Methods and Systems for Spectral Grading Assessment of Eutrophication in Rivers and Lakes Using Remote Sensing Images

Abstract

This invention discloses a method and system for spectral classification and assessment of eutrophication in river and lake water bodies using remote sensing images. The method includes the following steps: a shallow water bottom sediment spectral decoupling step based on radiative transfer mechanism, acquiring multispectral remote sensing images of the target water area, constructing a water body radiative transfer coupling model, and solving the model under constraints to extract the effective water body spectrum; an adaptive quantitative inversion step of chlorophyll concentration based on decoupled spectra, constructing a spectral band feature space, dynamically searching for the band combination most sensitive to changes in chlorophyll a concentration within this feature space, establishing a nonlinear mapping relationship, and obtaining the chlorophyll a concentration inversion value for each pixel; and a eutrophication evolution trend prediction step based on spatiotemporal continuity, classifying eutrophication levels according to the chlorophyll a concentration inversion values, constructing a spatiotemporal propagation model, and predicting the future eutrophication state of the water body. This method can solve the problem of chlorophyll concentration inversion accuracy distortion caused by bottom sediment exposure in grassy clear water areas.

Description

Technical Field

[0001] This invention relates to the field of remote sensing image processing technology, and in particular to a method and system for spectral classification and assessment of eutrophication in remote sensing images of rivers and lakes. Background Technology

[0002] Eutrophication is one of the most serious water environment problems facing the world today, especially for shallow lakes, where frequent cyanobacterial blooms seriously threaten drinking water safety, ecosystem balance, and regional socio-economic sustainable development. Remote sensing technology, due to its advantages of macroscopic observation, speed, and repeatability, has become an important means of dynamic monitoring of eutrophication in large-scale water bodies. Among these methods, inverting chlorophyll a concentration based on remote sensing reflectance is a core technical approach for assessing the nutrient status of water bodies.

[0003] However, in shallow lakes, especially grassy clear water areas with well-developed aquatic vegetation, existing remote sensing inversion methods face a key technical bottleneck: interference from the reflection of the bottom sediment. In these types of water bodies, the water transparency is high, and solar radiation can penetrate the water layer to reach the bottom. The reflected signals from the sediment (such as mud, sand, or submerged plants), after being attenuated by the water layer, can still be received by the sensor and superimposed with the volume scattering signals from the internal components of the water body, forming a mixed spectrum. Traditional chlorophyll a concentration inversion algorithms (such as the band ratio method and the three-band method) are mostly based on the optical characteristics of deep water bodies or high-turbidity water bodies. Their core assumption is that the water-free reflectance received by the sensor is mainly or entirely contributed by water components. When these algorithms are directly applied to grassy clear water areas, the sediment reflection signal is incorrectly attributed to the absorption and scattering characteristics of water components such as chlorophyll, leading to a systematic overestimation of chlorophyll a concentration, and even misjudging clean water bodies as eutrophic, seriously affecting the accuracy and reliability of remote sensing monitoring.

[0004] To address the aforementioned issues, some existing technologies have attempted to correct for the influence of sediment. For example, some studies have mitigated sediment interference by introducing prior information on sediment reflection or constructing a sediment insensitivity index based on multi-band combinations. However, most of these methods are empirical or semi-empirical corrections, lacking a complete description of the radiative transfer mechanism and making them difficult to adapt to real-world water scenarios with diverse sediment types and complex water depth variations. Specifically, existing methods typically face the following technical challenges: first, it is difficult to accurately separate the water body scattering signal from the sediment reflection signal at the physical mechanism level; second, they are heavily reliant on water depth information, and the acquisition of water depth data often depends on synchronous field measurements or high-precision topographic data, making it difficult to promote and apply in large-scale remote sensing monitoring; and third, they lack an adaptive judgment mechanism for different water body optical types (such as optically shallow water and optically deep water), leading to additional errors in deep water or high-turbidity areas due to ineffective water depth inversion.

[0005] Therefore, there is an urgent need to develop a spectral classification assessment method for eutrophication of river and lake water bodies using remote sensing images that can accurately remove bottom sediment interference signals from the perspective of radiation transmission mechanism, adapt to different water depths and bottom sediment types, and does not rely on prior water depth data. This method would solve the technical problem of chlorophyll concentration inversion accuracy distortion caused by bottom sediment exposure in grassy clear water areas, and provide reliable technical support for accurate monitoring and early warning of eutrophication in shallow lakes. Summary of the Invention

[0006] The purpose of this section is to outline some aspects of embodiments of the present invention and to briefly describe some preferred embodiments. Simplifications or omissions may be made in this section, as well as in the abstract and title of this application, to avoid obscuring the purpose of these documents; however, such simplifications or omissions should not be construed as limiting the scope of the invention.

[0007] To address the aforementioned technical problems, this invention provides the following technical solution: a method for spectral classification and assessment of eutrophication in river and lake water bodies using remote sensing images, comprising the following steps:

[0008] The shallow water bottom sediment spectral decoupling step based on the radiative transfer mechanism is used to acquire multispectral remote sensing images of the target water area. A water radiative transfer coupling model is constructed that includes the backscattering coefficient of water components and the contribution of bottom sediment reflectivity. Prior water depth information is introduced to constrain the solution of the model, so as to remove the interference signal affected by the bottom sediment from the original pixel spectrum, thereby extracting the effective water spectrum that only represents the inherent optical properties of the water body.

[0009] The adaptive quantitative inversion step of chlorophyll concentration based on decoupled spectrum involves constructing a spectral band feature space based on the extracted effective water body spectrum, dynamically searching for the band combination most sensitive to changes in chlorophyll a concentration within this feature space, and establishing a nonlinear mapping relationship based on the optimal band combination found to obtain the inversion value of chlorophyll a concentration for each pixel.

[0010] The eutrophication evolution trend prediction step based on spatiotemporal continuity involves classifying eutrophication levels according to the inversion value of chlorophyll a concentration, and constructing a spatiotemporal propagation model that integrates spatial neighborhood influences and time series change characteristics by combining multi-temporal remote sensing monitoring data to predict the future eutrophication state of water bodies.

[0011] As a preferred embodiment of the spectral classification assessment method for eutrophication of river and lake water bodies using remote sensing images according to the present invention, the shallow water sediment spectral decoupling step specifically includes:

[0012] Based on the theory of radiative transfer, the total water reflectance received by the sensor Volume scattering contributes to the optical properties of the water body Reflection from the substrate after water attenuation The sum of these is used to construct the objective function by introducing an optical closure condition;

[0013] The absorption coefficient and backscattering coefficient of water components are initialized using the spectrum of deep water regions in the image where the water depth is greater than the preset optical depth.

[0014] Based on this, and combining the relative water depth values ​​of pixels estimated by the band ratio method, the objective function is solved using a nonlinear iterative optimization algorithm. During the solution process, The effective water spectrum is obtained by successively stripping away from the total reflectance until the optical closure error threshold is met. .

[0015] As a preferred embodiment of the spectral classification assessment method for eutrophication of river and lake water bodies using remote sensing images described in this invention, the adaptive quantitative inversion step of chlorophyll concentration specifically includes:

[0016] Construct an initial spectral index pool consisting of all possible two-band ratios, three-band indices, and normalized difference indices;

[0017] Based on the effective water body spectrum and synchronously measured chlorophyll a concentration data, an adaptive pooling search strategy is adopted to traverse the spectral index pools and calculate the correlation coefficient between each index and chlorophyll a concentration.

[0018] The top N indices with the highest correlation coefficients were selected as a candidate set. Collinearity diagnosis was performed on the candidate set to eliminate redundant indices. Finally, a set of core spectral indices that are independent of each other and strongly correlated with chlorophyll concentration were retained.

[0019] A multivariate nonlinear regression model was constructed using this core spectral index, and the model structure parameters were determined using leave-one-out cross-validation.

[0020] As a preferred embodiment of the spectral classification assessment method for eutrophication of river and lake water bodies using remote sensing images according to the present invention, the step of predicting the eutrophication evolution trend specifically includes:

[0021] Based on the eutrophication levels obtained from the inversion values ​​of chlorophyll a concentration, a eutrophication level classification map for a continuous time series is constructed.

[0022] Based on the eutrophication level classification map of the continuous time series, a eutrophication level transition probability matrix is ​​constructed; at the same time, considering water flow and connectivity, a neighborhood spatial influence factor centered on the pixel is constructed.

[0023] The transition probability matrix and the neighborhood space influence factor are coupled into a cellular automata model. By iteratively simulating the diffusion, aggregation and evolution of eutrophication levels in space, a eutrophication spatial distribution prediction map for future preset time nodes is generated.

[0024] As a preferred embodiment of the spectral classification and assessment method for eutrophication of river and lake water bodies using remote sensing images described in this invention, the effective water body spectrum obtained after stripping the sediment signal using a nonlinear iterative optimization algorithm is described. The data is used as input to perform the adaptive spectral index search and chlorophyll inversion.

[0025] As a preferred embodiment of the spectral classification assessment method for eutrophication of river and lake water bodies using remote sensing images described in this invention, the relative water depth value of the pixel estimated by the band ratio method specifically adopts the ratio of the water body absorption band that is not sensitive to changes in the spectral shape of the bottom sediment to the water body weak absorption band that is sensitive to changes in water depth. Combined with the adaptive threshold judgment of optical depth, this ratio is used to estimate water depth only when the optical depth of the pixel is less than a preset threshold, so as to avoid invalid inversion states caused by the lack of bottom sediment signals in deep water areas or the interference of water scattering in high turbidity areas.

[0026] As a preferred embodiment of the spectral classification assessment method for eutrophication of river and lake water bodies using remote sensing images described in this invention, the multivariate nonlinear regression model is a Gaussian process regression model; when training the Gaussian process regression model using the core spectral index, its kernel function parameters are adaptively determined by maximizing the marginal likelihood estimation.

[0027] As a preferred embodiment of the spectral classification assessment method for eutrophication of river and lake water bodies using remote sensing images according to the present invention, the method further includes a step of identifying the type of the target water body after acquiring the multispectral remote sensing image of the target water body and before performing shallow water sediment spectral decoupling processing on the image:

[0028] Calculate the ratio of water reflectance in the green light band to that in the red light band. and reflectivity in the near-infrared band ;

[0029] When the Greater than the first preset threshold and the If the water level is less than the second preset threshold, the water area is determined to be a grass-type clear water area with aquatic vegetation development. It is determined that the water area has a risk of spectral interference due to the bottom sediment being exposed. When performing the shallow water bottom sediment spectral decoupling process, the bottom sediment contribution term in the water body radiation transfer coupling model is activated and the water depth prior information is introduced for constraint solution.

[0030] When the Less than or equal to the first preset threshold, or the If the value is greater than or equal to the second preset threshold, the water area is determined to be a non-grass-type clear water area, and chlorophyll inversion is performed using a standard water body optical model.

[0031] This invention also provides a spectral classification and assessment system for remote sensing images of eutrophication in rivers and lakes, applied to the aforementioned method for spectral classification and assessment of remote sensing images of eutrophication in rivers and lakes, comprising:

[0032] The image acquisition and preprocessing module is used to acquire multispectral images and perform geometric and radiometric corrections.

[0033] The water body type identification and decoupling activation module, connected to the image acquisition and preprocessing module, is used to calculate the reflectance ratio of the water body in the green light band and the red light band, as well as the reflectance in the near-infrared band, to identify the water body type. When it is determined to be a grassy clear water body, the bottom sediment contribution term of the water body radiation transfer coupling model in the bottom sediment interference decoupling module is activated and prior water depth information is introduced. When it is determined to be a non-grassy clear water body, the preprocessed image is directly transmitted to the adaptive spectral analysis and inversion module or processed using the standard model.

[0034] The sediment interference decoupling module has a built-in water radiative transfer model and nonlinear iterative solver, which is used to strip the sediment reflection signal and output the pure water spectrum.

[0035] The adaptive spectral analysis and inversion module includes a dynamic spectral index search unit and a Gaussian process regression unit. It is used to receive the pre-processed images of the pure water body spectrum or the non-grass-type clear water area, and generate a chlorophyll a concentration distribution map based on the core spectral indices selected by the dynamic spectral index search unit through Gaussian process regression.

[0036] The spatiotemporal evolution analysis and early warning module integrates Markov chain and cellular automata models to simulate the eutrophication spread trend and generate early warning information.

[0037] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described method for spectral classification and assessment of eutrophication of river and lake water bodies using remote sensing images.

[0038] The beneficial effects of this invention are:

[0039] 1. This invention constructs a coupled model of water body radiative transfer by combining the absorption backscattering coefficient of water components with the contribution of bottom sediment reflectivity. It introduces the optical closure principle to construct the objective function and combines prior water depth information for nonlinear iterative optimization. This achieves the physical mechanism-level precise removal of bottom sediment reflection interference signals from the original remote sensing pixel spectrum, obtaining a pure and physically consistent effective water body spectrum. This fundamentally solves the technical problem of chlorophyll concentration inversion accuracy distortion caused by bottom sediment exposure in shallow water areas, especially in grassy clear water areas, which is caused by traditional methods. This lays a reliable data foundation for subsequent high-precision eutrophication assessment.

[0040] 2. This invention constructs an initial spectral index pool covering all possible band combinations based on the decoupled pure spectrum, employs an adaptive pooling search strategy to traverse and calculate the correlation coefficient between each index and chlorophyll a concentration, and combines collinearity diagnosis to eliminate redundant indices, thereby automatically selecting core spectral indices that are independent of each other and strongly correlated with chlorophyll concentration. This effectively avoids the subjectivity and information redundancy problems caused by relying on experience to select fixed indices in traditional methods, and significantly improves the adaptability and accuracy of chlorophyll a concentration inversion.

[0041] 3. This invention constructs a continuous time series eutrophication level classification map and calculates the level transition probability matrix to characterize the evolution law in the time dimension. At the same time, it constructs a neighborhood spatial influence factor centered on the pixel to quantify the spatial diffusion effect of the surrounding water area. Then, it couples the two into a cellular automata model to establish a unified spatiotemporal transformation rule. Through iterative simulation, it realizes the joint modeling and quantitative prediction of the spatial diffusion path, aggregation area and evolution trend of eutrophication level, providing a scientific basis for early warning and proactive prevention and control of eutrophication risk. Attached Figure Description

[0042] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. 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. Wherein:

[0043] Figure 1 This is a flowchart illustrating the overall process of the remote sensing image spectral classification assessment method for eutrophication of rivers and lakes according to the present invention.

[0044] Figure 2 This is a flowchart of the shallow water sediment spectral decoupling process for the remote sensing image spectral classification assessment method for eutrophication of rivers and lakes according to the present invention.

[0045] Figure 3 This is a flowchart of the adaptive quantitative inversion process for chlorophyll concentration in the remote sensing image spectral classification assessment method for eutrophication of rivers and lakes according to the present invention.

[0046] Figure 4 This is a flowchart illustrating the spatiotemporal evolution prediction process of the eutrophication remote sensing image spectral classification assessment method for river and lake water bodies according to the present invention.

[0047] Figure 5 This is a flowchart illustrating the intelligent identification and diversion process of water body type in the remote sensing image spectral classification assessment method for eutrophication of rivers and lakes according to the present invention. Detailed Implementation

[0048] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0049] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0050] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0051] Secondly, the present invention is described in detail with reference to the schematic diagrams. When detailing the embodiments of the present invention, for ease of explanation, the cross-sectional views illustrating the device structure may be partially enlarged, not according to the usual scale. Furthermore, the schematic diagrams are merely examples and should not limit the scope of protection of the present invention. In addition, actual fabrication should include three-dimensional spatial dimensions of length, width, and depth.

[0052] Example 1

[0053] Reference Figure 1-5 This first embodiment of the invention provides a method for spectral classification and assessment of eutrophication in remote sensing images of rivers and lakes. This method is suitable for solving the problem of distorted accuracy in remote sensing inversion of eutrophication in shallow lakes, especially grassy clear water areas with well-developed aquatic vegetation, due to interference from bottom sediment reflection. Specifically, it includes the following steps:

[0054] S1: The shallow water bottom sediment spectral decoupling step based on the radiative transfer mechanism acquires multispectral remote sensing images of the target water area. By constructing a water body radiative transfer coupling model that includes the backscattering coefficient of water body components and the contribution of bottom sediment reflectivity, and introducing prior water depth information to constrain the solution of the model, the interference signal affected by the bottom sediment is extracted from the original pixel spectrum, thereby extracting the effective water body spectrum that only represents the inherent optical properties of the water body.

[0055] This step aims to remove sediment reflection interference from the original remote sensing image and extract the "effective water spectrum" determined solely by the inherent optical properties of the water body, laying the data foundation for subsequent high-precision chlorophyll inversion.

[0056] The method for acquiring multispectral remote sensing images of the target water area is as follows: Multispectral satellite remote sensing image products such as Sentinel-2 MSI or Landsat OLI covering the target river and lake area are acquired through ground receiving stations or public data distribution centers (such as the European Space Agency Copernicus Open Access Centre or USGS EarthExplorer). Temporal data with cloud cover below 10% and matching the ground-measured time window are preferred. The acquired images are then preprocessed with radiometric calibration, atmospheric correction, and geometric fine correction to obtain the true surface reflectance data.

[0057] Specifically, to address the common problem of sediment reflection interference in shallow water areas, a coupled model of water radiative transfer is constructed. This model, based on the radiative transfer mechanism, calculates the total water reflectance received by the sensor. Volume scattering contributes to the optical properties of the water body Reflection from the substrate attenuated by the water layer and reaching the water surface The sum, its mathematical expression is as follows:

[0058]

[0059] in:

[0060] Indicates wavelength;

[0061] : Represents the volume scattering component contributed by the internal optical properties of the water body (composed of the absorption coefficients of various components in the water body, including chlorophyll a, suspended particulate matter, yellow substances, etc.). and backscattering coefficient (To be jointly determined). In this field, those skilled in the art can calculate this component in various ways based on well-known radiative transfer theories.

[0062] In a preferred embodiment, the expression is of the following form: ,in, This is an empirical coefficient (dimensionless) related to illumination geometry and water phase function. Its value range is usually 0.09-0.15. The specific value can be determined by looking up a table or empirical formula based on the solar zenith angle and water type. and It is composed of the linear superposition of contributions from components such as pure water, chlorophyll, suspended matter, and yellow substances.

[0063] In another implementation, a more accurate numerical model of radiative transfer (such as Hydrolight) can be used to pre-generate a lookup table. Based on the angular information from remote sensing observations and the optical type of the water body, the lookup table can be used to determine the... and , The mapping relationship.

[0064] Regardless of the specific calculation method used, the core of this invention lies in subsequent steps to... and Separation is performed to obtain the pure water spectrum, which is determined solely by the fixed optical properties of the water body.

[0065] : represents the bottom sediment reflection component that reaches the sensor after attenuation through the water layer, and its expression can be written as: ; where represents the reflectance of the bottom substrate (such as mud, sand, or aquatic vegetation). This is the attenuation coefficient of downstream irradiance in water, which is related to the absorption coefficient of water components. The water is deep.

[0066] To obtain a unique and physically accurate effective water spectrum from the aforementioned underdetermined equations, this invention introduces the optical closure principle as a constraint. This principle requires that the inherent optical properties of the water obtained after decoupling, when re-substituted into the forward radiative transfer model, should optimally reconstruct the original observed spectrum, with the reconstruction error within physically permissible limits.

[0067] Based on this, we construct an objective function in the following form:

[0068]

[0069] Part One This is a data fitting term used to measure the spectrum simulated by the model. Compared with the actual observed spectrum of the sensor The differences between them; Part Two This is a regularization constraint term used to ensure the spectral smoothness and physical rationality (e.g., the absorption coefficient is non-negative) of the solved parameters (such as the water absorption coefficient and the bottom sediment reflectance). The regularization parameter is used to balance the goodness of fit of the data with the strength of the physical constraints. The essence of this objective function is to find the optimal balance between "fitting the observed data as closely as possible" and "conforming to the optical and physical laws of water", thereby achieving accurate signal separation.

[0070] Then, model initialization and water depth estimation are performed. First, the absorption coefficients of water components are calculated using the spectra of deep-water regions in the image where the water depth is greater than a preset optical depth (e.g., optical depth > 3, i.e., underwater signal attenuation exceeds 95%). and backscattering coefficient An initial estimation is performed. Next, for shallow water pixels requiring substrate decoupling, the relative water depth is estimated. In a preferred embodiment, considering the shape influence of different substrate types (e.g., muddy, sandy, grassy) on spectral reflectance, this invention uses a band ratio method to estimate the relative water depth of pixels. Specifically, the ratio of the water body absorption band (e.g., near-infrared band (B8a, 865nm) for Sentinel-2 MSI imagery), which is insensitive to changes in substrate spectral shape, to the water body weak absorption band (e.g., green band, B3, 560nm) that is sensitive to changes in water depth is used to construct a water depth inversion feature. This is then combined with a pre-built water depth-band ratio lookup table from the radiative transfer model for estimation.

[0071] More preferably, this invention introduces an adaptive threshold judgment mechanism for optical depth to avoid invalid inversion states caused by the lack of bottom sediment signals in deep water areas or interference from scattering in high-turbidity water areas. The specific steps are as follows:

[0072] Optical depth estimation: Based on the absorption coefficient of water in the red light band (e.g., 665 nm). and preliminary estimated water depth Calculate the optical depth of each pixel. Or, more precisely, using the diffuse attenuation coefficient With water depth The product of.

[0073] Adaptive threshold determination: Set a preset optical depth threshold. (As an optional embodiment, The value ranges from 0.5 to 1.5, preferably 1.0. This value corresponds to the depth at which approximately 63% of the light radiation below the water surface is absorbed or scattered, and is a typical boundary for distinguishing between "optically shallow water" and "optically deep water".

[0074] when < When a pixel is determined to belong to the "optically shallow water" region, the bottom sediment signal significantly contributes to the total water reflectance. In this case, the aforementioned band ratio method (such as B8a / B3) is used in conjunction with a pre-constructed water depth-band ratio lookup table for the radiative transfer model to estimate the water depth, obtaining the relative water depth value used for model solving. .

[0075] when ≥ At this point, the pixel is determined to belong to either an "optical deep water" region (where the sediment signal is negligible) or a high-turbidity water body (where water scattering is absolutely dominant). For such pixels, sediment decoupling is unnecessary; therefore, in subsequent model solutions, the sediment contribution term is directly included. Set to zero, considering only the water volume scattering component. This simplifies the calculation and avoids errors introduced by invalid water depth inversion.

[0076] Then, the objective function is solved using a nonlinear iterative optimization algorithm. Preferably, the Levenberg-Marquardt algorithm can be used to solve the objective function. Solve the problem.

[0077] During the iteration process, the optical parameters of the water components and (for) are gradually adjusted. < The reflectivity of the substrate (pixel) will The components are successively stripped from the total reflectance until the objective function is reached. The value converges to a preset optical closure error threshold (e.g., (or the number of iterations reaches its limit). At this point, the minimum... value corresponding That is, the effective water spectrum that ultimately characterizes only the inherent optical properties of the water body. .

[0078] It should be noted that, through the above-mentioned decoupling processing of coupled optical closure constraints and adaptive water depth threshold judgment for radiative transmission, this invention successfully eliminates the interference of bottom sediment reflection in shallow water areas, obtaining a pure and physically consistent effective water spectrum. This provides a high-quality, interference-free input data foundation for the adaptive chlorophyll a concentration inversion based on this pure spectrum in the subsequent step S2, thereby fundamentally ensuring the accuracy and reliability of eutrophication remote sensing assessment.

[0079] S2: The adaptive quantitative inversion step of chlorophyll concentration based on decoupled spectrum. Based on the extracted effective water body spectrum, a spectral band feature space is constructed. Within this feature space, the band combination most sensitive to changes in chlorophyll a concentration is dynamically searched. A nonlinear mapping relationship is established based on the searched optimal band combination to obtain the inversion value of chlorophyll a concentration for each pixel.

[0080] This step receives the effective water spectrum output from step S1. As input data. It is particularly important to emphasize that... The sediment interference has been eliminated, and the absorption and scattering information of the water components can be accurately reflected, which is a prerequisite for subsequent high-precision inversion.

[0081] Specifically, the adaptive quantitative inversion steps for chlorophyll concentration include:

[0082] Construct an initial spectral index pool consisting of all possible two-band ratios, three-band indices, and normalized difference indices;

[0083] Based on effective water body spectra and synchronously measured chlorophyll a concentration data, an adaptive pooling search strategy is adopted to traverse the spectral index pools and calculate the correlation coefficient between each index and chlorophyll a concentration.

[0084] The top N indices with the highest correlation coefficients were selected as a candidate set. Collinearity diagnosis was performed on the candidate set to eliminate redundant indices. Finally, a set of core spectral indices that are independent of each other and strongly correlated with chlorophyll concentration were retained.

[0085] A multivariate nonlinear regression model was constructed using this core spectral index, and the model structure parameters were determined using leave-one-out cross-validation.

[0086] In one specific implementation, constructing the initial spectral index pool includes: firstly, based on the multiple spectral bands contained in the remote sensing image, constructing an initial spectral index pool covering all possible band combinations. For a remote sensing image with m bands (such as the visible-near-infrared bands of Sentinel-2 MSI), the index pool includes, but is not limited to, the following forms:

[0087] Two-band ratio index: ,in and Reflectivity for different wavelength bands;

[0088] Three-band index: ( - )× These types of indices are often used to eliminate interference from suspended matter and yellow substances;

[0089] Normalized Difference Index: .

[0090] As an optional implementation, for Sentinel-2 MSI imagery, a configuration can be constructed that includes all possible pairwise band combinations (total...). The ratio of species and 45 normalized difference indices) and all possible three-band combinations (total) The index pool comprises over 800 candidate spectral indices (considering the order of bands). Those skilled in the art will understand that the larger the number of bands, the larger the index pool, but the core processing method remains the same.

[0091] In one specific implementation, the steps for screening the core spectral indices are as follows:

[0092] First, based on the effective water spectrum obtained in step S1 Along with synchronously measured chlorophyll a concentration data (e.g., data obtained by uniformly distributing 50-100 sampling points in the study area), an adaptive pooling search strategy is employed to traverse the entire initial spectral index pool and calculate the statistical correlation coefficient between each candidate spectral index and the measured chlorophyll a concentration. (Preferably, the Pearson correlation coefficient is used). After obtaining the correlation coefficients of all candidate indices, the top N indices with the highest absolute values ​​of their correlation coefficients are selected as the candidate set. In a preferred embodiment, N ranges from 5 to 15, preferably 10, to ensure that the candidate set is sufficiently representative without introducing too much redundant information.

[0093] Then, because these candidate indices may be highly correlated (i.e., collinearity), directly using them for multivariate nonlinear regression modeling would lead to model instability. Therefore, it is necessary to perform collinearity diagnosis on the candidate set to eliminate redundant indices. Specifically, the variance inflation factor among the indices in the candidate set is calculated:

[0094]

[0095] in, For the first The coefficient of determination of one index when performing linear regression on all other indices. As a preferred embodiment, the VIF threshold is set to 10, and indices with VIF greater than 10 are gradually eliminated, ultimately retaining a set of core spectral indices that are independent of each other (low collinearity) and strongly correlated with chlorophyll a concentration.

[0096] In one specific implementation, constructing a multivariate nonlinear regression model includes the following steps:

[0097] First, using the selected core spectral indices, a multivariate nonlinear regression model was constructed to establish the mapping relationship between the spectral indices and chlorophyll a concentration.

[0098] In a preferred embodiment, the multivariate nonlinear regression model is a Gaussian process regression model. Gaussian process regression is a nonparametric Bayesian method that can effectively handle small sample sizes, high dimensionality, and nonlinear regression problems. The basic form of the model can be expressed as:

[0099]

[0100] in, The input feature vector is composed of core spectral indices. This refers to the concentration of chlorophyll a. This is the mean function (usually set to 0). It is the covariance function (i.e., the kernel function). It is Gaussian noise. Representation function It follows a Gaussian process distribution, which is given by the mean function. Sum of covariance functions (kernel functions) Common definition.

[0101] Next, the model is trained. During the model training phase, a dataset consisting of n training samples is constructed, where each training sample contains two parts:

[0102] Input feature vector: composed of the core spectral indices selected in the preceding steps, for the th The feature vector of a sampling point can be represented as: ,in The number of core spectral indices;

[0103] Output label: Measured chlorophyll a concentration corresponding to this sampling point .

[0104] Based on this, leave-one-out cross-validation is used to evaluate the model performance. For a training set containing n samples, one sample is left out each time. The remaining n-1 samples serve as the validation set, and are used to train the Gaussian process regression model. This process is repeated n times, and the average prediction error is calculated. The kernel function parameters of the Gaussian process regression model are adaptively determined by maximizing the marginal likelihood estimation. A commonly used squared exponential kernel is used as an example:

[0105]

[0106] in, and These are the core spectral index feature vectors of two different samples. For signal variance, For length scale, Let the noise variance be denoted by maximizing the log-marginal likelihood.

[0107]

[0108] The optimal values ​​of the aforementioned hyperparameters can be adaptively determined, thereby achieving a balance between model complexity and data fit, effectively avoiding overfitting. Among these, The matrix consisting of the feature vectors of all training samples. This corresponds to the measured chlorophyll a concentration vector. The kernel matrix has the following elements. .

[0109] After the model is trained, the core spectral index of each water cell in the study area is calculated using the model. Substituting this into the trained Gaussian process regression model, the spatially continuous chlorophyll a concentration inversion value can be obtained.

[0110] In another implementation, other machine learning models such as support vector regression and random forest can also be used to construct a multivariate nonlinear regression model. The core of this model is that the input features are core spectral indices obtained after decoupling in step S1 and adaptive filtering in step S2, rather than the original spectra or fixed indices selected empirically.

[0111] It should be noted that, through the above-mentioned adaptive pooling search based on decoupled spectrum and Gaussian process regression modeling, the present invention obtains the spatial continuous inversion results of chlorophyll a concentration at the pixel scale of the water body in the study area. The inversion results can effectively characterize the spatial distribution characteristics of the eutrophication state of the water body, and provide a data basis for the eutrophication level classification and spatiotemporal evolution trend prediction based on the concentration inversion results in step S3.

[0112] S3: Prediction steps for eutrophication evolution trend based on spatiotemporal continuity. Eutrophication levels are classified according to the inversion value of chlorophyll a concentration. Combined with multi-temporal remote sensing monitoring data, a spatiotemporal propagation model integrating spatial neighborhood influence and time series change characteristics is constructed to predict the future eutrophication state of water bodies.

[0113] This step is based on the chlorophyll a concentration inversion value obtained in step S2 for subsequent analysis.

[0114] Specifically, the steps for predicting eutrophication evolution trends include:

[0115] Based on the eutrophication levels obtained from the inversion values ​​of chlorophyll a concentration, a eutrophication level classification map for continuous time series is constructed.

[0116] Based on the eutrophication level classification map of continuous time series, a eutrophication level transition probability matrix is ​​constructed; at the same time, considering water flow and connectivity, a neighborhood spatial influence factor centered on the pixel is constructed.

[0117] By coupling the transition probability matrix and the neighborhood spatial influence factor into the cellular automata model, the diffusion, aggregation and evolution of eutrophication levels in space are simulated iteratively to generate a predicted spatial distribution map of eutrophication at a future preset time node.

[0118] In one specific implementation, the eutrophication level classification method is as follows: based on a preset eutrophication evaluation standard, the chlorophyll a concentration inversion value obtained in step S2 is converted into eutrophication level information. In an optional implementation, a water trophic status evaluation standard based on chlorophyll a concentration (National Surface Water Environmental Quality Standard (GB3838-2002) or relevant industry standards for eutrophication evaluation based on chlorophyll a concentration) can be used to classify water trophic status into multiple levels such as oligotrophic, mesotrophic, slightly eutrophic, moderately eutrophic, and severely eutrophic. The chlorophyll a concentration inversion value of each water cell in the study area is determined pixel-by-pixel, thereby generating a single-period eutrophication level classification map reflecting the spatial distribution of the water trophic status at the current moment.

[0119] In one specific implementation, constructing a continuous time-series eutrophication classification map includes the following steps: acquiring multiple eutrophication classification maps covering the same target water area at different times to construct time-series data on eutrophication evolution. For example, continuous remote sensing monitoring results can be acquired at fixed time intervals. As a preferred implementation, classification maps for 12 consecutive months, one for each month, can be acquired to form a time-series dataset describing the eutrophication evolution process of the target water area. ,in This represents the hierarchical classification chart at time t. This represents the total number of periods.

[0120] In one specific implementation, the method for constructing the eutrophication level transition probability matrix based on the eutrophication level classification map of continuous time series is as follows:

[0121] For two adjacent time phases and Statistics of all pixels from level Transfer to level frequency Then the transition probability It can be calculated as:

[0122] ;

[0123] in, The total number of eutrophication levels (in this embodiment) By calculating the transition probabilities for all level combinations, we can obtain... The transition probability matrix This matrix depicts the evolution of eutrophication levels over time.

[0124] Considering the fluidity and connectivity of water bodies, the eutrophication state of a water body is influenced not only by its own historical state but also by the surrounding water bodies. Therefore, this invention constructs a neighborhood spatial influence factor centered on a single pixel. In one specific implementation, a neighborhood window, such as a 3×3 or 5×5 pixel window, is constructed centered on each water body pixel, and the distribution ratio of each eutrophication level within this neighborhood is statistically analyzed to constitute the neighborhood state vector of that pixel. ,in Indicates the level within the neighborhood. The percentage of pixels. This vector quantifies the spatial influence of the surrounding water area on the central pixel.

[0125] The above level transition probability matrix Influence factors of neighboring space Coupled to a cellular automata model, the spatiotemporal evolution of eutrophication levels is simulated and predicted.

[0126] In a specific implementation, the basic elements of the cellular automata model are defined as follows:

[0127] Cell: Each water cell within the study area is considered an independent cell.

[0128] State: The state of each cell is the eutrophication level of that cell at a certain moment, denoted as . ,in For the total number of levels, For pixel coordinates, For a moment.

[0129] Neighborhood: Defines the neighborhood type of a cell. As a preferred option, a Mohr's neighborhood is used, which is a 3×3 window containing the center cell and its eight surrounding neighboring cells. For boundary cells, only their effective neighboring cells are considered.

[0130] Transition Rule: The state of a cell at the next time step is determined by the transition rule, which also incorporates the evolutionary pattern in the time dimension (derived from the transition probability matrix). (Characteristics) and the diffusion effect in the spatial dimension (by the influence factor of the neighborhood space) (Description).

[0131] In a preferred embodiment, the transformation rule construction process is as follows:

[0132] First, based on the transition probability matrix For the current state is From the cells, we can obtain the probability vectors of their transitions to each level. ,in To be from the level Transfer to level The probability of.

[0133] Secondly, based on the current moment Given the neighborhood state, calculate the neighborhood space influence factor of the cell. ,in For neighborhood level The proportion of the number of pixels in the effective water body of the neighborhood.

[0134] Then, the two vectors are coupled to obtain the combined influence vector. :

[0135]

[0136] in, To start from the current level The probability vector for transitioning to each level, derived from the constructed transition probability matrix. Extract from; These are weighting coefficients used to balance the influence of temporal evolution and spatial diffusion on the state at the next time step. In a preferred embodiment... The value ranges from 0.4 to 0.6, and is more preferably 0.5, indicating that temporal evolution and spatial diffusion are of equal importance. Those skilled in the art can determine the value based on the hydrodynamic characteristics of the target water body and the dominant driving factors of eutrophication through sensitivity analysis or model calibration. The optimal value.

[0137] Finally, cell In the next moment The state is determined by the following formula:

[0138] ;

[0139] in, Comprehensive influence vector The Each component has a maximum overall impact. The level with the greatest combined influence is selected as the predicted state for the next time step.

[0140] In another implementation, to avoid the simulation results from deterministic rules becoming too rigid (i.e., once the system forms a certain distribution pattern, subsequent evolution will follow a fixed path, making it difficult to reflect possible fluctuations and disturbances in reality), probabilistic rules can also be used: the comprehensive influence vector... After normalization, it is considered as the probability distribution of each level being selected, and the state at the next moment is randomly determined by roulette. This method can better simulate the inherent randomness and uncertainty in natural ecosystems, making the prediction results closer to the real evolution process.

[0141] In one specific implementation, the spatial diffusion, accumulation, and evolution of eutrophication levels are iteratively simulated to generate a predicted spatial distribution map of eutrophication at a predetermined future time point. This includes using the last period (i.e., the current latest phase) of the aforementioned continuous time series classification map as the initial state. Applying the above transformation rules, calculate the next time step cell by cell. The state is obtained to get a preliminary classification map. Then, using this prediction map as the new initial state, the transformation rules are repeatedly applied to iterate and simulate period by period to generate a spatial distribution prediction map of eutrophication level for future preset time nodes (such as the next quarter, the next year, or any specified time).

[0142] As a preferred output format, it can generate dynamic evolution diagrams of multi-period prediction results, intuitively displaying the diffusion path, aggregation areas, and evolution trend of eutrophication levels in time and space.

[0143] It should be noted that, based on the eutrophication spatial distribution prediction map obtained from the iterative simulation of the aforementioned cellular automata model at a preset time point, and combined with preset risk thresholds (for example, classifying areas predicted as moderate and severe eutrophication as risk areas), an eutrophication risk warning area distribution map is automatically generated. This warning map can overlay and display the spatial distribution of eutrophication levels during the prediction period, hotspots where levels are continuously rising, and automatically identify key lake areas recommended for priority and intensive monitoring. This provides intuitive and quantitative decision support for ecological and environmental management departments to formulate differentiated water body governance strategies and dynamic monitoring plans.

[0144] In summary, this invention constructs a radiative transfer model that couples the absorption backscattering coefficient of water components with the contribution of bottom sediment reflectivity. It then introduces the optical closure principle to construct the objective function and combines prior water depth information with an adaptive optical depth threshold for nonlinear iterative optimization. This achieves precise physical mechanism-level removal of bottom sediment reflection interference signals from the original remote sensing pixel spectrum, resulting in a pure and physically consistent effective water spectrum. This invention fundamentally solves the technical bottleneck of traditional methods in grassy, ​​clear water areas where bottom sediment is visible, leading to overestimation or misjudgment of chlorophyll concentration. It provides a high-quality, interference-free input data foundation for subsequent inversion steps. Based on the decoupled, pure spectra, this invention constructs an initial spectral index pool covering all possible band combinations. It automatically selects core spectral indices that are independent of each other and strongly correlated with chlorophyll a concentration using an adaptive pooling search strategy combined with collinearity diagnosis. A multivariate nonlinear mapping relationship is constructed based on Gaussian process regression. Simultaneously, leave-one-out cross-validation and maximum marginal likelihood estimation are introduced to adaptively determine model parameters. This effectively avoids the subjectivity and overfitting risk of traditional empirical index methods, significantly improving the adaptability and spatially continuous distribution estimation accuracy of chlorophyll a concentration inversion. This invention constructs a continuous time-series eutrophication level classification map and calculates the level transition probability matrix to characterize the temporal evolution pattern. Simultaneously, it constructs a neighborhood spatial influence factor centered on a pixel to quantify the spatial diffusion effect of surrounding water areas. These two elements are then coupled into a cellular automata model to establish a unified spatiotemporal transformation rule. Through iterative simulation, it achieves joint modeling and quantitative prediction of the spatial diffusion path, aggregation area, and evolution trend of eutrophication levels. This overcomes the limitations of traditional methods that can only perform single-point time-series analysis or simple spatial interpolation, providing a scientific basis for early warning and proactive prevention of eutrophication risks. Through the aforementioned "decoupling-inversion-prediction" three-in-one technical architecture, this invention forms a complete intelligent assessment chain from raw remote sensing data to the spatiotemporal evolution trend of eutrophication. In complex water environments such as shallow lakes, especially grassy clear water areas, it achieves effective suppression of sediment disturbance, accurate inversion of chlorophyll concentration, and dynamic prediction of eutrophication diffusion trends, significantly improving the accuracy, efficiency, and intelligence level of remote sensing monitoring of eutrophication in rivers and lakes.

[0145] Example 2, as Figure 5 This is the second embodiment of the present invention. Based on embodiment 1, this embodiment further provides a preferred implementation scheme, the core of which lies in the addition of an intelligent identification and diversion processing mechanism for water area types. After acquiring multispectral remote sensing images of the target water area, before performing shallow water sediment spectral decoupling processing on the images, this mechanism can automatically determine whether the target water area belongs to a "grass-type clear water area" with a risk of sediment interference based on the spectral characteristics of the water body, and dynamically activate or simplify the subsequent decoupling processing flow accordingly, thereby significantly improving the overall computational efficiency while ensuring high-precision inversion.

[0146] Specifically, the method also includes a step of identifying the type of target water area:

[0147] Calculate the ratio of water reflectance in the green light band to that in the red light band. and reflectivity in the near-infrared band ;

[0148] when Greater than the first preset threshold and If the water level is less than the second preset threshold, the water area is determined to be a grass-type clear water area with aquatic vegetation development. It is determined that the water area has the risk of spectral interference caused by the bottom sediment being exposed. When performing shallow water bottom sediment spectral decoupling processing, the bottom sediment contribution term in the water body radiation transfer coupling model is activated and the water depth prior information is introduced for constraint solution.

[0149] when Less than or equal to the first preset threshold, or If the chlorophyll content is greater than or equal to the second preset threshold, the water body is determined to be a non-grass-type clear water body, and a standard water body optical model (such as the QAA algorithm) is used for chlorophyll inversion. Because the sediment disturbance in such water bodies is not significant or has been masked by other factors (such as high turbidity), there is no need for complex sediment decoupling treatment.

[0150] in, The reflectivity in the green light band (e.g., the B3 band of Sentinel-2 MSI, with a center wavelength of approximately 560 nm), This is the reflectance in the red light band (e.g., the B4 band, with a center wavelength of approximately 665 nm). This ratio is quite sensitive to the development of aquatic vegetation (especially submerged plants) in the water. When aquatic vegetation is well-developed, this ratio usually increases significantly due to the reflection peak in the green light band and the absorption trough in the red light band.

[0151] Among them, the reflectivity of the near-infrared band The reflectance was measured using the near-infrared band (B8a, center wavelength approximately 865 nm) of the Sentinel-2 MSI. Clean water exhibits strong absorption in the near-infrared band, resulting in typically low reflectance; however, when a large amount of suspended matter or aquatic vegetation is present above the water surface, the near-infrared reflectance increases significantly. Therefore, It can be used as an auxiliary indicator to determine the turbidity of water bodies or whether there is disturbance from emergent vegetation.

[0152] As an optional implementation method, statistical analysis is performed based on water body type survey data and synchronous spectral data from a large number of measured sampling points within the study area. Specifically, several sample points known as "grass-type clear water areas" and "non-grass-type clear water areas" can be collected, and their statistical analysis is conducted. and The distribution range is determined by analyzing the subject operating characteristic curve to identify the optimal first and second preset thresholds.

[0153] In a preferred embodiment, the first preset threshold value ranges from 2.0 to 3.0. For example, for Sentinel-2 MSI images, when When the temperature is >2.5, it can be preliminarily determined that there is an impact from aquatic vegetation.

[0154] In a preferred embodiment, the second preset threshold value ranges from 0.02 to 0.05 (reflectance value, dimensionless). For example, when When the value is less than 0.03, the water turbidity can be considered low, and the near-infrared band is not significantly disturbed.

[0155] Those skilled in the art will understand that the above thresholds are merely preferred examples based on typical situations, and in practical applications, the thresholds can be adaptively adjusted according to specific sensor band settings, the optical characteristics of the water body in the study area, and field survey data.

[0156] In summary, the intelligent water type identification and diversion processing mechanism added in Embodiment 2 enables the identification of water types and the processing of diversions before shallow water sediment spectral decoupling. and By jointly judging two key spectral features and their preset thresholds, the system accurately identifies "grass-type clear water areas" with potential bottom sediment interference risks and activates a complete radiative transfer coupling model for fine decoupling as needed. Meanwhile, for "non-grass-type clear water areas" without interference, the system automatically simplifies the processing flow and uses a standard optical model for rapid inversion. This significantly reduces the overall computational complexity and processing time while ensuring high-precision eutrophication assessment, achieving optimized allocation of computing resources and synergistic improvement in monitoring efficiency.

[0157] Example 3 is the third embodiment of the present invention. This embodiment differs from the previous two in that it provides a spectral classification and assessment system for eutrophication remote sensing images of rivers and lakes, applied to the aforementioned spectral classification and assessment method for eutrophication remote sensing images of rivers and lakes, including:

[0158] The image acquisition and preprocessing module is used to acquire multispectral images and perform geometric and radiometric corrections.

[0159] The water body type identification and decoupling activation module, connected to the image acquisition and preprocessing module, is used to calculate the reflectance ratio of the water body in the green and red bands and the reflectance in the near-infrared band to identify the water body type. When it is determined to be a grassy clear water body, the bottom sediment contribution term of the water body radiative transfer coupling model in the bottom sediment interference decoupling module is activated and the water depth prior information is introduced. When it is determined to be a non-grassy clear water body, the preprocessed image is directly transmitted to the adaptive spectral analysis and inversion module or processed using the standard model.

[0160] The sediment interference decoupling module has a built-in water radiative transfer model and nonlinear iterative solver, which is used to strip the sediment reflection signal and output the pure water spectrum.

[0161] The adaptive spectral analysis and inversion module includes a dynamic spectral index search unit and a Gaussian process regression unit. It is used to receive pre-processed images of pure water spectra or non-grass-type clear water areas. Based on the core spectral indices selected by the dynamic spectral index search unit, a chlorophyll a concentration distribution map is generated through Gaussian process regression.

[0162] The spatiotemporal evolution analysis and early warning module integrates Markov chain and cellular automata models to simulate the eutrophication spread trend and generate early warning information.

[0163] Example 4 is the fourth embodiment of the present invention. This embodiment provides a computer-readable storage medium storing a computer program thereon. When the program is executed by a processor, it implements the steps of the spectral classification assessment method for eutrophication of river and lake water bodies using remote sensing images as described in any one of Examples 1 to 2.

[0164] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for spectral grading evaluation of eutrophication of water bodies in rivers and lakes based on remote sensing images, characterized in that, Includes the following steps: The shallow water bottom sediment spectral decoupling step based on the radiative transfer mechanism is used to acquire multispectral remote sensing images of the target water area. A water radiative transfer coupling model is constructed that includes the backscattering coefficient of water components and the contribution of bottom sediment reflectivity. Prior water depth information is introduced to constrain the solution of the model, so as to remove the interference signal affected by the bottom sediment from the original pixel spectrum, thereby extracting the effective water spectrum that only represents the inherent optical properties of the water body. The adaptive quantitative inversion step of chlorophyll concentration based on decoupled spectrum involves constructing a spectral band feature space based on the extracted effective water body spectrum, dynamically searching for the band combination most sensitive to changes in chlorophyll a concentration within this feature space, and establishing a nonlinear mapping relationship based on the optimal band combination found to obtain the inversion value of chlorophyll a concentration for each pixel. The eutrophication evolution trend prediction step based on spatiotemporal continuity involves classifying eutrophication levels according to the inversion value of chlorophyll a concentration, and constructing a spatiotemporal propagation model that integrates spatial neighborhood influences and time series change characteristics by combining multi-temporal remote sensing monitoring data to predict the future eutrophication state of water bodies.

2. The method for remote sensing image spectral grading evaluation of eutrophication of river-lake water bodies according to claim 1, characterized in that: The shallow water sediment spectral decoupling step specifically includes: Based on the radiative transfer theory, the total water-leaving reflectance received by the sensor is expressed as the sum of the contributions of the volume scattering representing the internal optical properties of the water body and the reflectance of the bottom after attenuation through the water layer and the introduction of the optical closure condition to construct the objective function; The absorption coefficient and backscattering coefficient of water components are initialized using the spectrum of deep water regions in the image where the water depth is greater than the preset optical depth. On this basis, combined with the pixel relative water depth value estimated by the band ratio method, the objective function is solved by a nonlinear iterative optimization algorithm, and in the solving process The final effective water body spectrum is obtained by stripping from the total reflectance one by one until the optical closure error threshold is met .

3. The method for remote sensing image spectral grading evaluation of eutrophication of river-lake water bodies according to claim 1, characterized in that: The adaptive quantitative inversion step for chlorophyll concentration specifically includes: Construct an initial spectral index pool consisting of all possible two-band ratios, three-band indices, and normalized difference indices; Based on the effective water body spectrum and synchronously measured chlorophyll a concentration data, an adaptive pooling search strategy is adopted to traverse the spectral index pools and calculate the correlation coefficient between each index and chlorophyll a concentration. The top N indices with the highest correlation coefficients were selected as a candidate set. Collinearity diagnosis was performed on the candidate set to eliminate redundant indices. Finally, a set of core spectral indices that are independent of each other and strongly correlated with chlorophyll concentration were retained. A multivariate nonlinear regression model was constructed using this core spectral index, and the model structure parameters were determined using leave-one-out cross-validation.

4. The method for remote sensing image spectral grading evaluation of eutrophication of river-lake water bodies according to claim 1, characterized in that: The steps for predicting the eutrophication evolution trend specifically include: Based on the eutrophication levels obtained from the inversion values ​​of chlorophyll a concentration, a eutrophication level classification map for a continuous time series is constructed. Based on the eutrophication level classification map of the continuous time series, a eutrophication level transition probability matrix is ​​constructed; at the same time, considering water flow and connectivity, a neighborhood spatial influence factor centered on the pixel is constructed. The transition probability matrix and the neighborhood space influence factor are coupled into a cellular automata model. By iteratively simulating the diffusion, aggregation and evolution of eutrophication levels in space, a eutrophication spatial distribution prediction map for future preset time nodes is generated.

5. The method for spectral classification and assessment of eutrophication in river and lake water bodies using remote sensing images as described in claim 2 is characterized by: The effective water body spectrum obtained by stripping the substrate signal through a nonlinear iterative optimization algorithm As input data, for performing the adaptive spectral index search and chlorophyll retrieval.

6. The method for remote sensing image spectral hierarchical assessment of eutrophication of river-lake water bodies according to claim 2, characterized in that: The relative water depth value of the pixel estimated by the band ratio method specifically adopts the ratio of the water body absorption band that is not sensitive to changes in the shape of the bottom sediment spectrum to the water body weak absorption band that is sensitive to changes in water depth. Combined with the adaptive threshold judgment of optical depth, this ratio is used to estimate water depth only when the optical depth of the pixel is less than the preset threshold, so as to avoid invalid inversion state caused by the lack of bottom sediment signal in deep water area or the interference of water scattering in high turbidity area.

7. The method for remote sensing image spectral hierarchical assessment of eutrophication of river-lake water bodies as claimed in claim 3, characterized in that: The multivariate nonlinear regression model is a Gaussian process regression model; when training the Gaussian process regression model using the core spectral index, its kernel function parameters are adaptively determined by maximizing the marginal likelihood estimation.

8. The method for remote sensing image spectral hierarchical assessment of eutrophication of river-lake water bodies according to claim 1, characterized in that: After acquiring multispectral remote sensing images of the target water area, and before performing shallow water sediment spectral decoupling processing on the images, the method further includes a step of identifying the type of the target water area: The ratio of reflectance of the water body in the green light band and the red light band , and the reflectance in the near-infrared band ; When the Greater than the first preset threshold and the If the water level is less than the second preset threshold, the water area is determined to be a grass-type clear water area with aquatic vegetation development. It is determined that the water area has a risk of spectral interference due to the bottom sediment being exposed. When performing the shallow water bottom sediment spectral decoupling process, the bottom sediment contribution term in the water body radiation transfer coupling model is activated and the water depth prior information is introduced for constraint solution. When the Less than or equal to the first preset threshold, or the If the value is greater than or equal to the second preset threshold, the water area is determined to be a non-grass-type clear water area, and chlorophyll inversion is performed using a standard water body optical model.

9. A spectral classification and assessment system for eutrophication remote sensing images of rivers and lakes, applied to the spectral classification and assessment method for eutrophication remote sensing images of rivers and lakes as described in any one of claims 1-8, characterized in that, include: The image acquisition and preprocessing module is used to acquire multispectral images and perform geometric and radiometric corrections. The water body type identification and decoupling activation module, connected to the image acquisition and preprocessing module, is used to calculate the reflectance ratio of the water body in the green light band and the red light band, as well as the reflectance in the near-infrared band, to identify the water body type. When it is determined to be a grassy clear water body, the bottom sediment contribution term of the water body radiation transfer coupling model in the bottom sediment interference decoupling module is activated and prior water depth information is introduced. When it is determined to be a non-grassy clear water body, the preprocessed image is directly transmitted to the adaptive spectral analysis and inversion module or processed using the standard model. The sediment interference decoupling module has a built-in water radiative transfer model and nonlinear iterative solver, which is used to strip the sediment reflection signal and output the pure water spectrum. The adaptive spectral analysis and inversion module includes a dynamic spectral index search unit and a Gaussian process regression unit. It is used to receive the pre-processed images of the pure water body spectrum or the non-grass-type clear water area, and generate a chlorophyll a concentration distribution map based on the core spectral indices selected by the dynamic spectral index search unit through Gaussian process regression. The spatiotemporal evolution analysis and early warning module integrates Markov chain and cellular automata models to simulate the eutrophication spread trend and generate early warning information.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by the processor, the program implements the steps of the spectral classification assessment method for eutrophication of river and lake water bodies using remote sensing images as described in any one of claims 1 to 8.

Images