Optimization method for hyperspectral image atmospheric correction based on ground truthing data

Abstract

This invention provides a hyperspectral image atmospheric correction optimization method based on ground-based authenticity verification data, belonging to the field of remote sensing image processing technology. The method includes: acquiring raw satellite radiance data during satellite transit in the experimental area to obtain true values ​​of surface reflectance and atmospheric parameters; performing initial atmospheric correction to generate initial surface reflectance; quantifying spectral deviations through band-by-band comparison to identify systematic spectral biases; performing global optimization to output optimized atmospheric parameters and atmospheric transmittance; performing local error correction to obtain terrain correction parameters and mixed pixel correction parameters; and iteratively optimizing and inverting to output a high-precision surface reflectance product. This invention solves the technical problem of existing technologies using physical methods based on radiative transfer models, where correction accuracy heavily relies on the accuracy of prior atmospheric parameters and cannot accurately reflect the actual atmospheric conditions in local areas during satellite transit, leading to reduced reliability in mineral identification.

Description

Technical Field

[0001] This invention relates to the field of remote sensing image processing technology, and in particular to a method for atmospheric correction and optimization of hyperspectral images based on ground authenticity verification data. Background Technology

[0002] While traditional physical methods based on radiative transfer models are widely used, their correction accuracy heavily depends on the accuracy of prior atmospheric parameters. Conventional parameter acquisition methods often fail to accurately reflect the actual atmospheric conditions in local areas during satellite transit, especially with significant uncertainties in sensitive parameters such as aerosols and water vapor.

[0003] Meanwhile, in order to balance computational efficiency, traditional models often use fast approximation algorithms to replace precise line-by-line calculations in this band. This further introduces non-negligible systematic spectral bias and distortion, which weakens or distorts the diagnostic absorption characteristics of minerals such as clay and carbonates, directly affecting the reliability of mineral identification.

[0004] Furthermore, existing methods typically lack effective mechanisms for closed-loop, targeted error analysis and parameter optimization with synchronous ground-measured ground data. Their processing flow is mostly an open-loop fixed mode, making it difficult to automatically correct model biases. Moreover, they generally do not fully consider the effects of complex terrain and spatial scale errors caused by mixed pixels, resulting in inconsistent correction results under different landforms and ground features. This limits spatial fidelity and fails to meet the urgent needs of quantitative remote sensing applications such as geological exploration and precise mineral mapping for high-precision, high-consistency reflectance products.

[0005] It should be noted that the information disclosed in this background section is intended only to enhance the understanding of the overall background of the present invention, and should not be construed as an admission or in any way implying that the information constitutes prior art known to those skilled in the art. Summary of the Invention

[0006] In response to the above-mentioned deficiencies or improvement needs of existing technologies, this invention provides a hyperspectral image atmospheric correction optimization method based on ground authenticity verification data. This method solves the technical problem that existing technologies use physical methods based on radiative transfer models, where the correction accuracy heavily depends on the accuracy of prior atmospheric parameters and cannot accurately reflect the actual atmospheric conditions of local areas during satellite transit, leading to reduced reliability of mineral identification.

[0007] The specific technical solution is as follows:

[0008] This invention provides a method for atmospheric correction and optimization of hyperspectral images based on ground-based veracity verification data, the method comprising:

[0009] During satellite transit in the experimental area, raw satellite radiance data is acquired, and ground truth data acquisition is performed simultaneously to obtain true values ​​for surface reflectance and atmospheric parameters. A standard radiative transfer model is used to perform initial atmospheric correction on the raw satellite radiance data to generate initial surface reflectance. Within a preset mineral diagnostic band range, the true surface reflectance and initial surface reflectance are compared band by band to quantify spectral deviations and identify systematic spectral biases. Global optimization is performed based on these systematic spectral biases, outputting optimized atmospheric parameters and atmospheric transmittance. Spatial deviations between the initial surface reflectance and the true surface reflectance are locally corrected to obtain terrain correction parameters and mixed pixel correction parameters. The raw satellite radiance data, optimized atmospheric parameters, atmospheric transmittance, terrain correction parameters, and mixed pixel correction parameters are input into a hierarchical inversion model, and a high-precision surface reflectance product is output through iterative optimization and inversion.

[0010] In one implementation, after inputting the original satellite radiance data, optimized atmospheric parameters, atmospheric transmittance, terrain correction parameters, and hybrid pixel correction parameters into a hierarchical inversion model, a high-precision surface reflectance product is output through iterative optimization and inversion, including:

[0011] A hierarchical inversion model comprising a global optimization layer and a local correction layer is constructed. In the global optimization layer, the initial surface reflectance is inverted based on the original satellite radiance data, optimized atmospheric parameters, and atmospheric transmittance. In the local correction layer, the terrain correction parameters are applied to perform radiometric correction on the terrain shadow distortion area, and the mixed pixel correction parameters are applied to perform spectral unmixing correction on the mixed pixel distortion area to obtain the corrected reflectance. The terrain correction parameters and the mixed pixel correction parameters are iteratively adjusted, and the deviation between the corrected reflectance and the true value of the surface reflectance is calculated until the deviation meets a preset threshold, and the high-precision surface reflectance product is output.

[0012] In one embodiment, the true values ​​of the atmospheric parameters are measured using a solar photometer, wherein the true values ​​of the atmospheric parameters include the measured true values ​​of aerosol optical thickness and the measured true values ​​of water vapor column concentration.

[0013] In one implementation, global optimization is performed based on the systematic spectral bias, outputting optimized atmospheric parameters and atmospheric transmittance, including:

[0014] If the systematic spectral deviation indicates an atmospheric parameter deviation, the original model parameters of the standard radiative transfer model are replaced with the true atmospheric parameters to obtain the optimized atmospheric parameters. Within the mineral diagnostic band, the optimized atmospheric parameters are used as input conditions, and the LBL algorithm is used to replace the fast approximation algorithm in the standard radiative transfer model to calculate the atmospheric transmittance.

[0015] In one implementation, the original model parameters of the standard radiative transfer model are replaced using the true values ​​of the atmospheric parameters to obtain the optimized atmospheric parameters, including:

[0016] If the systematic spectral deviation manifests as multiple consecutive bands shifting in the same direction within the mineral diagnostic band range, and the shift amplitude is greater than a preset shift threshold, then the systematic spectral deviation is determined to indicate an atmospheric parameter deviation. Based on the spectral distribution characteristics of the systematic spectral deviation, the type of deviation parameter is analyzed and located. If the deviation parameter type is aerosol optical thickness, then the default aerosol optical thickness parameter in the standard radiative transfer model is replaced with the measured true value of aerosol optical thickness, and the model parameters are encapsulated as the optimized atmospheric parameter output. If the deviation parameter type is water vapor column concentration, then the default water vapor parameter in the standard radiative transfer model is replaced with the measured true value of water vapor column concentration, and the model parameters are encapsulated as the optimized atmospheric parameter output.

[0017] In one implementation, within the mineral diagnostic band range, the optimized atmospheric parameters are used as input conditions, and the LBL algorithm is used to replace the fast approximation algorithm in the standard radiative transfer model to calculate the atmospheric transmittance, including:

[0018] In the standard radiative transfer model, the fast approximation algorithm module for the mineral diagnostic band interval is turned off, and the bound LBL algorithm module is activated; within the mineral diagnostic band interval, real-time atmospheric profile data and the optimized atmospheric parameters are input into the LBL algorithm module, atmospheric absorption and scattering transmittance are calculated wavelength by wavelength, and the atmospheric transmittance is output.

[0019] In one embodiment, the spatial deviation between the initial surface reflectance and the true surface reflectance is locally corrected to obtain terrain correction parameters and hybrid pixel correction parameters, including:

[0020] The relative deviation between the initial surface reflectance and the true surface reflectance is calculated pixel by pixel to generate a spatial deviation distribution map. DEM data of the experimental area is overlaid on the spatial deviation distribution map to identify high-deviation pixels in the shaded areas and mark the spatial distribution of terrain shadow distortion zones. An NDVI map of the experimental area is overlaid on the spatial deviation distribution map to identify high-deviation pixels in heterogeneous areas, thereby marking the spatial distribution of mixed pixel distortion zones. A radiation topography correction model is introduced to perform local spatial correction on the spatial distribution of the terrain shadow distortion zones, obtaining the topography correction parameters. A linear spectral unmixing model is introduced to perform local spatial correction on the spatial distribution of the mixed pixel distortion zones, obtaining the mixed pixel correction parameters.

[0021] In one implementation, it further includes:

[0022] Ground targets and a standard reference plate are deployed within the experimental area; the surface radiance of the ground targets and the standard radiance of the standard reference plate are simultaneously measured using a ground spectrometer; the true value of the surface reflectance is calculated based on the surface radiance and the standard radiance.

[0023] Beneficial effects of the embodiments of the present invention:

[0024] During satellite transit, ground data, including true values ​​of surface reflectance and atmospheric parameters, are systematically and synchronously collected. This provides accurate and reliable reference data for subsequent atmospheric correction, improving the fidelity of mineral diagnostic features. Preliminary atmospheric correction of satellite radiance data using a standard radiative transfer model provides initial results for surface reflectance inversion. Although the preliminary inversion contains some errors, it lays the foundation for subsequent optimization, ensuring that a generally accurate reflectance value can still be obtained even without a precise true surface value. By comparing the initial surface reflectance with the true ground values ​​band by band, error analysis is performed on the mineral diagnostic band. This more clearly identifies distortion problems in the mineral diagnostic band caused by traditional methods, providing a clear direction for subsequent optimization. Based on the identified deviations, global optimization is performed, resulting in optimized atmospheric parameters that are more realistic and avoid distortions caused by atmospheric parameters. Spectral distortion caused by numerical errors allows for better preservation of the diagnostic absorption characteristics of minerals, improving the accuracy of mineral identification. By locally correcting the spatial deviation between the initial reflectance and the ground truth, errors caused by terrain shading or mixed pixels are identified and corrected. This process effectively reduces the mixing effect of terrain and land cover, ensuring the accuracy of reflectance data in different land cover and terrain regions, laying the foundation for the final output of high-precision reflectance products. Through a hierarchical inversion model, combined with optimized atmospheric parameters, atmospheric transmittance, terrain correction parameters, and mixed pixel correction parameters, an iterative optimization method is used to accurately invert surface reflectance, achieving high-precision surface reflectance extraction, especially with high fidelity in the mineral diagnostic band. Finally, through iterative optimization, the output high-precision reflectance data can accurately reflect the spectral characteristics of the surface, significantly improving identification accuracy.

[0025] Of course, implementing any product or method of the present invention does not necessarily require achieving all of the advantages described above at the same time. Attached Figure Description

[0026] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0027] Figure 1This invention illustrates a flowchart of the hyperspectral image atmospheric correction optimization method based on ground authenticity verification data provided by the present invention.

[0028] Figure 2 The diagram illustrates the process of outputting a high-precision surface reflectance product in the hyperspectral image atmospheric correction optimization method based on ground authenticity verification data provided by the present invention. Detailed Implementation

[0029] To facilitate understanding of the present invention, a more complete description of the invention will be given below with reference to the accompanying drawings, which illustrate preferred embodiments of the invention. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein; rather, these embodiments are provided to enable a more thorough and complete understanding of the disclosure of the invention.

[0030] Furthermore, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0031] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "clockwise," and "counterclockwise," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0032] Unless otherwise expressly stated, throughout the specification and claims, the term "comprising" or its variations such as "including" or "comprises" shall be understood to include the stated elements or components without excluding other elements or other components.

[0033] The present invention provides a hyperspectral image atmospheric correction optimization method based on ground authenticity verification data, which solves the technical problem that the existing technology uses a physical method based on a radiative transfer model, the correction accuracy of which is heavily dependent on the accuracy of prior atmospheric parameters and cannot accurately reflect the actual atmospheric conditions of local areas when the satellite passes over, thus reducing the reliability of mineral identification.

[0034] See Figure 1 The present invention provides a method for atmospheric correction and optimization of hyperspectral images based on ground-based verification data, the method comprising:

[0035] Y100: When the satellite passes over the experimental area, acquire the satellite's raw radiance data and simultaneously perform ground truth acquisition to obtain the true values ​​of surface reflectance and atmospheric parameters.

[0036] During the satellite's transit window, hyperspectral image data to be calibrated, i.e., raw satellite radiance data, is collected. Radiance is a measure of the amount of light reflected from the ground that reaches the sensor after traveling through the atmosphere. During the satellite's transit, typical ground target objects, such as different types of land features or geological samples, are deployed in the experimental area. The true surface reflectance of these targets is simultaneously measured using a ground-based spectrometer. Through ground observation, surface reflectance data at the same time and location as the satellite imagery is obtained. In addition to surface reflectance, ground-based equipment such as a solar photometers are used to simultaneously measure real atmospheric parameters, such as aerosol optical thickness, reflecting the concentration of aerosols in the atmosphere, and water vapor column concentration, indicating the amount of water vapor in the atmosphere.

[0037] Y200: The original radiance data of the satellite is initially atmospherically corrected using a standard radiative transfer model to generate the initial surface reflectance.

[0038] A standard radiative transfer model, such as FLAASH, is selected to perform preliminary radiometric correction on the satellite imagery data. The FLAASH model uses standard atmospheric parameters, such as aerosols and gas composition, to invert satellite radiance and preliminarily estimate surface reflectance. Using this standard radiative transfer model, an initial surface reflectance is calculated. This reflectance is the result obtained by inverting satellite radiance data and input standard atmospheric parameters. However, when using the standard radiative transfer model, this inversion result may contain errors.

[0039] Y300: Within a preset mineral diagnostic band range, the true value of the surface reflectance and the initial surface reflectance are compared band by band to quantify the spectral deviation, so as to identify systematic spectral deviations.

[0040] The initial surface reflectance is compared band by band with the true surface reflectance, especially in the mineral diagnostic band (2000-2500 nm), which contains key information about the characteristic absorption of minerals such as clay and carbonates. By comparing this band with ground-measured data, deviations in the initial surface reflectance are identified. Methods such as root mean square error and systematic bias are used to quantify the deviation in each band. By calculating these deviations, the accuracy of the preliminary atmospheric correction can be quantitatively assessed. The spatial distribution and spectral characteristics of the deviations are analyzed to identify systematic deviations in specific bands, especially the mineral diagnostic band. For example, large deviations in certain bands indicate inaccurate atmospheric parameters or problems with approximate calculations in the correction algorithm.

[0041] Y400: Performs global optimization based on the aforementioned systematic spectral bias, outputting optimized atmospheric parameters and atmospheric transmittance.

[0042] Based on the calculated deviations, it is determined which atmospheric parameters caused the spectral bias, such as aerosol optical thickness and water vapor column concentration. Systematic spectral biases indicate errors in atmospheric parameters, especially in the mineral diagnostic band, where the spectral characteristics are highly sensitive to atmospheric parameters. Global optimization is performed to adjust atmospheric parameters to reduce these spectral biases. For example, if the bias indicates a deviation in aerosol optical thickness parameters, the default aerosol parameters in the model can be replaced with the actual measured aerosol optical thickness values. In this way, the optimization process makes the atmospheric parameters more closely resemble the actual environment. In atmospheric transport calculations, after optimizing the atmospheric parameters, the LBL (Line-by-Line) algorithm is used to calculate atmospheric transmittance. The LBL algorithm calculates atmospheric absorption and scattering transmittance wavelength by wavelength, replacing the fast approximation algorithm in traditional models. The LBL algorithm provides more accurate atmospheric transmittance calculations, especially in key areas of the mineral diagnostic band. The optimized atmospheric parameters and atmospheric transmittance are then passed as input to subsequent steps to ensure more accurate surface reflectance retrieval.

[0043] Y500: Perform local error correction on the spatial deviation between the initial surface reflectance and the true value of surface reflectance to obtain terrain correction parameters and mixed pixel correction parameters.

[0044] The relative deviation between the initial surface reflectance and the true surface reflectance value is calculated pixel by pixel, generating a spatial deviation distribution map that reflects reflectance errors in different regions. By overlaying DEM (Digital Elevation Model) data of the experimental area, terrain-shadowed areas, such as valleys and areas with steep slopes, are identified, as these areas will exhibit significant deviations due to shading effects. Furthermore, by combining NDVI (Normalized Difference Vegetation Index) maps, heterogeneous regions, such as mixed-pixel areas, are identified. The surface reflectance of these areas is influenced by multiple land features, leading to substantial deviations.

[0045] Radiometric correction is performed on areas with topographic shadow distortion. Topographic shadow distortion is caused by differences in illumination due to topographic variations. A radiometric topographic correction model can be used to locally correct the reflectance of these areas, thus obtaining topographic correction parameters. For heterogeneous areas, i.e., mixed pixel areas, spectral unmixing is performed. The pixels in these areas contain information about multiple land features, so the reflectance values ​​are a mixture of multiple land features. A linear spectral unmixing model is used to unmix the spectra of these areas, obtaining mixed pixel correction parameters to ensure that the inversion results accurately reflect the actual reflectance of each land feature. Through the above process, the topographic correction parameters and mixed pixel correction parameters are obtained, providing the necessary correction basis for subsequent inversion processes.

[0046] Y600: After inputting the original satellite radiance data, optimized atmospheric parameters, atmospheric transmittance, terrain correction parameters, and hybrid pixel correction parameters into the hierarchical inversion model, a high-precision surface reflectance product is output through iterative optimization inversion.

[0047] A hierarchical inversion model consisting of a global optimization layer and a local correction layer is constructed. The global optimization layer is based on the original satellite radiance data, optimized atmospheric parameters and atmospheric transmittance to invert the preliminary surface reflectance. The local correction layer combines terrain correction parameters and mixed pixel correction parameters to further adjust the inversion results and solve the bias problem in local areas, especially terrain shadow and mixed pixel areas.

[0048] Satellite radiance data, optimized atmospheric parameters, atmospheric transmittance, topographic correction parameters, and mixed pixel correction parameters are input into a hierarchical inversion model. Inversion is performed through an iterative optimization process, progressively adjusting surface reflectance until the optimal inversion result is obtained. During the iteration, the deviation between the inverted surface reflectance and the true ground value is calculated until this deviation meets a preset threshold. Through continuous adjustment and optimization, the final inverted surface reflectance is ensured to be as close as possible to the true ground reflectance. Ultimately, this optimization and correction process outputs high-precision surface reflectance data products. These data exhibit higher accuracy, particularly in the mineral diagnostic band, significantly improving the ability to extract subtle spectral features of ground objects.

[0049] Further, see Figure 2 After inputting the original satellite radiance data, optimized atmospheric parameters, atmospheric transmittance, terrain correction parameters, and hybrid pixel correction parameters into the hierarchical inversion model, a high-precision surface reflectance product is output through iterative optimization and inversion, including:

[0050] Y610: Construct the hierarchical inversion model comprising a global optimization layer and a local correction layer; Y620: In the global optimization layer, invert the initial surface reflectance based on the original satellite radiance data, optimized atmospheric parameters, and atmospheric transmittance; Y630: In the local correction layer, apply the terrain correction parameters to perform radiometric correction on the terrain shadow distortion area, and apply the hybrid pixel correction parameters to perform spectral unmixing correction on the hybrid pixel distortion area to obtain the corrected reflectance; Y640: Iteratively adjust the terrain correction parameters and the hybrid pixel correction parameters, and calculate the deviation between the corrected reflectance and the true value of the surface reflectance until the deviation meets a preset threshold, and output the high-precision surface reflectance product.

[0051] The global optimization layer processes raw satellite radiance data and optimizes atmospheric parameters and transmittance. The goal of global optimization is to retrieve the initial surface reflectance based on global atmospheric conditions and remote sensing data. This layer involves global adjustments to atmospheric parameters to address reflectance biases caused by varying atmospheric conditions. The local correction layer further refines the retrieval results, particularly for specific terrain and land cover categories. Local correction focuses on specific areas, such as terrain shadow areas and mixed pixel areas, improving retrieval accuracy through refined correction strategies. This layer includes terrain shadow correction and mixed pixel correction to address biases in these areas. The hierarchical retrieval model consists of both the global optimization layer and the local correction layer, ensuring a balance between large-scale and local accuracy.

[0052] In the global optimization layer, the initial surface reflectance is retrieved using raw satellite radiance data, optimized atmospheric parameters, and atmospheric transmittance. Specifically, a standard radiative transfer model, such as FLAASH, or a custom optimized model, is used to retrieve the initial surface reflectance based on the input data. The role of the global optimization layer is to improve the accuracy of the initial reflectance by optimizing atmospheric parameters and transmittance, minimizing errors caused by atmospheric influences. The initial surface reflectance generated in this process is a preliminary estimate at the global level, combining satellite observation data and optimized atmospheric parameters. However, errors may still exist due to factors such as terrain and surface feature complexity. Therefore, the retrieved initial reflectance needs further optimization and correction, especially in complex areas.

[0053] In the local correction layer, radiometric correction and spectral unmixing correction are performed on topographic shadow distortion areas and mixed pixel distortion areas. Specifically, in areas with complex terrain, especially where shadow effects exist, ground reflectance deviates due to the influence of illumination angle and terrain. These shadow areas are corrected using a radiometric topographic correction model with terrain correction parameters. Topographic shadow distortion areas are more pronounced in areas with steep slopes or valleys. The radiometric correction model can reduce the radiometric errors caused by terrain factors in these areas, restoring a more accurate surface reflectance. In some areas, such as areas combining vegetation, soil, rocks, and other land features (i.e., mixed pixels), the spectral signal exhibits complex characteristics due to the mixing of multiple land features, leading to reflectance deviations. To address this issue, spectral unmixing technology is used to correct these mixed areas using mixed pixel correction parameters. A linear spectral unmixing model is used in this process, decomposing the reflectance of each pixel into the spectral combination of its constituent land features to solve the spectral mixing problem. This process provides a more accurate estimate of the reflectance of each land feature. After completing the above two corrections, the corrected reflectance is obtained. This corrected reflectance is closer to the true surface reflectance after resolving the effects of terrain shadows and mixed pixels.

[0054] The corrected reflectance is compared with the true surface reflectance, and the deviation between the two is calculated. This deviation is quantified using indicators such as root mean square error or systematic bias. Based on the reflectance deviation, topographic correction parameters and blending pixel correction parameters are adjusted to reduce the deviation. The topographic correction parameters primarily adjust for errors caused by topographic shading effects, while the blending pixel correction parameters address spectral errors introduced by ground feature mixing. This adjustment process is iterative: parameters are adjusted based on the current deviation, then a new reflectance is calculated, the deviation is recalculated, and parameters are adjusted again until the deviation meets a preset threshold—that is, the deviation no longer changes significantly, indicating sufficient accuracy in the correction. Finally, an optimized and adjusted high-precision surface reflectance product is output. This reflectance data more accurately reflects the actual surface reflectance characteristics, especially in the mineral diagnostic band, such as 2000-2500 nm, where it can better extract the spectral characteristics of ground features.

[0055] Furthermore, Y110: The true values ​​of the atmospheric parameters are measured using a solar photometer, wherein the true values ​​of the atmospheric parameters include the measured true values ​​of aerosol optical thickness and the measured true values ​​of water vapor column concentration.

[0056] A sunphotometer is an instrument that estimates atmospheric composition by measuring changes in sunlight as it passes through the atmosphere. It provides highly accurate true values ​​for atmospheric parameters, including aerosol optical thickness and water vapor column concentration. Aerosol optical thickness describes the concentration of aerosols in the atmosphere, affecting light scattering and absorption. By measuring the attenuation of solar radiation in the atmosphere with a sunphotometer, the aerosol concentration can be estimated, yielding the true value of aerosol optical thickness. Water vapor column concentration represents the amount of water vapor in the atmosphere, affecting atmospheric absorption characteristics. A sunphotometer can measure the optical thickness of water vapor, and then calculate the true value of the measured water vapor column concentration. These measured true values ​​of atmospheric parameters serve as input to atmospheric correction models, particularly in optimizing atmospheric transmittance and adjusting reflectance, providing more realistic parameters to reduce errors in the atmospheric correction process.

[0057] Furthermore, based on the aforementioned systematic spectral bias, global optimization is performed to output optimized atmospheric parameters and atmospheric transmittance, including:

[0058] Y410: If the systematic spectral deviation indicates an atmospheric parameter deviation, then the original model parameters of the standard radiative transfer model are replaced with the true values ​​of the atmospheric parameters to obtain the optimized atmospheric parameters; Y420: Within the mineral diagnostic band interval, the optimized atmospheric parameters are used as input conditions, and the LBL algorithm is used to replace the fast approximation algorithm in the standard radiative transfer model to calculate the atmospheric transmittance.

[0059] If a systematic spectral bias indicates errors in atmospheric parameters, then these parameters need to be optimized. For example, if the bias is large in certain bands, it means that the atmospheric parameters in these bands do not accurately reflect the actual situation, leading to biased calculation results. In this case, replacing the original parameters in the standard radiative transfer model with the true atmospheric parameters, which are closer to actual atmospheric conditions, can improve the model's accuracy. The replaced atmospheric parameters will be used as optimized atmospheric parameters for subsequent radiative transfer model calculations, effectively eliminating atmospheric parameter errors and improving the accuracy of reflectance calculations.

[0060] Mineral diagnostic bands contain key mineral characteristic absorption bands, which are crucial for identifying ground features. In these bands, atmospheric absorption and scattering effects significantly impact radiative transfer, necessitating more accurate atmospheric transmittance calculations. The Line-by-Line (LBL) algorithm is a wavelength-by-wavelength method for calculating atmospheric transmittance. It is more accurate than traditional fast approximation algorithms because it incorporates detailed atmospheric spectral characteristics, particularly within narrow bands, accurately describing variations in spectral absorption and scattering. In this step, optimized atmospheric parameters are used as input to the LBL algorithm. Within the mineral diagnostic band, traditional radiative transfer models use simplified fast approximation algorithms to calculate atmospheric transmittance, but this method has limitations in accuracy, especially near strong absorption bands such as the water vapor and carbon dioxide bands. By using the LBL algorithm, atmospheric absorption and scattering transmittance can be calculated wavelength-by-wavelength, yielding more accurate atmospheric transmittance results and improving the accuracy of surface reflectance inversion. Atmospheric transmittance calculated using the LBL algorithm more realistically reflects the actual impact of the atmosphere on radiation.

[0061] Furthermore, the original model parameters of the standard radiative transfer model are replaced using the true values ​​of the atmospheric parameters to obtain the optimized atmospheric parameters, including:

[0062] Y411: If the systematic spectral deviation manifests as multiple consecutive bands shifting in the same direction within the mineral diagnostic band range, and the shift amplitude is greater than a preset shift threshold, then the systematic spectral deviation is determined to indicate an atmospheric parameter deviation; Y412: Based on the spectral distribution characteristics of the systematic spectral deviation, analyze and locate the type of deviation parameter; Y413: If the deviation parameter type is aerosol optical thickness, then replace the default aerosol optical thickness parameter in the standard radiative transfer model with the measured true value of aerosol optical thickness, encapsulate the model parameters, and use them as the optimized atmospheric parameter output; Y414: If the deviation parameter type is water vapor column concentration, replace the default water vapor parameter in the standard radiative transfer model with the measured true value of water vapor column concentration, encapsulate the model parameters, and use them as the optimized atmospheric parameter output.

[0063] Within the mineral diagnostic band range, if a systematic spectral bias shows a continuous unidirectional shift across multiple bands, and the magnitude of these shifts exceeds a preset shift threshold, then the bias can be determined to be related to errors in atmospheric parameters. In this case, the next steps will focus on analyzing and optimizing these atmospheric parameters to improve the accuracy of the correction results.

[0064] Among them, the same-direction offset means that the reflectance deviation in multiple consecutive bands is in the same direction, that is, it is either too high or too low. This phenomenon is usually caused by errors in atmospheric parameters. For example, factors such as aerosol concentration and gas absorption can have similar effects on multiple bands, resulting in the same direction of deviation. The preset offset threshold is a standard value determined based on experience and actual data. It is used to judge whether the deviation is significant. If the offset exceeds this threshold, it means that the deviation is not caused by random error, but is related to errors in atmospheric parameters.

[0065] Based on the spectral distribution characteristics of systematic spectral biases, the types of bias parameters are analyzed to determine which specific atmospheric parameter caused these biases. For example, if the bias mainly occurs in the 2000-2500 nm mineral diagnostic band and its characteristics match the absorption characteristics of water vapor or carbon dioxide in the atmosphere, it may be caused by errors in water vapor column concentration or carbon dioxide concentration. If the bias occurs in bands such as 1000-1400 nm, it may be caused by errors in aerosol optical thickness. The spectral distribution of the bias will exhibit specific patterns, such as abrupt changes in certain bands or stable changes within a wavelength range. These patterns help to identify the source of the bias and thus optimize the atmospheric correction process. By analyzing the spectral distribution characteristics, directions for optimizing atmospheric parameters are provided. For example, if the analysis results indicate an error in aerosol optical thickness, this parameter can be optimized; if the bias mainly comes from errors in water vapor concentration, the water vapor column concentration parameter can be optimized.

[0066] If the systematic spectral bias is primarily manifested in aerosol-related bands, such as the visible to near-infrared region, and the bias matches the variation pattern of aerosol optical thickness, then the bias is determined to originate from errors in aerosol optical thickness. In standard radiative transfer models, default aerosol optical thickness parameters are typically used. These default values ​​may not be entirely accurate, especially under different regional and climatic conditions. Replacing the default aerosol optical thickness parameters in the standard model with measured true aerosol optical thickness values ​​as optimized atmospheric parameters ensures that subsequent calculations use more accurate atmospheric parameters.

[0067] If a systematic spectral bias occurs in the characteristic bands of water vapor absorption, and the distribution pattern of the bias matches the spectral characteristics of water vapor, then the bias is determined to originate from an error in the water vapor column concentration. The standard radiative transfer model includes default water vapor parameters to simulate the effect of water vapor on radiation; however, due to significant variations in water vapor concentration across different regions and weather conditions, the default values ​​may be inaccurate. Replacing the default water vapor concentration parameters in the standard model with the measured true values ​​of the water vapor column concentration as the optimized atmospheric parameter output ensures that the water vapor concentration data used in the model matches the actual situation, thereby improving correction accuracy.

[0068] Furthermore, within the mineral diagnostic band range, using the optimized atmospheric parameters as input conditions, the LBL algorithm is employed to replace the fast approximation algorithm in the standard radiative transfer model to calculate the atmospheric transmittance, including:

[0069] Y421: In the standard radiative transfer model, disable the fast approximation algorithm module for the mineral diagnostic band interval and activate the bound LBL algorithm module; Y422: Within the mineral diagnostic band interval, input real-time atmospheric profile data and the optimized atmospheric parameters into the LBL algorithm module, calculate atmospheric absorption and scattering transmittance wavelength by wavelength, and output the atmospheric transmittance.

[0070] In standard radiative transfer models, fast approximation algorithms, such as simplified calculation methods in MODTRAN or 6S models, are used to quickly estimate atmospheric transmittance. These approximation algorithms are effective when the atmosphere is relatively simple and the computational accuracy requirements are not high. However, for complex mineral diagnostic bands, such as the 2000-2500 nm range, these algorithms lack sufficient accuracy and can lead to biases. Higher precision calculations are needed within the mineral diagnostic band; therefore, the fast approximation algorithm module should be disabled to avoid using simplified calculation methods. The LBL (Line-by-Line) algorithm is an accurate method for calculating atmospheric transmittance. It calculates the absorption and scattering processes of light in the atmosphere wavelength by wavelength, thus providing a more accurate transmittance estimate. In this step, the LBL algorithm module is activated to replace the fast approximation algorithm. The LBL algorithm can handle the absorption characteristics of different gases and aerosols in the atmosphere, providing more accurate results for transmittance calculations in the mineral diagnostic band. By enabling the LBL algorithm module, the standard radiative transfer model can perform accurate atmospheric transmittance calculations within the mineral diagnostic band, especially near strong absorption bands, ensuring high accuracy of the inversion results.

[0071] Real-time atmospheric profile data describes the composition and distribution of gases (such as temperature, humidity, and aerosol concentration) at different altitudes in the atmosphere. This data is crucial for accurately calculating atmospheric transmittance because the distribution of atmospheric components directly affects light scattering and absorption. Real-time atmospheric profile data is input into the LBL algorithm module to ensure that atmospheric transmittance calculations are based on the latest changes in atmospheric conditions. Simultaneously, optimized atmospheric parameters are input into the LBL algorithm. By replacing default parameters with measured atmospheric parameters, the LBL algorithm can accurately calculate transmittance within the mineral diagnostic band.

[0072] The LBL algorithm simulates the propagation of light in the atmosphere through wavelength-by-wavelength calculations, including phenomena such as light absorption and scattering. Within the mineral diagnostic band, the LBL algorithm can accurately handle subtle absorption characteristics, such as the strong absorption bands of water vapor or carbon dioxide, and provides transmittance calculations for each wavelength. After calculation, the LBL algorithm outputs atmospheric transmittance, ensuring that the inversion results more closely approximate the spectral characteristics of real ground features.

[0073] Furthermore, the spatial deviation between the initial surface reflectance and the true surface reflectance is locally corrected to obtain terrain correction parameters and hybrid pixel correction parameters, including:

[0074] Y510: Calculate the relative deviation between the initial surface reflectance and the true surface reflectance value pixel by pixel to generate a spatial deviation distribution map; Y520: Overlay the experimental area DEM data onto the spatial deviation distribution map to identify high-deviation pixels in the shadow area and mark the spatial distribution of terrain shadow distortion areas; Y530: Overlay the experimental area NDVI map onto the spatial deviation distribution map to identify high-deviation pixels in heterogeneous areas to mark the spatial distribution of mixed pixel distortion areas; Y540: Introduce a radiation topography correction model to perform local spatial correction on the spatial distribution of the terrain shadow distortion areas to obtain the topography correction parameters; Y550: Introduce a linear spectral unmixing model to perform local spatial correction on the spatial distribution of the mixed pixel distortion areas to obtain the mixed pixel correction parameters.

[0075] The initial surface reflectance is obtained through preliminary calculations using a radiative transfer model, while the true surface reflectance is the actual reflectance obtained from ground observations. For each pixel, the relative deviation between the initial surface reflectance and the true surface reflectance is calculated. The result is the reflectance deviation value for each pixel, which can be expressed as a percentage or other error index. The deviation value for each pixel is visualized to generate a spatial deviation distribution map. This map shows the deviation distribution in different regions, serving as the basis for subsequent corrections, particularly for identifying high-deviation areas requiring special processing.

[0076] DEM (Digital Elevation Model) data provides topographic height information for each pixel within the experimental area. By overlaying the spatial deviation distribution map and DEM data, topographic information can be combined with reflectance deviation information to identify illumination and shadow effects caused by topographic changes (such as valleys, slopes, etc.). In high-deviation areas, especially in topographically shadowed areas such as ridges and valleys where topographic changes are significant, reflectance may deviate considerably due to the influence of topography on illumination. In these areas, the reflectance values ​​of pixels are affected by shadows, occlusion, or other changes in illumination, resulting in larger deviations. By analyzing the deviations in these areas, high-deviation pixels affected by topography can be identified. The spatial distribution of high-deviation pixels is marked in the topographic shadow distortion area and labeled on the spatial deviation distribution map. This clearly shows which areas have significant deviations due to topographic shadow effects.

[0077] NDVI (Normalized Difference Vegetation Index) maps are calculated from the red and near-infrared bands of remote sensing imagery. They reflect the degree of vegetation cover on the land surface. Higher NDVI values ​​generally indicate abundant vegetation, while lower NDVI values ​​represent non-vegetated areas or bare soil. By overlaying the NDVI map with a spatial bias distribution map, areas with significant reflectance deviations can be identified, especially areas where vegetation and non-vegetation are mixed. Heterogeneous regions, or mixed-memory regions, refer to areas containing multiple different land cover components within a single pixel, such as grassland, soil, and rock. Due to the mixing of these various land cover components, the reflectance of the remote sensing signal is affected by these different land cover components, often resulting in significant deviations in reflectance calculations. Identifying these high-bias areas on the NDVI map allows for the precise location of mixed-memory distortion areas. The spatial bias distribution map marks the spatial distribution of mixed-memory distortion areas. These areas require spectral unmixing correction to correct the reflectance deviations caused by mixed pixels.

[0078] The radiation topographic correction model is a model specifically designed to correct for topographic shading effects. It adjusts pixel reflectance values ​​by incorporating the influence of topography to compensate for deviations caused by changes in illumination angle or topographic shading. This model utilizes topographic information from DEM data, such as height, slope, and aspect, as well as the angle of sunlight, to calculate the radiation correction factor for each pixel, performing local spatial correction on topographically shaded areas. After processing by the radiation topographic correction model, topographic correction parameters are obtained, which are used to correct the reflectance values ​​of distorted topographically shaded areas.

[0079] Linear spectral unmixing is a method for decomposing the reflectance of individual land cover components in a mixed pixel. This model decomposes the mixed reflectance of each pixel into combinations of reflectance from different land cover components. By assigning a set of land cover endmembers to each pixel and calculating the relative proportion of each land cover component based on the reflectance data of each endmember, the linear spectral unmixing model estimates the independent reflectance of each land cover component. After processing by the linear spectral unmixing model, mixed pixel correction parameters are obtained, characterizing the proportion and features of different land cover components in each pixel. These mixed pixel correction parameters are used for subsequent reflectance correction, ensuring that in pixels containing multiple land cover components, the reflectance accurately represents the characteristics of each land cover, rather than merely reflecting the mixing effect.

[0080] Furthermore, it also includes:

[0081] Y120: Deploy ground target objects and a standard reference plate within the experimental area; Y130: Simultaneously measure the surface radiance of the ground target objects and the standard radiance of the standard reference plate using a ground spectrometer; Y140: Calculate the true value of the surface reflectance based on the surface radiance and the standard radiance.

[0082] Ground feature targets refer to different types of ground features selected within the experimental area, such as rocks, vegetation, and soil, and observation points set up on these features. The selection of targets is based on the research objectives, choosing representative ground feature types to ensure that the reflectance of different ground feature types can be accurately measured. The placement of ground feature targets needs to ensure the accuracy of their location and that the targets are within the image range of the satellite passing overhead when remote sensing images are captured.

[0083] A standard reference plate is a device or material used to provide a known reflectance value. It is typically an object with high reflectance (such as a white board) or known reflective properties. The purpose of the reference plate is to provide a known standard reflectance, facilitating the calibration of measurement results. The standard reference plate should be placed at a known location within the experimental area and observed synchronously with the ground-based spectrometer measurements.

[0084] Ground-based spectrometers are used to measure the radiance of ground features or reference surfaces. Radiance refers to the radiant energy emitted per unit area within a unit solid angle. By measuring radiance, reflectivity can be further calculated. Ground-based spectrometers cover different wavelengths, such as visible light and near-infrared, and provide radiance data for each wavelength band.

[0085] The ground-based spectrometer simultaneously measures the surface radiance of different ground objects at various target locations, recording the radiance of each target. The radiance of different targets reflects the reflectivity of different ground objects. Simultaneously, the ground-based spectrometer also measures the radiance of a standard reference plate. As a known standard of reflectivity, the radiance measurement of the reference plate is used to calibrate the measurement results of the ground objects, ensuring the accuracy of the measurements.

[0086] Reflectivity refers to the ratio of the intensity of light reflected from a ground surface to the total radiant energy received by that surface. Based on measured surface radiance and standard radiance, the true value of surface reflectivity is calculated using the following formula: True surface reflectivity = (Surface radiance / Standard radiance) * Standard reflectivity, where the standard reflectivity is known and represents the reflectivity of a reference board, such as a white board with a reflectivity of over 90%. The calculated true surface reflectivity represents the actual reflectivity of a ground object in a specific wavelength band, providing accurate reference data for atmospheric correction processes.

[0087] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

[0088] The foregoing description of specific exemplary embodiments of the invention is for illustrative and explanatory purposes. These descriptions are not intended to limit the invention to the precise forms disclosed, and it will be apparent that many changes and variations can be made in accordance with the foregoing teachings. The exemplary embodiments were chosen and described in order to explain the specific principles of the invention and its practical application, thereby enabling those skilled in the art to implement and utilize various different exemplary embodiments of the invention, as well as various different choices and variations. The scope of the invention is intended to be defined by the claims and their equivalents.

Claims

1. A method for atmospheric correction and optimization of hyperspectral images based on ground authenticity verification data, characterized in that, The method includes: During the satellite's transit test area, the raw radiance data of the satellite was acquired, and ground truth data acquisition was performed simultaneously to obtain the true values ​​of surface reflectance and atmospheric parameters. The original radiance data of the satellite were initially atmospherically corrected using a standard radiative transfer model to generate the initial surface reflectance. Within a preset mineral diagnostic band range, the true value of the surface reflectance and the initial surface reflectance are compared band by band to quantify the spectral deviation in order to identify systematic spectral deviations. Global optimization is performed based on the aforementioned systematic spectral bias, outputting optimized atmospheric parameters and atmospheric transmittance; Local error correction is performed on the spatial deviation between the initial surface reflectance and the true surface reflectance to obtain terrain correction parameters and hybrid pixel correction parameters. After inputting the original satellite radiance data, optimized atmospheric parameters, atmospheric transmittance, terrain correction parameters, and hybrid pixel correction parameters into the hierarchical inversion model, a high-precision surface reflectance product is output through iterative optimization inversion.

2. The hyperspectral image atmospheric correction and optimization method based on ground authenticity verification data as described in claim 1, characterized in that, After inputting the original satellite radiance data, optimized atmospheric parameters, atmospheric transmittance, terrain correction parameters, and hybrid pixel correction parameters into the hierarchical inversion model, a high-precision surface reflectance product is output through iterative optimization and inversion, including: Construct the hierarchical inversion model that includes a global optimization layer and a local correction layer; In the global optimization layer, the initial surface reflectance is retrieved based on the original satellite radiance data, optimized atmospheric parameters, and atmospheric transmittance. In the local correction layer, the terrain correction parameters are applied to perform radiometric correction on the terrain shadow distortion area, and the mixed pixel correction parameters are applied to perform spectral unmixing correction on the mixed pixel distortion area to obtain the corrected reflectance; The terrain correction parameters and hybrid pixel correction parameters are iteratively adjusted, and the deviation between the corrected reflectance and the true value of the surface reflectance is calculated until the deviation meets the preset threshold. The high-precision surface reflectance product is then output.

3. The atmospheric correction and optimization method for hyperspectral images based on ground authenticity verification data as described in claim 1, characterized in that, The true values ​​of the atmospheric parameters are measured using a solar photometer, wherein the true values ​​of the atmospheric parameters include the measured true values ​​of aerosol optical thickness and the measured true values ​​of water vapor column concentration.

4. The atmospheric correction and optimization method for hyperspectral images based on ground authenticity verification data as described in claim 3, characterized in that, Global optimization is performed based on the aforementioned systematic spectral bias, outputting optimized atmospheric parameters and atmospheric transmittance, including: If the systematic spectral deviation indicates an atmospheric parameter deviation, then the original model parameters of the standard radiative transfer model are replaced with the true values ​​of the atmospheric parameters to obtain the optimized atmospheric parameters; Within the mineral diagnostic band range, the optimized atmospheric parameters are used as input conditions, and the LBL algorithm is used to replace the fast approximation algorithm in the standard radiative transfer model to calculate the atmospheric transmittance.

5. The atmospheric correction and optimization method for hyperspectral images based on ground authenticity verification data as described in claim 4, characterized in that, The optimized atmospheric parameters are obtained by replacing the original model parameters of the standard radiative transfer model with the true values ​​of the atmospheric parameters, including: If the systematic spectral deviation is manifested as multiple consecutive bands shifting in the same direction within the mineral diagnostic band range, and the shift amplitude is greater than a preset shift threshold, then the systematic spectral deviation is determined to indicate an atmospheric parameter deviation. Based on the spectral distribution characteristics of the systematic spectral deviation, the types of positioning deviation parameters are analyzed; If the deviation parameter type is aerosol optical thickness, then the default aerosol optical thickness parameter in the standard radiative transfer model is replaced with the measured true value of aerosol optical thickness, and the model parameters are encapsulated as the output of the optimized atmospheric parameters. If the deviation parameter type is water vapor column concentration, the default water vapor parameter in the standard radiative transfer model is replaced with the true value of the measured water vapor column concentration, and then the model parameters are encapsulated and used as the output of the optimized atmospheric parameters.

6. The atmospheric correction and optimization method for hyperspectral images based on ground authenticity verification data as described in claim 4, characterized in that, Within the mineral diagnostic band range, using the optimized atmospheric parameters as input conditions, the LBL algorithm is employed to replace the fast approximation algorithm in the standard radiative transfer model to calculate the atmospheric transmittance, including: In the standard radiative transfer model, the fast approximation algorithm module for the mineral diagnostic band interval is turned off, and the bound LBL algorithm module is activated. Within the mineral diagnostic band range, real-time atmospheric profile data and the optimized atmospheric parameters are input into the LBL algorithm module to calculate atmospheric absorption and scattering transmittance wavelength by wavelength and output the atmospheric transmittance.

7. The atmospheric correction and optimization method for hyperspectral images based on ground authenticity verification data as described in claim 1, characterized in that, Local error correction is performed on the spatial deviation between the initial surface reflectance and the true surface reflectance value to obtain terrain correction parameters and hybrid pixel correction parameters, including: The relative deviation between the initial surface reflectance and the true value of surface reflectance is calculated pixel by pixel to generate a spatial deviation distribution map. By overlaying the experimental area DEM data onto the spatial deviation distribution map, high-deviation pixels in the shadowed areas are identified, and the spatial distribution of terrain shadow distortion areas is marked. By overlaying the experimental area NDVI map onto the spatial deviation distribution map, high-deviation pixels in heterogeneous regions are identified to mark the spatial distribution of mixed pixel distortion areas. A radiation topography correction model is introduced to perform local spatial correction on the spatial distribution of the topographic shadow distortion area, and the topography correction parameters are obtained. A linear spectral unmixing model is introduced to perform local spatial correction on the spatial distribution of the distorted region of the mixed pixel, thereby obtaining the correction parameters of the mixed pixel.

8. The atmospheric correction and optimization method for hyperspectral images based on ground authenticity verification data as described in claim 1, characterized in that, Also includes: Ground targets and standard reference boards were deployed within the experimental area; The surface radiance of the ground target and the standard radiance of the standard reference plate were simultaneously measured using a ground-based spectrometer. The true value of the surface reflectance is calculated based on the surface radiance and the standard radiance.

Images