A method for closed-loop simulation design based on satellite computer maintenance mode

Abstract

The application discloses a kind of based on satellite computer maintenance mode closed loop simulation design method, it is related to satellite remote sensing simulation algorithm verification technical field, including ground end construction scene dynamic simulation platform, establish the multidimensional feature sample library of real observation data extraction multispectral texture feature, geometric structure information is stored.Through mathematical modeling tool, sample library information parameterization decomposition is generated, and the feature element and feature structure element that can be independently edited are generated.Set dynamic evolution parameter set to drive primitive according to preset space-time evolution logic combination evolution, real-time generation contains detailed feature details and dynamic characteristics of continuous time sequence simulation scene.According to the target satellite load parameter simulation imaging generation time sequence simulation remote sensing image data flow, complete satellite computer on-orbit processing algorithm verification.The method realizes feature primitive independent editing, dynamic scene continuous generation, improves simulation scene authenticity and algorithm verification matching degree.

Description

Technical Field

[0001] This invention belongs to the field of spaceborne remote sensing simulation algorithm verification technology, specifically a closed-loop simulation design method based on spaceborne computer maintenance mode. Background Technology

[0002] Currently, in the field of closed-loop simulation for spaceborne computer maintenance, static, fixed ground feature models are mostly used for simulation operations. Remote sensing simulation data is generated based on pre-set, fixed ground feature resources. The multispectral texture features and geometric structure information of the ground features exist as a whole model without being processed by mathematical modeling tools. This type of simulation can only generate single frames or fixed sequences of simulation scenes, relying on fixed parameters to complete imaging simulations, and cannot adapt to the dynamic changes of on-orbit scenarios. The holistic form of the ground feature model means that features and structures cannot be adjusted individually. The generation logic of the simulation scene lacks a spatiotemporal evolution-driven mechanism, resulting in low detail matching between simulated images and actual on-orbit observation data. Furthermore, the verification process of on-orbit processing algorithms for spaceborne computers cannot accurately reflect the actual on-orbit operating conditions.

[0003] During the construction of spaceborne simulation scenarios, it is necessary to parametrically decompose the multispectral texture features and geometric structure information in the multidimensional ground feature sample library to form independently editable ground feature feature primitives and ground feature structure primitives. These primitives need to be driven by a dynamic evolution parameter set to complete their combination and evolution according to a preset spatiotemporal evolution logic, generating a continuous temporal simulation scenario in real time containing detailed ground feature details and dynamic features. This process completes the generation of temporal simulation remote sensing image data streams and the verification of on-orbit processing algorithms. Summary of the Invention

[0004] This invention aims to solve at least one of the technical problems existing in the prior art;

[0005] Therefore, this invention proposes a closed-loop simulation design method based on onboard computer maintenance mode, comprising:

[0006] A scene dynamic simulation platform is built on the ground, which is used to generate highly realistic dynamic simulation data;

[0007] In the scene dynamic simulation platform, a multi-dimensional land cover sample library is established, which stores multispectral texture features and geometric structure information of various land cover types extracted based on real observation data;

[0008] Using mathematical modeling tools, the multispectral texture features and geometric structure information of various land cover types stored in the multidimensional land cover sample library are parametrically decomposed to generate independently editable land cover feature primitives and land cover structure primitives.

[0009] In the scene dynamic simulation platform, a dynamic evolution parameter set is set, and the dynamic evolution parameter set is called to drive the independently editable ground feature primitives and ground structure primitives to combine and evolve according to the preset spatiotemporal evolution logic, thereby generating a continuous temporal simulation scene containing detailed ground feature details and dynamic features in real time.

[0010] From the continuous time-series simulation scenario, the imaging process is simulated based on the payload parameters of the target satellite, and a time-series simulation remote sensing image data stream corresponding to the target satellite is generated.

[0011] The on-orbit processing algorithm of the spaceborne computer was verified based on the time-series simulated remote sensing image data stream.

[0012] Furthermore, the step of parametrically decomposing the multispectral texture features and geometric structure information of various land cover types stored in the multidimensional land cover sample database using mathematical modeling tools to generate independently editable land cover feature primitives and land cover structure primitives includes:

[0013] Principal component analysis was performed on the multispectral texture features of each land cover type in the multidimensional land cover sample library to extract the set of core feature vectors that characterize the spectral response of the corresponding land cover type.

[0014] A linear transformation is performed on the core feature vector set to decouple the reflectivity basis vector, which is independent of the illumination conditions, and the modulation coefficient, which is related to the illumination conditions.

[0015] The geometric structure information of each land cover type in the multidimensional land cover sample library is segmented into three-dimensional point clouds to extract the basic geometric voxel set that constitutes the corresponding land cover type.

[0016] For each geometric voxel in the set of basic geometric voxels, associate it with its corresponding reflectivity basis vector that is independent of illumination conditions to form the feature primitive of the ground object with spectral and geometric attributes;

[0017] Define the spatial connection rules and combination constraints between the aforementioned feature primitives to form the structural primitives of the land features that describe their macroscopic morphology.

[0018] Furthermore, by invoking the dynamic evolution parameter set, the independently editable feature primitives and structure primitives are driven to combine and evolve according to a preset spatiotemporal evolution logic, generating in real time a continuous temporal simulation scene containing detailed feature details and dynamic characteristics, including:

[0019] The dynamic evolution parameter set includes illumination conditions, meteorological conditions, seasonal change parameters, spatial distribution probability of land features, and rules for land feature state changes.

[0020] Based on the seasonal change parameters in the dynamic evolution parameter set, calculate the solar altitude angle, solar azimuth angle, and typical surface temperature at the current simulation moment;

[0021] Based on the calculated solar altitude angle, solar azimuth angle, and illumination conditions in the dynamic evolution parameter set, the radiance value of each pixel in the independently editable ground feature primitive is calculated;

[0022] Based on the meteorological conditions in the dynamic evolution parameter set, atmospheric effects of clouds, fog, and haze are superimposed in the simulation scenario, and the radiance value is corrected for atmospheric path radiation and attenuation.

[0023] Based on the spatial distribution probability of ground features and the rules for changes in ground feature state in the dynamic evolution parameter set, the spatial location, orientation, and phenological state of the independently editable ground feature structure primitive at the simulation time are determined.

[0024] The radiance value, corrected for atmospheric effects, is assigned to the independently editable ground feature primitives that are in the corresponding phenological state and spatial location, rendering a single frame of the continuous temporal simulation scene at the current simulation moment.

[0025] Furthermore, based on the spatial distribution probability of land features and the rules governing land feature state changes in the dynamic evolution parameter set, the spatial location, orientation, and phenological state of the independently editable land feature structure primitives at the simulation time are determined, including:

[0026] The spatial distribution probability of the land features is read, which defines the probability distribution of the occurrence of different land feature types in the simulation area in the form of a probability density field;

[0027] Monte Carlo sampling is performed based on the probability density field to determine the type of the independently editable ground feature structure primitive and its initial spatial location coordinates in the scene at the current simulation time;

[0028] According to the rules for changes in the state of land features, the standard phenological state code of the land features of the aforementioned sampled and determined type is queried at the current simulation time. The standard phenological state code defines the nominal values ​​of leaf area index, vegetation coverage, and environmental humidity parameters.

[0029] A random perturbation term is introduced to fine-tune the nominal values ​​of the parameters in the standard phenological state code, and the orientation angle of the structure is affected by the surrounding land features according to the spatial connection rules between the independently editable land feature structural elements, thereby generating the final phenological state and orientation with individual differences.

[0030] Furthermore, from the continuous time-series simulation scenario, based on the payload parameters of the target satellite, the imaging process is simulated to generate a time-series simulated remote sensing image data stream corresponding to the target satellite, including:

[0031] The payload parameters include imaging band, spatial resolution, signal-to-noise ratio, and imaging geometry;

[0032] Based on the imaging geometry in the payload parameters of the target satellite, including satellite orbit parameters, sensor viewpoint and imaging time, calculate the ground observation geometry corresponding to each pixel in the continuous time-series simulation scenario, including the observation zenith angle and the observation azimuth angle.

[0033] Using the atmospheric radiative transfer model, and combining the imaging bands, the observation zenith angle, and the observation azimuth angle in the payload parameters, the theoretical radiance of each band at the sensor entrance pupil is calculated.

[0034] Based on the spatial resolution in the load parameters, the continuous temporal simulation scene is resampled, and the high-resolution scene information is aggregated into a pixel grid that matches the sensor spatial resolution.

[0035] Gaussian noise that matches the signal-to-noise ratio characteristics in the load parameters is added to the resampled radiance data to simulate sensor electronic noise.

[0036] The radiance data of each band after noise addition is quantized into digital quantized values, and then packaged according to the time series to form the time-series simulated remote sensing image data stream corresponding to the target satellite.

[0037] Furthermore, it also includes the step of correcting the model of the scene dynamic simulation platform using actual observation data:

[0038] To acquire actual remote sensing images of a specific area observed by a target satellite at a specific time;

[0039] From the scene dynamic simulation platform, extract the corresponding frames from the time-series simulation remote sensing image data stream generated at the same specific time and for the same specific area, and use them as simulation images;

[0040] The actual observed remote sensing images and the simulated images are subjected to strict geometric registration and radiometric normalization processing to ensure that the actual observed remote sensing images and the simulated images are under the same spatial reference and radiometric scale.

[0041] The differences in pixel values ​​between the registered actual remote sensing image and the simulated image in each band and in each land cover region are calculated to generate a residual distribution map.

[0042] Analyze the statistical characteristics and spatial patterns of the residual distribution map to identify land cover types or regions where systematic deviations exceed a preset threshold;

[0043] Based on the identified deviation results, the characteristic parameters of relevant land cover types in the multidimensional land cover sample library are adjusted in reverse, or the relevant evolution rules in the dynamic evolution parameter set are corrected, thus completing one model correction iteration of the scene dynamic simulation platform.

[0044] Furthermore, by analyzing the statistical characteristics and spatial patterns of the residual distribution map, land cover types or regions with systematic deviations exceeding a preset threshold are identified, including:

[0045] The residual distribution map is segmented into land cover types, and the residual pixels are classified into the corresponding land cover types based on the land cover classification layer in the scene dynamic simulation platform.

[0046] For each land cover type, the mean and standard deviation of the residuals of all its associated pixels are calculated.

[0047] The mean residual is compared with a preset deviation tolerance threshold, and land cover types whose absolute value of the mean residual is consistently greater than the deviation tolerance threshold are marked as land cover types with systematic spectral deviation.

[0048] In the spatial dimension, cluster analysis is performed on the residual distribution map to identify anomalous regions that are spatially continuous and have the same sign for residual values.

[0049] The abnormal region blocks are compared with the dynamic evolution parameter logs in the scene dynamic simulation platform to identify suspected erroneous evolution parameters or rules that cause the occurrence of spatially continuous abnormal region blocks with the same residual value.

[0050] Furthermore, the step of adjusting the characteristic parameters of relevant land cover types in the multidimensional land cover sample library in reverse, or correcting the relevant evolution rules in the dynamic evolution parameter set, based on the identified deviation results, includes:

[0051] For land cover types marked as having systematic spectral bias, the component values ​​of the reflectance basis vector, which are independent of illumination conditions, are scaled proportionally from the corresponding independently editable land cover feature primitives. The scaling ratio is determined by the sign and magnitude of the residual mean.

[0052] For abnormal region blocks associated with suspected erroneous evolution parameters or rules, locate the specific parameter items or rule entries in the dynamic evolution parameter set;

[0053] The gradient descent optimization algorithm is used to minimize the sum of squared residuals of anomalous region blocks associated with suspected erroneous evolution parameters or rules, and the values ​​of the located parameter or rule entries are fine-tuned.

[0054] The scaled reflectance basis vector and the fine-tuned parameter terms or rule entries are then used to update the multidimensional land cover sample library and the dynamic evolution parameter set.

[0055] The scene dynamic simulation platform was rerun using the updated parameters to generate new simulation images and calculate new residuals to verify the correction effect.

[0056] Furthermore, the on-orbit processing algorithm of the spaceborne computer is verified based on the time-series simulated remote sensing image data stream, including:

[0057] The time-series simulated remote sensing image data stream is injected into the on-orbit processing algorithm test environment of the onboard computer;

[0058] The on-orbit processing algorithm is controlled to continuously process the injected time-series simulated remote sensing image data stream and output a sequence of target recognition results, land cover classification results, or change detection results.

[0059] The real scene annotation information used by the scene dynamic simulation platform when generating the time-series simulation remote sensing image data stream is recorded synchronously. The real scene annotation information includes the real category, location and status of all ground features in each frame of the simulation image.

[0060] After the algorithm test is completed, the result sequence output by the on-orbit processing algorithm is compared frame by frame with the corresponding real scene annotation information.

[0061] Based on the comparison results, the detection rate, false alarm rate, and positioning accuracy of the on-orbit processing algorithm are statistically analyzed, and an algorithm performance evaluation report is generated.

[0062] Furthermore, the time-series simulated remote sensing image data stream is injected into the on-orbit processing algorithm test environment of the spaceborne computer, including:

[0063] In a host environment that includes a spaceborne computer prototype or simulator, deploy a virtual data injection channel that conforms to the real satellite data interface protocol.

[0064] The time-series simulated remote sensing image data stream is continuously sent to the memory buffer of the onboard computer through the virtual data injection channel according to the preset frame rate and data packet format.

[0065] Simulate the attitude, orbit, and timing data of the satellite platform to generate auxiliary data packets that match the time-series simulated remote sensing image data stream for each frame;

[0066] After aligning the auxiliary data packet with the corresponding image data packet in time, they are injected into the memory buffer of the onboard computer, triggering the on-orbit processing algorithm to begin the processing flow.

[0067] Compared with the prior art, the beneficial effects of the present invention are:

[0068] A multidimensional land cover sample library is constructed based on real observation data. Mathematical modeling tools are used to parametrically decompose the multispectral texture features and geometric structure information of various land cover types in the sample library, forming independently editable land cover feature primitives and land cover structure primitives. Land cover features and geometric structures are decomposed from the overall model into independently adjustable primitive units. The parameter attributes of each primitive can be modified individually. The restored form of multispectral texture features and geometric structure information conforms to the extraction standards of real observation data. The detailed representation of land cover simulation is no longer limited by the overall fixed model, and the adjustment methods of land cover features and structures are independently controllable.

[0069] In the scene dynamic simulation platform, a dynamic evolution parameter set is set, which is then used to drive independently editable ground feature primitives and ground structure primitives to perform combination and evolution operations according to preset spatiotemporal evolution logic. The simulation scene is generated in real time in a continuous temporal sequence, synchronously preserving detailed ground feature details and dynamic change characteristics within the scene. The generation process of the temporal simulation scene closely matches the spatiotemporal change patterns of the on-orbit scene. Based on the remote sensing image data stream generated by simulating imaging according to the target satellite payload parameters, it can match the actual data form processed by the on-orbit computer and is suitable for the closed-loop simulation verification process in the on-orbit computer maintenance mode. Attached Figure Description

[0070] Figure 1 This is a flowchart illustrating the steps of a closed-loop simulation design method based on a spaceborne computer maintenance mode as described in this invention.

[0071] Figure 2 A flowchart for generating a continuous temporal simulation scenario to drive the combination and evolution of primitives;

[0072] Figure 3 This is a graph showing the geometric temporal changes observed by satellite.

[0073] Figure 4 A trend chart showing the improvement in accuracy of the spaceborne simulation platform model;

[0074] Figure 5 This is a time-series performance evaluation diagram of the spaceborne remote sensing change detection algorithm. Detailed Implementation

[0075] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0076] See Figure 1This invention provides a closed-loop simulation design method based on a spaceborne computer maintenance mode. This method constructs a scene dynamic simulation platform on the ground, whose core function is to generate highly realistic dynamic simulation data. To achieve this goal, a multi-dimensional ground feature sample library is first established within the scene dynamic simulation platform. This library stores multispectral texture features and geometric structure information of various ground feature types extracted from real observation data. Mathematical modeling tools are used to parametrically decompose the information in the multi-dimensional ground feature sample library, generating independently editable ground feature feature primitives and ground feature structure primitives. A dynamic evolution parameter set is set within the scene dynamic simulation platform. By calling this parameter set, the generated independently editable ground feature feature primitives and ground feature structure primitives are driven to combine and evolve according to a preset spatiotemporal evolution logic, thereby generating a continuous temporal simulation scene containing detailed ground feature details and dynamic features in real time. The imaging process is simulated based on the payload parameters of the target satellite, generating a temporal simulation remote sensing image data stream corresponding to the target satellite from this continuous temporal simulation scene. The on-orbit processing algorithm of the spaceborne computer is verified based on this temporal simulation remote sensing image data stream.

[0077] In one embodiment of the present invention, during the generation of independently editable feature primitives and structure primitives for land cover, principal component analysis is performed on the multispectral texture features of each land cover type in the multidimensional land cover sample library to extract a set of core feature vectors characterizing the spectral response of the corresponding land cover type. A linear transformation is performed on this core feature vector set to decouple the reflectance basis vector, which is independent of illumination conditions, and the modulation coefficient, which is related to illumination conditions. Simultaneously, three-dimensional point cloud segmentation is performed on the geometric structure information of each land cover type in the multidimensional land cover sample library to extract a set of basic geometric voxels constituting the corresponding land cover type. For each geometric voxel in the set of basic geometric voxels, its corresponding reflectance basis vector, which is independent of illumination conditions, is associated to form a feature primitive with spectral and geometric attributes. Spatial connection rules and combination constraints between feature primitives are defined to constitute structure primitives describing the macroscopic morphology of land cover.

[0078] In practice, the process of generating independently editable feature primitives and structure primitives is based on a multidimensional land cover sample library. This library stores multispectral texture features and geometric structure information of various land cover types, such as urban buildings, farmland, woodland, and water bodies, extracted from real observation data. Taking woodland as an example, principal component analysis is performed on the multispectral texture features of woodland in the multidimensional land cover sample library. The multispectral texture features include reflectance values ​​of woodland in multiple bands (blue, green, red, near-infrared, and shortwave infrared) extracted from real satellite imagery. Principal component analysis extracts a set of core feature vectors characterizing the spectral response of woodland from these high-dimensional data. This set of core feature vectors contains several feature vectors sorted by contribution rate, among which the first three feature vectors can explain more than 95% of the variance information of the woodland spectral response.

[0079] In practice, a linear transformation is performed on the core feature vector set. The purpose of this linear transformation is to decouple the reflectivity basis vectors, which are independent of illumination conditions, from the modulation coefficients, which are dependent on illumination conditions. This decoupling process can be understood as being achieved by constructing a transformation matrix. Assuming the core feature vector set of forest land types obtained from principal component analysis is a matrix... , dimension ,in It is the number of feature vectors. It refers to the number of bands. This can be understood as a linear transformation achieved by solving the coefficient matrix. and basis vector matrix To achieve decomposition, so that In this decomposition, the basis vector matrix Each row represents a reflectivity basis vector independent of illumination conditions, and the coefficient matrix... Each column of elements represents a modulation coefficient related to illumination conditions. For forest types, the reflectance basis vector can characterize its inherent spectral reflectance properties, while the modulation coefficients are used to dynamically adjust them in subsequent simulations based on solar angle, atmospheric conditions, and illumination conditions.

[0080] In the specific implementation, the geometric structure information of forest land types in the multidimensional land cover sample database is segmented into three-dimensional point clouds. The geometric structure information of forest land types in the multidimensional land cover sample database comes from point cloud data obtained by airborne lidar scanning. The three-dimensional point cloud segmentation extracts the basic geometric voxel set that constitutes the forest land type from the point cloud data. The basic geometric voxel set includes various voxels such as cylinders, cones, and irregular polyhedra. Cylinders are used to simulate tree trunks, cones are used to simulate tree crowns, and irregular polyhedra are used to simulate litter on the ground. In the specific implementation, each geometric voxel in the basic geometric voxel set is associated with its corresponding reflectance basis vector, which is independent of illumination conditions. Each cylindrical geometric voxel is associated with the reflectance basis vector of the tree trunk and bark, and each conical geometric voxel is associated with the reflectance basis vector of the tree crown and leaves, thereby forming forest land cover feature primitives with spectral and geometric attributes. Define the spatial connection rules and combination constraints between forest land feature feature primitives to form forest land feature structure primitives that describe the macroscopic morphology of forest land. The spatial connection rules include stipulating that the tree crown cone must be located at the top of the tree trunk cylinder. The combination constraints include limiting the minimum distance between adjacent trees and the distribution ratio of trees of different heights. These rules together constitute forest land feature structure primitives.

[0081] In one embodiment of the present invention, when driving the combination and evolution of independently editable feature primitives and structure primitives, the dynamic evolution parameter set includes illumination conditions, meteorological conditions, seasonal change parameters, spatial distribution probability of land features, and rules governing land feature state changes. (See also...) Figure 2 Based on the seasonal change parameters in the dynamic evolution parameter set, the solar altitude angle, solar azimuth angle, and typical surface temperature at the current simulation moment are calculated. Based on the calculated solar altitude angle, solar azimuth angle, and illumination conditions in the dynamic evolution parameter set, the radiance value of each pixel in the independently editable ground feature primitives is calculated. Based on the meteorological conditions in the dynamic evolution parameter set, atmospheric effects of clouds, fog, and haze are superimposed on the simulation scene, and the radiance values ​​are corrected for atmospheric path radiation and attenuation. Based on the spatial distribution probability of ground features and the rules of ground feature state change in the dynamic evolution parameter set, the spatial location, orientation, and phenological state of the independently editable ground feature structure primitives at the simulation moment are determined. The radiance values ​​corrected for atmospheric effects are assigned to the independently editable ground feature structure primitives at the corresponding phenological state and spatial location, rendering a single frame of the continuous temporal simulation scene at the current simulation moment.

[0082] In practical implementation, independently editable feature and structure primitives are combined and evolved according to a preset spatiotemporal evolution logic. A dynamic evolution parameter set defines complete driving rules, including illumination conditions defined by latitude, longitude, and time functions; meteorological conditions including cloud cover, visibility, and aerosol optical thickness parameters; seasonal transformation parameters describing the sun-earth relationship and surface temperature model; spatial distribution probabilities of different features appearing regularly within the simulation area; and rules depicting the growth, withering, and state transition processes of features. Taking a simulated mixed scene of farmland and town at 10 AM on a summer morning in the mid-latitudes of the Northern Hemisphere as an example, based on the seasonal transformation parameters in the dynamic evolution parameter set—specifically, date and latitude / longitude information—the solar altitude angle, solar azimuth angle, and typical surface temperature derived from a heat balance model at the current simulation moment are calculated.

[0083] In some embodiments, based on the calculated solar altitude angle, solar azimuth angle, and illumination conditions defined in a dynamic evolution parameter set, including the solar constant and the atmospheric top spectrum, the radiance value of each pixel in the independently editable land feature primitive is calculated. Optionally, the calculation process considers the surface bidirectional reflectance distribution function; for independently editable land feature primitives in farmland areas, the radiance value of its pixels is... formula:

[0084]

[0085] in: This indicates the radiance of a specific band in the direction observed by the sensor. This represents the solar spectral irradiance at the top of the atmosphere. Indicates the zenith angle of the sun. Indicates atmospheric transmittance. This indicates the directional reflectance of independently editable ground feature primitives in this band. This represents atmospheric path radiation. Based on meteorological conditions set by a dynamic evolution parameter set, such as clear skies with thin cirrus clouds, the atmospheric effects of clouds are superimposed on the simulation scenario. The calculated radiance values ​​of farmland and building rooftops are then corrected for atmospheric path radiation and attenuation. The presence of clouds dynamically affects... and The value.

[0086] In practical implementation, based on the spatial distribution probability of land features and the rules governing land feature state changes in the dynamic evolution parameter set, the spatial location, orientation, and phenological state of independently editable land feature structural elements at the simulation time are determined. The spatial distribution probability of land features specifies the probability of adjacency between farmland and residential areas / roads, while the rules governing land feature state changes define the range of leaf area index for summer maize and the weathering state of roof materials for urban buildings. Based on these rules, the row and column positions, plant orientation, and specific values ​​of the leaf area index of independently editable land feature structural elements for maize plants in the field are determined. Simultaneously, the spatial coordinates, roof orientation, and surface aging parameters of independently editable land feature structural elements for reinforced concrete buildings are determined. It is understandable that the radiance value corrected by atmospheric effects is assigned to the independently editable ground structure primitives that are in the corresponding phenological state and spatial location. The corresponding spectral radiance is assigned to the geometric facets of the leaves and stems of each corn plant, and the corresponding spectral radiance is assigned to the roof and facade of each building. Through the synthesis of the graphics rendering engine, a single frame of the continuous temporal simulation scene containing the growing cornfield and the staggered building group at the current simulation moment is rendered.

[0087] In one embodiment of the present invention, when determining the spatial location, orientation, and phenological state of independently editable ground feature structural elements, the spatial distribution probability of ground features, defined in the form of a probability density field, is read. Monte Carlo sampling is performed based on this probability density field to determine the type of independently editable ground feature structural elements at the current simulation time and their initial spatial location coordinates in the scene. According to the ground feature state transition rules, the standard phenological state code of the aforementioned sampled and determined type of ground feature at the current simulation time is queried. This code defines the nominal values ​​of leaf area index, vegetation cover, and environmental humidity parameters. A random perturbation term is introduced to fine-tune the nominal values ​​of the parameters in the standard phenological state code, and the orientation angle affected by surrounding ground features is calculated according to the spatial connection rules between independently editable ground feature structural elements, generating the final phenological state and orientation with individual differences. When generating a time-series simulation remote sensing image data stream from a continuous time-series simulation scene, the payload parameters of the target satellite include imaging band, spatial resolution, signal-to-noise ratio, and imaging geometry. Based on the imaging geometry in the payload parameters, including satellite orbit parameters, sensor viewing angle, and imaging time, the ground observation geometry corresponding to each pixel in the continuous time-series simulation scene is calculated, including the observation zenith angle and observation azimuth angle. Using an atmospheric radiative transfer model, combined with the imaging band, observation zenith angle, and observation azimuth angle, the theoretical radiance of each band at the sensor entrance pupil is calculated. According to the spatial resolution, the continuous time-series simulation scene is resampled, aggregating high-resolution scene information into a pixel grid matching the sensor's spatial resolution. Gaussian noise with signal-to-noise ratio characteristics is added to the resampled radiance data to simulate sensor electronic noise. The radiance data of each band after adding noise is quantized into digital quantized values, encapsulated according to the time series, and forms a time-series simulated remote sensing image data stream corresponding to the target satellite.

[0088] In practical implementation, when determining the spatial location, orientation, and phenological state of independently editable feature structure primitives at the simulation time, the spatial distribution probability of features, defined in the form of a probability density field, is read. This spatial distribution probability is stored in the form of grid data, defining the probability of occurrence of various feature types such as farmland, woodland, water bodies, and buildings on each pixel within a 100-square-kilometer simulation area of ​​a plain. Monte Carlo sampling is performed based on the grid defined by the spatial distribution probability of features to determine the type of independently editable feature structure primitives at the current simulation time and their initial spatial coordinates in the scene. For example, in one sampling, it is determined that an independently editable farmland feature structure primitive is placed at coordinates (X=1250 meters, Y=3870 meters), and another independently editable farmland feature structure primitive is placed at coordinates (X=1320 meters, Y=3900 meters). Based on the rules governing land cover state changes, the standard phenological state codes of the aforementioned sampled land cover types are queried at the current simulation time. These standard phenological state codes are stored in a lookup table, defining the nominal values ​​of leaf area index, vegetation cover, and environmental humidity parameters for different land cover types on different dates. A random perturbation term is introduced to fine-tune the nominal values ​​of the parameters in the standard phenological state codes. This random perturbation term is extracted from a preset statistical distribution and, based on the spatial connection rules between independently editable land cover structural units, calculates the orientation angle of each independently editable land cover structural unit influenced by surrounding land cover features. This generates the final phenological state and orientation with individual differences. For example, two adjacent trees may have different canopy tilt orientations due to competition for sunlight. When generating a time-series simulated remote sensing image data stream from a continuous time-series simulation scenario, the target satellite's payload parameters include imaging bands, spatial resolution, signal-to-noise ratio (SNR), and imaging geometry. The imaging bands include four specific wavelength ranges: blue, green, red, and near-infrared. The spatial resolution is 10 meters, and the SNR is 100 under typical illumination. The imaging geometry includes the satellite orbital six-element number and the sensor side-swing angle. Based on the imaging geometry in the payload parameters, including satellite orbital parameters, sensor viewing angle, and imaging time, the ground observation geometry corresponding to each pixel in the continuous time-series simulation scenario is calculated, including the observation zenith angle and observation azimuth angle. The calculation process involves coordinate system transformation and line-of-sight vector intersection. Using the atmospheric radiative transfer model, combined with the imaging bands, observation zenith angle, and observation azimuth angle, the theoretical radiance of each band at the sensor's entrance pupil is calculated. The atmospheric radiative transfer model inputs atmospheric profile parameters and aerosol data. Based on the spatial resolution in the payload parameters, the continuous time-series simulation scenario is resampled, aggregating high-resolution scene information into a pixel grid matching the sensor's spatial resolution. The resampling process uses a pixel aggregation averaging method. Gaussian noise matching the signal-to-noise ratio characteristics in the load parameters is added to the resampled radiance data to simulate sensor electronic noise.The radiance data of each band after noise addition is quantized into digital quantized values, packaged according to time series, and formed a time-series simulated remote sensing image data stream corresponding to the target satellite.

[0089] In some embodiments, the nominal values ​​of the standard phenological state codes are stored in a structured database for easy querying. Referring to Table 1, an example fragment of a standard phenological state code for some land cover types is shown. It can be understood that the nominal values ​​of the parameters in the standard phenological state codes are derived from statistical summaries of historical observation data and serve as the baseline state for simulation.

[0090] Table 1: Example Table of Standard Phenological State Codes

[0091] winter wheat 2026-04-15 3.5 0.85 0.6 corn 2026-06-20 2.8 0.75 0.5 Deciduous woodland 2026-05-10 5.2 0.95 0.7

[0092] In practice, the process of fine-tuning the nominal values ​​of parameters in the standard phenological state coding by introducing random perturbation terms is achieved through an additive model. For example, for the leaf area index parameter, its final value... formula:

[0093]

[0094] in: This represents the final leaf area index after introducing individual differences. This represents the nominal leaf area index value obtained from the standard phenological state code. This represents a random perturbation value sampled from a specific distribution with a mean of zero and a variance determined by the land cover type. Optionally, when calculating the orientation angle of an independently editable land cover element affected by surrounding land cover elements, the spatial location and height difference between adjacent independently editable land cover elements are considered. For example, a taller building independently editable land cover element may block sunlight, causing a shift in the sun-facing orientation angle of a shorter tree independently editable land cover element.

[0095] In some embodiments, the process of resampling a continuous time-series simulation scene involves aggregating sub-meter-level high-resolution scene raster data to a ten-meter resolution, whereby for each ten-meter output pixel, its radiance value... The radiance values ​​of all high-resolution pixels falling within its boundary The average is obtained as follows:

[0096]

[0097] in: This represents the total number of high-resolution pixels falling within the current output pixel. It can be understood that when Gaussian noise is added to the resampled radiance data, the standard deviation of the noise... According to the signal-to-noise ratio specified in the load parameters and the local average radiance of the current pixel Confirmed, relationship is Optionally, when quantizing to digital values, continuous radiance values ​​are mapped to integer numerical values ​​based on the sensor's radiation dynamic range and quantization bit depth. For example, the range from zero to maximum radiance is linearly mapped to an integer range from zero to 1023.

[0098] See Figure 3 This is a time-series graph showing the changes in the observed zenith angle and azimuth angle at different times of day. The observed zenith angle initially decreases and then increases, reaching its minimum value (approximately 15°) at 12:00, when the satellite's observation direction is closest to the zenith, resulting in optimal ground resolution and imaging geometry. The observed azimuth angle monotonically increases with time, gradually increasing from 120° at 08:00 to 240° at 16:00, reflecting the continuous change in the relative position of the sun and the satellite. A smaller zenith angle indicates a shorter atmospheric path, less radiation attenuation and scattering, and higher image quality. Therefore, midday is the golden window for remote sensing observation. The continuous change in azimuth angle can be used to invert the surface BRDF (two-way reflectance distribution function), improving the accuracy of ground feature classification and change detection. This curve provides direct temporal constraints for imaging planning, attitude control, and data preprocessing by the onboard computer.

[0099] In one embodiment of the present invention, when correcting the model of a scene dynamic simulation platform using actual observation data, actual remote sensing images of a target satellite over a specific area at a specific time are acquired. Corresponding frames from the time-series simulated remote sensing image data stream generated at the same specific time and for the same specific area are extracted from the scene dynamic simulation platform as simulated images. Strict geometric registration and radiometric normalization are performed on the actual observation remote sensing images and the simulated images to ensure they are under the same spatial reference and radiometric scale. The pixel value differences between the registered actual observation remote sensing images and the simulated images in each band and for each land cover region are calculated to generate a residual distribution map. The statistical characteristics and spatial patterns of the residual distribution map are analyzed to identify land cover types or regions with systematic deviations exceeding a preset threshold. Land cover type segmentation is performed on the residual distribution map, and residual pixels are classified into corresponding land cover types based on the land cover classification layer in the scene dynamic simulation platform. For each land cover type, the mean residual and standard deviation of all assigned pixels are statistically analyzed. The residual mean is compared with a preset deviation tolerance threshold. Land cover types whose absolute residual mean value is consistently greater than the tolerance threshold are marked as having systematic spectral bias. Cluster analysis is performed on the residual distribution map spatially to identify spatially continuous anomalous regions with the same residual sign. These anomalous regions are compared with the dynamic evolution parameter logs in the scene dynamic simulation platform to identify suspected erroneous evolution parameters or rules that led to their appearance. Based on the identified deviations, the feature parameters of relevant land cover types in the multidimensional land cover sample library are adjusted in reverse, or the relevant evolution rules in the dynamic evolution parameter set are corrected. For land cover types marked with systematic spectral bias, the reflectivity basis vector components, independent of illumination conditions, are scaled proportionally from their corresponding independently editable land cover feature primitives. The scaling ratio is determined by the sign and magnitude of the residual mean. For anomalous regions associated with suspected erroneous evolution parameters or rules, specific parameter items or rule entries are located in the dynamic evolution parameter set. A gradient descent optimization algorithm is employed to minimize the sum of squared residuals in anomalous regions, fine-tuning the values ​​of the located parameters or rule entries. The scaled reflectance basis vectors and the fine-tuned parameters or rule entries are then used to update the multidimensional ground feature sample library and the dynamic evolution parameter set. The scene dynamic simulation platform is then rerun using the updated parameters to generate new simulation images and calculate new residuals to verify the correction effect.

[0100] In the specific implementation, when using actual observation data to correct the model of the scene dynamic simulation platform, actual remote sensing images of a specific area observed by the target satellite at a specific time are acquired. The specific time is set to 10:30 AM on April 15, 2026, and the specific area is a mixed area in a plain containing farmland, water bodies, and towns. Corresponding frames from the time-series simulated remote sensing image data stream generated at the same specific time for the same specific area are extracted from the scene dynamic simulation platform as simulated images. Rigorous geometric registration and radiometric normalization are performed on the actual remote sensing images and the simulated images. Geometric registration uses a polynomial correction model based on control points, and radiometric normalization unifies the digital quantization values ​​of both into the physical quantity of atmospheric top reflectivity, ensuring that the actual remote sensing images and the simulated images are under the same spatial reference and radiometric scale. The pixel value differences between the registered actual remote sensing images and the simulated images in each band and for each land cover area are calculated. In the near-infrared band, the reflectivity value of the simulated images is generally lower than that of the actual remote sensing images, generating a residual distribution map with pixel values ​​as reflectivity differences.

[0101] In practice, the statistical characteristics and spatial patterns of the residual distribution map are analyzed to identify land cover types or regions where systematic deviations exceed a preset threshold. The residual distribution map is segmented by land cover type. Based on the land cover classification layer in the scene dynamic simulation platform, residual pixels are categorized into corresponding land cover types. The land cover classification layer includes multiple categories such as winter wheat, corn, woodland, water bodies, and building roofs. For each land cover type, the mean and standard deviation of the residuals for all assigned pixels are calculated. The mean residual is compared with a preset deviation tolerance threshold, which is two percent of the absolute difference in reflectance. Land cover types with a consistently greater absolute mean residual than the deviation tolerance threshold are marked as having systematic spectral deviations. Spatially, cluster analysis is performed on the residual distribution map using a clustering algorithm based on connected pixels to identify spatially continuous anomalous regions with the same residual sign. The abnormal region blocks are compared with the dynamic evolution parameter logs in the scene dynamic simulation platform. The dynamic evolution parameter logs record all parameters used at each simulation moment during the simulation process. The suspected erroneous evolution parameters or rules that cause the occurrence of spatially continuous abnormal region blocks with the same residual value are correlated. For example, a continuous negative residual abnormal block in a building area may be related to the rules on the aging rate of the building roof in the dynamic evolution parameter set.

[0102] In some embodiments, referring to Table 2, partial results of residual statistics for each major land cover type are shown for the near-infrared band. It is understood that the calculation of the residual mean and residual standard deviation is based on all valid pixels of that land cover type within the scene.

[0103] Table 2: Residual Statistics of Major Land Cover Types in the Near-Infrared Band

[0104] In practice, the feature parameters of relevant land cover types in the multidimensional land cover sample library or the relevant evolution rules in the dynamic evolution parameter set are adjusted in reverse based on the identified deviation results. For land cover types marked as having systematic spectral deviations, such as building roofs, the component values ​​of the reflectance basis vector, which are independent of illumination conditions, are scaled proportionally from the independently editable land cover feature primitives corresponding to the building roofs. The scaling ratio is determined by the sign and magnitude of the residual mean. A negative residual mean for building roofs in the near-infrared band indicates that the simulated reflectance is too low; therefore, the coefficients of the reflectance basis vector in this band are scaled proportionally. For anomalous regions associated with suspected erroneous evolution parameters or rules, specific parameter items or rule entries are located in the dynamic evolution parameter set, such as the parameter item "interannual decay rate of building roof reflectance". A gradient descent optimization algorithm is used to fine-tune the values ​​of the located parameter items or rule entries with the optimization objective of minimizing the sum of squared residuals of anomalous regions associated with suspected erroneous evolution parameters or rules. It is understandable that the scaled reflectance basis vector and the fine-tuned parameter terms or rule entries are updated back to the multidimensional ground feature sample library and dynamic evolution parameter set. The scene dynamic simulation platform is rerun using the updated parameters to generate a new simulation image of the same area on April 15, 2026, and the new residuals are calculated. In the new residual distribution map, the absolute value of the mean residual of the building roof area is reduced to less than 0.5%, verifying the correction effect.

[0105] Optionally, for land cover types with systematic spectral bias, the reflectance basis vector component values ​​are scaled from their corresponding independently editable land cover feature primitives. The calculation can be expressed as:

[0106]

[0107] in: Indicates the first The reflectivity basis vector component values ​​after band scaling adjustment This represents the reflectivity basis vector component values ​​before scaling adjustment. This represents a preset scaling factor used to control the adjustment range. This indicates that the land cover type is in the first The mean residual value for each band. In some embodiments, when fine-tuning the values ​​of parameter items or rule entries in the dynamic evolution parameter set, the gradient descent optimization algorithm iteratively updates the parameter values. In each iteration, the gradient of the sum of squared residuals with respect to the parameter is calculated, and the parameter value is updated along the negative gradient direction. Optionally, after rerunning the scene dynamic simulation platform with the updated parameters to generate new simulation images, not only are new residuals calculated, but the residual statistics and deviation identification steps are also re-executed to evaluate whether the correction has made the mean residual value for all land cover types fall within the deviation tolerance threshold.

[0108] See Figure 4 This is a trend chart showing the improvement in model correction accuracy of the spaceborne simulation platform, intuitively illustrating the changes in spectral accuracy, spatial accuracy, and overall accuracy from before correction to the final correction stage. All three types of accuracy increase monotonically with each correction step, verifying the effectiveness of the model correction process. The differences in accuracy among the three types are greatest before correction. After the final correction, the three types of accuracy converge significantly, indicating that the model has achieved balanced optimization in both the spectral and spatial dimensions. After each correction, the accuracy improvement is approximately 5% to 7%, demonstrating the stable iterative optimization capability of the correction strategy. This curve directly proves the feasibility of the closed-loop process of "actual observation data → residual analysis → model correction → simulation verification," providing quantitative evidence for on-orbit algorithm verification of the spaceborne computer. Throughout the entire process, spectral accuracy is consistently slightly higher than spatial accuracy, reflecting that this correction focuses more on optimizing spectral features. After the final correction, the overall accuracy reaches 94%, meeting the engineering requirements of high-fidelity remote sensing simulation.

[0109] In one embodiment of the present invention, when verifying the on-orbit processing algorithm of a spaceborne computer based on a time-series simulated remote sensing image data stream, the time-series simulated remote sensing image data stream is injected into the on-orbit processing algorithm test environment of the spaceborne computer. The on-orbit processing algorithm is controlled to continuously process the injected time-series simulated remote sensing image data stream, outputting a sequence of target recognition results, ground feature classification results, or change detection results. Simultaneously, the real scene annotation information used by the scene dynamic simulation platform when generating the time-series simulated remote sensing image data stream is recorded. This information includes the real category, location, and status of all ground features in each frame of the simulated image. After the algorithm test is completed, the result sequence output by the algorithm is compared frame by frame with the corresponding real scene annotation information. Based on the comparison results, the detection rate, false alarm rate, and positioning accuracy indicators of the on-orbit processing algorithm are statistically analyzed to generate an algorithm performance evaluation report. When injecting the time-series simulated remote sensing image data stream into the on-orbit processing algorithm test environment, a virtual data injection channel consistent with the real satellite data interface protocol is deployed in the host environment containing the spaceborne computer prototype or simulator. The time-series simulated remote sensing image data stream is continuously sent to the onboard computer's memory buffer through a virtual data injection channel according to a preset frame rate and data packet format. The attitude, orbit, and timing data of the simulated satellite platform generate auxiliary data packets that match each frame of the time-series simulated remote sensing image data stream. After time alignment of the auxiliary data packets with the corresponding image data packets, they are injected into the onboard computer's memory buffer, triggering the on-orbit processing algorithm to begin the processing flow.

[0110] In practical implementation, the on-orbit processing algorithm of the spaceborne computer is verified based on the time-series simulated remote sensing image data stream. The time-series simulated remote sensing image data stream is injected into the on-orbit processing algorithm test environment of the spaceborne computer. The on-orbit processing algorithm test environment includes a host environment built based on real spaceborne computer hardware or instruction set simulators. The on-orbit processing algorithm in the on-orbit processing algorithm test environment of the spaceborne computer continuously processes the injected time-series simulated remote sensing image data stream. The time-series simulated remote sensing image data stream simulates the continuous 14-day transit observation of a certain area of ​​a plain by a target satellite. The on-orbit processing algorithm outputs a sequence of farmland change patches monitored during this period as a change detection result sequence. The real scene annotation information used by the scene dynamic simulation platform when generating the time-series simulated remote sensing image data stream is recorded synchronously. The real scene annotation information is stored in the form of a time-stamped vector layer, containing the real category, location and status of all land features in each frame of the simulated image, especially recording the occurrence time and boundary of real change events such as farmland being plowed and construction occupation. After the algorithm testing is completed, the change detection result sequence output by the on-orbit processing algorithm is compared frame by frame with the corresponding real scene annotation information. During the comparison process, spatial overlay analysis is performed on the change patches output by the algorithm and the real change patches recorded in the real scene annotation information. Based on the comparison results, the detection rate, false alarm rate, and positioning accuracy of the on-orbit processing algorithm are statistically analyzed, and an algorithm performance evaluation report is generated. The positioning accuracy is measured by the intersection-union ratio (IUGR).

[0111] In some embodiments, time-series simulated remote sensing image data streams are injected into the on-orbit processing algorithm test environment of the onboard computer. In a host environment containing an onboard computer prototype or simulator, a virtual data injection channel consistent with the real satellite data interface protocol is deployed. The virtual data injection channel simulates a high-speed serial data bus protocol. The time-series simulated remote sensing image data streams are continuously sent to the onboard computer's memory buffer through the virtual data injection channel according to a preset frame rate and data packet format. The preset frame rate is one frame per second, and the data packet format includes a frame header, image data payload, and checksum. Simulating the satellite platform's attitude, orbit, and timing data, auxiliary data packets matching each frame of the time-series simulated remote sensing image data stream are generated. These auxiliary data packets contain the satellite's three-axis attitude angles, orbital six-axis numbers, and precise timing information. After time alignment of the auxiliary data packets with the corresponding image data packets, they are injected into the onboard computer's memory buffer. Time alignment ensures that the timestamps of the auxiliary data packets and the image data packets are strictly synchronized, triggering the on-orbit processing algorithm to begin processing.

[0112] In practical implementation, the on-orbit processing algorithm, pre-installed in the on-orbit computer's non-volatile memory, continuously processes the injected data within the on-orbit computer's testing environment. The algorithm sequentially decodes, radiometrically calibrates, and geometrically corrects each frame of input image data and its auxiliary data, then performs pixel-level comparisons with historical reference images to identify regions where reflectance changes exceed a threshold and clusters them into change patches. Finally, it outputs a sequence of change detection results containing the boundary coordinates and change confidence scores of these patches. The synchronously recorded real-scene annotation information provides a ground-based reference for subsequent ground truth determination of each change patch in the algorithm's output sequence, classifying it as a real change or a false alarm. During frame-by-frame comparison, the algorithm performance evaluation report is presented in statistical tables, such as the detection rate. The calculation formula is:

[0113]

[0114] in: This indicates the change detection rate of the on-orbit processing algorithm. This indicates the number of real change patches correctly detected by the on-orbit processing algorithm. This represents the total number of real-world changed patches recorded in the real-scene annotation information. Optionally, the false alarm rate is the proportion of changed patches incorrectly reported by the on-orbit processing algorithm to the total number of reported patches. The positioning accuracy is measured by calculating the average intersection-union ratio (IUU) between the changed patches output by the algorithm and the real-world changed patches.

[0115] See Figure 5 This is a time-series performance evaluation chart of a spaceborne remote sensing change detection algorithm, showing the variation patterns of the number of real change patches, detected change patches, and false alarm patches over a 14-day observation period. The number of real change patches and the number of detected change patches both increase monotonically and synchronously, with the difference between them consistently remaining between 2 and 3 patches, indicating that the algorithm has a stable high detection rate and can effectively capture real surface changes. The number of detected change patches is consistently slightly lower than the number of real change patches, reflecting a small number of missed detections, but overall performance is excellent. The number of false alarm patches initially fluctuates slightly upwards during the observation period, then continuously decreases, reaching a peak of 9 on day 10, and eventually dropping to 4. The significant decrease in the number of false alarms in the later stages indicates that after iterative optimization or adaptive learning, the algorithm effectively suppresses false alarms and improves robustness. The overall performance trend shows that as the number of observation days increases, real changes accumulate, the detection algorithm continuously captures changes, and false alarms are controlled, demonstrating the value of time-series remote sensing data and the long-term stability of the algorithm.

[0116] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. A closed-loop simulation design method based on onboard computer maintenance mode, characterized in that, The method includes: A scene dynamic simulation platform is built on the ground, which is used to generate highly realistic dynamic simulation data; In the scene dynamic simulation platform, a multi-dimensional land cover sample library is established, which stores multispectral texture features and geometric structure information of various land cover types extracted based on real observation data; Using mathematical modeling tools, the multispectral texture features and geometric structure information of various land cover types stored in the multidimensional land cover sample library are parametrically decomposed to generate independently editable land cover feature primitives and land cover structure primitives. In the scene dynamic simulation platform, a dynamic evolution parameter set is set, and the dynamic evolution parameter set is called to drive the independently editable ground feature primitives and ground structure primitives to combine and evolve according to the preset spatiotemporal evolution logic, thereby generating a continuous temporal simulation scene containing detailed ground feature details and dynamic features in real time. From the continuous time-series simulation scenario, the imaging process is simulated based on the payload parameters of the target satellite, and a time-series simulation remote sensing image data stream corresponding to the target satellite is generated. The on-orbit processing algorithm of the spaceborne computer was verified based on the time-series simulated remote sensing image data stream.

2. The closed-loop simulation design method based on onboard computer maintenance mode according to claim 1, characterized in that, The process involves using mathematical modeling tools to parametrically decompose the multispectral texture features and geometric structure information of various land cover types stored in the multidimensional land cover sample database, generating independently editable land cover feature primitives and land cover structure primitives, including: Principal component analysis was performed on the multispectral texture features of each land cover type in the multidimensional land cover sample library to extract the set of core feature vectors that characterize the spectral response of the corresponding land cover type. A linear transformation is performed on the core feature vector set to decouple the reflectivity basis vector, which is independent of the illumination conditions, and the modulation coefficient, which is related to the illumination conditions. The geometric structure information of each land cover type in the multidimensional land cover sample library is segmented into three-dimensional point clouds to extract the basic geometric voxel set that constitutes the corresponding land cover type. For each geometric voxel in the set of basic geometric voxels, associate it with its corresponding reflectivity basis vector that is independent of illumination conditions to form the feature primitive of the ground object with spectral and geometric attributes; Define the spatial connection rules and combination constraints between the aforementioned feature primitives to form the structural primitives of the land features that describe their macroscopic morphology.

3. The closed-loop simulation design method based on onboard computer maintenance mode according to claim 2, characterized in that, The dynamic evolution parameter set is invoked to drive the independently editable feature primitives and structure primitives to combine and evolve according to a preset spatiotemporal evolution logic, generating a continuous temporal simulation scene in real time containing detailed feature details and dynamic features, including: The dynamic evolution parameter set includes illumination conditions, meteorological conditions, seasonal change parameters, spatial distribution probability of land features, and rules for land feature state changes. Based on the seasonal change parameters in the dynamic evolution parameter set, calculate the solar altitude angle, solar azimuth angle, and typical surface temperature at the current simulation moment; Based on the calculated solar altitude angle, solar azimuth angle, and illumination conditions in the dynamic evolution parameter set, the radiance value of each pixel in the independently editable ground feature primitive is calculated; Based on the meteorological conditions in the dynamic evolution parameter set, atmospheric effects of clouds, fog, and haze are superimposed in the simulation scenario, and the radiance value is corrected for atmospheric path radiation and attenuation. Based on the spatial distribution probability of ground features and the rules for changes in ground feature state in the dynamic evolution parameter set, the spatial location, orientation, and phenological state of the independently editable ground feature structure primitive at the simulation time are determined. The radiance value, corrected for atmospheric effects, is assigned to the independently editable ground feature primitives that are in the corresponding phenological state and spatial location, rendering a single frame of the continuous temporal simulation scene at the current simulation moment.

4. The closed-loop simulation design method based on onboard computer maintenance mode according to claim 3, characterized in that, Based on the spatial distribution probability of land features and the rules governing land feature state changes in the dynamic evolution parameter set, the spatial location, orientation, and phenological state of the independently editable land feature structure primitives at the simulation time are determined, including: The spatial distribution probability of the land features is read, which defines the probability distribution of the occurrence of different land feature types in the simulation area in the form of a probability density field; Monte Carlo sampling is performed based on the probability density field to determine the type of the independently editable ground feature structure primitive and its initial spatial location coordinates in the scene at the current simulation time; According to the rules for changes in the state of land features, the standard phenological state code of the land features of the aforementioned sampled and determined type is queried at the current simulation time. The standard phenological state code defines the nominal values ​​of leaf area index, vegetation coverage, and environmental humidity parameters. A random perturbation term is introduced to fine-tune the nominal values ​​of the parameters in the standard phenological state code, and the orientation angle of the structure is affected by the surrounding land features according to the spatial connection rules between the independently editable land feature structural elements, thereby generating the final phenological state and orientation with individual differences.

5. The closed-loop simulation design method based on onboard computer maintenance mode according to claim 4, characterized in that, From the continuous time-series simulation scenario, based on the payload parameters of the target satellite, the imaging process is simulated to generate a time-series simulated remote sensing image data stream corresponding to the target satellite, including: The payload parameters include imaging band, spatial resolution, signal-to-noise ratio, and imaging geometry; Based on the imaging geometry in the payload parameters of the target satellite, including satellite orbit parameters, sensor viewpoint and imaging time, calculate the ground observation geometry corresponding to each pixel in the continuous time-series simulation scenario, including the observation zenith angle and the observation azimuth angle; Using the atmospheric radiative transfer model, and combining the imaging bands, the observation zenith angle, and the observation azimuth angle in the payload parameters, the theoretical radiance of each band at the sensor entrance pupil is calculated. Based on the spatial resolution in the load parameters, the continuous temporal simulation scene is resampled, and the high-resolution scene information is aggregated into a pixel grid that matches the sensor spatial resolution. Gaussian noise that matches the signal-to-noise ratio characteristics in the load parameters is added to the resampled radiance data to simulate sensor electronic noise. The radiance data of each band after noise addition is quantized into digital quantized values, and then packaged according to the time series to form the time-series simulated remote sensing image data stream corresponding to the target satellite.

6. The closed-loop simulation design method based on onboard computer maintenance mode according to claim 5, characterized in that, It also includes the step of correcting the model of the scene dynamic simulation platform using actual observation data: To acquire actual remote sensing images of a specific area observed by a target satellite at a specific time; From the scene dynamic simulation platform, extract the corresponding frames from the time-series simulation remote sensing image data stream generated at the same specific time and for the same specific area, and use them as simulation images; The actual observed remote sensing images and the simulated images are subjected to strict geometric registration and radiometric normalization processing to ensure that the actual observed remote sensing images and the simulated images are under the same spatial reference and radiometric scale. The differences in pixel values ​​between the registered actual remote sensing image and the simulated image in each band and in each land cover region are calculated to generate a residual distribution map. Analyze the statistical characteristics and spatial patterns of the residual distribution map to identify land cover types or regions where systematic deviations exceed a preset threshold; Based on the identified deviation results, the characteristic parameters of relevant land cover types in the multidimensional land cover sample library are adjusted in reverse, or the relevant evolution rules in the dynamic evolution parameter set are corrected, thus completing one model correction iteration of the scene dynamic simulation platform.

7. The closed-loop simulation design method based on onboard computer maintenance mode according to claim 6, characterized in that, Analyzing the statistical characteristics and spatial patterns of the residual distribution map, we identify land cover types or regions where systematic deviations exceed a preset threshold, including: The residual distribution map is segmented into land cover types, and the residual pixels are classified into the corresponding land cover types based on the land cover classification layer in the scene dynamic simulation platform. For each land cover type, the mean and standard deviation of the residuals of all its associated pixels are calculated. The mean residual is compared with a preset deviation tolerance threshold, and land cover types whose absolute value of the mean residual is consistently greater than the deviation tolerance threshold are marked as land cover types with systematic spectral deviation. In the spatial dimension, cluster analysis is performed on the residual distribution map to identify anomalous regions that are spatially continuous and have the same sign for residual values. The abnormal region blocks are compared with the dynamic evolution parameter logs in the scene dynamic simulation platform to identify suspected erroneous evolution parameters or rules that cause the occurrence of spatially continuous abnormal region blocks with the same residual value.

8. The closed-loop simulation design method based on onboard computer maintenance mode according to claim 7, characterized in that, The step of adjusting the characteristic parameters of relevant land cover types in the multidimensional land cover sample library, or correcting the relevant evolution rules in the dynamic evolution parameter set, based on the identified deviation results, includes: For land cover types marked as having systematic spectral bias, the component values ​​of the reflectance basis vector, which are independent of illumination conditions, are scaled proportionally from the corresponding independently editable land cover feature primitives. The scaling ratio is determined by the sign and magnitude of the residual mean. For abnormal region blocks associated with suspected erroneous evolution parameters or rules, locate the specific parameter items or rule entries in the dynamic evolution parameter set; The gradient descent optimization algorithm is used to minimize the sum of squared residuals of anomalous region blocks associated with suspected erroneous evolution parameters or rules, and the values ​​of the located parameter or rule entries are fine-tuned. The scaled reflectance basis vector and the fine-tuned parameter terms or rule entries are then used to update the multidimensional land cover sample library and the dynamic evolution parameter set. The scene dynamic simulation platform was rerun using the updated parameters to generate new simulation images and calculate new residuals to verify the correction effect.

9. The closed-loop simulation design method based on onboard computer maintenance mode according to claim 8, characterized in that, Verification of on-orbit processing algorithms for spaceborne computers based on the aforementioned time-series simulated remote sensing image data stream includes: The time-series simulated remote sensing image data stream is injected into the on-orbit processing algorithm test environment of the onboard computer; The on-orbit processing algorithm is controlled to continuously process the injected time-series simulated remote sensing image data stream and output a sequence of target recognition results, land cover classification results, or change detection results. The real scene annotation information used by the scene dynamic simulation platform when generating the time-series simulation remote sensing image data stream is recorded synchronously. The real scene annotation information includes the real category, location and status of all ground features in each frame of the simulation image. After the algorithm test is completed, the result sequence output by the on-orbit processing algorithm is compared frame by frame with the corresponding real scene annotation information. Based on the comparison results, the detection rate, false alarm rate, and positioning accuracy of the on-orbit processing algorithm are statistically analyzed, and an algorithm performance evaluation report is generated.

10. The closed-loop simulation design method based on onboard computer maintenance mode according to claim 9, characterized in that, Injecting the time-series simulated remote sensing image data stream into the on-orbit processing algorithm test environment of the spaceborne computer includes: In a host environment that includes a spaceborne computer prototype or simulator, deploy a virtual data injection channel that conforms to the real satellite data interface protocol. The time-series simulated remote sensing image data stream is continuously sent to the memory buffer of the onboard computer through the virtual data injection channel according to the preset frame rate and data packet format. Simulate the attitude, orbit, and timing data of the satellite platform to generate auxiliary data packets that match the time-series simulated remote sensing image data stream for each frame; After aligning the auxiliary data packet with the corresponding image data packet in time, they are injected into the memory buffer of the onboard computer, triggering the on-orbit processing algorithm to begin the processing flow.

Images