Terrain radiance correction method for remote sensing image based on full scene simulation and machine learning

Abstract

The application relates to a remote sensing image terrain radiation correction method based on full-scene simulation and machine learning. A full-scene imaging geometry combination including imaging geometry parameter groups of several scenes is acquired. For each imaging geometry parameter group, a three-dimensional radiation transmission model is used to simulate corresponding slope surface feature reflectivity and planar feature reflectivity under ideal conditions, and target correction parameters of each terrain parameter under the illumination-observation parameter group are obtained accordingly. A training data set is constructed by using the terrain, illumination and observation parameter groups and the corresponding target correction parameters, and a preset model is trained to obtain a correction parameter prediction model. Finally, in actual application, after real terrain, illumination and observation parameters of a remote sensing image to be corrected are acquired, the parameters are input into the correction parameter prediction model, so that terrain correction parameters of the remote sensing image can be quickly and accurately obtained, and terrain radiation correction is carried out based on the parameters.

Description

Technical Field

[0001] This application relates to the field of remote sensing technology, and in particular to a method for topographic radiometric correction of remote sensing images based on full-scene simulation and machine learning. Background Technology

[0002] In the field of optical remote sensing imaging, complex terrain has a significant impact on the imaging process. As the spatial resolution of remote sensing images continues to improve, the radiation distortion problem caused by terrain effects is becoming increasingly prominent, which severely restricts the widespread application of optical remote sensing data in mountainous and complex terrain conditions.

[0003] Specifically, topographic distortion has numerous adverse effects. In ecological monitoring and agricultural management, topographic distortion can lead to errors in the calculation of various vegetation indices, thus affecting decision-making in related fields. For example, inaccurate vegetation indices may result in a lack of scientific basis for ecological protection measures, deviations in agricultural planting plans, and impacts crop yield and quality. In ecosystem service assessment, the surface reflectance inversion results of remote sensing data from mountainous shadow areas are generally lower due to topographic influence, leading to decreased reliability of biomass estimation and an inability to accurately assess ecosystem service functions, which is detrimental to the protection and rational utilization of the ecological environment. In the field of land resource management and planning, topographic distortion in mountainous remote sensing images results in low accuracy of land cover classification and inaccurate land use change analysis, restricting the rational development and utilization of land resources and potentially leading to land waste or irrational development.

[0004] Currently, topographic radiometric correction is a crucial step in data preprocessing and is essential for obtaining the true reflectance of ground features. However, existing methods for topographic radiometric correction of remote sensing images have certain limitations, making it difficult to effectively address the various problems caused by radiometric distortion in complex terrains and failing to meet the requirements of practical applications for the accuracy and reliability of remote sensing data. Summary of the Invention

[0005] Based on this, the purpose of this application is to provide a method for topographic radiometric correction of remote sensing images based on full-scene simulation and machine learning. This method fully considers the influence of terrain and illumination observation conditions on radiometric distortion and utilizes model simulation and machine learning to map the relationship between terrain parameters and illumination-observation parameter sets and correction parameters, thereby achieving more accurate topographic radiometric correction of remote sensing images.

[0006] The remote sensing image topographic radiometric correction method based on full-scene simulation and machine learning described in this application includes the following steps:

[0007] A full-scene imaging geometry combination is obtained, which includes several imaging geometry parameter groups for several scenes; wherein, the imaging geometry parameter group includes several terrain parameters and several illumination-observation parameter groups; each illumination-observation parameter group contains illumination parameters and observation parameters, and the illumination parameters and / or observation parameters in different illumination-observation parameter groups are different;

[0008] For each of the imaging geometric parameter sets, based on the several terrain parameters and the illumination-observation parameter set, the slope reflectance of the corresponding scene is simulated by a terrain grid, the slope reflectance of the corresponding scene includes the slope reflectance of each terrain parameter; based on the preset ideal terrain parameters and the illumination-observation parameter set, the planar reflectance of the reference plane of the corresponding scene is simulated.

[0009] For each of the imaging geometry parameter groups, based on the reflectivity of the slope features in the terrain mesh and the reflectivity of the planar features in the reference plane, the target correction parameters for each terrain parameter under each of the illumination-observation parameter groups are obtained.

[0010] Based on the terrain parameters, illumination-observation parameter sets, and target correction parameters of each terrain parameter under each illumination-observation parameter set of the several scenes, a training dataset is constructed by traversing the entire scene parameter space; the training dataset is input into a preset model for training to obtain a trained correction parameter prediction model; the preset model is used to learn the entire scene correction mapping relationship from terrain parameters and illumination-observation parameter sets to correction parameters;

[0011] The terrain parameters, illumination parameters, and observation parameters of the remote sensing image to be corrected are obtained. The terrain parameters, illumination parameters, and observation parameters of the remote sensing image to be corrected are input into the correction parameter prediction model to obtain the correction parameters of the remote sensing image to be corrected. Based on the correction parameters of the remote sensing image to be corrected, the remote sensing image to be corrected is subjected to full-scene adaptive terrain radiometric correction.

[0012] This application's embodiments acquire a full-scene imaging geometry combination containing rich terrain parameters and diverse illumination-observation parameter sets. For each imaging geometry parameter set, it simulates the reflectivity of slope features in a terrain grid and the reflectivity of planar features on a reference plane based on preset ideal terrain parameters. This process comprehensively and meticulously considers the impact of complex terrain on remote sensing image radiometry under different illumination observation conditions, covering various possible combinations of terrain and illumination observation scenarios. Based on the two simulated reflectivities, the target correction parameters for each terrain parameter under different illumination-observation parameter sets are calculated. These parameters are then used to construct a training dataset through a full-scene parameter space traversal, and a preset model is trained to obtain a correction parameter prediction model. This model deeply learns the full-scene correction mapping relationship from terrain parameters and illumination-observation parameter sets to correction parameters, capturing the complex and subtle intrinsic connections between terrain, illumination observation, and correction parameters. The accurate establishment of this mapping relationship is the core key to achieving accurate correction of remote sensing images. In practical applications, after acquiring the terrain parameters, illumination parameters, and observation parameters of the remote sensing image to be corrected, inputting them into the trained correction parameter prediction model allows for the rapid and accurate acquisition of the correction parameters for the remote sensing image. Based on these correction parameters, adaptive topographic radiometric correction across the entire scene can effectively eliminate radiometric distortion caused by complex terrain. Overall, the embodiments of this application, through the organic combination of full-scene simulation and machine learning, achieve comprehensive and accurate correction of topographic radiometric distortion in remote sensing images, significantly improving the quality and usability of remote sensing images. This effectively solves the problem of existing technologies' inability to comprehensively and accurately handle radiometric distortion under complex terrain, and has significant practical value and broad application prospects in multiple remote sensing application fields.

[0013] To better understand and implement this application, the following detailed description is provided in conjunction with the accompanying drawings. Attached Figure Description

[0014] Figure 1 This is a flowchart illustrating the remote sensing image topographic radiometric correction method based on full-scene simulation and machine learning, according to an embodiment of this application.

[0015] Figure 2 This is a schematic diagram illustrating the steps for obtaining terrain parameters and illumination-observation parameter sets in an embodiment of this application;

[0016] Figure 3 A schematic diagram illustrating the steps for calculating the reflectance of ground features in different terrains and ideal terrains according to embodiments of this application;

[0017] Figure 4 This is a schematic diagram illustrating the steps of calculating target correction parameters for shady slope areas and non-shady slope areas in an embodiment of this application. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings. Wherein, when the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements.

[0019] It should be understood that the embodiments described below do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0020] The terminology used in this application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. The singular forms “a,” “the,” and “the” used in this application are also intended to include the plural forms unless the context clearly indicates otherwise. Furthermore, in the description of this application, unless otherwise stated, “a plurality” means two or more. It should also be understood that the term “and / or” as used herein refers to and includes any or all possible combinations of one or more associated listed items, for example, A and / or B, which can represent: A alone, A and B together, and B alone; the character “ / ” generally indicates that the preceding and following objects are in an “or” relationship.

[0021] It should be understood that although the terms first, second, third, etc., may be used in this application to describe various information, this information should not be limited to these terms, and these terms are only used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence, nor should they be construed as indicating or implying relative importance. Those skilled in the art can understand the specific meaning of the above terms in this application according to the specific circumstances. Depending on the context, the word "if" as used in this application can be interpreted as "when," "when," or "in response to determination."

[0022] In the field of optical remote sensing imaging, complex terrain has a significant impact on the imaging process. As the spatial resolution of remote sensing images continues to improve, the radiation distortion problem caused by terrain effects is becoming increasingly prominent, which severely restricts the widespread application of optical remote sensing data in mountainous and complex terrain conditions.

[0023] Specifically, topographic distortion has numerous adverse effects. In ecological monitoring and agricultural management, topographic distortion can lead to errors in the calculation of various vegetation indices, thus affecting decision-making in related fields. For example, inaccurate vegetation indices may result in a lack of scientific basis for ecological protection measures, deviations in agricultural planting plans, and impacts crop yield and quality. In ecosystem service assessment, the surface reflectance inversion results of remote sensing data from mountainous shadow areas are generally lower due to topographic influence, leading to decreased reliability of biomass estimation and an inability to accurately assess ecosystem service functions, which is detrimental to the protection and rational utilization of the ecological environment. In the field of land resource management and planning, topographic distortion in mountainous remote sensing images results in low accuracy of land cover classification and inaccurate land use change analysis, restricting the rational development and utilization of land resources and potentially leading to land waste or irrational development.

[0024] Currently, topographic radiometric correction is a crucial step in data preprocessing and is essential for obtaining the true reflectance of ground features. However, existing methods for topographic radiometric correction of remote sensing images have certain limitations, making it difficult to effectively address the various problems caused by radiometric distortion in complex terrains and failing to meet the requirements of practical applications for the accuracy and reliability of remote sensing data.

[0025] This application proposes a method for topographic radiometric correction of remote sensing images based on full-scene simulation and machine learning. By comprehensively considering multiple topographic parameters and illumination-observation parameter sets, and using a machine learning model to learn the complex mapping relationship between them and correction parameters, the method can more accurately predict correction parameters under different conditions, thereby achieving more accurate topographic radiometric correction of remote sensing images.

[0026] Please refer to Figure 1 The remote sensing image topographic radiometric correction method based on full-scene simulation and machine learning described in this application includes the following steps:

[0027] S101: Obtain the full-scene imaging geometry combination, which includes several scene imaging geometry parameter groups; wherein, the imaging geometry parameter group includes several terrain parameters and several illumination-observation parameter groups; each illumination-observation parameter group contains illumination parameters and observation parameters, and the illumination parameters and / or observation parameters in different illumination-observation parameter groups are different.

[0028] S102: For each of the imaging geometric parameter groups, based on the several terrain parameters and the illumination-observation parameter group, simulate the slope reflectance of the corresponding scene in a terrain gridded manner, wherein the slope reflectance of the terrain gridded manner includes the slope reflectance of each terrain parameter; based on the preset ideal terrain parameters and the illumination-observation parameter group, simulate the planar reflectance of the reference plane of the corresponding scene.

[0029] S103: For each of the imaging geometric parameter groups, based on the reflectivity of the slope features in the terrain gridding and the reflectivity of the planar features in the reference plane, the target correction parameters for each terrain parameter under each of the illumination-observation parameter groups are obtained.

[0030] S104: Based on the terrain parameters, illumination-observation parameter groups, and target correction parameters of each terrain parameter under each illumination-observation parameter group of the imaging geometric parameter group of the several scenes, a training dataset is constructed by traversing the full scene parameter space; the training dataset is input into a preset model for training to obtain a trained correction parameter prediction model; the preset model is used to learn the full scene correction mapping relationship from terrain parameters and illumination-observation parameter groups to correction parameters;

[0031] S105: Obtain the terrain parameters, illumination parameters, and observation parameters of the remote sensing image to be corrected; input the terrain parameters, illumination parameters, and observation parameters of the remote sensing image to be corrected into the correction parameter prediction model to obtain the correction parameters of the remote sensing image to be corrected; perform full-scene adaptive terrain radiometric correction on the remote sensing image to be corrected based on the correction parameters of the remote sensing image to be corrected.

[0032] This application's embodiments acquire a full-scene imaging geometry combination containing rich terrain parameters and diverse illumination-observation parameter sets. For each imaging geometry parameter set, it simulates the reflectivity of slope features in a terrain grid and the reflectivity of planar features on a reference plane based on preset ideal terrain parameters. This process comprehensively and meticulously considers the impact of complex terrain on remote sensing image radiometry under different illumination observation conditions, covering various possible combinations of terrain and illumination observation scenarios. Based on the two simulated reflectivities, the target correction parameters for each terrain parameter under different illumination-observation parameter sets are calculated. These parameters are then used to construct a training dataset through a full-scene parameter space traversal, and a preset model is trained to obtain a correction parameter prediction model. This model deeply learns the full-scene correction mapping relationship from terrain parameters and illumination-observation parameter sets to correction parameters, capturing the complex and subtle intrinsic connections between terrain, illumination observation, and correction parameters. The accurate establishment of this mapping relationship is the core key to achieving accurate correction of remote sensing images. In practical applications, after acquiring the terrain parameters, illumination parameters, and observation parameters of the remote sensing image to be corrected, inputting them into the trained correction parameter prediction model allows for the rapid and accurate acquisition of the correction parameters for the remote sensing image. Based on these correction parameters, adaptive topographic radiometric correction across the entire scene can effectively eliminate radiometric distortion caused by complex terrain. Overall, the embodiments of this application, through the organic combination of full-scene simulation and machine learning, achieve comprehensive and accurate correction of topographic radiometric distortion in remote sensing images, significantly improving the quality and usability of remote sensing images. This effectively solves the problem of existing technologies' inability to comprehensively and accurately handle radiometric distortion under complex terrain, and has significant practical value and broad application prospects in multiple remote sensing application fields.

[0033] The remote sensing image topographic radiometric correction method based on full-scene simulation and machine learning described in this application uses a computer as the execution subject. The following is a detailed description of each step.

[0034] For step S101, a full-scene imaging geometry combination is obtained, which includes several imaging geometry parameter groups for several scenes; wherein, the imaging geometry parameter group includes several terrain parameters and several illumination-observation parameter groups; each illumination-observation parameter group contains illumination parameters and observation parameters, and the illumination parameters and / or observation parameters in different illumination-observation parameter groups are different.

[0035] Among them, terrain parameters are various types of data used to describe terrain features, such as slope and aspect. These parameters can comprehensively characterize the geometric and spatial properties of the terrain.

[0036] The illumination-observation parameter set includes illumination parameters and observation parameters. Illumination parameters generally cover solar zenith angle, solar azimuth angle, etc., which reflect illumination conditions. Observation parameters, such as the sensor observation zenith angle and observation azimuth angle of a remote sensing satellite, determine the observation conditions during remote sensing image acquisition. Different illumination-observation parameter sets contain different illumination parameters and / or observation parameters, thus covering a variety of different illumination and observation scenarios.

[0037] Please refer to Figure 2 In one embodiment, the imaging geometry parameter set in step S101 is obtained in the following manner:

[0038] Step S1011: Obtain the digital elevation model of the target terrain area;

[0039] Among them, digital elevation model is a physical ground model that uses a set of ordered numerical arrays to represent ground elevation. It can accurately describe the topography and store the height information of each location on the ground in digital form.

[0040] Step S1012: Obtain the terrain parameters of each grid region from the digital elevation model to obtain several terrain parameters;

[0041] To facilitate terrain analysis and processing, the target terrain area is divided into several regular or irregular small regions, which are called grid regions. By gridding the entire region, terrain information for each local area can be obtained more systematically.

[0042] Step S1013: Obtain several illumination parameters and observation parameters of the target terrain area; Based on the several illumination parameters and observation parameters, obtain several illumination-observation parameter sets;

[0043] The illumination parameters of the target terrain area in this step can be simulated according to different scenarios, or obtained by acquiring the illumination parameters input by the user, so as to fully simulate the illumination conditions under different scenarios.

[0044] The observation parameters for the target terrain area can be simulated based on different scenarios, or obtained by acquiring user-inputted observation parameters to fully simulate observation conditions under different scenarios. Observation parameters generally include the attitude parameters of the satellite sensor, i.e., the position and attitude information of the satellite sensor in space, including the sensor's pitch angle (the angle of rotation of the sensor around the horizontal axis), roll angle (the angle of rotation of the sensor around the forward and backward axes), and yaw angle (the angle of rotation of the sensor around the vertical axis). These attitude parameters reflect the sensor's observation angle and direction relative to the Earth's surface. Based on the satellite sensor's attitude parameters, combined with satellite orbit information, observation parameters such as the sensor's observation zenith angle and observation azimuth angle can be obtained through geometric calculation models. The sensor's observation zenith angle and observation azimuth angle affect the observation perspective of ground features, thus affecting the radiometric information of ground features in the acquired remote sensing image.

[0045] Combining illumination parameters and observation parameters is beneficial because different illumination and observation conditions have different effects on remote sensing images. Combining them into illumination-observation parameter sets can cover a variety of different illumination and observation scenarios, providing comprehensive data support for subsequent analysis of the impact of terrain effects on the radiometric properties of remote sensing images.

[0046] Step S1014: Based on the aforementioned terrain parameters and the aforementioned illumination-observation parameter sets, the imaging geometric parameter set is obtained.

[0047] This embodiment acquires a digital elevation model of the target terrain region and extracts detailed terrain parameters from this data using a gridding method. This comprehensively and accurately describes the terrain features of the target terrain region, providing rich and accurate basic data for analyzing terrain effects. Regarding the acquisition of illumination-observation parameter sets, illumination and observation parameters under different conditions are obtained. These illumination-observation parameter sets can be simulated as needed, fully considering the actual illumination and observation conditions during imaging. This ensures that the acquired illumination-observation parameter sets truly reflect the environmental conditions at the time of remote sensing image acquisition. This comprehensive approach to acquiring terrain parameters and illumination-observation parameter sets lays a solid foundation for subsequent accurate analysis of the impact of terrain effects on remote sensing image radiometry.

[0048] In one embodiment, step S1012, which involves obtaining terrain parameters from each grid region of the digital elevation model to obtain several terrain parameters, includes:

[0049] Step S10121: Extract the slope parameters and aspect parameters of each grid region from the digital elevation model;

[0050] In Geographic Information System (GIS) software, digital elevation models can be processed to calculate slope and aspect parameters. Slope refers to the angle between the tangent plane at any point on the terrain surface and the horizontal plane; it reflects the steepness of the terrain at that point. For example, in mountainous areas, a larger slope value indicates steep terrain, while in plains areas, a smaller slope value indicates flatter terrain.

[0051] Slope aspect parameters refer to the angle between the projection of the normal to the tangent plane at any point on the terrain surface onto the horizontal plane and the direction of true north, usually measured in degrees, ranging from 0° to 360°. Slope aspect determines the direction a point faces; different slope aspects receive different amounts of solar radiation. For example, south-facing slopes in the Northern Hemisphere typically receive more solar radiation and have relatively higher temperatures, while north-facing slopes receive less solar radiation and have relatively lower temperatures. Using terrain analysis tools in Geographic Information System (GIS) software, slope and aspect parameters for each grid area can be accurately extracted from a digital elevation model.

[0052] Step S10122: Based on the slope parameters and aspect parameters of each grid region, several terrain parameters are obtained.

[0053] After obtaining the slope and aspect parameters for each grid region, these parameters collectively constitute crucial information describing the topographic features of that region. Slope and aspect parameters reflect topographic characteristics from different perspectives; combining them provides a more comprehensive description. For example, when analyzing the impact of topography on climate, hydrology, and ecology, these three parameters need to be considered comprehensively. Slope parameters affect processes such as water flow and material transport, while aspect parameters are closely related to solar radiation and vegetation distribution. Integrating these parameters yields several topographic parameters that accurately characterize the topographic features of each grid region, providing detailed foundational data for subsequent analysis of the impact of topographic effects on remote sensing image radiometrics.

[0054] In summary, this embodiment, by extracting slope and aspect parameters from the digital elevation model, can accurately characterize the steepness and orientation of the terrain in each grid area. This is crucial for understanding the impact of terrain on natural factors such as sunlight and airflow. By combining slope and aspect, the topographic features of the target terrain area are comprehensively and in detail described.

[0055] For step S102, for each of the imaging geometric parameter groups, based on the several terrain parameters and the illumination-observation parameter group, the slope reflectance of the corresponding scene is simulated and obtained by terrain meshing, the slope reflectance of the corresponding scene includes the slope reflectance of each terrain parameter; based on the preset ideal terrain parameters and the illumination-observation parameter group, the planar reflectance of the reference plane of the corresponding scene is simulated and obtained by planar reflectance.

[0056] Ideal terrain is usually assumed to be flat and unobstructed. Therefore, parameters such as elevation, slope, and aspect in ideal terrain are all set to 0.

[0057] This step, for each set of illumination-observation parameters, combined with several acquired topographic parameters, uses a specific three-dimensional radiative transfer model to calculate the slope reflectivity under each topographic parameter. Slope reflectivity reflects the ability of objects to reflect sunlight under different actual topographic conditions. The remote sensing radiative transfer model, based on physical principles, considers the influence of topography, illumination, and other factors on the reflected light from objects, simulating the propagation process of light between objects and the atmosphere. Simultaneously, based on preset ideal topographic parameters, using the same three-dimensional radiative transfer model and the same set of illumination-observation parameters, the planar reflectivity under ideal topographic conditions is calculated. Planar reflectivity represents the reflectivity of objects under ideal topographic conditions.

[0058] Please refer to Figure 3 In one embodiment, step S102, which involves simulating the reflectivity of slope features in a terrain-gridized manner based on the plurality of terrain parameters and the illumination-observation parameter set, includes:

[0059] Step S1021: Input each of the terrain parameters and the illumination-observation parameter group into the three-dimensional radiative transfer model to obtain the full-spectrum reflectance data of the simulated ground corresponding to each of the terrain parameters.

[0060] The three-dimensional radiative transfer model is a mathematical model used to simulate the propagation, scattering, and absorption of light in three-dimensional space. In the field of remote sensing, it can consider the influence of various factors such as the three-dimensional features of the terrain, illumination conditions, and the atmosphere on the reflection of light on the Earth's surface. Based on the principles of physical optics, this model calculates the reflection of light on the Earth's surface by establishing a series of equations governing the interaction between photons and surface materials. Various terrain parameters, such as slope and aspect, as well as illumination-observation parameters such as solar zenith angle, solar azimuth angle, sensor observation zenith angle, and observation azimuth angle, are input into the three-dimensional radiative transfer model. The model then simulates the reflection of light on a simulated ground surface under different terrain conditions, thereby obtaining full-spectrum reflectance data for each terrain parameter. This full-spectrum reflectance data covers surface reflectance information across multiple wavelengths, from visible to infrared light, comprehensively reflecting the surface's reflectance characteristics at different wavelengths.

[0061] Step S1022: Perform spectral response function convolution processing on the full-spectrum reflectance data corresponding to each of the terrain parameters to obtain the slope reflectance of each terrain parameter.

[0062] When receiving light reflected from the Earth's surface, satellite sensors are not equally sensitive to all wavelengths; they exhibit specific spectral response characteristics. The spectral response function describes the sensor's response to different wavelengths of light. Convolving the full-spectrum reflectance data for each terrain parameter with the spectral response function involves mathematically performing a operation between the full-spectrum reflectance data and the satellite sensor's spectral response function. This process converts the full-spectrum reflectance data into reflectance data that the satellite sensor can actually receive—that is, the reflectance of slope features under each terrain parameter.

[0063] For the convolution processing of the spectral response function in this embodiment, please refer to the following formula:

[0064]

[0065] in, To simulate the reflectivity of an image, Indicates the band. It is the spectral response function corresponding to each band of the satellite sensor. The satellite sensor reflectivity is obtained based on the integral of the measured reflectivity.

[0066] This step makes the simulated ground reflectance more consistent with the actual observations of satellite sensors, providing more accurate data for subsequent analysis and correction.

[0067] Step S102, which involves simulating the planar reflectance of the reference plane of the corresponding scene based on preset ideal terrain parameters and the illumination-observation parameter set, includes:

[0068] Step S1023: Input the ideal terrain parameters and the illumination-observation parameter set into the three-dimensional radiative transfer model to obtain the full-spectrum reflectance data of the simulated ground corresponding to the ideal terrain parameters;

[0069] Ideal terrain parameters are an idealized description of terrain features, typically assuming flat terrain, a 0° slope, and no specific aspect. By inputting these ideal terrain parameters along with illumination-observation parameters into a three-dimensional radiative transfer model, the model, also based on physical optics principles, simulates the reflection of light under ideal terrain conditions, yielding full-spectrum reflectance data for the simulated ground surface with the ideal terrain parameters. This simulation provides a benchmark for subsequent comparison with reflectance under actual terrain conditions, aiding in the analysis of the impact of terrain effects on the reflectance of ground features.

[0070] Step S1024: Perform spectral response function convolution processing on the full-spectrum reflectance data corresponding to the ideal terrain parameters to obtain the planar reflectance of the reference plane of the corresponding scene.

[0071] Similar to step S1022, the full-spectrum reflectance data corresponding to the ideal terrain parameters is convolved with the spectral response function. This convolution operation yields the planar reflectance of the ground features under the ideal terrain. This planar reflectance represents the actual reflectance of ground features that the satellite sensor can receive when there are no terrain effects (i.e., the terrain is ideally flat). It provides important reference data for subsequent assessment and elimination of the impact of terrain effects on actual remote sensing images.

[0072] In summary, this embodiment, by introducing a three-dimensional radiative transfer model, can fully consider the complex influence of actual terrain and illumination observation conditions on the light reflection process on the land surface, accurately simulating full-spectrum reflectance data under different terrain parameters. Then, through spectral response function convolution processing, these simulated data are converted into the reflectance of slope features that can actually be received by satellite sensors, making the data closer to actual observation conditions. Simultaneously, using preset ideal terrain parameters, simulation and convolution processing are performed under the same illumination observation conditions to obtain the reflectance of planar features under ideal terrain, providing a benchmark reference for subsequent analysis of terrain effects.

[0073] In one embodiment, step S1022, which involves performing spectral response function convolution processing on the full-spectrum reflectance data corresponding to each of the terrain parameters to obtain the slope reflectance corresponding to each of the terrain parameters, includes:

[0074] Step S10221: Obtain the spectral response function corresponding to each observation signal band of the satellite sensor;

[0075] When satellite sensors receive light signals reflected from the Earth's surface, they are not equally sensitive to all wavelengths of light; instead, they observe specific bands. Each observed signal band has a corresponding spectral response function, which describes the sensor's response to different wavelengths within that band. For example, for a specific band of the Landsat satellite, the spectral response function might show a strong response within a particular wavelength range (e.g., 0.5–0.6 micrometers) and a weaker response in other wavelength ranges. Obtaining these spectral response functions is fundamental for subsequent convolution processing; these functions are typically available from the satellite sensor's official documentation, technical manuals, or relevant remote sensing data platforms.

[0076] Step S10222: According to the wavelength range of each observation signal band of the satellite sensor, the full spectrum reflectance data corresponding to each terrain parameter is segmented to obtain the reflectance data corresponding to each observation signal band.

[0077] Full-spectrum reflectance data encompasses surface reflectance information across multiple bands, from visible to infrared light, and is a continuous spectral curve. Satellite sensors, however, observe in specific bands; therefore, the full-spectrum reflectance data needs to be segmented according to the wavelength range of each observed signal band. For example, if a satellite sensor has three observation bands with wavelength ranges of band 1 (0.4-0.5 micrometers), band 2 (0.5-0.6 micrometers), and band 3 (0.6-0.7 micrometers), then the reflectance data for each of these three bands is extracted from the full-spectrum reflectance data to obtain the full-spectrum reflectance data corresponding to each observed signal band.

[0078] Step S10223: Perform a weighted integral operation on the reflectance data corresponding to each observation signal band and the spectral response function corresponding to the observation signal band to obtain the reflectance of the slope surface features corresponding to each terrain parameter.

[0079] The purpose of weighted integration is to simulate the actual reception of reflected light by satellite sensors within each observed signal band. Since the spectral response function describes the sensor's response to different wavelengths of light, weighted integration of the full-spectrum reflectance data corresponding to each observed signal band with the spectral response function of that band effectively considers the sensor's sensitivity to different wavelengths of reflected light within that band. Specifically, for each band, the reflectance values ​​of each wavelength within that band are multiplied by the corresponding spectral response function value, and then the product is integrated over the entire band. The result is the slope reflectance value corresponding to that terrain parameter in that observed signal band. In this way, the full-spectrum reflectance data is converted into reflectance data that the satellite sensor can actually receive, making the data more consistent with actual observations.

[0080] Step S1024, which involves performing spectral response function convolution processing on the full-spectrum reflectance data corresponding to the ideal terrain parameters to obtain the planar reflectance of the reference plane of the corresponding scene, includes:

[0081] Step S10241: According to the wavelength range of each observation signal band of the satellite sensor, the full spectrum reflectance data corresponding to the ideal terrain parameters is segmented to obtain the reflectance data corresponding to each observation signal band.

[0082] Similar to step S10222, the full-spectrum reflectance data corresponding to the ideal terrain parameters is also a continuous spectral curve. In order to simulate the observation of reflected light from ground objects under ideal terrain by satellite sensors, it is necessary to segment the data according to the wavelength range of each observation signal band of the satellite sensor to obtain the full-spectrum reflectance data corresponding to each observation signal band.

[0083] Step S10242: Perform a weighted integral operation on the full-spectrum reflectance data corresponding to each observation signal band and the spectral response function corresponding to the observation signal band to obtain the planar reflectance of the reference plane of the corresponding scene.

[0084] Similar to step S10223, the full-spectrum reflectance data for each observation signal band corresponding to the ideal terrain parameters is weighted and integrated with the spectral response function of that band. This calculation simulates the actual reception of reflected light by the satellite sensor in each observation signal band under ideal terrain conditions, yielding the reflectance of planar features in that band. This planar feature reflectance represents the actual reflectance that the satellite sensor can receive when there are no terrain effects (i.e., the terrain is ideally flat), providing important reference data for subsequent assessment and elimination of the impact of terrain effects on actual remote sensing images.

[0085] In summary, this embodiment obtains the spectral response functions corresponding to each observation signal band of the satellite sensor, and segments the full-spectrum reflectance data according to the wavelength range of the observation signal bands, so that the data corresponds to the actual observation bands of the satellite sensor. The weighted integral operation fully considers the sensor's response to different wavelengths of light, converting the full-spectrum reflectance data into reflectance data that the satellite sensor can actually receive, namely, the reflectance of slope features and the reflectance of planar features. This method based on spectral response function convolution processing can more accurately and comprehensively obtain reflectance information of features under different terrain conditions, providing a solid data foundation for subsequently eliminating the influence of terrain effects on remote sensing image radiometrics. This helps improve the accuracy of terrain radiometric correction of remote sensing images, enabling the corrected remote sensing images to more realistically reflect the characteristics of ground features.

[0086] For step S103, for each of the imaging geometric parameter groups, based on the reflectivity of the slope features in the terrain gridding and the reflectivity of the planar features in the reference plane, the target correction parameters of each terrain parameter under each of the illumination-observation parameter groups are obtained.

[0087] For each set of imaging geometric parameters, the ratio of slope reflectance to planar reflectance under each topographic parameter is calculated. This ratio reflects the influence of actual topography on reflectance relative to ideal topography, and is used as the target correction parameter for each topographic parameter under that illumination-observation parameter set. The target correction parameter is used to quantify the radiometric distortion caused by topographic effects, providing a crucial basis for subsequent topographic radiometric correction.

[0088] For the calculation of the target correction parameters in this embodiment, please refer to the following formula:

[0089]

[0090] in, , These represent the simulated plane reflectance and slope reflectance, respectively. These are the solar zenith angle under a flat surface, the sensor-observed zenith angle, the solar azimuth angle, and the sensor-observed azimuth angle, respectively. The incident zenith angle, the observed zenith angle, the incident azimuth angle, and the observed azimuth angle under slope geometry; This includes the slope, aspect, and corresponding waveband.

[0091] Please refer to Figure 4 In one embodiment, the terrain parameters include aspect parameters and slope parameters, and the illumination parameters include solar zenith angle and solar azimuth angle;

[0092] Step S103, which describes obtaining the target correction parameters for each terrain parameter under each illumination-observation parameter group for each imaging geometric parameter group, based on the reflectivity of the slope features in the terrain mesh and the reflectivity of the planar features in the reference plane, includes:

[0093] Step S1031: For each of the imaging geometric parameter groups, the first correction parameter of each terrain parameter under each of the illumination-observation parameter groups is obtained based on the ratio of the reflectance of the slope surface to the reflectance of the planar surface under each of the terrain parameters.

[0094] For each set of illumination-observation parameters, the ratio of slope reflectance under each topographic parameter to planar reflectance under ideal topography is calculated. This ratio reflects the degree of difference in reflectance between actual and ideal flat topography. For example, if the slope reflectance is 0.3 under a certain topographic parameter and the planar reflectance is 0.25 under ideal topography, then the first correction parameter for that topographic parameter under this set of illumination-observation parameters is 0.3 / 0.25 = 1.2. This first correction parameter initially reflects the influence of topography on reflectance, providing basic data for further processing and determining the target correction parameter.

[0095] Step S1032: Based on the solar zenith angle and solar azimuth angle in the illumination-observation parameter group and the slope aspect parameter and slope parameter in the terrain parameters, obtain the slope solar incidence angle of the grid area corresponding to the terrain parameters;

[0096] The solar incidence angle is the angle between the sun's rays and the ground normal. Based on the solar zenith angle and solar azimuth angle in the illumination-observation parameter set, and the slope aspect and slope parameters in the terrain parameters, the solar incidence angle of the slope in the corresponding grid area can be calculated. The solar zenith angle is the angle between the sun's rays and the local zenith direction, while the slope aspect parameter describes the orientation of the terrain. Through geometric calculations, the actual angle of incidence of sunlight in that terrain grid area can be determined. For example, given a solar zenith angle of 30°, a slope of 30°, a solar azimuth angle of 270°, and a slope aspect of 180°, the solar incidence angle of the slope can be calculated to be approximately 41°.

[0097] Step S1033: If the solar incidence angle on the slope is greater than the solar incidence angle on the plane, the grid area corresponding to the terrain parameter is determined to be a shaded slope area; if the solar incidence angle on the slope is less than or equal to the solar incidence angle on the plane, the grid area corresponding to the terrain parameter is determined to be a non-shaded slope area.

[0098] The calculated solar incidence angle on the slope is compared with the solar incidence angle on the plane. If the solar incidence angle on the slope is greater than the solar incidence angle on the plane, it indicates that the angle between the sunlight and the ground is larger, and the grid area corresponding to this terrain parameter receives relatively less solar radiation, thus being identified as a shaded slope area. If the solar incidence angle on the slope is less than or equal to the solar incidence angle on the plane, it indicates that the angle between the sunlight and the ground is smaller, and the grid area receives relatively more solar radiation, thus being identified as a non-shaded slope area. For example, if the solar zenith angle is 60°, when the solar incidence angle is 70°, the grid area is a shaded slope area; when the solar incidence angle is 50°, the grid area is a non-shaded slope area.

[0099] Step S1034: When the grid area corresponding to the terrain parameter is a non-shaded slope area, the first correction parameter of the terrain parameter under the illumination-observation parameter group is determined as the target correction parameter of the terrain parameter under the illumination-observation parameter group.

[0100] When the grid area corresponding to the terrain parameters is determined to be a non-shaded slope area, the terrain effect has a relatively direct and significant impact on the reflectivity of ground features in this area because the solar radiation received in non-shaded slope areas is relatively abundant. Therefore, the first correction parameter of the terrain parameter under this illumination-observation parameter set is directly determined as the target correction parameter. This means that in this case, the first correction parameter can well reflect the correction requirements of the terrain on the reflectivity of ground features, without the need for additional processing.

[0101] Step S1035: When the grid area corresponding to the terrain parameter is a shady slope area, determine several non-shady slope areas adjacent to the shady slope area, calculate the mean value of the first correction parameter corresponding to the several non-shady slope areas, and determine the mean value of the first correction parameter corresponding to the several non-shady slope areas as the target correction parameter of the terrain parameter under the illumination-observation parameter group.

[0102] When the grid area corresponding to the terrain parameters is determined to be a shaded slope area, the impact of terrain effects on the reflectivity of features in this area may be influenced by various complex factors due to the lower solar radiation received. Therefore, the first correction parameter may not accurately reflect the actual terrain correction requirements. Thus, several non-shaded slope areas adjacent to the shaded slope area are identified, and the mean of the first correction parameters for these non-shaded slope areas is calculated. This mean comprehensively considers the terrain effects of the surrounding non-shaded slope areas and is used as the target correction parameter for the terrain parameters of the shaded slope area under this illumination-observation parameter set. For example, if four non-shaded slope areas are found around the shaded slope area, and their first correction parameters are calculated to be 1.1, 1.2, 1.15, and 1.25 respectively, then the mean of these four first correction parameters is (1.1 + 1.2 + 1.15 + 1.25) / 4 = 1.175. This mean is used as the target correction parameter for the shaded slope area.

[0103] In summary, this embodiment, by comprehensively considering topographic and illumination parameters, can more accurately determine the target correction parameters for different topographic regions, such as shaded and non-shaded slopes. Specifically, the calculation of the first correction parameter initially reflects the influence of topography on the reflectivity of ground features, while the calculation of the solar incidence angle and the identification of shaded slope regions further distinguish the differences in illumination conditions among different topographic regions. For non-shaded slope regions, the first correction parameter is directly used as the target correction parameter, ensuring the accuracy of the correction; for shaded slope regions, the average value of the first correction parameter of the surrounding non-shaded slope regions is used as the target correction parameter, fully considering the spatial correlation between topographic features and the complexity of illumination conditions, avoiding the problem of inaccurate correction parameters due to the special illumination conditions of shaded slope regions. This comprehensive processing method based on topographic and illumination parameters can improve the accuracy of topographic radiometric correction of remote sensing images, enabling the corrected remote sensing images to more realistically reflect the characteristics of ground features.

[0104] For step S104, a training dataset is constructed by traversing the entire scene parameter space based on the terrain parameters, illumination-observation parameter groups, and target correction parameters of each terrain parameter under each illumination-observation parameter group of the imaging geometric parameter group of the several scenes; the training dataset is input into a preset model for training to obtain a trained correction parameter prediction model; the preset model is used to learn the entire scene correction mapping relationship from terrain parameters and illumination-observation parameter groups to correction parameters.

[0105] This step combines several terrain parameters, several sets of illumination-observation parameters, and target correction parameters for each terrain parameter under each illumination-observation parameter set for various scenes to form a training dataset. This training dataset contains rich correspondences between terrain, illumination observation conditions, and correction parameters. The training dataset is then input into a pre-defined model, such as a neural network model, for training. During training, the model continuously adjusts its parameters to learn the mapping relationship from terrain parameters and illumination-observation parameter sets to correction parameters. After multiple iterations of training, when the model's prediction error reaches a preset threshold, the trained correction parameter prediction model is obtained.

[0106] This step constructs feature engineering based on slope, aspect, and illumination geometry, and uses the CatBoost gradient boosting algorithm to generate a correction parameter prediction model. This model enables band-adaptive terrain correction.

[0107] In one embodiment, the training dataset includes several training samples, each training sample including the terrain parameters and the illumination-observation parameter set, and the terrain parameters and / or the illumination-observation parameter set are different in different training samples; the training label of each training sample is the corresponding target correction parameter;

[0108] Step S104, which involves inputting the training dataset into a preset model for training to obtain a trained corrected parameter prediction model, includes:

[0109] Step S1041: Input the training dataset into the preset model to obtain the prediction information corresponding to each training sample; calculate the loss value between the prediction information of each training sample and the corresponding training label.

[0110] The training samples include terrain parameters and illumination-observation parameter sets, and different training samples differ in terrain parameters and / or illumination-observation parameter sets. This design makes the training dataset rich in diversity, covering various combinations of terrain and illumination conditions, which helps the model learn more comprehensive and accurate patterns. At the same time, each training sample has a corresponding target correction parameter as a training label, which is an important basis for subsequent model training and evaluation.

[0111] A training dataset containing several training samples is input into a pre-defined model. The model processes each training sample and, based on its internal structure and algorithm, generates prediction information for each sample. This prediction information consists of the target correction parameters predicted by the model based on the input terrain parameters and illumination-observation parameter set.

[0112] To measure the accuracy of model predictions, it is necessary to calculate the loss value between the predicted information of each training sample and the corresponding training label (target correction parameter). The loss value is an indicator that measures the difference between the model's prediction result and the actual result. Common loss functions include mean squared error and cross-entropy loss. By calculating the loss value, we can intuitively understand the magnitude of the model's prediction error on the current training samples.

[0113] Step S1042: Adjust the weight parameters of the preset model according to the loss value until the loss value is less than the preset loss threshold or the total number of weight parameter adjustments reaches the preset number of adjustments, and obtain the trained correction parameter prediction model.

[0114] Based on the calculated loss value, the weight parameters of the preset model are adjusted. Weight parameters are key parameters in the model that determine the degree of influence of each input feature on the output result. If the loss value is large, it indicates a significant deviation between the model's prediction and the actual result. In this case, it is necessary to adjust the weight parameters to change the model's processing method of the input features, thereby reducing the prediction error. The process of adjusting weight parameters typically employs optimization algorithms such as gradient descent. By calculating the gradient of the loss function with respect to the weight parameters, the values ​​of the weight parameters are gradually adjusted along the direction of gradient descent.

[0115] The process of adjusting the weight parameters described above continues until one of the following two conditions is met: First, the loss value is less than a preset loss threshold, which means that the model's prediction error has been reduced to an acceptable range, and the model has achieved a good training effect; second, the total number of weight parameter adjustments reaches a preset number of adjustments. This is to prevent the model from overfitting or getting stuck in local optima during training. By setting a maximum number of adjustments, the training process is limited to ensure the efficiency of model training. When either of these conditions is met, the model can be considered to have completed training, and a trained, corrected parameter prediction model is obtained.

[0116] In summary, this embodiment provides the model with rich training samples by constructing a training dataset containing diverse combinations of terrain and illumination conditions. This helps the model learn to accurately predict target correction parameters under different terrain and illumination conditions. During model training, the loss value between the predicted information and the training labels is calculated, and the model's weight parameters are adjusted based on the loss value, allowing the model to continuously optimize its prediction performance. Simultaneously, setting a preset loss threshold and a preset number of adjustments as criteria for training completion ensures both the accuracy of model training and avoids overfitting and excessively long training times. The resulting fully trained correction parameter prediction model can quickly and accurately predict target correction parameters based on the input terrain parameters and illumination-observation parameter set, providing an efficient and reliable solution for subsequent applications such as remote sensing image topographic radiometric correction.

[0117] For step S105, the terrain parameters, illumination parameters, and observation parameters of the remote sensing image to be corrected are obtained. The terrain parameters, illumination parameters, and observation parameters of the remote sensing image to be corrected are input into the correction parameter prediction model to obtain the correction parameters of the remote sensing image to be corrected. Based on the correction parameters of the remote sensing image to be corrected, the remote sensing image to be corrected is subjected to full-scene adaptive terrain radiometric correction.

[0118] In this step, the topographic parameters, illumination parameters, and observation parameters of the remote sensing image to be corrected are obtained through the image's metadata or relevant geographic information system data. These parameters are then input into a trained correction parameter prediction model. Based on the previously learned mapping relationships, the model outputs the correction parameters for the image. Finally, based on the obtained correction parameters, a specific topographic radiometric correction algorithm, such as a correction algorithm based on a radiative transfer model, is used to perform topographic radiometric correction on the image, eliminating radiometric distortion caused by topographic effects.

[0119] Please refer to the following formula, which is used in this embodiment to achieve topographic radiation correction:

[0120]

[0121] in, The CatBoost model obtained after training. and This represents the reflectance before and after correction.

[0122] The above embodiments are merely illustrative of several implementation methods of this application, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make several modifications and improvements without departing from the concept of this application, and this application also intends to include these modifications and variations.

Claims

1. A method for terrain radiometric correction of remote sensing images based on full scene simulation and machine learning, characterized in that, The method comprises the following steps: obtaining a full-scene imaging geometry combination comprising a plurality of imaging geometry parameter sets of scenes; wherein each imaging geometry parameter set comprises a plurality of terrain parameters and a plurality of light-observation parameter sets; each light-observation parameter set comprises light parameters and observation parameters, and the light parameters and / or observation parameters in different light-observation parameter sets are different; for each imaging geometry parameter set, based on the plurality of terrain parameters and the light-observation parameter sets, simulating terrain grid slope surface feature reflectance of a corresponding scene, wherein the terrain grid slope surface feature reflectance comprises slope surface feature reflectance of each terrain parameter; based on preset ideal terrain parameters and the light-observation parameter sets, simulating plane surface feature reflectance of a reference plane of the corresponding scene; for each imaging geometry parameter set, based on the terrain grid slope surface feature reflectance and the plane surface feature reflectance of the reference plane, obtaining target correction parameters of each terrain parameter under each light-observation parameter set; based on the terrain parameters, the light-observation parameter sets, and the target correction parameters of each terrain parameter under each light-observation parameter set of the imaging geometry parameter sets of the plurality of scenes, constructing a training data set through full-scene parameter space traversal; inputting the training data set into a preset model for training to obtain a trained correction parameter prediction model; the preset model is used to learn a full-scene correction mapping relationship from terrain parameters and light-observation parameter sets to correction parameters; obtaining terrain parameters, light parameters, and observation parameters of a to-be-corrected remote sensing image, inputting the terrain parameters, light parameters, and observation parameters of the to-be-corrected remote sensing image into the correction parameter prediction model to obtain correction parameters of the to-be-corrected remote sensing image; and performing full-scene adaptive terrain radiation correction on the to-be-corrected remote sensing image based on the correction parameters of the to-be-corrected remote sensing image.

2. The remote sensing image terrain radiation correction method based on full-scene simulation and machine learning according to claim 1, wherein the imaging geometry parameter set is obtained by the following method: obtaining a digital elevation model of a target terrain region; obtaining terrain parameters of each grid region from the digital elevation model to obtain a plurality of terrain parameters; obtaining a plurality of light parameters and observation parameters of the target terrain region; and obtaining a plurality of light-observation parameter sets according to the plurality of light parameters and observation parameters; obtaining the imaging geometry parameter set according to the plurality of terrain parameters and the plurality of light-observation parameter sets.

3. The remote sensing image terrain radiation correction method based on full-scene simulation and machine learning according to claim 2, wherein the step of obtaining terrain parameters of each grid region from the digital elevation model to obtain a plurality of terrain parameters comprises: extracting slope parameters and aspect parameters of each grid region from the digital elevation model; obtaining a plurality of terrain parameters according to the slope parameters and aspect parameters of each grid region. ​ ​ 4. The full-scene simulation and machine learning based remote sensing image terrain radiometric correction method according to claim 2, wherein the terrain parameters include slope aspect parameters and slope gradient parameters, and the illumination parameters include solar zenith angle and solar azimuth angle. The step of obtaining the target correction parameter of each terrain parameter under each illumination-observation parameter group based on the slope surface terrain object reflectivity of the terrain grid and the planar terrain object reflectivity of the reference plane for each imaging geometry parameter group comprises: For each imaging geometry parameter group, a first correction parameter of each terrain parameter under each illumination-observation parameter group is obtained according to the ratio of the slope surface terrain object reflectivity under each terrain parameter to the planar terrain object reflectivity of the reference plane; The slope surface solar incident angle of the grid region corresponding to the terrain parameter is obtained according to the solar zenith angle, the solar azimuth angle in the illumination-observation parameter group, and the slope aspect parameter and the slope gradient parameter in the terrain parameter; If the slope surface solar incident angle is greater than the planar solar incident angle, the grid region corresponding to the terrain parameter is determined as a shady region; if the slope surface solar incident angle is less than or equal to the planar solar incident angle, the grid region corresponding to the terrain parameter is determined as a non-shady region; When the grid region corresponding to the terrain parameter is a non-shady region, the first correction parameter of the terrain parameter under the illumination-observation parameter group is determined as the target correction parameter of the terrain parameter under the illumination-observation parameter group; When the grid region corresponding to the terrain parameter is a shady region, a plurality of non-shady regions adjacent to the shady region are determined, and the average of the first correction parameters corresponding to the plurality of non-shady regions is calculated; the average of the first correction parameters corresponding to the plurality of non-shady regions is determined as the target correction parameter of the terrain parameter under the illumination-observation parameter group.

5. The full-scene simulation and machine learning based remote sensing image terrain radiometric correction method according to claim 1, wherein the step of simulating the slope surface terrain object reflectivity of the terrain grid corresponding to the scene based on the plurality of terrain parameters and the illumination-observation parameter group comprises: The full-wave spectrum reflectivity data corresponding to the simulation ground of each terrain parameter is obtained by inputting each terrain parameter and the illumination-observation parameter group into a three-dimensional radiative transfer model; The slope surface terrain object reflectivity under each terrain parameter is obtained by performing spectral response function convolution processing on the full-wave spectrum reflectivity data corresponding to each terrain parameter; The step of simulating the planar terrain object reflectivity of the reference plane corresponding to the scene based on the preset ideal terrain parameter and the illumination-observation parameter group comprises: The full-wave spectrum reflectivity data corresponding to the simulation ground of the ideal terrain parameter is obtained by inputting the ideal terrain parameter and the illumination-observation parameter group into a three-dimensional radiative transfer model; The planar terrain object reflectivity of the reference plane corresponding to the scene is obtained by performing spectral response function convolution processing on the full-wave spectrum reflectivity data corresponding to the ideal terrain parameter. ​ ​ 6. The full-scene simulation and machine learning based remote sensing image terrain radiation correction method according to claim 5, wherein the step of performing spectral response function convolution processing on the full-spectrum reflectivity data corresponding to each terrain parameter to obtain the slope surface object reflectivity corresponding to each terrain parameter comprises: obtaining spectral response functions corresponding to each observation signal band of the satellite sensor; performing segmented processing on the full-spectrum reflectivity data corresponding to each terrain parameter according to the wavelength range of each observation signal band of the satellite sensor to obtain reflectivity data corresponding to each observation signal band; performing weighted integral operation on the reflectivity data corresponding to each observation signal band and the spectral response function corresponding to the observation signal band to obtain the slope surface object reflectivity corresponding to each terrain parameter. The step of performing spectral response function convolution processing on the full-spectrum reflectivity data corresponding to the ideal terrain parameter to obtain the planar object reflectivity of the reference plane of the corresponding scene comprises: performing segmented processing on the full-spectrum reflectivity data corresponding to the ideal terrain parameter according to the wavelength range of each observation signal band of the satellite sensor to obtain reflectivity data corresponding to each observation signal band; performing weighted integral operation on the reflectivity data corresponding to each observation signal band and the spectral response function corresponding to the observation signal band to obtain the planar object reflectivity of the reference plane of the corresponding scene.

7. The full-scene simulation and machine learning based remote sensing image terrain radiation correction method according to claim 1, wherein the training data set comprises a plurality of training samples, each training sample comprising the terrain parameter and the illumination-observation parameter group, and the terrain parameter and / or the illumination-observation parameter group in different training samples are different; and the training label of each training sample is the corresponding target correction parameter. The step of inputting the training data set into a preset model for training to obtain a trained correction parameter prediction model comprises: inputting the training data set into the preset model to obtain prediction information corresponding to each training sample; calculating the loss value between the prediction information of each training sample and the corresponding training label; adjusting the weight parameters of the preset model according to the loss value until the loss value is less than a preset loss threshold or the total number of weight parameter adjustments reaches a preset adjustment number, to obtain the trained correction parameter prediction model. ​ ​

Images