Hyperspectral data quality evaluation method for water color remote sensing parameter inversion

Abstract

The invention provides a water color remote sensing parameter inversion-oriented hyperspectral data quality evaluation method, which relates to the technical field of optical remote sensing and comprises the following steps of: dividing a plurality of hydrological response unit sub-graphs from a hyperspectral remote sensing image of a target area by taking a preset multi-scale biological optical fingerprint wave band as a reference; performing water color noise transmission analysis on the plurality of hydrological response unit sub-graphs, and outputting a plurality of groups of water color inversion signal-to-noise ratio fingerprints; performing atmospheric correction distortion field diagnosis on the hyperspectral remote sensing image, and outputting a near-infrared negative distortion field and a spectral fidelity error field; dynamically fusing to generate a real-time quality entropy field; and confidence domain convergence mapping of the real-time quality entropy field is carried out, and a water color parameter inversion uncertainty thermodynamic diagram is output. The technical problems that in the prior art, a targeted diagnosis mechanism for distortion in the atmospheric correction process is lacked, systematic interference of atmospheric correction on water color parameter inversion precision cannot be captured, and consequently a final water quality monitoring result is unreliable and errors are large are solved.

Description

Technical Field

[0001] This invention relates to the field of optical remote sensing technology, and in particular to a method for assessing the quality of hyperspectral data for water color remote sensing parameter inversion. Background Technology

[0002] Existing methods for assessing the quality of hyperspectral data have significant limitations. The core problem is that the assessment framework is strongly coupled with the terrestrial scene and cannot adapt to the special needs of water color parameter inversion.

[0003] Specifically, on the one hand, existing methods only focus on the apparent quality of the raw data, completely ignoring the reliability of the weak spectral information remaining after atmospheric correction of water signals; on the other hand, existing technologies lack a targeted diagnostic mechanism for distortions in the atmospheric correction process. They neither utilize the physical characteristics of clean water in the near-infrared band to detect non-physical negative value anomalies after correction, nor assess the degree of distortion of the true water signal through spectral shape consistency analysis, resulting in the inability to capture the systematic interference of atmospheric correction on the accuracy of water color parameter inversion.

[0004] Furthermore, the existing assessment results are completely disconnected from the reliability of the final inversion products. They are neither linked to the specific band combinations on which the water color algorithm depends, nor are predictive quality indicators constructed to quantify the uncertainty of the inversion. This makes it difficult for data users to predict the actual credibility of water quality products, ultimately causing decision-making risks in the application of water color remote sensing 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 data quality assessment method for water color remote sensing parameter inversion. This method solves the technical problems of existing technologies lacking a targeted diagnostic mechanism for atmospheric correction process distortion, failing to capture the systematic interference of atmospheric correction on the accuracy of water color parameter inversion, and resulting in unreliable and large errors in the final water quality monitoring results.

[0007] The specific technical solution is as follows:

[0008] A method for assessing the quality of hyperspectral data for water color remote sensing parameter inversion is provided. The method includes: dividing multiple hydrological response unit sub-maps from the hyperspectral remote sensing image of the target area based on a pre-set multi-scale bio-optical fingerprint band, wherein the hyperspectral remote sensing image is an atmospheric path-corrected hyperspectral image; performing water color noise propagation analysis on the multiple hydrological response unit sub-maps to output multiple sets of water color inversion signal-to-noise ratio fingerprints; performing atmospheric correction distortion field diagnosis on the hyperspectral remote sensing image to output a near-infrared negative distortion field and a spectral fidelity error field; dynamically fusing the near-infrared negative distortion field, the spectral fidelity error field, and the multiple sets of water color inversion signal-to-noise ratio fingerprints to generate a real-time quality entropy field; and using the environment-driven ground-state uncertainty field as a reference manifold to perform confidence region convergence mapping of the real-time quality entropy field to output a heatmap of water color parameter inversion uncertainty.

[0009] In one implementation, the environment-driven ground-state uncertainty field is used as a reference manifold to perform a confidence region convergence mapping of the real-time mass entropy field, outputting a water color parameter inversion uncertainty heatmap, including:

[0010] By performing phase tracking of environmental driving factors in the target region, a ground-state uncertainty field is constructed; using the ground-state uncertainty field as a reference manifold, the confidence region convergence mapping of the real-time mass entropy field is performed, and a calibrated mass confidence field is output; according to a preset uncertainty thermodynamic topology rule, the calibrated mass confidence field is mapped into a water color parameter inversion uncertainty thermogram.

[0011] In one implementation, a ground-state uncertainty field is constructed by performing phase tracking of environmental driving factors in the target region, including:

[0012] Multiple temporal hyperspectral images and multiple environmental driving factor data of the target region are loaded, wherein the environmental driving factor data includes solar elevation angle, aerosol optical thickness, and water surface flare index; the target region is divided into a grid cell array using a preset spatial resolution; the multiple temporal hyperspectral images and multiple environmental driving factor data are segmented using the grid cell array to obtain multiple sets of temporal hyperspectral sub-images and multiple sets of environmental driving factor values ​​corresponding to multiple regional grid cells in the grid cell array; multiple sets of temporal mass entropy fields are generated based on the multiple sets of temporal hyperspectral sub-images; according to the multi-dimensional binning attributes of the multiple sets of environmental driving factor values, the multiple sets of temporal mass entropy fields are binned and aggregated to construct the ground state uncertainty field.

[0013] In one implementation, based on the multi-dimensional binning attributes of the multiple sets of environmental driving factor values, the multiple sets of time-series quality entropy fields are binned and aggregated to construct the ground-state uncertainty field, including:

[0014] The multiple sets of environmental driving factor values ​​are encoded using multi-dimensional environmental binning to obtain multiple sets of environmental binning combinations. Based on the mapping relationship between the multiple sets of environmental driving factor values ​​and the multiple sets of environmental binning combinations, the multiple sets of time-series mass entropy fields are statistically aggregated using multi-modal coupling to obtain multiple sets of mass ground state values. Based on the grid cell array, the spatial environmental coupling field of the multiple sets of mass ground state values ​​and the multiple sets of environmental binning combinations is reconstructed to obtain the ground state uncertainty field.

[0015] In one implementation, water color noise propagation analysis is performed on the plurality of hydrological response unit sub-maps to output multiple sets of water color inversion signal-to-noise ratio fingerprints, including:

[0016] Noise propagation entropy analysis is performed on the multiple hydrological response unit sub-graphs to output multiple intrinsic noise reflectivity vectors and multiple local signal-to-noise ratio tensors; twin-generation noise coupling analysis is performed on the multiple intrinsic noise reflectivity vectors and multiple local signal-to-noise ratio tensors to output multiple sets of water color inversion signal-to-noise ratio fingerprints.

[0017] In one implementation, a dual-noise coupling analysis is performed on the plurality of intrinsic noise reflectivity vectors and the plurality of local signal-to-noise ratio tensors to output multiple sets of water color inversion signal-to-noise ratio fingerprints, including:

[0018] Based on the multiple water color inversion targets corresponding to the multi-scale bio-optical fingerprint bands, an inversion noise propagation coupling matrix is ​​constructed. The multiple intrinsic noise reflectivity vectors and multiple local signal-to-noise ratio tensors are respectively loaded into the inversion noise propagation coupling matrix to obtain a water color parameter-sensitive ground state matrix and a noise distortion propagation matrix. Ground state response features are decoupled based on the water color parameter-sensitive ground state matrix to obtain multiple parameter sensitivity feature vectors. Distortion spectrum extraction is performed based on the noise distortion propagation matrix to obtain multiple noise distortion feature spectra. The multiple parameter sensitivity feature vectors and multiple noise distortion feature spectra constitute the multiple sets of water color inversion signal-to-noise ratio fingerprints.

[0019] In one implementation, the near-infrared negative distortion field, the spectral fidelity error field, and multiple sets of water color inversion signal-to-noise ratio fingerprints are dynamically fused to generate a real-time quality entropy field, including:

[0020] Dimensionless standardization is performed on the near-infrared negative distortion field and the spectral fidelity error field to generate a physical distortion field. Field-based recombination is performed on the multiple sets of water color inversion signal-to-noise ratio fingerprints to obtain parameter sensitivity feature matrices corresponding to the multiple parameter sensitivity feature vectors and noise distortion spectrum tensors corresponding to the multiple noise distortion feature spectra. Fusion weights are configured according to real-time environmental driving factors, and the contribution ratios of the physical distortion field, feature matrix, and distortion spectrum tensor are adjusted. Then, a nonlinear multi-field coupling function is invoked to perform bio-optical weighted geometric fusion, outputting a preliminary quality index field. The preliminary quality index field is mapped to the spatially continuously distributed real-time quality entropy field through information entropy conversion.

[0021] In one implementation, using the ground-state uncertainty field as a reference manifold, a confidence region convergence mapping of the real-time mass entropy field is performed to output a calibrated mass confidence field, including:

[0022] The real-time environment binning code is retrieved based on the real-time environment driving factors; using the real-time environment binning code, a two-dimensional spatial reference manifold field is extracted from the ground-state uncertainty field; an absolute deviation field is calculated for the two-dimensional spatial reference manifold field and the real-time mass entropy field; based on the ground-state level interval of the two-dimensional spatial reference manifold field, a confidence threshold field for spatial variation is dynamically set; based on the spatial position comparison results of the confidence threshold field and the absolute deviation field, a manifold convergence mapping operation is performed to generate a preliminary calibration field; a terrain-constrained spatial regularization operation is performed on the preliminary calibration field to output a calibration quality confidence field.

[0023] In one embodiment, atmospheric correction distortion field diagnosis is performed on the hyperspectral remote sensing image to output a near-infrared negative distortion field and a spectral fidelity error field, including:

[0024] Based on the near-infrared reflectance, clean water pixels are selected from the hyperspectral remote sensing image to generate a clean water mask. After extracting the absolute value of the negative reflectance in the near-infrared band of the clean water mask, the negative near-infrared distortion field is generated by spatial interpolation. The measured visible light spectrum of the clean water mask is extracted. The spectral fidelity error field is generated by calculating the spectral angular distance between the measured visible light spectrum and a preset standard clean water spectral library.

[0025] Beneficial effects of the embodiments of the present invention:

[0026] By using pre-defined multi-scale bio-optical fingerprint bands as a benchmark, multiple hydrological response unit sub-maps are divided from the hyperspectral remote sensing image of the target area. This ensures that the signal of each water body response unit is accurately identified in the hyperspectral data analysis, laying the foundation for subsequent noise analysis and signal-to-noise ratio (SNR) calculation. Water color noise propagation analysis is performed on multiple hydrological response unit sub-maps, outputting multiple sets of water color inversion SNR fingerprints. This quantifies the signal-to-noise ratio in different water body regions. By accurately evaluating the water color inversion SNR fingerprints, potential risks caused by noise can be better captured, thus achieving more precise data quality monitoring during the inversion process. Atmospheric correction distortion field diagnosis is performed on the hyperspectral remote sensing image, outputting near-infrared negative distortion field and spectral fidelity error field. By revealing these non-physical errors, direct quantitative indicators can be provided for subsequent data quality control and optimization, preventing these errors from affecting the reliability of water color inversion. By dynamically fusing the near-infrared negative distortion field, the spectral fidelity error field, and multiple sets of water color inversion signal-to-noise ratio fingerprints, a real-time quality entropy field is generated. This field can intuitively display the quality changes in different regions. The real-time quality entropy field can serve as a dynamic monitoring tool for data quality control, assessing the current quality status of remote sensing data in real time and providing effective prediction and guidance for future inversion accuracy. By using the environment-driven ground-state uncertainty field as the benchmark manifold, the confidence region convergence mapping of the real-time quality entropy field is performed, ultimately outputting a heatmap of water color parameter inversion uncertainty. This directly links data quality with the credibility of the inversion results, showcasing the inversion uncertainty of water color remote sensing data through a visual heatmap. This helps users intuitively judge the accuracy and reliability of water color parameter inversion in different water areas, providing an operable prior reliability assessment tool for water quality remote sensing applications and significantly reducing decision-making risks caused by data quality issues.

[0027] 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

[0028] 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.

[0029] Figure 1 This invention illustrates a flowchart of a hyperspectral data quality assessment method for water color remote sensing parameter inversion provided by the present invention.

[0030] Figure 2The diagram illustrates the process of outputting a heatmap of uncertainty in water color parameter inversion in the hyperspectral data quality assessment method for water color remote sensing parameter inversion provided by the present invention. Detailed Implementation

[0031] 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.

[0032] 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.

[0033] 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.

[0034] 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.

[0035] The hyperspectral data quality assessment method for water color remote sensing parameter inversion provided by this invention addresses the technical problem that existing technologies lack a targeted diagnostic mechanism for atmospheric correction process distortion, are unable to capture the systematic interference of atmospheric correction on the accuracy of water color parameter inversion, and thus lead to unreliable and large errors in the final water quality monitoring results.

[0036] See Figure 1 The flowchart of the hyperspectral data quality assessment method for water color remote sensing parameter inversion provided in this embodiment of the invention includes:

[0037] Y100: Based on the preset multi-scale bio-optical fingerprint band, multiple hydrological response unit sub-maps are divided from the hyperspectral remote sensing image of the target area, wherein the hyperspectral remote sensing image is a hyperspectral image after atmospheric path correction.

[0038] Based on the bio-optical characteristics of water bodies, such as the reflectance spectral features of chlorophyll a, suspended matter, and colored soluble organic matter (CDOM), a series of key bands are selected and defined to form a multi-scale bio-optical fingerprint band. For example, the core bands for chlorophyll a include 440nm, 550nm, 670nm, and 700nm; the inversion bands for suspended matter are 550nm to 650nm; and the inversion bands for CDOM are 350nm to 450nm. By spatially dividing the hyperspectral remote sensing image, the target area is divided into multiple regions with similar spectral characteristics, forming multiple hydrological response unit sub-maps. Each unit sub-map represents a spatial range, indicating the water body characteristics and response of that region. The hyperspectral remote sensing image is an atmospheric path-corrected hyperspectral image, meaning that all remote sensing data has been freed from atmospheric scattering and absorption effects.

[0039] Y200: Perform water color noise propagation analysis on the multiple hydrological response unit sub-maps and output multiple sets of water color inversion signal-to-noise ratio fingerprints.

[0040] The equivalent noise reflectance of each hydrological response unit submap is calculated, i.e., the noise level in each band of each submap. This assesses how this noise is transmitted and potentially amplified through band ratios. For band ratio calculations inverting chlorophyll a or suspended matter, noise may be significantly amplified through these ratios, thus affecting the reliability of the final inversion. Based on the noise transmission analysis, the local signal-to-noise ratio (SNR) is calculated, and the water color inversion SNR fingerprint for each hydrological response unit submap is output. This provides the SNR characteristics for each region, reflecting the reliability of the water color inversion for that region. The SNR fingerprint refers to the SNR characteristics of each submap in each band, which directly affects the accuracy of water color parameter inversion.

[0041] Y300: Performs atmospheric correction distortion field diagnosis on the hyperspectral remote sensing image, and outputs near-infrared negative distortion field and spectral fidelity error field.

[0042] Clean water bodies, such as deep lakes and unpolluted water bodies, should have near-zero reflectance in the near-infrared band due to water's absorption characteristics in this region. Therefore, clean water body pixels are first selected from hyperspectral images. These pixels should have near-zero reflectance in the 780nm to 900nm band. If atmospheric correction is flawed, non-physical negative reflectance values ​​will appear in these areas. By detecting negative reflectance values ​​in the near-infrared band, the effectiveness of atmospheric correction can be evaluated. Too many or too large negative values ​​indicate serious problems with atmospheric correction, leading to a decline in data quality.

[0043] For pixels in clean water bodies, the measured spectra in the visible light band (400-600nm) are extracted and then compared with a standard clean water body spectral library. Methods such as spectral angle or root mean square error are used to quantify the degree of distortion in the corrected spectrum. If the corrected spectrum differs significantly from the standard spectrum, it indicates distortion during atmospheric correction. By calculating the difference between the measured and standard spectra, a spectral fidelity error field is generated. This field quantifies the degree of distortion of the true spectral signal of the water body during atmospheric correction and serves as an indicator for subsequent quality assessment.

[0044] Y400: Dynamically fuses the near-infrared negative distortion field, spectral fidelity error field, and multiple sets of water color inversion signal-to-noise ratio fingerprints to generate a real-time quality entropy field.

[0045] These three different quality indicators each represent different aspects of quality, and their effects are interrelated. For example, regions with high signal-to-noise ratios (SNR) may also perform well in spectral correction, while negative distortion fields may affect spectral fidelity and SNR. Therefore, dynamically fusing these three indicators and combining their respective weights yields a comprehensive real-time quality entropy field, used to quantify information uncertainty. In remote sensing image quality analysis, a higher entropy value indicates higher inversion uncertainty in that region; conversely, a lower entropy value indicates higher inversion quality.

[0046] Y500: Using the environment-driven ground-state uncertainty field as the reference manifold, perform confidence domain convergence mapping of the real-time mass entropy field, and output a thermogram of uncertainty inversion of water color parameters.

[0047] The ground-state uncertainty field is a field constructed based on the impact of environmental factors on the quality of remote sensing data. Environmental factors, such as climate conditions, water temperature, weather changes, and illumination conditions, have a significant impact on the acquisition and inversion of remote sensing data, especially the uncertainty in water color inversion, which manifests as spatial and temporal inhomogeneities. For example, in hyperspectral imagery, sunny or cloudy weather leads to different illumination variations, thus affecting the image's reflectance characteristics. These environmental factors constitute the ground-state uncertainty field, which serves as a baseline manifold for quality control and is used to quantify the impact of these environmental changes on the uncertainty of water color inversion.

[0048] Based on the ground-state uncertainty field, a confidence region convergence mapping is performed to combine the real-time mass entropy field with the ground-state uncertainty field, quantifying the uncertainty and determining the confidence interval for each hydrological response unit. The confidence region is calculated using statistical methods, such as Monte Carlo simulation and Bayesian methods, to determine the range of uncertainty in the inversion results for each hydrological response unit under different environmental conditions. The convergence mapping uses the convergence of the confidence region to determine which regions have relatively certain water color parameter inversion results and which regions have results with large errors or uncertainties.

[0049] Finally, based on the results of the confidence region convergence mapping, a heat map of the uncertainty of water color parameter inversion is generated. This heat map displays the uncertainty of water color inversion in each region through color coding. The change of color value reflects the reliability of the inversion results in different regions. The darker the color, the more uncertain the inversion results in that region; conversely, the lighter the color, the more reliable the inversion results.

[0050] In one implementation, see Figure 2 Using the environment-driven ground-state uncertainty field as the reference manifold, the confidence region convergence mapping of the real-time mass entropy field is performed, and a water color parameter inversion uncertainty heatmap is output, including:

[0051] Y510: Construct a ground-state uncertainty field by performing phase tracking of environmental driving factors in the target region; Y520: Using the ground-state uncertainty field as a reference manifold, perform confidence domain convergence mapping of the real-time mass entropy field to output a calibrated mass confidence field; Y530: Map the calibrated mass confidence field into a water color parameter inversion uncertainty heatmap according to a preset uncertainty thermodynamic topology rule.

[0052] In remote sensing water color inversion, environmental driving factors refer to external factors affecting the spectral reflectance characteristics of water bodies. These factors include: meteorological factors, such as light intensity, cloud cover, air temperature, humidity, and wind speed; hydrological factors, such as water temperature, flow velocity, dissolved oxygen content, and plankton concentration; and geographical factors, such as land cover type and topography surrounding the water body. Temporal tracking of environmental driving factors for the target area involves monitoring and recording the changes of these environmental factors at different time points. This tracking is based on time-series data and can be obtained through historical data, real-time monitoring, or meteorological models to understand the patterns of environmental factor changes. Ground-state uncertainty fields can be modeled using statistical methods, such as time series analysis, to model the impact of environmental factors and their changes. This model, combined with remote sensing data, can then be used to estimate the uncertainty in water color inversion. For example, changes in air temperature affect the spectral reflectance characteristics of water bodies, causing some uncertainty; cloud cover can lead to the loss of spectral information acquired by sensors, further increasing uncertainty.

[0053] The ground-state uncertainty field is a benchmark model that considers the impact of environmental factors on the quality of water color inversion. When calibrating the quality reliability, this benchmark model is used as a reference, and the quality is corrected by fusing it with the real-time mass entropy field. The real-time mass entropy field reflects the initial uncertainty of the water color inversion results, while the ground-state uncertainty field reflects the long-term impact of environmental driving factors on water color inversion. The process of confidence region convergence mapping involves combining the ground-state uncertainty field and the real-time mass entropy field to calculate their confidence intervals. Through statistical analysis, such as Bayesian methods or Monte Carlo simulations, the confidence regions for each region are calculated to assess the reliability range of the water color inversion results for each hydrological response unit under different environmental conditions. Based on the confidence region mapping results, a calibration quality reliability field is generated. Each region in this field represents the reliability of the water color inversion results; regions with higher reliability are less affected by environmental factors, while those with lower reliability have greater uncertainty.

[0054] Uncertainty thermal topology rules are predefined rules that define how to convert different levels of quality confidence into colors on a heatmap; for example, green indicates high confidence and red indicates low confidence. Based on these rules, each region of the calibration quality confidence field is mapped to a corresponding color on the heatmap. This mapping generates an uncertainty heatmap for the water color parameter inversion for each hydrological response unit. This heatmap visually displays the uncertainty in water color inversion, helping researchers, policymakers, or environmental monitors identify which regions have more reliable water color inversion results and which require further analysis or processing.

[0055] In one implementation, a ground-state uncertainty field is constructed by performing phase tracking of environmental driving factors in the target region, including:

[0056] Y511: Load multiple temporal hyperspectral images and multiple environmental driving factor data of the target region, wherein the environmental driving factor data includes solar elevation angle, aerosol optical thickness, and water surface flare index; Y512: Divide the target region into a grid cell array using a preset spatial resolution; Y513: Segment the multiple temporal hyperspectral images and multiple environmental driving factor data using the grid cell array to obtain multiple sets of temporal hyperspectral sub-images and multiple sets of environmental driving factor values ​​corresponding to multiple regional grid cells in the grid cell array; Y514: Generate multiple sets of temporal mass entropy fields based on the multiple sets of temporal hyperspectral sub-images; Y515: Perform binning and aggregation of the multiple sets of temporal mass entropy fields according to the multi-dimensional binning attributes of the multiple sets of environmental driving factor values ​​to construct the ground state uncertainty field.

[0057] Temporal hyperspectral imagery, sourced from satellite remote sensing platforms, covers multiple time points within the target area, spanning periods such as several months to several years, to monitor temporal changes in water bodies. This rich spectral information allows for more accurate water color retrieval. Environmental driving factors refer to external factors influencing the spectral reflectance characteristics of water bodies. These factors require dynamic tracking and modeling during water color retrieval. Among these, the solar altitude angle (the angle between the sun and the horizon) directly affects the angle of solar radiation incidence, thus influencing the reflectance spectrum; aerosol optical thickness represents the degree of light scattering and absorption by aerosols in the atmosphere, affecting the spectral information received by the sensor; and the water surface flare index represents the intensity of water surface reflection, typically occurring when the reflection angle between the water surface and sunlight is close, leading to irregularities in the spectral signal and affecting the accuracy of water color retrieval.

[0058] Based on the required analytical precision and data characteristics, the spatial resolution of the raster cells is set, for example, choosing 30m × 30m or a finer resolution, depending on the image resolution and research objectives. According to the set spatial resolution, the target area is divided into small raster cells. These raster cells serve as analysis units, making subsequent data processing, analysis, and modeling more flexible. Each raster cell corresponds to a specific geographic area, ensuring that these small cells can represent different water body characteristics, environmental factors, and changes within the region. By dividing the data into raster cells, data from different regions can be processed independently, and spatial analysis of the impact of different environmental factors on the inversion results is also facilitated.

[0059] Multiple temporal hyperspectral images are spatially segmented according to a raster cell array. For each raster cell, a temporal hyperspectral sub-image of that region is extracted, i.e., the image data corresponding to each time point. This provides a temporal dataset for subsequent analysis, enabling the detection of the impact of environmental factors changing over time on the water color inversion results. The same operation is performed to spatially segment multiple environmental driving factor data. For each raster cell, the corresponding environmental driving factor values ​​are extracted. The temporal changes of these factors will affect the water color inversion results of each raster cell; therefore, accurate matching of the corresponding environmental data is essential.

[0060] Based on multiple sets of temporal hyperspectral sub-images of multiple regional raster units, quality analysis is performed, similar to the previous process. Methods such as water color noise transmission analysis, atmospheric correction distortion field diagnosis, and signal-to-noise ratio fingerprint extraction are used to calculate the quality entropy field using these temporal data. That is, through noise and distortion analysis and other techniques, the quality fluctuation and uncertainty of the image data at each moment are quantified. This will reflect the changing trend of each raster unit over time and its quality changes. This step is similar to the previous water color noise transmission analysis and atmospheric correction distortion field diagnosis. By analyzing the quality of each temporal image, the corresponding temporal quality entropy field is generated.

[0061] Based on different dimensions of multiple sets of environmental driving factor values, these factor values ​​are binned, that is, divided into different groups or intervals according to their range or category. These binning attributes reflect the impact of environmental factors on time-series data under different categories. For example, the accuracy of water color inversion may vary significantly under different solar altitude angles. By binning these factors, water color inversion errors under different environmental conditions can be captured more accurately. The time-series mass entropy field of each raster cell is binned and aggregated, that is, its data under different intervals of environmental factors are aggregated using weighted averaging or other statistical methods. The aggregation result reflects the overall trend of the mass entropy field under the influence of different environmental driving factors, helping to identify areas of high uncertainty that may occur under specific environmental conditions. Through the above aggregation and binning processes, a ground-state uncertainty field is finally constructed, that is, a spatial distribution map characterizing the uncertainty in the water color inversion process, providing reference information for subsequent water color parameter inversion and quality control.

[0062] In one implementation, based on the multi-dimensional binning attributes of the multiple sets of environmental driving factor values, the multiple sets of time-series quality entropy fields are binned and aggregated to construct the ground-state uncertainty field, including:

[0063] Y5151: Perform multi-dimensional environmental binning encoding on the multiple sets of environmental driving factor values ​​to obtain multiple sets of environmental binning combinations; Y5152: Based on the mapping relationship between the multiple sets of environmental driving factor values ​​and the multiple sets of environmental binning combinations, perform multi-modal coupled ground state statistical aggregation on the multiple sets of time-series quality entropy fields to obtain multiple sets of quality ground state values; Y5153: Based on the grid cell array, reconstruct the spatial environmental coupled field of the multiple sets of quality ground state values ​​and the multiple sets of environmental binning combinations to obtain the ground state uncertainty field.

[0064] Each set of environmental driving factor values ​​is binned according to certain rules, such as equidistant binning or equal-frequency binning, and divided into multiple intervals. For example, the solar altitude angle is divided into three intervals: low, medium, and high, and the aerosol optical thickness is divided into three intervals: small, medium, and large. Combining the intervals of different factors yields different environmental binning combinations, each of which represents a specific environmental configuration.

[0065] Based on the environmental bin combinations, each time-series mass entropy field is mapped to its corresponding environmental driving factor value. For example, the environmental factor value corresponding to a certain moment in the first set of time-series hyperspectral sub-maps, such as a solar altitude angle of 45° and an aerosol thickness of 0.2, is assigned to a specific environmental bin combination. For each environmental bin combination, ground-state statistical aggregation is performed under multiple time-series mass entropy fields, such as calculating the median, mean, and standard deviation under that combination. These statistical values ​​reflect the quality changes and uncertainties in water color inversion under specific environmental conditions. The ground-state quality value of each grid cell is obtained, which is the expected value of the water color inversion quality under specific environmental conditions.

[0066] By using multiple sets of obtained ground-state mass values ​​and corresponding environmental bin combinations, the inversion quality of each grid cell under different environmental conditions is coupled. The ground-state mass value of each grid cell is spatially reconstructed based on the historical environmental factor combination corresponding to that cell. This process reveals the quality and reliability of water color inversion under specific environmental patterns. The ground-state mass values ​​of each grid cell under different environmental factor conditions are aggregated to obtain a three-dimensional probabilistic model, namely the ground-state uncertainty field. This field statistically aggregates the long-term mass entropy values ​​of the target water area gridded cells under bin combinations of multiple environmental driving factors (solar altitude angle, aerosol optical thickness, and water surface flare index), such as calculating the median, to form a joint mapping function between spatial location and environmental pattern. This function represents the historical expected value of the water color inversion mass entropy of each grid cell under specific illumination, atmospheric, and water surface conditions, thereby revealing the systematic influence of environmental factors on inversion accuracy and providing a dynamic benchmark for real-time quality assessment.

[0067] In one implementation, water color noise propagation analysis is performed on the plurality of hydrological response unit sub-maps to output multiple sets of water color inversion signal-to-noise ratio fingerprints, including:

[0068] Y210: Perform noise propagation entropy analysis on the multiple hydrological response unit sub-maps, and output multiple intrinsic noise reflectivity vectors and multiple local signal-to-noise ratio tensors; Y220: Perform twin-generation noise coupling analysis on the multiple intrinsic noise reflectivity vectors and multiple local signal-to-noise ratio tensors, and output multiple sets of water color inversion signal-to-noise ratio fingerprints.

[0069] Noise propagation entropy analysis calculates the noise propagation effect between different bands in each hydrological response unit sub-map, identifying and quantifying noise sources in each band. It can be understood that the noise propagation entropy describes how the uncertainty of a signal changes from input to output. Through noise propagation entropy analysis, the reflectivity signal of each hydrological response unit sub-map is deconstructed into an intrinsic noise reflectivity vector. This vector contains the magnitude of the noise's impact on water reflectivity and the noise distribution characteristics within the target area. Simultaneously, it outputs a local signal-to-noise ratio tensor for each sub-map, describing the signal-to-noise ratio ratio within the local area of ​​each grid cell. This tensor is a multi-dimensional matrix characterizing the signal-to-noise ratio variation across different grid cells.

[0070] The coupled analysis process for twin noise sources treats the intrinsic noise reflectivity vector and the local signal-to-noise ratio tensor as two dimensions of noise, performing coupled analysis. This coupling implies that noise affects not only the water reflectivity but also the signal clarity. These two noise sources are analyzed simultaneously using mathematical models such as multiple regression analysis and principal component analysis to reveal their mutual influence and combined effects on noise patterns, including how noise propagates across different bands and the impact of noise from different grid cells on the inversion results. Through coupled analysis, multiple sets of water color inversion signal-to-noise ratio fingerprints are ultimately output. These fingerprints contain information on the signal-to-noise ratio changes caused by noise in different hydrological response units. Each fingerprint can be considered a digital feature representing the water color inversion quality of a specific water body or region; they are characteristic expressions of the noise impact in water color inversion under different regions and conditions.

[0071] In one implementation, a dual-noise coupling analysis is performed on the plurality of intrinsic noise reflectivity vectors and the plurality of local signal-to-noise ratio tensors to output multiple sets of water color inversion signal-to-noise ratio fingerprints, including:

[0072] Y221: Construct an inversion noise propagation coupling matrix based on multiple water color inversion targets corresponding to the multi-scale bio-optical fingerprint bands; Y222: Load the multiple intrinsic noise reflectivity vectors and multiple local signal-to-noise ratio tensors into the inversion noise propagation coupling matrix to obtain a water color parameter-sensitive ground state matrix and a noise distortion propagation matrix; Y223: Decouple the ground state response features based on the water color parameter-sensitive ground state matrix to obtain multiple parameter sensitivity feature vectors; Y224: Extract the distortion spectrum based on the noise distortion propagation matrix to obtain multiple noise distortion feature spectra, wherein the multiple parameter sensitivity feature vectors and the multiple noise distortion feature spectra constitute the multiple sets of water color inversion signal-to-noise ratio fingerprints.

[0073] Multi-scale bio-optical fingerprint bands are key bands for water color inversion in remote sensing imagery, especially relevant to biological targets in water bodies. Each water color inversion target, such as chloroplasts, phytoplankton, and zooplankton, has different responses to reflection or absorption in different bands. By identifying these bio-optical fingerprint bands, suitable bands can be selected for water color inversion more effectively. For example, chloroplasts have strong absorption characteristics in the red-edge and near-infrared bands, making these bands more sensitive for chloroplast inversion. Other water color targets, such as suspended matter and dissolved organic matter, have different spectral characteristics in different bands.

[0074] During the inversion process, noise not only affects a single band but also propagates between multiple bands, especially when different bands correspond to different water color targets. To quantify this propagation process, an inversion noise propagation coupling matrix is ​​constructed, which describes how noise propagates and couples between bands. This matrix links the noise impact of each band to the propagation relationships between other bands, enabling accurate simulation and prediction of noise interactions between bands in subsequent inversion processes.

[0075] By loading the intrinsic noise reflectivity vector and the local signal-to-noise ratio tensor into the inversion noise propagation coupling matrix, the propagation characteristics of the inverted noise can be correlated with the noise characteristics and signal-to-noise ratio of each band, resulting in a more accurate noise propagation model. The water color parameter-sensitive ground state matrix, combining the spectral characteristics of the water color inversion target, the noise reflectivity of each band, and the signal-to-noise ratio tensor, describes the changes in the sensitivity of each band to water color parameters during the water color inversion process. This helps to understand the impact of different bands on the water color inversion results. The noise distortion propagation matrix quantifies the propagation effect of noise between different bands. Noise propagation during the inversion process affects the inversion results of multiple bands; therefore, it is necessary to model the noise propagation to correct and optimize the final results.

[0076] Decoupling refers to breaking down complex, multivariate response characteristics into simpler, independent features. Here, mathematical methods, such as principal component analysis and factor analysis, are used to decouple highly correlated parameters in the water color parameter sensitivity ground-state matrix. For example, if a water color inversion target is influenced by multiple environmental factors, ground-state response feature decoupling can extract individual features from these influencing factors, identifying the dominant response modes of each water color inversion target. The decoupled matrix yields multiple parameter sensitivity eigenvectors, representing the sensitivity of different water color inversion targets to different spectral bands. Each eigenvector provides the response intensity of the water color target in different spectral bands.

[0077] Distortion spectrum extraction refers to extracting the propagation path of noise and its impact on signals in different frequency bands from the noise distortion propagation matrix. Typically, noise propagation is uneven, potentially causing significant signal shifts or distortions in certain bands. Distortion spectrum extraction, by analyzing the noise propagation matrix, reveals the influence patterns of noise on the inversion results in each band. Signal processing methods, such as filtering and Fourier transform, can be used to extract the characteristics of noise propagation from the matrix. Through noise distortion spectrum extraction, multiple noise distortion feature spectra are obtained. These feature spectra represent the distortion effects of noise on different frequency bands, revealing the spatial and frequency characteristics of noise under different environments.

[0078] The parameter sensitivity eigenvector describes the contribution of each band to the water color parameters during the water color inversion process, while the noise distortion eigenspectrum describes the propagation of noise and its distortion effect in each band. Combining these two results in the water color inversion signal-to-noise ratio fingerprint, which can serve as a dynamic indicator of the water color inversion quality and be used to monitor the reliability of the inversion in real time.

[0079] In one implementation, the near-infrared negative distortion field, the spectral fidelity error field, and multiple sets of water color inversion signal-to-noise ratio fingerprints are dynamically fused to generate a real-time quality entropy field, including:

[0080] Y410: Perform dimensionless standardization on the near-infrared negative distortion field and spectral fidelity error field to generate a physical distortion field; Y420: Perform field-based recombination on the multiple sets of water color inversion signal-to-noise ratio fingerprints to obtain parameter sensitivity feature matrices corresponding to the multiple parameter sensitivity feature vectors and noise distortion spectrum tensors corresponding to the multiple noise distortion feature spectra; Y430: Configure fusion weights according to real-time environmental driving factors, adjust the contribution ratios of the physical distortion field, feature matrix, and distortion spectrum tensor, and then call a nonlinear multi-field coupling function to perform bio-optical weighted geometric fusion to output a preliminary quality index field; Y440: Map the preliminary quality index field to the spatially continuously distributed real-time quality entropy field through information entropy conversion.

[0081] The near-infrared negative distortion field and the spectral fidelity error field are the results generated previously. They reflect the negative effects and spectral fidelity errors that occur in the near-infrared band during atmospheric correction, respectively. These distortion fields and error fields contain error values ​​at different spatial locations, and may be unable to be directly compared or fused due to inconsistent data dimensions, such as pixel values ​​and units.

[0082] Dimensionless standardization refers to the normalization of data, converting them into unitless standardized values ​​so that they have the same dimensions for unified processing and analysis. For example, the Z-score standardization method transforms data into standardized values ​​with a mean of 0 and a variance of 1 by subtracting the mean and dividing by the standard deviation. This standardization process eliminates dimensional differences between different error sources, bringing the distortion field and error field values ​​to the same scale for more effective comparison and fusion. The standardized distortion field and error field yield a new quantization result, the physical distortion field, which reflects the physical errors or distortions caused by atmospheric and spectral responses during water color inversion.

[0083] Field-based recombination refers to reorganizing originally scattered datasets, such as feature vectors and feature spectra of various bands, into a more structured and directly applicable form, represented as a matrix or tensor. The parameter sensitivity feature matrix, corresponding to multiple parameter sensitivity feature vectors, has each column representing the sensitivity feature of a band or combination of bands. Each row in the matrix corresponds to a grid cell or a spatiotemporal location, representing the water color inversion sensitivity of each band at that location. The noise distortion spectrum tensor, corresponding to the multiple noise distortion feature spectra, is a recombination of noise distortion features in multidimensional space. It represents the distortion effect of noise at different spatial locations and in different bands. For example, one dimension of the tensor is spatial location, another is band, and the third dimension can be the noise type, such as background noise, scattering noise, etc.

[0084] The real-time environmental driving factors are the current solar elevation angle, aerosol optical thickness, and water surface flare index. The fusion weight is based on the real-time environmental driving factors to adjust the contribution of different data sources. This means that the influence of physical distortion field, feature matrix and distortion spectrum tensor on the final quality assessment result is optimized according to the current environmental conditions. By adjusting the weight of each factor, the response to different environmental factors can be more accurate.

[0085] Nonlinear multi-field coupling functions are used to fuse multiple environmental driving factors with physical distortion fields, characteristic matrices, and distortion spectral tensors. The relationships between these factors and the data are not nonlinear but rather influenced by multiple complex factors, such as meteorological changes and topographic features. Multi-field coupling means that multiple independent fields are jointly processed through nonlinear functions, comprehensively considering their mutual influences. This coupling method can more accurately capture the complex nonlinear relationships in environmental and water color inversion processes. Bio-optical weighted geometric fusion refers to combining bio-optical principles to geometrically fuse data from various fields in a weighted manner. This helps optimize the contribution of different environmental driving factors to the water color inversion quality, making the water color inversion results for different regions more consistent with reality. After nonlinear multi-field coupling processing, a preliminary quality index field is generated. This field contains water color inversion quality assessment information adjusted by environmental factors, reflecting the inversion quality of different regions under different environmental conditions.

[0086] Information entropy is an indicator that measures information uncertainty. In remote sensing data processing, information entropy can be used to measure the uncertainty or information density of water color inversion results for a region. A higher entropy value indicates that the information in that region is more uncertain or complex, while a lower entropy value indicates that the inversion quality of that region is more stable and reliable. Information entropy transformation converts the quality information of each pixel in the initial quality index field into an entropy value. Through entropy value transformation, a spatially continuous distribution reflecting quality uncertainty can be obtained, making the quality assessment of the entire region clearer. Mapping the transformed entropy values ​​into space generates a real-time quality entropy field. This entropy field is spatially continuous, with each grid cell corresponding to a quality entropy value, representing the stability and reliability of the water color inversion quality in that region.

[0087] In one implementation, using the ground-state uncertainty field as a reference manifold, a confidence region convergence mapping of the real-time mass entropy field is performed to output a calibrated mass confidence field, including:

[0088] Y521: Retrieve the real-time environment binning code based on the real-time environment driving factors; Y522: Extract the two-dimensional spatial reference manifold field from the ground-state uncertainty field using the real-time environment binning code; Y523: Calculate the absolute deviation field between the two-dimensional spatial reference manifold field and the real-time mass entropy field; Y524: Dynamically set the confidence threshold field for spatial variation based on the ground-state level interval of the two-dimensional spatial reference manifold field; Y525: Perform a manifold convergence mapping operation to generate a preliminary calibration field based on the spatial position comparison results of the confidence threshold field and the absolute deviation field; Y526: Perform a terrain-constrained spatial regularization operation on the preliminary calibration field to output a calibration quality confidence field.

[0089] Real-time environmental driving factors change with time and geographical location, affecting the quality of water color inversion. For example, the solar altitude angle determines illumination conditions, aerosol optical thickness affects atmospheric scattering, and the water surface flare index reflects the intensity of water surface reflection. Binning coding is a method of dividing continuous environmental factors into discrete categories according to certain rules. For example, it can be divided into several intervals based on the solar altitude angle, such as high, medium, and low angles, and similar divisions can be made based on other factors. Real-time retrieval refers to finding the corresponding environmental binning code based on the environmental factor value at the current moment. These codes represent a specific combination of environmental states and can be used as background data for further analysis of water color inversion.

[0090] The ground-state uncertainty field is a three-dimensional probabilistic model representing the inherent reliability and quality of water color inversion under different environmental driving factors. Constructed based on historical data, this field reflects the expected quality of water color inversion under varying environmental conditions. It incorporates a joint mapping between spatial location and environmental patterns, quantifying the uncertainties arising from environmental changes during the water color inversion process. The two-dimensional spatial baseline manifold field is a specific slice extracted from the three-dimensional ground-state uncertainty field, representing the spatial distribution corresponding to the current environmental binning encoding. This field captures the spatial distribution characteristics of water color inversion under specific environmental conditions. In this way, more accurate quality assessments can be provided for different regions based on real-time environmental conditions.

[0091] The two-dimensional spatial reference manifold field is a spatial distribution pattern generated based on the binning combination of environmental factors and ground-state uncertainties. It provides the expected quality of the water color inversion results under specific environmental conditions. The real-time quality entropy field is the result of a quantitative assessment of the current water color inversion quality. It maps the preliminary quality index field to a spatially continuous quality assessment field through information entropy transformation, representing the stability and reliability of the water color inversion quality. The absolute deviation refers to the absolute value of the difference between the two. By calculating the absolute deviation between the two-dimensional spatial reference manifold field and the real-time quality entropy field, the difference between the inversion results and the expected quality can be quantified. This calculation process can reveal the deviation between the water color inversion results and the theoretical benchmark under real-time environmental conditions.

[0092] The ground state level range represents the range of variation of the ground state uncertainty field under specific environmental conditions, reflecting the possible fluctuation range of the inversion quality. The confidence threshold field represents the range of variation of the inversion quality under specific environmental conditions, quantifying the impact of environmental conditions on the fluctuation of water color inversion quality. It provides a dynamic standard for subsequent calibration operations, indicating under what conditions calibration is required.

[0093] By comparing the relationship between the absolute deviation value and the confidence threshold field, it can be determined whether calibration is needed and how to adjust it. Specifically, when the absolute deviation value is less than or equal to the confidence threshold, the corresponding value of the two-dimensional spatial reference manifold field is taken as the calibration value; when the absolute deviation value is greater than the confidence threshold and the real-time mass entropy field value is greater than the reference manifold field value, the absolute deviation increment is adjusted by a compression factor to generate a calibration value; when the absolute deviation value is greater than the confidence threshold and the real-time mass entropy field value is less than the reference manifold field value, the absolute deviation reduction is adjusted by a decay factor to generate a calibration value. Through the above mapping operations, a preliminary calibration field is generated. This field provides a preliminary calibration value based on the current environmental conditions and inversion results for subsequent optimization and verification.

[0094] Topographic constraints refer to incorporating geographic information into the calibration process to ensure that the calibration field is not only based on environmental drivers and differences in inversion quality, but also matches actual topography and regional characteristics. For example, topographic features such as mountains, oceans, and rivers can affect water color inversion results. Spatial regularization smooths and adjusts the initial calibration field to conform to actual topography and regional characteristics. The regularization process reduces noise caused by local environmental fluctuations, ensuring the stability of calibration values. After regularization, the calibration field becomes smoother and more consistent with regional characteristics, facilitating accurate evaluation of water color inversion quality in different topographic regions. The calibration quality reliability field is the final, topographically constrained, adjusted, and optimized quality assessment field. It provides a reliable quality assessment standard for subsequent water color inversion, ensuring the reliability and accuracy of inversion results under different environmental and topographic conditions.

[0095] In one implementation, atmospheric correction distortion field diagnosis is performed on the hyperspectral remote sensing image, outputting a near-infrared negative distortion field and a spectral fidelity error field, including:

[0096] Y310: Based on the near-infrared band reflectance, clean water pixels are selected from the hyperspectral remote sensing image to generate a clean water mask; Y320: After extracting the absolute value of the negative reflectance in the near-infrared band of the clean water mask, the negative near-infrared distortion field is generated by spatial interpolation; Y330: The measured visible light spectrum of the clean water mask is extracted; Y340: The spectral fidelity error field is generated by calculating the spectral angular distance between the measured visible light spectrum and the preset standard clean water spectral library.

[0097] The reflectivity of water bodies varies across different spectral bands. The near-infrared band is particularly sensitive to water bodies because their reflectivity is typically low, while other surface features, such as vegetation and soil, have higher reflectivity. Selecting clean water pixels means utilizing the reflectivity characteristics of the near-infrared band in hyperspectral remote sensing imagery to identify pixels with low reflectivity as water body pixels, especially those representing clean water bodies. These selected clean water pixels are then marked to create a clean water mask, which serves as the basis for further analysis.

[0098] In remote sensing data processing, near-infrared reflectance may be negative due to atmospheric effects, sensor errors, or the characteristics of the water body itself. For pixels of clean water bodies, the absolute value of the negative reflectance is first extracted. Then, the negative reflectance data is smoothed using spatial interpolation methods, such as inverse distance weighting and Kriging interpolation, to generate a near-infrared negative distortion field. This distortion field reveals the reflectance deviation in remote sensing images caused by factors such as atmosphere, illumination, and sensors.

[0099] The spectral data in the visible light band contains various information such as water bodies and land features. By extracting the visible light band data of corresponding pixels from the clean water body mask, the actual reflectance characteristics of the water body in the visible light range can be obtained. Since this spectral data is extracted from the clean water body mask, it can represent the typical reflectance characteristics of clean water bodies in the visible light band, which is helpful for subsequent spectral comparison and error analysis.

[0100] Spectral angular distance is a spectral similarity metric that calculates the angular difference between two spectra. By calculating the angular difference between the measured spectrum and the spectra in a pre-set library of standard clean water spectra, the spectral fidelity of clean water in the visible light band can be quantified. By calculating the spectral angular distance of all clean water pixels, a spectral fidelity error field is obtained, reflecting the degree of difference between the spectral characteristics of each pixel and the standard clean water spectrum. This error field can be used to assess the spectral accuracy during the water retrieval process.

[0101] 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.

[0102] 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 hyperspectral data quality assessment for water color remote sensing parameter retrieval, characterized in that, The method comprises: Divide a plurality of hydrological response unit subgraphs from a hyperspectral remote sensing image of a target area based on a preset multiscale biological optical fingerprint band, wherein the hyperspectral remote sensing image is an atmospheric path corrected hyperspectral image; Perform water color noise transfer analysis on the plurality of hydrological response unit subgraphs, and output a plurality of groups of water color inversion signal-to-noise ratio fingerprints; Perform atmospheric correction distortion field diagnosis on the hyperspectral remote sensing image, and output a near-infrared negative value distortion field and a spectral fidelity error field; Dynamically fuse the near-infrared negative value distortion field, the spectral fidelity error field and the plurality of groups of water color inversion signal-to-noise ratio fingerprints to generate a real-time quality entropy field; Take an environment-driven ground state uncertainty field as a reference manifold to perform confidence domain convergence mapping of the real-time quality entropy field, and output a water color parameter inversion uncertainty heat map.

2. The method for hyperspectral data quality assessment for water color remote sensing parameter retrieval according to claim 1, characterized in that, Take an environment-driven ground state uncertainty field as a reference manifold to perform confidence domain convergence mapping of the real-time quality entropy field, and output a water color parameter inversion uncertainty heat map, comprising: Performing temporal tracking of environment-driven factors in the target area to construct a ground state uncertainty field; Taking the ground state uncertainty field as a reference manifold, performing confidence domain convergence mapping of the real-time quality entropy field, and outputting a calibrated quality confidence field; According to a preset uncertainty heat topology rule, the calibrated quality confidence field is mapped to a water color parameter inversion uncertainty heat map.

3. The method for hyperspectral data quality assessment for water color remote sensing parameter retrieval according to claim 2, characterized in that, Performing temporal tracking of environment-driven factors in the target area to construct a ground state uncertainty field, comprising: Loading a plurality of time series hyperspectral images and a plurality of environment-driven factor data of the target area, wherein the environment-driven factor data includes solar elevation angle, aerosol optical depth and water surface flare index; Divide the target area into a grid cell array using a preset spatial resolution; Divide the plurality of time series hyperspectral images and the plurality of environment-driven factor data using the grid cell array to obtain a plurality of groups of time series hyperspectral subgraphs and a plurality of groups of environment-driven factor values corresponding to a plurality of regional grid cells in the grid cell array; Based on the plurality of groups of time series hyperspectral subgraphs, a plurality of groups of time series quality entropy fields are generated; According to the multi-dimensional binning properties of the plurality of groups of environment-driven factor values, the plurality of groups of time series quality entropy fields are binned and aggregated to construct the ground state uncertainty field.

4. The method for hyperspectral data quality assessment for water color remote sensing parameter retrieval according to claim 3, characterized in that, According to the multi-dimensional binning properties of the plurality of groups of environment-driven factor values, the plurality of groups of time series quality entropy fields are binned and aggregated to construct the ground state uncertainty field, comprising: Multi-dimensional environment binning encoding is performed on the plurality of groups of environment-driven factor values to obtain a plurality of groups of environment binning combinations; According to the mapping relationship between the plurality of groups of environment-driven factor values and the plurality of groups of environment binning combinations, multi-modal coupled ground state statistical aggregation is performed on the plurality of groups of time series quality entropy fields to obtain a plurality of groups of quality ground state values; According to the grid cell array, spatial environment coupling field reconstruction is performed on the plurality of groups of quality ground state values and the plurality of groups of environment binning combinations to obtain the ground state uncertainty field.

5. The method for hyperspectral data quality assessment for water color remote sensing parameter retrieval according to claim 2, characterized in that, Performing water color noise transfer analysis on the plurality of hydrological response unit subgraphs to output a plurality of groups of water color inversion signal-to-noise ratio fingerprints, comprising: Performing noise transfer entropy analysis on the plurality of hydrological response unit subgraphs outputs a plurality of intrinsic noise albedo vectors and a plurality of local signal-to-noise ratio tensors; Performing twin noise coupling analysis on the plurality of intrinsic noise albedo vectors and the plurality of local signal-to-noise ratio tensors outputs a plurality of groups of water color inversion signal-to-noise ratio fingerprints.

6. The method for hyperspectral data quality assessment for water color remote sensing parameter retrieval according to claim 5, characterized in that, Performing twin noise coupling analysis on the plurality of intrinsic noise albedo vectors and the plurality of local signal-to-noise ratio tensors outputs a plurality of groups of water color inversion signal-to-noise ratio fingerprints, including: According to a plurality of water color inversion targets corresponding to the plurality of multi-scale bio-optical fingerprint bands, an inversion noise transfer coupling matrix is constructed; The plurality of intrinsic noise albedo vectors and the plurality of local signal-to-noise ratio tensors are loaded into the inversion noise transfer coupling matrix respectively to obtain a water color parameter sensitive ground state matrix and a noise distortion propagation matrix; Based on the water color parameter sensitive ground state matrix, ground state response feature decoupling is performed to obtain a plurality of parameter sensitivity feature vectors; Based on the noise distortion propagation matrix, distortion spectral series extraction is performed to obtain a plurality of noise distortion feature spectra, wherein the plurality of parameter sensitivity feature vectors and the plurality of noise distortion feature spectra constitute the plurality of groups of water color inversion signal-to-noise ratio fingerprints.

7. The method for hyperspectral data quality assessment for water color remote sensing parameter retrieval according to claim 6, characterized in that, The near-infrared negative value distortion field, the spectral fidelity error field and the plurality of groups of water color inversion signal-to-noise ratio fingerprints are dynamically fused to generate a real-time quality entropy field, including: The near-infrared negative value distortion field and the spectral fidelity error field are subjected to dimensionless normalization to generate a physical distortion field; The plurality of groups of water color inversion signal-to-noise ratio fingerprints are subjected to field reorganization to obtain a parameter sensitivity feature matrix corresponding to the plurality of parameter sensitivity feature vectors and a noise distortion spectral tensor corresponding to the plurality of noise distortion feature spectra; After configuring a fusion weight according to a real-time environmental driving factor, the contribution proportions of the physical distortion field, the feature matrix and the distortion spectral tensor are regulated, a nonlinear multi-field coupling function is called to perform bio-optical weighted geometric fusion, and a preliminary quality index field is output; The preliminary quality index field is mapped to the spatially continuous distribution of the real-time quality entropy field through information entropy conversion.

8. The method for hyperspectral data quality assessment for water color remote sensing parameter retrieval according to claim 7, characterized in that, The ground state uncertainty field is taken as a reference manifold to perform confidence domain convergence mapping of the real-time quality entropy field, and a calibrated quality confidence field is output, including: Real-time environmental binning codes are retrieved according to the real-time environmental driving factor; A two-dimensional spatial reference manifold field is extracted from the ground state uncertainty field using the real-time environmental binning codes; An absolute deviation field is calculated for the two-dimensional spatial reference manifold field and the real-time quality entropy field; A confidence threshold field of spatial variation is dynamically set based on the ground state level interval of the two-dimensional spatial reference manifold field; A manifold convergence mapping operation is performed according to the spatial position comparison results of the confidence threshold field and the absolute deviation field to generate a preliminary calibration field; The preliminary calibration field is subjected to a terrain-constrained spatial regularization operation to output a calibrated quality confidence field.

9. The method for hyperspectral data quality assessment for water color remote sensing parameter retrieval according to claim 1, wherein, Performing atmospheric correction distortion field diagnosis on the hyperspectral remote sensing image outputs a near-infrared negative value distortion field and a spectral fidelity error field, including: Based on near-infrared band reflectance, clean water body pixels are screened in the hyperspectral remote sensing image to generate a clean water body mask; extracting the absolute value of the negative reflectivity of the near-infrared band of the clean water body mask, and generating the near-infrared negative value distortion field through spatial interpolation; extracting the measured spectrum of the visible light band of the clean water body mask; generating the spectrum fidelity error field by calculating the spectrum angle distance between the measured spectrum of the visible light band and the preset standard clean water body spectrum library.

Images