A water quality parameter hyperspectral analysis method and system based on a machine learning model

Abstract

This invention relates to the field of hyperspectral analysis technology, specifically disclosing a hyperspectral analysis method and system for water quality parameters based on a machine learning model, comprising the following steps: Step S1: Select several water quality parameter samples, and classify them into sensitive and non-sensitive water quality parameter samples according to the spectral curves of each water quality parameter sample in different signal-to-noise ratio spectral bands; Step S2: Obtain water samples from the pollution source of the water body, and obtain the concentration and source spectral curve of the non-sensitive water quality parameters in the water samples; Step S3: Perform hyperspectral analysis at the collection point of the water body to obtain a hyperspectral image, and select non-sensitive pixels according to the spectral curves of each pixel in the hyperspectral image; Step S4: Extract all non-sensitive curves and compare them with the source spectral curves to find the target curve with the highest similarity, and then predict the concentration of the non-sensitive water quality parameters at the collection point.

Description

Technical Field

[0001] This invention relates to the field of hyperspectral analysis technology, specifically to a hyperspectral analysis method and system for water quality parameters based on a machine learning model. Background Technology

[0002] Hyperspectral analysis of water quality parameters is a method that uses hyperspectral imaging technology to acquire and analyze the optical properties of water bodies, thereby assessing water quality. It includes the following steps: data acquisition, data preprocessing, model building, and water quality parameter inversion.

[0003] In existing technologies, different water quality parameters exhibit varying sensitivities to hyperspectral analysis. Specifically, some water quality parameters show significant responses in hyperspectral data and are easily identified and quantified accurately through their spectral characteristics; while others may be difficult to detect or measure accurately in hyperspectral analysis due to insufficiently significant spectral characteristics or susceptibility to interference from other factors. For water quality parameters with lower sensitivity, if their concentration or changes in the water body do not reach the detection threshold of hyperspectral analysis, these parameters may be missed or misjudged, thus failing to provide comprehensive and accurate information for water quality assessment. Summary of the Invention

[0004] The purpose of this invention is to provide a hyperspectral analysis method and system for water quality parameters based on a machine learning model, thereby solving the above-mentioned technical problems.

[0005] The objective of this invention can be achieved through the following technical solutions:

[0006] A hyperspectral analysis method for water quality parameters based on a machine learning model includes the following steps:

[0007] Step S1: Select several water quality parameter samples and obtain water quality parameter sample solutions; set a standard signal-to-noise ratio Sn0, and set several signal-to-noise ratio gradient values ​​according to the standard signal-to-noise ratio, perform hyperspectral analysis on the water quality parameter sample solutions with spectral bands of different signal-to-noise ratio gradient values, and obtain the spectral curves of the water quality parameter sample solutions, wherein the spectral curves are the reflectance at different wavelengths;

[0008] Based on each signal-to-noise ratio gradient value and the spectral curve corresponding to the water quality parameter sample solution at each signal-to-noise ratio gradient value, the reflectance change rate of the water quality parameter sample is obtained; water quality parameter samples with a reflectance change rate lower than a preset reflectance change threshold are recorded as non-sensitive water quality parameters.

[0009] Step S2: The location of the pollution source in the water body is recorded as the pollution source point; water samples are collected at the pollution source point to obtain the concentration c0 of the non-sensitive water quality parameter; the water sample is subjected to hyperspectral analysis at a standard signal-to-noise ratio spectral band to obtain a hyperspectral image, and the spectral curve of the pixel where the non-sensitive water quality parameter is located is obtained in the hyperspectral image and recorded as the source spectral curve.

[0010] Step S3: Record any other location in the water body as a collection point. When performing hyperspectral analysis at the collection point, obtain hyperspectral images under spectral bands with different signal-to-noise ratio gradient values, and record them as the current hyperspectral image. Record the spectral curves within each pixel of the current hyperspectral image as the current spectral curve.

[0011] In the current hyperspectral image, obtain all current spectral curves of the same pixel in spectral bands under different signal-to-noise ratio gradient values, and obtain the difference value:

[0012] Where n is the total number of signal-to-noise ratio gradient values, f i (λ) represents the current spectral curve under the spectral band of the i-th signal-to-noise ratio gradient value, where λ is the wavelength and [λ0, λ1] represents the wavelength range of the spectral band; pixels with a difference value less than a preset difference threshold are selected and recorded as non-sensitive pixels.

[0013] Step S4: Obtain all current spectral curves of the non-sensitive pixels, denoted as non-sensitive curves, and select the non-sensitive curve with the highest similarity to the source spectral curve, denoted as the target curve; obtain the signal-to-noise ratio gradient value corresponding to the target curve, denoted as Sn. min Based on the signal-to-noise ratio gradient value and the standard signal-to-noise ratio, the predicted concentration of the non-sensitive water quality parameter at the sampling point is obtained as Pc = c0 + μ(Sn). min -Sn0), where μ is a preset correction parameter and μ>0, with units of mol / (L*dB).

[0014] As a further aspect of the present invention: the process of setting the signal-to-noise ratio gradient value includes:

[0015] Set a signal-to-noise ratio (SNR) interval threshold, and use the standard SNR as the starting value. Increase or decrease the starting value according to the SNR interval threshold to obtain several SNR gradient values.

[0016] As a further aspect of the present invention: the process of obtaining the reflection rate includes:

[0017] Several reference points are selected on each spectral curve, and the average reflectance is obtained based on the reflectance at each reference point on the spectral curve. Where k represents the k-th spectral curve, P xThis represents the reflectance at the x-th reference point, where m is the total number of reference points;

[0018] A coordinate system is established with signal-to-noise ratio as the abscissa and reflectance as the ordinate. A standard line y = Mr is obtained within this coordinate system, where y represents the ordinate of the coordinate system. All spectral curves are placed within this coordinate system, with the starting point of each spectral curve coinciding with the origin. The difference between each spectral curve and the standard line is then obtained within this coordinate system. Then the rate of change of reflection is obtained. Where Sn i D represents the i-th signal-to-noise ratio gradient value. i This represents the difference in the spectral curves corresponding to the i-th signal-to-noise ratio gradient value.

[0019] As a further aspect of the present invention: water quality parameter samples whose reflectance change rate is greater than or equal to a preset reflectance change threshold are denoted as sensitive water quality parameters.

[0020] As a further aspect of the present invention: the process of obtaining the spectral curve of the pixel containing the non-sensitive water quality parameter in the hyperspectral image includes:

[0021] All sensitive water quality parameters in the water sample are obtained, and the spectral curves of the sensitive water quality parameters are obtained and recorded as sensitive spectral curves. In the hyperspectral image, all pixels where the sensitive spectral curves are located are selected and recorded as sensitive pixels. All sensitive pixels are then removed, and the spectral curves of all remaining pixels in the hyperspectral image are recorded as source spectral curves.

[0022] As a further aspect of the present invention: pixels with a difference value greater than or equal to a preset difference threshold are designated as sensitive pixels.

[0023] As a further aspect of the present invention, the process of selecting the non-sensitive curve with the highest similarity to the source spectral curve includes:

[0024] By selecting several comparison points at equal intervals on the source spectrum curve and the insensitive curve, the similarity between the source spectrum curve and the insensitive curve is obtained. Where G(t) represents the t-th comparison point on the source spectrum curve, T(t) represents the t-th comparison point on the insensitive curve, and J is the total number of comparison points.

[0025] As a further aspect of the present invention: a hyperspectral analysis system for water quality parameters based on a machine learning model, comprising:

[0026] Sensitive segmentation module: Select several water quality parameter samples to obtain water quality parameter sample solutions; set a standard signal-to-noise ratio Sn0, and set several signal-to-noise ratio gradient values ​​according to the standard signal-to-noise ratio; perform hyperspectral analysis on the water quality parameter sample solutions with spectral bands of different signal-to-noise ratio gradient values ​​to obtain spectral curves of the water quality parameter sample solutions, wherein the spectral curves are reflectance at different wavelengths;

[0027] Based on each signal-to-noise ratio gradient value and the spectral curve corresponding to the water quality parameter sample solution at each signal-to-noise ratio gradient value, the reflectance change rate of the water quality parameter sample is obtained; water quality parameter samples with a reflectance change rate lower than a preset reflectance change threshold are recorded as non-sensitive water quality parameters.

[0028] Pollution source analysis module: The location of the pollution source in the water body is recorded as the pollution source point; water samples are collected at the pollution source point to obtain the concentration c0 of the non-sensitive water quality parameter; the water sample is subjected to hyperspectral analysis at a standard signal-to-noise ratio spectral band to obtain a hyperspectral image; the spectral curve of the pixel where the non-sensitive water quality parameter is located is obtained in the hyperspectral image and recorded as the source spectral curve.

[0029] Collection point analysis module: Record any other location in the water body as a collection point. When performing hyperspectral analysis at the collection point, obtain hyperspectral images under spectral bands with different signal-to-noise ratio gradient values, record them as the current hyperspectral image, and record the spectral curves within each pixel of the current hyperspectral image as the current spectral curve.

[0030] In the current hyperspectral image, obtain all current spectral curves of the same pixel in spectral bands under different signal-to-noise ratio gradient values, and obtain the difference value:

[0031] Where n is the total number of signal-to-noise ratio gradient values, f i (λ) represents the current spectral curve under the spectral band of the i-th signal-to-noise ratio gradient value, where λ is the wavelength and [λ0, λ1] represents the wavelength range of the spectral band; pixels with a difference value less than a preset difference threshold are selected and recorded as non-sensitive pixels.

[0032] Concentration prediction module: Acquires all current spectral curves of the non-sensitive pixels, denoted as non-sensitive curves, and selects the non-sensitive curve with the highest similarity to the source spectral curve, denoted as the target curve; acquires the signal-to-noise ratio gradient value corresponding to the target curve, denoted as Sn. min Based on the signal-to-noise ratio gradient value and the standard signal-to-noise ratio, the predicted concentration of the non-sensitive water quality parameter at the sampling point is obtained as Pc = c0 + μ(Sn). min -Sn0), where μ is a preset correction parameter and μ>0, with units of mol / (L*dB).

[0033] The beneficial effects of this invention are:

[0034] This invention utilizes a machine learning model to perform hyperspectral analysis on water quality parameters, enabling the identification and differentiation of non-sensitive water quality parameters and improving analytical accuracy. By leveraging spectral bands at different signal-to-noise ratio gradients, it can more accurately acquire reflectance variation information from water samples, thereby predicting the concentration of non-sensitive water quality parameters. Water samples are collected at pollution sources, and the concentrations of non-sensitive water quality parameters and their corresponding hyperspectral images are obtained. Then, hyperspectral analysis is performed at the collection points to acquire hyperspectral images at different signal-to-noise ratio gradients, and non-sensitive curves are extracted. Finally, by comparing the non-sensitive curves from two locations and combining them with preset correction parameters, the concentration of non-sensitive water quality parameters at the collection point can be predicted. This method not only improves the efficiency of water quality parameter retrieval but also possesses strong adaptability and generalization capabilities, making it applicable to different pollution sources and collection points. Attached Figure Description

[0035] The invention will now be further described with reference to the accompanying drawings.

[0036] Figure 1 This is a schematic flowchart of a hyperspectral analysis method for water quality parameters based on a machine learning model, according to the present invention. Detailed Implementation

[0037] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0038] Please see Figure 1 As shown, this invention is a hyperspectral analysis method for water quality parameters based on a machine learning model, comprising the following steps:

[0039] Step S1: Select several water quality parameter samples and obtain water quality parameter sample solutions; set a standard signal-to-noise ratio Sn0, and set several signal-to-noise ratio gradient values ​​according to the standard signal-to-noise ratio, perform hyperspectral analysis on the water quality parameter sample solutions with spectral bands of different signal-to-noise ratio gradient values, and obtain the spectral curves of the water quality parameter sample solutions, wherein the spectral curves are the reflectance at different wavelengths;

[0040] Based on each signal-to-noise ratio gradient value and the spectral curve corresponding to the water quality parameter sample solution at each signal-to-noise ratio gradient value, the reflectance change rate of the water quality parameter sample is obtained; water quality parameter samples with a reflectance change rate lower than a preset reflectance change threshold are recorded as non-sensitive water quality parameters.

[0041] It should be noted that the water quality parameter samples include physical or chemical components such as dissolved oxygen, metal element content, and ammonia nitrogen; the water quality parameter sample solution is a solution formed by dissolving the selected water quality parameter samples in an appropriate solvent; and the standard signal-to-noise ratio is a determined benchmark signal-to-noise ratio value used as the standard for analysis.

[0042] Understandably, hyperspectral analysis is performed on water quality parameter sample solutions using spectral bands with different signal-to-noise ratio gradient values ​​to obtain the reflectance of each sample at different wavelengths, generating spectral curves. Based on each signal-to-noise ratio gradient value and the corresponding spectral curve, the reflectance change rate of each water quality parameter sample is calculated. The reflectance change rate reflects the change in the spectral response of the sample under different signal-to-noise ratio conditions. Water quality parameter samples with reflectance change rates lower than a preset reflectance change threshold are marked as non-sensitive water quality parameters, meaning that these parameters change less under different signal-to-noise ratio conditions and are not easily detected.

[0043] It is worth noting that by calculating the reflectance change under different signal-to-noise ratio conditions, the sensitivity of water quality parameters to signal-to-noise ratio changes can be assessed; a low reflectance change rate means that the parameter does not change much under different conditions, i.e., it is insensitive.

[0044] In a preferred embodiment of the present invention, the process of setting the signal-to-noise ratio gradient value includes:

[0045] Set a signal-to-noise ratio (SNR) interval threshold, take the standard SNR as the starting value, and increase or decrease the starting value according to the SNR interval threshold to obtain several SNR gradient values;

[0046] Understandably, the signal-to-noise ratio (SNR) interval threshold is used to control the interval between SNR gradient values, ensuring that each gradient value has an appropriate distance so that changes under different conditions can be captured in subsequent hyperspectral analysis. The standard SNR serves as a starting point, providing a benchmark value, and all gradient values ​​are obtained by incrementing or decrementing based on this benchmark value. Multiple SNR gradient values ​​are generated by incrementing or decrementing, and these gradient values ​​will be used in different experimental conditions or analytical steps to ensure the diversity and comprehensiveness of the data.

[0047] In a preferred embodiment of the present invention, the process of obtaining the reflection rate includes:

[0048] Several reference points are selected on each spectral curve, and the average reflectance is obtained based on the reflectance at each reference point on the spectral curve. Where k represents the k-th spectral curve, P x This represents the reflectance at the x-th reference point, where m is the total number of reference points;

[0049] A coordinate system is established with signal-to-noise ratio as the abscissa and reflectance as the ordinate. A standard line y = Mr is obtained within this coordinate system, where y represents the ordinate of the coordinate system. All spectral curves are placed within this coordinate system, with the starting point of each spectral curve coinciding with the origin. The difference between each spectral curve and the standard line is then obtained within this coordinate system. Then the rate of change of reflection is obtained. Where Sn i D represents the i-th signal-to-noise ratio gradient value. i This represents the difference in the spectral curves corresponding to the i-th signal-to-noise ratio gradient value;

[0050] Understandably, by selecting multiple reference points and calculating their average reflectance, a baseline value Mr is obtained for subsequent comparisons; establishing a coordinate system with signal-to-noise ratio as the abscissa and reflectance as the ordinate helps to intuitively display the changes in the spectral curves; by calculating the difference D between each spectral curve and the standard line, the degree of deviation of each curve from the baseline value can be quantified; the final reflectance change rate R is calculated by comparing the difference changes under different signal-to-noise ratio gradients, reflecting the sensitivity of water quality parameter samples under different conditions;

[0051] In a preferred embodiment of the present invention, the water quality parameter sample whose reflectance change rate is greater than or equal to a preset reflectance change threshold is denoted as a sensitive water quality parameter.

[0052] Step S2: The location of the pollution source in the water body is recorded as the pollution source point; water samples are collected at the pollution source point to obtain the concentration c0 of the non-sensitive water quality parameter; the water sample is subjected to hyperspectral analysis at a standard signal-to-noise ratio spectral band to obtain a hyperspectral image, and the spectral curve of the pixel where the non-sensitive water quality parameter is located is obtained in the hyperspectral image and recorded as the source spectral curve.

[0053] Understandably, hyperspectral analysis of water samples is performed using spectral bands with standard signal-to-noise ratios to generate hyperspectral images; these hyperspectral images contain rich spectral information that can be used for further analysis; spectral curves of pixels containing non-sensitive water quality parameters are extracted from the hyperspectral images, and these curves reflect the reflectance of non-sensitive water quality parameters at different wavelengths;

[0054] In a preferred embodiment of the present invention, the process of obtaining the spectral curve of the pixel containing the non-sensitive water quality parameter in the hyperspectral image includes:

[0055] All sensitive water quality parameters in the water sample are obtained, and the spectral curves of the sensitive water quality parameters are obtained and recorded as sensitive spectral curves. In the hyperspectral image, all pixels where the sensitive spectral curves are located are selected and recorded as sensitive pixels. All sensitive pixels are removed, and the spectral curves of all remaining pixels in the hyperspectral image are recorded as source spectral curves.

[0056] It is understandable that after removing sensitive pixels, the spectral curves of the remaining pixels mainly reflect the characteristics of non-sensitive water quality parameters.

[0057] Step S3: Record any other location in the water body as a collection point. When performing hyperspectral analysis at the collection point, obtain hyperspectral images under spectral bands with different signal-to-noise ratio gradient values, and record them as the current hyperspectral image. Record the spectral curves within each pixel of the current hyperspectral image as the current spectral curve.

[0058] In the current hyperspectral image, obtain all current spectral curves of the same pixel in spectral bands under different signal-to-noise ratio gradient values, and obtain the difference value:

[0059] Where n is the total number of signal-to-noise ratio gradient values, f i (λ) represents the current spectral curve under the spectral band of the i-th signal-to-noise ratio gradient value, where λ is the wavelength and [λ0, λ1] represents the wavelength range of the spectral band; pixels with a difference value less than a preset difference threshold are selected and recorded as non-sensitive pixels.

[0060] Understandably, sampling and analysis are conducted at different locations within the water body to ensure the representativeness and comprehensiveness of the data; hyperspectral analysis is performed using spectral bands with different signal-to-noise ratio gradients to obtain detailed spectral information; the spectral curve of each pixel is recorded for subsequent difference value calculation; the degree of change is quantified by calculating the difference in the spectral curves of the same pixel under different signal-to-noise ratio gradients; pixels with difference values ​​less than a preset threshold are considered to have small changes and are therefore considered non-sensitive.

[0061] In a preferred embodiment of the present invention, pixels with a difference value greater than or equal to a preset difference threshold are recorded as sensitive pixels.

[0062] Step S4: Obtain all current spectral curves of the non-sensitive pixels, denoted as non-sensitive curves, and select the non-sensitive curve with the highest similarity to the source spectral curve, denoted as the target curve; obtain the signal-to-noise ratio gradient value corresponding to the target curve, denoted as Sn. min Based on the signal-to-noise ratio gradient value and the standard signal-to-noise ratio, the predicted concentration of the non-sensitive water quality parameter at the sampling point is obtained as Pc = c0 + μ(Sn). min-Sn0), where μ is a preset correction parameter and μ>0, with units of mol / (L*dB);

[0063] It is understood that the non-sensitive curve is a spectral curve obtained from non-sensitive pixels, representing the reflectance change under different signal-to-noise ratio gradients; by using the difference between known concentration and signal-to-noise ratio gradient values, combined with the correction parameter μ, the predicted concentration of non-sensitive water quality parameters at the sampling point is calculated.

[0064] In a preferred embodiment of the present invention, the process of selecting the non-sensitive curve with the highest similarity to the source spectral curve includes:

[0065] By selecting several comparison points at equal intervals on the source spectrum curve and the insensitive curve, the similarity between the source spectrum curve and the insensitive curve is obtained. Where G(t) represents the t-th comparison point on the source spectrum curve, T(t) represents the t-th comparison point on the insensitive curve, and J is the total number of comparison points;

[0066] Understandably, the similarity between two curves is quantified by calculating the differences between the comparison points. The smaller the similarity value, the more similar the two curves are.

[0067] A hyperspectral analysis system for water quality parameters based on a machine learning model includes:

[0068] Sensitive segmentation module: Select several water quality parameter samples to obtain water quality parameter sample solutions; set a standard signal-to-noise ratio Sn0, and set several signal-to-noise ratio gradient values ​​according to the standard signal-to-noise ratio; perform hyperspectral analysis on the water quality parameter sample solutions with spectral bands of different signal-to-noise ratio gradient values ​​to obtain spectral curves of the water quality parameter sample solutions, wherein the spectral curves are reflectance at different wavelengths;

[0069] Based on each signal-to-noise ratio gradient value and the spectral curve corresponding to the water quality parameter sample solution at each signal-to-noise ratio gradient value, the reflectance change rate of the water quality parameter sample is obtained; water quality parameter samples with a reflectance change rate lower than a preset reflectance change threshold are recorded as non-sensitive water quality parameters.

[0070] Pollution source analysis module: The location of the pollution source in the water body is recorded as the pollution source point; water samples are collected at the pollution source point to obtain the concentration c0 of the non-sensitive water quality parameter; the water sample is subjected to hyperspectral analysis at a standard signal-to-noise ratio spectral band to obtain a hyperspectral image; the spectral curve of the pixel where the non-sensitive water quality parameter is located is obtained in the hyperspectral image and recorded as the source spectral curve.

[0071] Collection point analysis module: Record any other location in the water body as a collection point. When performing hyperspectral analysis at the collection point, obtain hyperspectral images under spectral bands with different signal-to-noise ratio gradient values, record them as the current hyperspectral image, and record the spectral curves within each pixel of the current hyperspectral image as the current spectral curve.

[0072] In the current hyperspectral image, obtain all current spectral curves of the same pixel in spectral bands under different signal-to-noise ratio gradient values, and obtain the difference value:

[0073] Where n is the total number of signal-to-noise ratio gradient values, f i (λ) represents the current spectral curve under the spectral band of the i-th signal-to-noise ratio gradient value, where λ is the wavelength and [λ0, λ1] represents the wavelength range of the spectral band; pixels with a difference value less than a preset difference threshold are selected and recorded as non-sensitive pixels.

[0074] Concentration prediction module: Acquires all current spectral curves of the non-sensitive pixels, denoted as non-sensitive curves, and selects the non-sensitive curve with the highest similarity to the source spectral curve, denoted as the target curve; acquires the signal-to-noise ratio gradient value corresponding to the target curve, denoted as Sn. min Based on the signal-to-noise ratio gradient value and the standard signal-to-noise ratio, the predicted concentration of the non-sensitive water quality parameter at the sampling point is obtained as Pc = c0 + μ(Sn). min -Sn0), where μ is a preset correction parameter and μ>0, with units of mol / (L*dB).

[0075] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A hyperspectral analysis method for water quality parameters based on a machine learning model, characterized in that, Includes the following steps: Step S1: Select several water quality parameter samples and obtain water quality parameter sample solutions; A standard signal-to-noise ratio Sn0 is set, and several signal-to-noise ratio gradient values ​​are set according to the standard signal-to-noise ratio. Hyperspectral analysis is performed on the water quality parameter sample solution with different signal-to-noise ratio gradient values ​​to obtain the spectral curve of the water quality parameter sample solution. The spectral curve is the reflectance at different wavelengths. Based on each signal-to-noise ratio gradient value and the spectral curve corresponding to the water quality parameter sample solution at each signal-to-noise ratio gradient value, the reflectance change rate of the water quality parameter sample is obtained; water quality parameter samples with a reflectance change rate lower than a preset reflectance change threshold are recorded as non-sensitive water quality parameters. Step S2: The location of the pollution source in the water body is recorded as the pollution source point; water samples are collected at the pollution source point to obtain the concentration c0 of the non-sensitive water quality parameter; the water sample is subjected to hyperspectral analysis at a standard signal-to-noise ratio spectral band to obtain a hyperspectral image, and the spectral curve of the pixel where the non-sensitive water quality parameter is located is obtained in the hyperspectral image and recorded as the source spectral curve. Step S3: Record any other location in the water body as a collection point. When performing hyperspectral analysis at the collection point, obtain hyperspectral images under spectral bands with different signal-to-noise ratio gradient values, and record them as the current hyperspectral image. Record the spectral curves within each pixel of the current hyperspectral image as the current spectral curve. In the current hyperspectral image, obtain all current spectral curves of the same pixel in spectral bands under different signal-to-noise ratio gradient values, and obtain the difference value: Where n is the total number of signal-to-noise ratio gradient values, f i (λ) represents the current spectral curve under the spectral band of the i-th signal-to-noise ratio gradient value, where λ is the wavelength and [λ0, λ1] represents the wavelength range of the spectral band; pixels with a difference value less than a preset difference threshold are selected and recorded as non-sensitive pixels. Step S4: Obtain all current spectral curves of the non-sensitive pixels, denoted as non-sensitive curves, and select the non-sensitive curve with the highest similarity to the source spectral curve, denoted as the target curve; obtain the signal-to-noise ratio gradient value corresponding to the target curve, denoted as Sn. min Based on the signal-to-noise ratio gradient value and the standard signal-to-noise ratio, the predicted concentration of the non-sensitive water quality parameter at the sampling point is obtained as Pc = c0 + μ(Sn). min -Sn0), where μ is a preset correction parameter and μ>0, with units of mol / (L*dB).

2. The hyperspectral analysis method for water quality parameters based on a machine learning model according to claim 1, characterized in that, In step S1, the process of setting the signal-to-noise ratio gradient value includes: Set a signal-to-noise ratio (SNR) interval threshold, and use the standard SNR as the starting value. Increase or decrease the starting value according to the SNR interval threshold to obtain several SNR gradient values.

3. The hyperspectral analysis method for water quality parameters based on a machine learning model according to claim 1, characterized in that, In step S1, the process of obtaining the reflection rate of change includes: Several reference points are selected on each spectral curve, and the average reflectance is obtained based on the reflectance at each reference point on the spectral curve. Where k represents the k-th spectral curve, P x This represents the reflectance at the x-th reference point, where m is the total number of reference points; A coordinate system is established with signal-to-noise ratio as the abscissa and reflectance as the ordinate. A standard line y = Mr is obtained within this coordinate system, where y represents the ordinate of the coordinate system. All spectral curves are placed within this coordinate system, with the starting point of each spectral curve coinciding with the origin. The difference between each spectral curve and the standard line is then obtained within this coordinate system. Then the rate of change of reflection is obtained. Where Sn i D represents the i-th signal-to-noise ratio gradient value. i This represents the difference in the spectral curves corresponding to the i-th signal-to-noise ratio gradient value.

4. The hyperspectral analysis method for water quality parameters based on a machine learning model according to claim 1, characterized in that, In step S1, water quality parameter samples whose reflectance change rate is greater than or equal to a preset reflectance change threshold are denoted as sensitive water quality parameters.

5. The hyperspectral analysis method for water quality parameters based on a machine learning model according to claim 4, characterized in that, In step S2, the process of obtaining the spectral curve of the pixel containing the non-sensitive water quality parameter in the hyperspectral image includes: All sensitive water quality parameters in the water sample are obtained, and the spectral curves of the sensitive water quality parameters are obtained and recorded as sensitive spectral curves. In the hyperspectral image, all pixels where the sensitive spectral curves are located are selected and recorded as sensitive pixels. All sensitive pixels are then removed, and the spectral curves of all remaining pixels in the hyperspectral image are recorded as source spectral curves.

6. The hyperspectral analysis method for water quality parameters based on a machine learning model according to claim 1, characterized in that, In step S3, pixels with a difference value greater than or equal to a preset difference threshold are recorded as sensitive pixels.

7. The hyperspectral analysis method for water quality parameters based on a machine learning model according to claim 1, characterized in that, In step S4, the process of selecting the non-sensitive curve with the highest similarity to the source spectral curve includes: By selecting several comparison points at equal intervals on the source spectrum curve and the insensitive curve, the similarity between the source spectrum curve and the insensitive curve is obtained. Where G(t) represents the t-th comparison point on the source spectrum curve, T(t) represents the t-th comparison point on the insensitive curve, and J is the total number of comparison points.

8. A hyperspectral analysis system for water quality parameters based on a machine learning model, characterized in that, include: Sensitive classification module: Select several water quality parameter samples and obtain water quality parameter sample solutions; A standard signal-to-noise ratio Sn0 is set, and several signal-to-noise ratio gradient values ​​are set according to the standard signal-to-noise ratio. Hyperspectral analysis is performed on the water quality parameter sample solution with different signal-to-noise ratio gradient values ​​to obtain the spectral curve of the water quality parameter sample solution. The spectral curve is the reflectance at different wavelengths. Based on each signal-to-noise ratio gradient value and the spectral curve corresponding to the water quality parameter sample solution at each signal-to-noise ratio gradient value, the reflectance change rate of the water quality parameter sample is obtained; water quality parameter samples with a reflectance change rate lower than a preset reflectance change threshold are recorded as non-sensitive water quality parameters. Pollution source analysis module: The location of the pollution source in the water body is recorded as the pollution source point; water samples are collected at the pollution source point to obtain the concentration c0 of the non-sensitive water quality parameter; the water sample is subjected to hyperspectral analysis at a standard signal-to-noise ratio spectral band to obtain a hyperspectral image; the spectral curve of the pixel where the non-sensitive water quality parameter is located is obtained in the hyperspectral image and recorded as the source spectral curve. Collection point analysis module: Record any other location in the water body as a collection point. When performing hyperspectral analysis at the collection point, obtain hyperspectral images under spectral bands with different signal-to-noise ratio gradient values, record them as the current hyperspectral image, and record the spectral curves within each pixel of the current hyperspectral image as the current spectral curve. In the current hyperspectral image, obtain all current spectral curves of the same pixel in spectral bands under different signal-to-noise ratio gradient values, and obtain the difference value: Where n is the total number of signal-to-noise ratio gradient values, f i (λ) represents the current spectral curve under the spectral band of the i-th signal-to-noise ratio gradient value, where λ is the wavelength and [λ0, λ1] represents the wavelength range of the spectral band; pixels with a difference value less than a preset difference threshold are selected and recorded as non-sensitive pixels. Concentration prediction module: Acquires all current spectral curves of the non-sensitive pixels, denoted as non-sensitive curves, and selects the non-sensitive curve with the highest similarity to the source spectral curve, denoted as the target curve; acquires the signal-to-noise ratio gradient value corresponding to the target curve, denoted as Sn. min Based on the signal-to-noise ratio gradient value and the standard signal-to-noise ratio, the predicted concentration of the non-sensitive water quality parameter at the sampling point is obtained as Pc = c0 + μ(Sn). min -Sn0), where μ is a preset correction parameter and μ>0, with units of mol / (L*dB).

Images