbital attitude data of space targets, orbital attitude data of spacecraft, and simulation configuration parameters.

[0079] The data processing layer is primarily responsible for texture image preprocessing, simulation data management, and mission scenario data processing required for target characteristic association. It is used for preprocessing texture images needed for target characteristic association, managing simulation data based on imaging mission planning documents and satellite payload test data, and generating scenario data files based on imaging mission planning documents.

[0080] The algorithm and model layer is used to calculate imaging geometry, radiative transfer, and impact analysis of spacecraft platforms. It also includes models for mapping and labeling feature components.

[0081] The service scheduling layer is used to implement imaging simulation annotation tasks by calling the algorithm and model layers based on load balancing, according to the simulation calculation requests.

[0082] The application layer is used to load the model and data from the basic data layer based on the target data file, associate the target characteristics, construct the spatial target scene, and combine ray tracing to generate and annotate specific simulation images by calling the service scheduling layer, thereby obtaining spatial target recognition samples.

[0083] The application layer, based on the space target imaging simulation framework, conducts imaging simulations under various conditions and obtains simulation results.

[0084] The spatial target recognition sample generation system based on imaging simulation includes modules such as... Figure 2 As shown, it includes: image preprocessing module, simulation data management module, mission scenario data processing module, spatial scene organization module, target characteristic association module, target pose calculation module, aircraft pose calculation module, ray tracing module, imaging geometry module, radiative transfer module, sensor module, platform influence module, feature component annotation module, computing service interaction module, computing service scheduling module, and computing service computing node module. The specific divisions are as follows:

[0085] The basic data layer includes 3D models of space targets, spectral data of typical space targets, orbital attitude data of space targets, orbital attitude data of space vehicles, and simulation configuration parameters; the orbital attitude data of space targets and the orbital attitude data of space vehicles include historical orbital attitude data, real-time received orbital attitude data, and simulated orbital attitude data.

[0086] The data processing layer includes a texture image preprocessing module, a simulation data management module, and a task scenario data processing module.

[0087] The algorithm and model layer includes a target pose calculation module, an aircraft pose calculation module, an imaging geometry module, a radiation transfer module, a sensor module, a platform influence module, and a feature component annotation module.

[0088] The service scheduling layer includes a computing service interaction module, a computing service scheduling module, and a computing service computing node module.

[0089] The application layer includes a target characteristic association module, a spatial scene organization module, and a ray tracing module.

[0090] The workflow of the space target recognition sample generation system based on imaging simulation is as follows: Figure 3 As shown.

[0091] The main functions of each module are described below:

[0092] The space scene organization module sets up and deploys the initial situation according to the experimental mission intent of the simulation experimenters, forming a computer-recognizable simulation scenario file. This file includes all instance models, interface relationships, and event association information required for the simulation experiment, forming a simulation experiment planning file or simulation scenario script file. Simultaneously, it provides basic scene data selection and preparation for the simulation process. This is mainly achieved by parsing the simulation scenario script, loading the simulation model, calling the simulation calculation model, and generating the simulation scene for the scenario task, which drives the software to conduct high-precision, high-realism imaging simulations. Furthermore, a KD-tree is used to organize and manage the entire imaging space scene, including space targets, satellites, the sun, and the space background.

[0093] Target characteristic association module: Collects the spectral reflectance and directional reflectance of typical targets and stores them in the spectral database; classifies the texture images of 3D models using automatic and semi-automatic image classification algorithms and saves the classification results; performs texture mapping, assigns a material type from the spectral database to each texture image classification result, and generates the corresponding material texture mapping file (*.mtm file), realizing the association between the geometric information and spectral information of 3D models in the scene.

[0094] Target pose calculation module: Calculates the position and attitude of spatial targets, as well as the spatial position and attitude of star backgrounds such as the sun, moon, and planets, based on the established spatial scene relationships.

[0095] Aircraft pose calculation module: Calculates the spatial position and attitude of sensors on the aircraft platform. Since the sensors are mounted on the aircraft platform, the observation direction will rotate with the rotation of the platform. Therefore, the position and attitude of the sensor is the accumulation of the sensor's pose relative to the aircraft platform and the aircraft platform's pose in the world coordinate system.

[0096] Ray tracing module: Calculates the transmission process of light rays between the light source, spatial target, and sensor. The ray tracing module allows the system to generate images at different resolutions. For each pixel in the final image (an imaging cell), ray tracing generates one or more light rays (oversampling). The final pixel result is formed by simulating the optical characteristics of these light rays in the simulated scene. Therefore, configuring the pixel size in the ray tracing module allows for high-precision or low-precision imaging simulation in the software.

[0097] Imaging geometry module: This module calculates the positional relationship between image points and corresponding spatial targets. It involves coordinate transformations between the geocentric equatorial coordinate system, the spacecraft platform's body coordinate system, and the camera's image space coordinate system.

[0098] Radiative Transfer Module: Based on radiometry and photoelectric imaging theory, this module calculates the radiance at the entrance pupil of the camera's optical system, generating a radiance image. The module needs to calculate on-orbit radiation based on the spacecraft's orbit and attitude data, taking into account real star catalog data for calculating radiation from point sources in the starry sky, and solar system scenarios for calculating direct solar radiation and reflected radiation from the Earth. When there are other external active illumination sources such as visible light, the module also needs to calculate the on-orbit radiation under active illumination conditions.

[0099] Sensor Module: This module simulates the sensor imaging process, including light energy radiation transmission, photoelectric conversion models, information modulation models, and special effects simulation models. Imaging process simulation requires setting the simulated camera to the parameters of a real on-orbit camera, including aperture, focal length, and shutter speed. Using the corresponding camera's spectral response curve, it realistically simulates the characteristics of the optical camera, generating image data in various formats, such as full-spectrum images and grayscale images with 8 to 16 bits of grayscale. Special effects simulation simulates image defects caused by inherent problems in the camera's optical or electronic systems (such as overexposure, defocus, motion blur, and smear effects), generating images with non-uniform defects.

[0100] Platform Influence Module: This module models the effects of linear vibration, high-frequency jitter, and low-frequency sinusoidal disturbances of the platform on imaging.

[0101] The Computing Service Interaction Module: Based on the simulation scenario scripts and basic simulation scenarios provided by the Spatial Scenario Organization Module, it forms a simulation computing process and provides users with an interactive configuration management interface for functions such as computing service management, computing process management, and computing cluster management.

[0102] The computing service scheduling module is responsible for receiving call requests from the computing service interaction module, allocating computing tasks to computing nodes with sufficient computing resources based on the status of each computing node, and adjusting the number of computing tasks on each computing node to achieve load balancing across the entire system. Furthermore, the computing service scheduling module centrally stores computing results and can determine whether a computing task can utilize historical computing results based on the parameters in the computing task, thereby reducing the duplication of computing resources through result reuse.

[0103] The computing service computing node module is both the executor of specific computing services and a sub-control center. When a computing service computing sub-node receives a computing task from the computing service scheduling module or other computing nodes, it parses the computing process, initial parameters, progress, and intermediate results from the task to determine the sub-computing task to execute. It then executes the corresponding computing service, saves the results, and if the entire computing process is complete, it notifies the computing service scheduling module that the task is finished. If there are still subsequent sub-computing tasks, it assigns them to itself or other computing nodes based on their resource usage and status.

[0104] Feature component annotation module: This is achieved through an automatic feature component annotation method based on full-chain simulation. In the data preparation phase, the names of the feature components to be annotated are obtained. Based on the original 3D spatial target model, each feature component is assigned a different color, and a mapping table between feature component names and colors (called a color mapping table) is established. The processed 3D spatial target model (called the target annotated 3D model) is stored. In the simulation phase, the target annotated 3D model replaces the original 3D spatial target model in the imaging simulation scene, and the entire simulation process is executed without changing other parameters to form a simulation image (called the target annotated simulation image). In the annotation processing phase, regions with similar colors are extracted from the target annotated simulation image. The bounding rectangles, rotated rectangles, or enclosing polygons of each region are calculated and compared with the color mapping table to find the image pixel regions where each feature component is located, forming feature component region templates. Since the size and resolution of the imaging simulation image and the target annotated simulation image are consistent, the corresponding feature components can be annotated by mapping the feature component region templates onto the imaging simulation image. Finally, the parameter configuration information from the image generation process is retrieved, including auxiliary information such as the orbital attitude (including relative pose) and sensor parameters of the aircraft and target related to the image. This information, along with the annotated simulation image, is then used to form the final annotation file according to the sample data description format. The module's workflow is as follows: Figure 4 As shown.

[0105] Example 2

[0106] Embodiment 2 of the present invention proposes a method for generating spatial target recognition samples based on imaging simulation, which is based on the system implementation of Embodiment 1 and specifically includes:

[0107] (1) Based on the three-dimensional model of the space target and the spectral data of typical targets, the multi-characteristic modeling of the target is realized through the target characteristic association module.

[0108] (2) Based on the target multi-characteristic modeling results, the simulation scene is generated using the spatial scene organization module.

[0109] (3) The position and attitude data of the space target and the aircraft are obtained through the target pose calculation module and the aircraft pose calculation module, respectively.

[0110] (4) Run the ray tracing module to obtain the correspondence between camera pixels and the imaging scene.

[0111] (5) Submit a simulation calculation request to the computing service scheduling module through the computing service interaction module;

[0112] (6) The computing service scheduling module generates computing service computing nodes based on the simulation computing request after considering load balancing. Each computing service computing node can process the imaging calculation of one pixel of the simulation image.

[0113] (7) The conversion from scene coordinates to image coordinates is realized through the imaging geometry module.

[0114] (8) Run the radiative transfer module to calculate the radiance image at the camera entrance pupil.

[0115] (9) The simulated image output by the camera is obtained through the sensor module and the platform influence module.

[0116] (10) Based on the feature component pixel information recorded by the feature component annotation module during the simulation process, classify, mark and annotate the feature components in the simulation image.

[0117] Technical effects:

[0118] The technical advantages of the space target recognition sample generation method and system based on imaging simulation are mainly reflected in the generation rate of low-precision simulation labeled images (512x512 resolution) of no less than 4 frames / s, the generation time of each frame of high-precision simulation labeled images (5120x5120 resolution) of no more than 10s, and the similarity between the generated simulation images and the real images taken by the on-orbit spacecraft is greater than 90%, as detailed below.

[0119] (1) Experiment on low-precision simulated labeled images

[0120] The experiment was conducted using 10 servers in a high-concurrency execution mode, sending 10 simulated annotation image calculation requests consecutively. Each request generated 13 images with a resolution of 512x512, resulting in a total of 130 low-precision images generated concurrently. The experimental results are shown in the table below.

[0121] Table 1

[0122]

[0123] High-precision simulation labeled image experiment

[0124] The experiment was conducted using 10 servers in a high-concurrency execution mode, sending 10 simulated annotation image calculation requests consecutively. Each request generated 10 images with a resolution of 5120x5120, resulting in a total of 100 high-precision images generated concurrently. The experimental results are shown in the table below.

[0125] Table 2

[0126]

[0127] Simulated image similarity evaluation

[0128] To verify the accuracy of the simulation method and results, instead of using traditional image quality evaluation methods that focus on the simulated images themselves, a consistency evaluation method between the simulated images and the actual images captured by the aircraft was adopted. Based on actual images captured by optical cameras during real space escort missions, imaging simulations were conducted using the same parameter settings as when the aircraft images were acquired. Simulated images and actual aircraft images of 5120×5120 resolution were obtained under three different observation conditions. The consistency between the simulated images and the actual aircraft images was quantitatively evaluated from four aspects: radiometric consistency, resolution similarity, geometric distortion similarity, and structural similarity.

[0129] a) Radiation Consistency Assessment

[0130] The relative error between the radiation intensity value (Isim) of a typical material within the simulated image region and the radiation intensity value (Isat) of a typical material within the same region in the actual aircraft image can be expressed as:

[0131]

[0132] The radiometric consistency was 95.58% under observation condition 1, 96.55% under observation condition 2, and 95.50% under observation condition 3. The average result was a similarity score of 95.88%.

[0133] b) Resolution similarity analysis

[0134] The precise dimensions of each object on the aircraft are measured separately. The number of pixels on both the longer and shorter sides of the same target in the actual and simulated images of the aircraft are calculated to obtain the resolution of the two components. The average value is then used to obtain the actual image resolution. The similarity between the resolutions of the satellite image and the simulated image is calculated using the following formula:

[0135]

[0136] The average resolution similarity obtained under the three observation conditions was 95.68%.

[0137] c) Geometric distortion similarity analysis

[0138] m random control points were selected from the simulated image and the actual image taken by the aircraft. For each pair of control points, the distance ratio was calculated. The image length distortion values ​​under the three observation conditions were 1.79%, 4.05%, and 5.36%, respectively. The average length distortion value was 3.74%. Therefore, the geometric distortion similarity was approximately 96.26%.

[0139] d) Structural similarity assessment

[0140] Structural similarity (SSIM) primarily reflects image quality distortion through the correlation between pixels. A value closer to 1 indicates a higher similarity between two images.

[0141]

[0142] That is, given two images x and y, where μ x μ is the average pixel value in image x. y It is the average pixel value in image y. It is the pixel variance of image x. σ is the pixel variance of image y. xy Let c1 be the pixel covariance of the image x and y, where c1 = (k1L). 2 c2 = (k2L) 2 These are constants used to maintain parameter stability, where L represents the dynamic range of pixel values, and k1 and k2 are typically set to 0.01 and 0.03 based on experience. The average structural similarity under the three sets of observation conditions is 98.65%.

[0143] Based on comprehensive radiation consistency assessment, resolution similarity analysis, geometric distortion similarity analysis, and structural similarity analysis, the similarity between the simulated image and the actual image of the aircraft is greater than 95%.

[0144] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to the embodiments, those skilled in the art should understand that modifications or equivalent substitutions to the technical solutions of the present invention do not depart from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A space object identification sample generation system based on imaging simulation, characterized by, The system comprises: a basic data layer for receiving and storing spatial target three-dimensional models, typical spatial target spectral data, orbit attitude data and simulation configuration parameters; a data processing layer for pre-processing texture images required for target characteristic association, for simulation data management according to imaging task planning files and satellite load test data, and for generating scenario data files according to imaging task planning files; an algorithm and model layer for implementing imaging geometric relationship calculation, for implementing radiation transmission calculation, for implementing vehicle platform influence analysis, and for implementing feature component mapping labeling; a service scheduling layer for implementing imaging simulation labeling tasks by calling the algorithm and model layer based on load balancing according to simulation calculation request conditions; and an application layer for target characteristic association by loading models and data of the basic data layer according to scenario data files, constructing a spatial target scene, combining ray tracing, and generating and labeling specific simulation images by calling the service scheduling layer, so as to obtain spatial target recognition samples. The algorithm and model layer comprises a feature component labeling module for labeling simulation images to generate sample images based on a feature component automatic labeling method of full-chain simulation, and the processing process specifically comprises: acquiring feature component information to be labeled; assigning different colors to feature components to be labeled of an original spatial target three-dimensional model according to the feature component information, to obtain a target labeled three-dimensional model and a color mapping table; performing simulation calculation on the target labeled three-dimensional model to obtain a target labeled simulation image; extracting different color regions of the target labeled simulation image to form region description information; looking up the color mapping table to establish a feature component region template; mapping the feature component region template to an imaging simulation image to label the corresponding feature component; and retrieving parameter configuration information in the image generation process, including orbit attitude and sensor parameters of the vehicle and the target related to the image, and forming a sample image description file according to the sample data description format.

2. The space object identification sample generation system based on imaging simulation according to claim 1, characterized in that, The basic data layer comprises spatial target three-dimensional models, typical spatial target spectral data, spatial target orbit attitude data, vehicle orbit attitude data and simulation configuration parameters; and the spatial target orbit attitude data and the vehicle orbit attitude data comprise historical orbit attitude data, real-time received orbit attitude data and simulation orbit attitude data.

3. The space object identification sample generation system based on imaging simulation according to claim 1, characterized in that, The data processing layer comprises: a texture image preprocessing module for collecting corresponding spectral reflectance and directional reflectance according to typical spatial target spectral data, storing them in a spectral database, and using automatic and semi-automatic image classification algorithms to classify texture images of spatial target three-dimensional models and save the classification results; a simulation data management module for generating deduction data according to satellite load test data and imaging task planning files, and generating a simulation report through statistical analysis; and a task scenario data processing module for setting a scenario time and a task type according to an imaging task planning file, and generating a scenario file.

4. The space object identification sample generation system based on imaging simulation according to claim 1, characterized in that, The algorithm and model layer comprises: A target pose calculation module is configured to calculate the position and pose of a space target and the position and pose of a starry sky background including the sun, the moon and planets according to space target orbit attitude data; An aircraft pose calculation module is configured to calculate the space position and pose of a sensor on an aircraft platform according to aircraft orbit attitude data; An imaging geometry relationship module is configured to realize conversion from scene coordinates to image coordinates through coordinate conversion among a geodetic coordinate system, an aircraft platform body coordinate system and a camera image space coordinate system; A radiation transmission module is configured to calculate radiance reaching an entrance pupil of a camera optical system and form a radiance image; A sensor module is configured to realize sensor imaging process simulation and generate a sensor simulation image; and A platform influence module is configured to model the influence of linear vibration, high-frequency jitter and low-frequency sinusoidal disturbance of a platform on imaging and obtain a simulation image output by the camera.

5. The imaging simulation-based space object identification sample generation system of claim 4, wherein, The sensor module processing process specifically includes: A simulation camera is set to be a real camera in orbit, including an aperture, a focal length and a shutter time, a corresponding camera spectral response curve is used to simulate optical camera characteristics, and multiple format image data are generated, including full-spectrum images and 8-16 bit gray scale gray images; Image defects caused by inherent problems of camera optics or electronic systems, including overexposure, defocus, motion blur and Smear effect, are simulated to generate images with uneven defects.

6. The space object identification sample generation system based on imaging simulation according to claim 4, characterized in that, The service scheduling layer includes: A computing service interaction module is configured to perform computing service management, computing service control, computing process management and computing process arrangement, submit a simulation computing request to a computing service scheduling module, submit a computing service call record, a computing service exception log, a computing process call record and a computing process exception log to the computing service scheduling module for query, and save and query a computing result; A computing service scheduling module is configured to send a simulation computing request to a computing service computing node module with sufficient computing resources based on load balancing according to the simulation computing request and scheduling feedback information of the computing service computing node module to realize calling, judge whether an existing historical computing result can be used, and send scheduling state information, a computing service call record, a computing service exception log, a computing process call record and a computing process exception log to the computing service interaction module; and A computing service computing node module is configured to complete imaging calculation of a simulation image pixel according to scheduling information of the computing service scheduling module, and send node state information to the computing service interaction module. The processing process of the computing service computing node module specifically includes: A computing task from the computing service scheduling module is received, parsed, and a computing process, a computing initial parameter, a computing progress and an intermediate computing result in the computing task are obtained; Sub-computing tasks are decomposed according to the computing process, corresponding computing services are executed, and a computing result is saved, if the entire computing process is completed, the computing service scheduling module is informed that the computing task is completed, and if there are still subsequent sub-computing tasks, the sub-computing tasks are fed back to the computing service scheduling module according to the resource usage and state of the computing service computing node module.

7. The imaging simulation-based space object identification sample generation system of claim 6, wherein, The application layer comprises: a target characteristic association module, configured to perform texture mapping according to a spatial target three-dimensional model of the basic data layer and typical spatial target spectral data, assign a material type in the typical spatial target spectral data to each texture image classification result, generate a corresponding material texture mapping file, obtain target characteristic data, and realize association of geometric information and spectral information of the spatial target three-dimensional model in a scene; a spatial scene organization module, configured to set and deploy an initial situation according to an experimental task intention, obtain position and attitude data of targets and vehicles from the algorithm and model layer, and generate a simulation scene in combination with the target characteristic data; and a ray tracing module, configured to calculate a transmission process of a ray between a light source, a spatial target and a sensor, and obtain a corresponding relationship between a camera pixel and an imaging scene.

8. The imaging simulation-based space object identification sample generation system of claim 7, wherein, The processing process of the spatial scene organization module specifically comprises: setting and deploying the initial situation according to the simulation experimental task intention, forming a scenario file of an imaging simulation task, containing spatial target three-dimensional models, interface relationships and event association information required by the simulation experiment, and forming a simulation scenario script file; parsing the simulation scenario script file, calling a calculation model of the algorithm and model layer, and generating a simulation scene of the scenario task; and simultaneously using a KD tree to organize and manage the entire imaging spatial scene including spatial targets, satellites, the sun and a spatial background. 9.A method for generating a spatial target recognition sample based on imaging simulation, implemented based on the system of claim 8, the method comprising: realizing target multi-characteristic modeling by a target characteristic association module based on a spatial target three-dimensional model and typical target spectral data; generating a simulation scene by a spatial scene organization module according to a target multi-characteristic modeling result; obtaining position and attitude data of spatial targets and vehicles respectively by a target pose calculation module and a vehicle pose calculation module; obtaining a corresponding relationship between a camera pixel and an imaging scene by a ray tracing module; submitting a simulation calculation request to a calculation service scheduling module by a calculation service interaction module; generating calculation service calculation nodes based on load balancing by the calculation service scheduling module according to the simulation calculation request, each calculation service calculation node being capable of processing imaging calculation of one simulation image pixel; realizing conversion from a scene coordinate to an image coordinate by an imaging geometric relationship module; calculating a radiance image at a camera entrance pupil by a radiation transmission module; obtaining a simulation image output by a camera by a sensor module and a platform influence module; annotating the simulation image to generate a sample image based on a feature component automatic annotation method of full-chain simulation by a feature component annotation module, specifically comprising: obtaining feature component information to be annotated; assigning different colors to feature components to be annotated of an original spatial target three-dimensional model according to the feature component information, obtaining a target annotated three-dimensional model and a color mapping table; performing simulation calculation on the target annotated three-dimensional model to obtain a target annotated simulation image; extracting different color regions of the target annotated simulation image to form region description information; looking up the color mapping table to establish a feature component region template; The feature part region template is mapped into the imaging simulation image to realize labeling of the corresponding feature part; Parameter configuration information in the image generation process is retrieved, including the orbit attitude of the aircraft and the target and the sensor parameter related to the image, and a sample image description file is formed according to a sample data description format.

Images