Method for analyzing and characterizing water body eutrophication components based on hyperspectral feature inversion

Abstract

The application relates to the technical field of water environment optical monitoring, and discloses a water body eutrophication component analysis and characterization method based on hyperspectral characteristic inversion, which comprises the following steps: calling a pure water intrinsic absorption spectrum sequence as a physical constraint benchmark, calculating the ratio of a to-be-measured hyperspectral reflectivity to the benchmark to generate a modulation vector, and determining a sequence of fractional differential orders related to a sampling wavelength; trough characteristics are extracted through morphological baseline correction, fractional differential transformation is completed in combination with the sequence of orders, feature trough depth and skewness parameters are extracted, and chlorophyll a and colored soluble organic matter concentrations are output by using an unmixing model; the benchmark is used to drive differential operator dynamic evolution, topological stripping of different waveband scattering backgrounds and trace component absorption characteristics is realized, local waveform distortion caused by background scattering heterogeneity is inhibited, and component inversion precision under a complex matrix is improved.

Description

Technical Field

[0001] This invention relates to a method for analyzing and characterizing eutrophication components in water bodies based on hyperspectral feature inversion, belonging to the field of optical monitoring technology for water environment. Background Technology

[0002] Currently, the mainstream method for water environment monitoring is to collect water reflectance sequences using hyperspectral remote sensing technology to analyze the concentration of components such as chlorophyll a and colored soluble organic matter. In the physical-optical analysis of eutrophic water bodies, photons are subjected to Mie scattering by suspended particulate matter, resulting in energy attenuation, and the scattering phase function dynamically evolves with wavelength.

[0003] In complex water environments such as inland shallow lakes, high concentrations of non-organic suspended matter cause nonlinear elevation of the spectral baseline, severely masking the intrinsic absorption signals of trace components in the red and near-infrared bands. Conventional baseline stripping schemes often use integer derivatives or fractional derivative algorithms based on scalar assumptions to process spectral sequences. Because they ignore the heterogeneity of water's optical properties across wavelengths, this global processing logic, which treats the spectral sequence as a homogeneous space, easily induces topological distortions of the physical signal in the short-wavelength range and struggles to effectively eliminate scattering background interference in the long-wavelength range. Existing spectral data processing methods have shortcomings at the algorithmic level. For example, the method described in publication CN117892633A... A Chinese invention patent application discloses a hyperspectral water quality parameter inversion model based on FOD and optimal spectral features. It uses fractional-order differential processing and Pearson correlation coefficient to screen characteristic bands and uses a deep forest model for inversion. This scheme belongs to a data-driven modeling logic. The underlying algorithm relies on the statistical correlation of a specific sample set. Under the working conditions of inland turbid water bodies, strong scattering background and absorption signals severely overlap in the spectral dimension. The optimal bands selected by simple mathematical statistical laws contain spurious correlation components and lack physical uniqueness. The global fixed-order differential transformation adopted does not consider the heterogeneity of water body optical properties with wavelength. When dealing with dynamic evolution nonlinear baseline drift, it cannot simultaneously improve the signal-to-noise ratio and preserve the physical characteristics.

[0004] Therefore, the technical problem to be solved by this invention is how to provide a method for analyzing eutrophication components in water bodies that can suppress strong scattering background interference and faithfully extract the intrinsic absorption characteristics of trace components. Summary of the Invention

[0005] To address the problems mentioned in the background art, the technical solution of the present invention is as follows: A method for analyzing and characterizing eutrophication components of water bodies based on hyperspectral feature inversion, comprising the following steps: Step 101: Obtain the hyperspectral reflectance sequence of the water body to be tested, and retrieve the pre-stored intrinsic absorption spectrum sequence of pure water containing discrete data nodes. Step 102: Traverse the wavelength sampling points in the hyperspectral reflectance sequence, calculate the ratio between the reflectance value of each wavelength sampling point and the value of the discrete data node of the corresponding wavelength sampling point in the pre-stored pure water intrinsic absorption spectrum sequence, and generate a modulation vector characterizing the local spectral response shift rate. Step 103: Map the modulation vector to a preset initial order interval within the range of 0 to 2 to generate a fractional differential order sequence with the same dimension as the hyperspectral reflectance sequence and wavelength compliance. Step 104: Perform morphological baseline correction on the hyperspectral reflectance sequence to extract absorption trough features, and combine the fractional derivative sequence with the morphologically baseline-corrected sequence to perform fractional derivative transformation on the sequence to generate a fractional characteristic spectral sequence. Step 105: Define the first characteristic trough interval corresponding to chlorophyll a and the second characteristic trough interval corresponding to colored soluble organic matter in the fractional-order characteristic spectral sequence, and extract the maximum absorption depth parameter and the peak skewness parameter characterizing the trough waveform deformation. Step 106: Input the maximum absorption depth parameter and the peak skewness parameter into the pre-calibrated partial least squares unmixing model, and output the mass concentration of chlorophyll a and colored soluble organic matter in the water body to be tested.

[0006] Preferably, generating the modulation vector in step 102 includes: mapping the hyperspectral reflectance sequence to the spectral feature space to extract the baseline offset component; performing normalized sensitivity analysis on the baseline offset component using a pre-stored pure water intrinsic absorption spectrum sequence to determine the energy attenuation ratio of the hyperspectral reflectance sequence at each wavelength, and determining the energy attenuation ratio as the weighting coefficient of the modulation vector at the corresponding wavelength sampling point.

[0007] Preferably, generating the fractional-order differential sequence in step 103 includes: establishing a linear mapping relationship between the modulation vector and the differential order; according to the linear mapping relationship, when the modulation vector reflects a decrease in the local spectral response offset rate, decreasing the differential order at the corresponding wavelength position; when the modulation vector reflects an increase in the local spectral response offset rate, increasing the differential order at the corresponding wavelength position, so as to form a fractional-order differential sequence distributed between 0 and 2.

[0008] Preferably, the morphological baseline correction in step 104 includes: performing morphological opening and closing operations on the hyperspectral reflectance sequence using structuring elements to extract the trend background curve of the hyperspectral reflectance sequence; subtracting the trend background curve from the hyperspectral reflectance sequence to eliminate the interference of water surface flares on feature extraction.

[0009] Preferably, the eigenvalue S(λ) at the wavelength sampling point λ in the fractional-order characteristic spectral sequence satisfies the following logical rule: ,in, R(λ) is a differential operator determined based on the order value V(λ) in the fractional-order differential order sequence, R(λ) is the reflectance value after morphological baseline correction, and M(λ) is the amplitude of the modulation vector at the wavelength sampling point λ.

[0010] Preferably, step 105, defining the first characteristic valley interval and the second characteristic valley interval, includes: performing second-order differential zero-point detection on the fractional characteristic spectral sequence to identify the extreme value position of the spectral curve; determining the interval from wavelength 665nm to 685nm as the first characteristic valley interval, and determining the interval from wavelength 410nm to 440nm as the second characteristic valley interval.

[0011] Preferably, the extraction of the peak skewness parameter includes: calculating the third central moment of the spectral curve within the first characteristic trough interval or the second characteristic trough interval; performing a ratio operation between the third central moment and the cube of the standard deviation of the spectral values ​​within the corresponding interval to obtain a dimensionless parameter reflecting the deviation of the absorption characteristics from the symmetry.

[0012] Preferably, the partial least squares unmixing model is calibrated in the following way: obtaining a standard water sample spectral dataset with known concentrations; performing steps 101 to 105 on the standard water sample spectral dataset to obtain training feature vectors; mapping the training feature vectors to a linearly separable space using a kernel function; and determining the weight matrix of the partial least squares unmixing model by minimizing the sum of squared residuals between the predicted concentration and the true value.

[0013] Preferably, after outputting the mass concentrations of chlorophyll a and colored soluble organic matter in the water body to be tested, the method further includes step 107: calculating the trophic status index, which characterizes the degree of water pollution, based on the mass concentrations of chlorophyll a and colored soluble organic matter in the water body to be tested; and outputting an early warning command when the trophic status index exceeds a preset risk threshold.

[0014] Preferably, when obtaining the hyperspectral reflectance sequence in step 101, the raw radiance data is collected using an imaging spectrometer deployed above the water surface; reflectance calibration is performed by introducing synchronously observed standard plate irradiance data to eliminate the interference of light environment fluctuations on spectral inversion and ensure that the input reflectance value is within the physical range of 0 to 1.

[0015] Compared with the prior art, the beneficial effects of the present invention are: 1. In the analysis of eutrophication components in water bodies, the intrinsic absorption of pure water is used as a physical constraint benchmark to construct a fractional-order array that dynamically evolves with wavelength. This mechanism replaces the traditional globally unified mathematical transformation method and implements dimensional distortion compensation for the differences in the physical penetration depth of photons in different bands. It effectively separates the scattering interference of broadband and the narrow-band absorption characteristics of trace components in the topological dimension, avoids local feature distortion caused by the heterogeneity of scattering intensity spectrum, and ensures the fidelity of the extraction of intrinsic optical signals of materials in complex matrix environments.

[0016] 2. By introducing morphological filtering based on envelope constraints before performing topological transformation, this invention constructs a preprocessing chain that balances background smoothing and feature preservation. This step works in conjunction with fractional derivative operations to filter out random noise and surface flare interference while maintaining the overall geometric topological shape of the absorption valley. This mechanistic combined effect breaks the inherent contradiction between signal truncation and noise amplification in traditional derivative processing, providing a feature input with a high signal-to-noise ratio for subsequent nonlinear demixing of high-dimensional data, and enhancing the monitoring stability of the system in extremely turbid environments.

[0017] 3. By extracting the relative absorption depth of characteristic valleys and the asymmetric skewness parameter of peaks, a multi-dimensional topological parameter joint characterization mechanism is formed. By utilizing the asymmetric deformation characteristics of absorption peaks caused by the overlap of material components, deep decoupling of components with severely overlapping spectral features is achieved. This logical shift from single-spectrum inversion to multi-dimensional morphological characterization solves the nonlinear interference and optical crosstalk problems existing in the inversion of multi-component water bodies, and improves the stability and environmental adaptability of simultaneous quantitative analysis of multi-components in complex fluid environments. Attached Figure Description

[0018] Figure 1 This is a flowchart of the quantitative analysis process for eutrophication components in water bodies derived from hyperspectral feature inversion according to the present invention. Figure 2 This is a diagram of the morphological baseline correction and fractional derivative processing architecture for hyperspectral reflectance of the present invention.

[0019] The objectives, features, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0020] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0021] A method for analyzing and characterizing eutrophication components in water bodies based on hyperspectral feature inversion includes the following steps: Step 101: Obtain the hyperspectral reflectance sequence of the water body to be tested, and retrieve the pre-stored intrinsic absorption spectrum sequence of pure water containing discrete data nodes. Step 102: Traverse the wavelength sampling points in the hyperspectral reflectance sequence, calculate the ratio between the reflectance value of each wavelength sampling point and the value of the discrete data node of the corresponding wavelength sampling point in the pre-stored pure water intrinsic absorption spectrum sequence, and generate a modulation vector characterizing the local spectral response shift rate. Step 103: Map the modulation vector to a preset initial order interval within the range of 0 to 2 to generate a fractional differential order sequence with the same dimension as the hyperspectral reflectance sequence and wavelength compliance. Step 104: Perform morphological baseline correction on the hyperspectral reflectance sequence to extract absorption trough features, and combine the fractional derivative sequence with the morphologically baseline-corrected sequence to perform fractional derivative transformation on the sequence to generate a fractional characteristic spectral sequence. Step 105: Define the first characteristic trough interval corresponding to chlorophyll a and the second characteristic trough interval corresponding to colored soluble organic matter in the fractional-order characteristic spectral sequence, and extract the maximum absorption depth parameter and the peak skewness parameter characterizing the trough waveform deformation. Step 106: Input the maximum absorption depth parameter and the peak skewness parameter into the pre-calibrated partial least squares unmixing model, and output the mass concentration of chlorophyll a and colored soluble organic matter in the water body to be tested.

[0022] Preferably, generating the modulation vector in step 102 includes: mapping the hyperspectral reflectance sequence to the spectral feature space to extract the baseline offset component; performing normalized sensitivity analysis on the baseline offset component using a pre-stored pure water intrinsic absorption spectrum sequence to determine the energy attenuation ratio of the hyperspectral reflectance sequence at each wavelength, and determining the energy attenuation ratio as the weighting coefficient of the modulation vector at the corresponding wavelength sampling point.

[0023] Preferably, generating the fractional-order differential sequence in step 103 includes: establishing a linear mapping relationship between the modulation vector and the differential order; according to the linear mapping relationship, when the modulation vector reflects a decrease in the local spectral response offset rate, decreasing the differential order at the corresponding wavelength position; when the modulation vector reflects an increase in the local spectral response offset rate, increasing the differential order at the corresponding wavelength position, so as to form a fractional-order differential sequence distributed between 0 and 2.

[0024] Preferably, the morphological baseline correction in step 104 includes: performing morphological opening and closing operations on the hyperspectral reflectance sequence using structuring elements to extract the trend background curve of the hyperspectral reflectance sequence; subtracting the trend background curve from the hyperspectral reflectance sequence to eliminate the interference of water surface flares on feature extraction.

[0025] Preferably, the eigenvalue S(λ) at the wavelength sampling point λ in the fractional-order characteristic spectral sequence satisfies the following logical rule: ,in, R(λ) is a differential operator determined based on the order value V(λ) in the fractional-order differential order sequence, R(λ) is the reflectance value after morphological baseline correction, and M(λ) is the amplitude of the modulation vector at the wavelength sampling point λ.

[0026] Preferably, step 105, defining the first characteristic valley interval and the second characteristic valley interval, includes: performing second-order differential zero-point detection on the fractional characteristic spectral sequence to identify the extreme value position of the spectral curve; determining the interval from wavelength 665nm to 685nm as the first characteristic valley interval, and determining the interval from wavelength 410nm to 440nm as the second characteristic valley interval.

[0027] Preferably, the extraction of the peak skewness parameter includes: calculating the third central moment of the spectral curve within the first characteristic trough interval or the second characteristic trough interval; performing a ratio operation between the third central moment and the cube of the standard deviation of the spectral values ​​within the corresponding interval to obtain a dimensionless parameter reflecting the deviation of the absorption characteristics from the symmetry.

[0028] Preferably, the partial least squares unmixing model is calibrated in the following way: obtaining a standard water sample spectral dataset with known concentrations; performing steps 101 to 105 on the standard water sample spectral dataset to obtain training feature vectors; mapping the training feature vectors to a linearly separable space using a kernel function; and determining the weight matrix of the partial least squares unmixing model by minimizing the sum of squared residuals between the predicted concentration and the true value.

[0029] Preferably, after outputting the mass concentrations of chlorophyll a and colored soluble organic matter in the water body to be tested, the method further includes step 107: calculating the trophic status index, which characterizes the degree of water pollution, based on the mass concentrations of chlorophyll a and colored soluble organic matter in the water body to be tested; and outputting an early warning command when the trophic status index exceeds a preset risk threshold.

[0030] Preferably, when obtaining the hyperspectral reflectance sequence in step 101, the raw radiance data is collected using an imaging spectrometer deployed above the water surface; reflectance calibration is performed by introducing synchronously observed standard plate irradiance data to eliminate the interference of light environment fluctuations on spectral inversion and ensure that the input reflectance value is within the physical range of 0 to 1.

[0031] Example 1: In the specific inland large shallow lake monitoring conditions of this invention, due to the high turbidity interference caused by rainstorm runoff, the total suspended solids concentration of the water body being tested reaches more than 100 mg / L. At this time, the original hyperspectral reflectance sequence in the 600 nm to 700 nm band experiences nonlinear baseline elevation due to Mie scattering by suspended particles. This elevation phenomenon masks the characteristic absorption signal of chlorophyll a near 675 nm. Traditional fixed-order differential processing, when eliminating such broadband scattering baselines, often induces noise amplification in the short-wavelength range and in the long-wavelength range. The loss of effective absorption characteristics due to intervals makes it difficult to maintain the inversion accuracy of trace components in complex matrices. The system acquires the hyperspectral reflectance sequence R(λ) of the water body to be tested and retrieves the intrinsic absorption spectrum sequence of pure water pre-stored in the storage unit. It traverses each discrete wavelength sampling point λ in the hyperspectral reflectance sequence R(λ), calculates the ratio between the reflectance value of the sampling point and the corresponding intrinsic absorption spectrum sequence value of pure water, and generates a modulation vector M(λ) characterizing the local spectral response shift rate. Based on this modulation vector M(λ), a linear relationship is established between the differential order and the [missing information]. The mapping relationship is as follows: when the modulation vector M(λ) reflects a decrease in the local spectral response shift rate, the system decreases the differential order at the corresponding wavelength position; when the modulation vector M(λ) reflects an increase in the local spectral response shift rate, the system increases the differential order at the corresponding wavelength position. This generates a fractional differential order sequence V(λ) distributed between 0 and 2 with wavelength dependence. The dimension of this sequence V(λ) is consistent with the hyperspectral reflectance sequence R(λ). By introducing the inherent physical absorption of pure water as a constraint benchmark, the order evolution of the differential operator is transformed by pure mathematics. The process is transformed into a dynamic compensation process for the differences in the physical penetration depth of photons in different wavelength bands. Before the transformation, the system uses structuring elements to perform morphological opening and closing operations on the hyperspectral reflectance sequence R(λ), extracts the background curve reflecting the trend, and subtracts the background curve from the hyperspectral reflectance sequence R(λ) to eliminate the influence of water surface flares on signal extraction. Combined with the fractional derivative sequence V(λ), the corrected sequence is subjected to fractional derivative transformation to generate a fractional characteristic spectral sequence S(λ). The fractional characteristic spectral sequence S(λ) satisfies the following logical rules: ,in, (λ) is the fractional eigenvalue at the wavelength sampling point λ, DV(λ) is the differential operator determined based on the order value V(λ) in the fractional differential order sequence, R(λ) is the reflectance value after morphological baseline correction, and M(λ) is the amplitude of the modulation vector at the wavelength sampling point λ.

[0032] After performing fractional-order differential transformation, the processor locates the trough position by finding the local minimum point of the fractional-order eigenvalue S(λ) within the characteristic band. The maximum absorption depth parameter is obtained by calculating the absolute difference between the fractional-order eigenvalue at the characteristic trough position and the fractional-order eigenvalue at the starting boundary of the band. This calculation process eliminates the vertical level drift caused by the residual scattering baseline, ensuring that the extracted depth parameter is modulated only by the absorption intensity of the target component's electronic transitions. This achieves linear anchoring of the chemical component concentration within the dimensionless transformation space. The system calculates the third central moment of the spectral curve within the first or second characteristic trough interval and performs a ratio operation with the cube of the standard deviation of the spectral values ​​within the corresponding interval to obtain the peak skewness parameter reflecting the deviation of the absorption feature from the symmetry. Before performing second-order differential zero-point detection, the processor activates a median filter operator with a window length of 11 sampling points to denoise the fractional-order characteristic spectral sequence. The zero-point identification logic is set as follows: when the numerical signs of three consecutive wavelength nodes change from zero to zero... When the slope at the intermediate node is greater than 0.0005 mV / nm, it is determined to be the physical absorption valley center. If the detected zero point is outside the 665nm to 685nm range, a secondary optimization is automatically triggered to find the node with the smallest eigenvalue in this range as the corrected characteristic valley point. This ensures that the correct absorption characteristic site can still be locked even under extremely harsh conditions with a signal-to-noise ratio of less than 20dB. The peak skewness parameter is used to capture the asymmetric deformation characteristics caused by the overlapping of spectra of different material components, thereby decoupling the overlapping characteristic components. The maximum absorption depth parameter and the peak skewness parameter are input into a pre-calibrated partial least squares unmixing model to output the mass concentration of chlorophyll a and colored soluble organic matter in the water body to be tested. This technical solution cuts off the nonlinear coupling between the apparent scattering effect and the intrinsic absorption signal through the synergy of physical constant constraints and dynamic order arrays, and realizes the accurate extraction of the distribution of target chemical components in complex optical matrices without increasing the underwater detection hardware.

[0033] Example 2: In the verification of eutrophication levels in high-turbidity landscape water bodies in inland semi-arid regions, the high concentration of mineral-derived suspended matter and humus in the water caused waveform distortion in the spectral energy distribution of the 400nm to 900nm band due to non-uniform scattering interference. This experiment was used to verify the stability of the experimental group using the method of this invention in determining the mass concentration of chlorophyll a when dealing with high-turbidity noise interference. The experimental platform used a device with hyperspectral data acquisition capabilities. The core sensor of this device adopted a holographic diffraction grating spectroscopic system, with a spectral range covering 400nm to 100nm. The sampling interval was set to 1.0 nm, the spectral resolution was better than 2.5 nm, and the signal-to-noise ratio was not less than 500 dB. The original reflectance data used in the experiment came from the measured sequence collected by the physical water quality monitoring platform, and the component content determined by the standard chemical analysis method was used as the true value. The sampling period was set based on the signal spectrum bandwidth analysis results. In order to balance the real-time performance of the data and the computational load of the system, the Nyquist sampling theorem was used to determine that the sampling frequency should be higher than twice the effective characteristic bandwidth of the signal. The sampling period was set to 120 ms, so as to suppress random readout noise while ensuring the real-time performance of the data.

[0034] The experiment constructs a complex working environment by actively superimposing additive white Gaussian noise with a signal-to-noise ratio of 25 dB and simulated water surface flare disturbances with an intensity of 15% onto the original reflectance sequence. The processor reads the hyperspectral reflectance sequence R(λ) of the water body under test and retrieves the intrinsic absorption spectrum sequence of pure water composed of discretized high-purity water absorption coefficients pre-stored in the storage unit. The system traverses the wavelength sampling points λ, calculates the ratio of the reflectance value of each sampling point to the corresponding position value of the intrinsic absorption spectrum sequence of pure water, and generates a modulation vector M(λ) characterizing the local spectral response shift rate. In this process, the amplitude distribution of the modulation vector M(λ) reflects the water scattering background as... The wavelength evolution trend shows that when the total suspended matter concentration increases from 10.2 mg / L to 152.5 mg / L, the modulation vector M(λ) at 675 nm increases from 1.15 to 4.62. Based on the established linear mapping relationship, the system converts the change of the modulation vector M(λ) into a fractional derivative order sequence V(λ) in real time. In the test sample with the highest suspended matter concentration, the derivative order V(λ) corresponding to the 600 nm to 700 nm band is automatically compensated from 0.22 to 1.78. Through this wavelength-related dynamic order adjustment mechanism, the filtering efficiency of the differential operator is focused on the band with strong scattering characteristics.

[0035] Data from the experimental group showed that after morphological baseline correction and transformation using a fractional-order differential sequence V(λ), the generated characteristic spectral sequence S(λ) exhibited a clear chlorophyll a absorption peak at 675 nm, with a signal contrast 3.25 times higher than the original sequence. By calculating the peak skewness parameter in the characteristic trough region, the asymmetric waveform shift caused by component overlap was captured. In the control group, the traditional first-order integer-order differential method failed to balance baseline elimination and characteristic preservation, resulting in chlorophyll a deficiency when the total suspended matter concentration exceeded 100.5 mg / L. The root mean square error (RMSE) of the inversion of element a mass concentration deteriorated from 1.48 mg / L to 8.92 mg / L, indicating a performance failure inflection point. In contrast, the RMSE of the sample group of this invention under the same operating conditions remained between 1.12 mg / L and 2.15 mg / L, and its inversion accuracy did not change by an order of magnitude with the increase of suspended solids concentration. The experimental results confirm that the differential operator evolution mechanism driven by the intrinsic absorption benchmark of pure water can offset the execution bias caused by background heterogeneity and realize the reliable extraction of the intrinsic absorption characteristics of the target component in the high turbidity optical matrix.

[0036] Example 3: When a water quality monitoring sensor has been deployed for over 180 days, the luminous efficiency of the light source component decays, and biological adhesion to the optical window causes a reference shift in the hyperspectral reflectance sequence. The system performs periodic calibration to reconstruct the intrinsic absorption spectrum sequence of pure water. The processor controls the sampling mechanism to acquire ultrapure water samples filtered through a 0.2 μm pore size membrane, and collects the hyperspectral reflectance of these ultrapure water samples, which is then used as the current physical constraint reference. During the generation of the fractional-order differential sequence V(λ), the system prepares standard scattering gradient samples with total suspended solids concentrations ranging from 0 mg / L to 200 mg / L under controlled conditions, measuring the modulation vector M(λ) of each sample across the entire wavelength range. Near the chlorophyll a intrinsic absorption band at 675 nm, the differential order V(λ) exhibits a certain variation with the modulation vector M(λ). Monotonically positively correlated, the processor determines the mapping parameters according to the following formula: V(λ)=αM(λ)+β, where V(λ) is the fractional derivative order at the wavelength sampling point λ, α is the mapping scaling factor with a value of 0.15; β is the basic offset with a value of 0.35; M(λ) is the amplitude of the modulation vector at the current sampling point. The mapping parameters between the fractional derivative order sequence V(λ) and the modulation vector M(λ) are determined. Standard gradient turbidity water samples with a range of 10NTU to 200NTU are selected, and the reflectance at a wavelength of 675nm is recorded. The local response offset rate is converted into an order control command. When the scattering background is enhanced, the derivative order is increased to suppress baseline broadening. The signal-to-background ratio of the 675nm absorption peak is calculated by traversing the test samples. The numerical group corresponding to the minimum fluctuation variance of the signal-to-background ratio at each turbidity level is selected as the sensor's fixed parameters.

[0037] This mapping method converts the physically detected local response offset rate into an order control command, enabling the system to increase the differential order to suppress baseline broadening when the scattering background is enhanced. When executing the fractional-order differential transform, the processor uses a discretized differential operator DV(λ). The system retrieves a pre-stored operator coefficient weight table from memory based on the real-time value of V(λ). This operator coefficient weight table is constructed based on the Granwald-Retnikov discrete definition. Specifically, the processor allocates a circular buffer of length 6 in memory to temporarily store the reflectance values ​​of the current wavelength sampling point and its five preceding neighboring points. Based on the differential order value of the current sampling point, it retrieves the values ​​from a lookup table with a preset step size of 0.01. Six weighting coefficients are selected, with the first coefficient corresponding to the current point and the remaining five coefficients corresponding to historical nodes. The reflectivity values ​​in the buffer are multiplied and accumulated point by point with the corresponding weighting coefficients. The result of the multiplication and accumulation is the fractional-order eigenvalue at that point. To ensure phase consistency, the weighting coefficients are normalized by energy conservation before calculation, ensuring that the sum of all coefficients equals 1.0 when the order is 0, thus avoiding numerical jumps during the order switching process. This table is generated by discrete sampling of the differential kernel function. In a sequence with a sampling step size of 1.0 nm, the differential operator covers the current sampling point and its five preceding neighboring points. The processor obtains the sampling point by calculating the weighted sum of the neighboring point values. The fractional eigenvalue S(λ) at the obtained maximum absorption depth parameter and peak skewness parameter are used for calculation by the partial least squares unmixing model, which employs a weight matrix trained from 200 known concentration samples. The processor combines these parameters into a two-dimensional input feature vector, and the mass concentration of the target component is obtained through mapping using the weight matrix. The partial least squares unmixing model used here integrates a Gaussian radial basis kernel function, whose distribution bandwidth parameter is set to 0.75, determined by scanning the residual distribution of 50 standard scattering samples during the power-on calibration phase. The specific mapping process is as follows: the real-time extracted maximum absorption depth parameter and the peak skewness parameter are combined with the weight matrix to obtain the mass concentration of the target component. Preprocessing is performed by subtracting the mean vector of the training set from the peak skewness parameter and dividing by the standard deviation vector. The squared Euclidean distance between the preprocessed vector and the center points of the 20 support vectors stored in the model is calculated. This squared value is divided by a negative exponent of 1.125 to obtain the kernel mapping output. Finally, it is multiplied by a 20x2 weight coefficient matrix to obtain the initial concentrations of chlorophyll a and colored soluble organic matter. When the system detects that the ambient temperature fluctuation exceeds 5°C, the bias term of the weight matrix is ​​corrected using the built-in temperature compensation function. The correction amount has a quadratic function relationship with the temperature change value, which is used to offset the interference of thermal noise on the extraction of weak absorption features.

[0038] The processor outputs the mass concentration of chlorophyll a. and the mass concentration of colored soluble organic matter Then, the nutrient status index is calculated using a preset nonlinear mapping function. The mass concentration value is mapped to a continuous evaluation range of 0 to 100 through logarithmic transformation. When the chlorophyll a mass concentration of the water body increases by 10 units, the corresponding trophic status index increases by 10 units. This index is used to classify the eutrophication status of the water body. When the TSI exceeds the critical threshold of 60, the processor identifies that the water body has entered the eutrophication state and sends an early warning command to the execution agency, completing the complete logical transfer from the underlying physical quantity of the spectrum to the water quality evaluation decision parameters. The partial least squares unmixing model has a built-in temperature compensation program. When the ambient temperature fluctuation exceeds 5°C, the processor calls the pre-stored weight correction matrix. The input vector is shifted and corrected, with the correction amount showing a quadratic function relationship with the temperature change value. This compensates for the slight drift of the spectral reference caused by the thermal noise of the photodetector. The trophic state index (TSI) evaluation threshold is set to 60. When the TSI exceeds the threshold three times consecutively, the sampling period is compressed from 120ms to 40ms to capture the characteristic pulse signal at the beginning of algal blooms. This ensures that the root mean square error of chlorophyll a mass concentration measurement is below 2.20mg / L when the sensor performance shows a gain attenuation of less than 10%. When the calculated trophic state index exceeds the preset risk threshold of 60 three times consecutively, the system triggers a high-frequency sampling mode, compressing the sampling period from 120ms to 40ms to capture the characteristic pulse signal at the beginning of algal blooms. Using the above parameter calibration and adaptive adjustment logic, the system maintains the root mean square error of chlorophyll a mass concentration measurement below 2.20mg / L even when the sensor performance shows a gain attenuation of less than 10%, ensuring the physical authenticity of the eutrophication component characterization results.

[0039] Example 4: In the initialization process of a newly built automatic water quality monitoring station, to determine the mapping parameters between the fractional-order differential sequence and the modulation vector, standard scattering samples with mass concentration gradients of 0 mg / L, 50 mg / L, 100 mg / L, 150 mg / L, and 200 mg / L were prepared using formazan standard material in a controlled environment. The hyperspectral detector collected the raw reflectance data of each sample in the 400 nm to 1000 nm band. The processor used the formula V(λ) = αM(λ) + β to fit a linear regression equation to the data under different scattering intensities, where V(λ) is the fractional-order differential at the wavelength sampling point λ, α is the mapping proportionality coefficient, M(λ) is the amplitude of the modulation vector at the wavelength sampling point λ, and β is the basic offset. After processing 500 sets of gradient test data, the processor determined that α is 0.15 and β is 0.35 under the optical structure of the sensor. The generated mapping relationship was stored in the processor's register as a weight allocation logic. The order boundary of the dynamic differential operator was determined using the scattering feedback of the standard samples.

[0040] When the system is deployed to a wetland water environment monitoring site with humic acid characteristics, the processor initiates an on-site pre-calibration process to correct the weight matrix of the partial least squares unmixing model. The sampling agency acquires three sets of in-situ water sample mixtures with different volume ratios. The system uses a chlorophyll a standard solution with a known mass concentration for standard addition titration, simultaneously acquiring the hyperspectral reflectance sequence during the titration process and extracting the corresponding maximum absorption depth parameter and peak skewness parameter. The processor compares the changes in parameter response intensity before and after the addition of the standard and calculates the weight correction factor under the current water matrix interference, compensating it into the input vector weight distribution of the partial least squares unmixing model. The corrected weight matrix is ​​used to verify the first set of hyperspectral feature data collected on-site. When the root mean square error of the inversion is detected to be less than 2.20 mg / L, the current operation operator weights are locked. The incremental feedback of the on-site measured data compensates for the characteristic spectral shift caused by matrix differences, maintaining the inversion stability of the system under the monitoring environment.

[0041] Example 5: In the case of estuary monitoring deployment under the influence of tidal wave dynamics and complex suspended matter composition, due to the differences in the background scattering patterns of water bodies at different tidal levels, the system determines the calculation parameters for morphological baseline correction through an adaptive calibration process based on structural element scale. It reads the hyperspectral reflectance sequence R(λ) of the water body under test and determines the full width at half maximum (FWHM) at the 675nm absorption peak using the first derivative extreme point detection method. The discrete length parameter of the structural element is determined based on a scale of three times the half-width. ,in Satisfying the formula ;in, represents the discrete length parameter of the structuring element. The full width at half maximum (FWHM) at the 675 nm absorption peak is used to eliminate the non-stationary baseline shift induced by water surface fluctuations through geometric constraints. The fractional eigenvalue S(λ) is calculated to satisfy the following conditions: Where S(λ) is the fractional eigenvalue at the wavelength sampling point λ. The differential operator is determined by a fractional-order differential sequence, R(λ) is the reflectance value after morphological baseline correction, and M(λ) is the amplitude of the modulation vector at the wavelength sampling point λ. The exponential term 1+M(λ) is introduced to mitigate the decrease in signal contrast under high scattering matrix environment.

[0042] During the initialization phase, the system iterates through preset length steps from 20 nm to 50 nm and calculates the baseline residual variance after morphological opening and closing operations. The value that minimizes the baseline residual variance is determined as the structural parameter of the current monitoring site. This quantized geometric constraint is used to eliminate non-stationary baseline shifts induced by water surface fluctuations and aerosol scattering. When the system detects that the mean intensity of the original reflectivity sequence across the entire band is lower than the preset signal-to-noise ratio criterion or the local gradient change rate of the modulation vector M(λ) exceeds the physical extreme threshold of 1.5, the processor uses an abnormal gain compensation mechanism to suppress the interference of signal saturation on feature extraction. The system dynamically adjusts the detector's integration time using a controlled attenuation operator, inputs the acquired correction sequence into a partial least squares unmixing model, and extracts the fluctuation characteristics of the output component concentration. It calculates the confidence probability value of the current inversion result using a variance distribution function established based on 500 sets of historical monitoring data. When the confidence probability value reaches 0.95 or higher, the system pushes the result to the characterization unit and updates the reference intrinsic absorption spectrum sequence in the storage unit. It compensates for the characteristic spectral shift caused by matrix differences through incremental feedback of field measured data, thus maintaining the stability of the system's dynamic response to chlorophyll a mass concentration measurement.

[0043] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0044] Finally, 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.

Claims

1. A method for analyzing and characterizing eutrophication components in water bodies based on hyperspectral feature inversion, characterized in that, Includes the following steps: Step 101: Obtain the hyperspectral reflectance sequence of the water body to be tested, and retrieve the pre-stored intrinsic absorption spectrum sequence of pure water containing discrete data nodes. Step 102: Traverse the wavelength sampling points in the hyperspectral reflectance sequence, calculate the ratio between the reflectance value of each wavelength sampling point and the value of the discrete data node of the corresponding wavelength sampling point in the pre-stored pure water intrinsic absorption spectrum sequence, and generate a modulation vector characterizing the local spectral response shift rate. Step 103: Map the modulation vector to a preset initial order interval within the range of 0 to 2 to generate a fractional differential order sequence with the same dimension as the hyperspectral reflectance sequence and wavelength compliance. Step 104: Perform morphological baseline correction on the hyperspectral reflectance sequence to extract absorption trough features, and combine the fractional derivative sequence with the morphologically baseline-corrected sequence to perform fractional derivative transformation on the sequence to generate a fractional characteristic spectral sequence. Step 105: Define the first characteristic trough interval corresponding to chlorophyll a and the second characteristic trough interval corresponding to colored soluble organic matter in the fractional-order characteristic spectral sequence, and extract the maximum absorption depth parameter and the peak skewness parameter characterizing the trough waveform deformation. Step 106: Input the maximum absorption depth parameter and the peak skewness parameter into the pre-calibrated partial least squares unmixing model, and output the mass concentration of chlorophyll a and colored soluble organic matter in the water body to be tested.

2. The method for analyzing and characterizing eutrophication components of water bodies based on hyperspectral feature inversion according to claim 1, characterized in that, Step 102, generating the modulation vector, includes: mapping the hyperspectral reflectance sequence to the spectral feature space to extract the baseline offset component; performing normalized sensitivity analysis on the baseline offset component using a pre-stored pure water intrinsic absorption spectrum sequence to determine the energy attenuation ratio of the hyperspectral reflectance sequence at each wavelength, and determining the energy attenuation ratio as the weighting coefficient of the modulation vector at the corresponding wavelength sampling point.

3. The method for analyzing and characterizing eutrophication components of water bodies based on hyperspectral feature inversion according to claim 1, characterized in that, Step 103, generating the fractional-order differential sequence, includes: establishing a linear mapping relationship between the modulation vector and the differential order; based on the linear mapping relationship, when the modulation vector reflects a decrease in the local spectral response offset rate, decreasing the differential order at the corresponding wavelength position; when the modulation vector reflects an increase in the local spectral response offset rate, increasing the differential order at the corresponding wavelength position, so as to form a fractional-order differential sequence distributed between 0 and 2.

4. The method for analyzing and characterizing eutrophication components of water bodies based on hyperspectral feature inversion according to claim 1, characterized in that, Step 104 involves performing morphological baseline correction, which includes: performing morphological opening and closing operations on the hyperspectral reflectance sequence using structuring elements to extract the trend background curve of the hyperspectral reflectance sequence; and subtracting the trend background curve from the hyperspectral reflectance sequence to eliminate the interference of water surface flares on feature extraction.

5. The method for analyzing and characterizing eutrophication components of water bodies based on hyperspectral feature inversion according to claim 1, characterized in that, The eigenvalue S(λ) at wavelength sampling point λ in a fractional-order characteristic spectral sequence satisfies the following logical rule: ,in, R(λ) is a differential operator determined based on the order value V(λ) in the fractional-order differential order sequence, R(λ) is the reflectance value after morphological baseline correction, and M(λ) is the amplitude of the modulation vector at the wavelength sampling point λ.

6. The method for analyzing and characterizing eutrophication components of water bodies based on hyperspectral feature inversion according to claim 1, characterized in that, Step 105 defines the first and second characteristic valley intervals, including: performing second-order differential zero-point detection on the fractional characteristic spectral sequence to identify the extreme value positions of the spectral curve; determining the interval from 665nm to 685nm as the first characteristic valley interval, and determining the interval from 410nm to 440nm as the second characteristic valley interval.

7. The method for analyzing and characterizing eutrophication components of water bodies based on hyperspectral feature inversion according to claim 1, characterized in that, The extraction of peak skewness parameters includes: calculating the third central moment of the spectral curve within the first or second characteristic trough interval; and performing a ratio operation between the third central moment and the cube of the standard deviation of the spectral values ​​within the corresponding interval to obtain a dimensionless parameter reflecting the deviation of the absorption characteristics from the symmetry.

8. The method for analyzing and characterizing eutrophication components of water bodies based on hyperspectral feature inversion according to claim 1, characterized in that, The partial least squares unmixing model is calibrated as follows: a standard water sample spectral dataset with known concentrations is obtained; steps 101 to 105 are performed on the standard water sample spectral dataset to obtain training feature vectors; the training feature vectors are mapped to a linearly separable space using a kernel function, and the weight matrix of the partial least squares unmixing model is determined by minimizing the sum of squared residuals between the predicted concentration and the true value.

9. The method for analyzing and characterizing eutrophication components of water bodies based on hyperspectral feature inversion according to claim 1, characterized in that, After outputting the mass concentrations of chlorophyll a and colored soluble organic matter in the water body to be tested, the method further includes step 107: calculating the trophic state index, which characterizes the degree of water pollution, based on the mass concentrations of chlorophyll a and colored soluble organic matter in the water body to be tested; and outputting an early warning command when the trophic state index exceeds a preset risk threshold.

10. The method for analyzing and characterizing eutrophication components of water bodies based on hyperspectral feature inversion according to claim 1, characterized in that, In step 101, when obtaining the hyperspectral reflectance sequence, the raw radiance data is collected using an imaging spectrometer deployed above the water surface. Reflectance calibration is performed by introducing synchronously observed standard plate irradiance data to eliminate the interference of light environment fluctuations on spectral inversion and ensure that the input reflectance value is within the physical range of 0 to 1.

Images