A polarized hyperspectral water quality monitoring system and method

Abstract

The application provides a polarized hyperspectral water quality monitoring system and method, which comprises: a tower base polarized hyperspectral imaging module, which is arranged on a building at the edge of a water area to be measured, and uses a polarized hyperspectral imager to perform a surface array staring scan on the water area to be measured to obtain a water body polarized hyperspectral image signal of the water area to be measured; an image spectral data processing module, which deducts dark current and stray signals from the water body polarized hyperspectral image signal, and combines a relationship model between a gray value and reflectivity constructed by using a gray scale calibration plate hyperspectral image signal to calculate water body spectral reflectivity of the water area to be measured according to the water body polarized hyperspectral image signal; and a water quality parameter spatial distribution inversion module, which is used for inverting a water quality parameter to be measured of the water area to be measured according to a water quality parameter inversion model and the water body spectral reflectivity. The technical scheme provided by the application has high integration and high automation, and can realize rapid deployment, all-day continuous unattended monitoring.

Description

Technical Field

[0001] This invention relates to the field of hyperspectral water quality monitoring technology, and in particular, to a polarization hyperspectral water quality monitoring system and method. Background Technology

[0002] With rapid economic development and population growth, industrial and domestic water consumption has been increasing, leading to water pollution. To address this issue, effective monitoring of the water environment is urgently needed. Currently, water quality monitoring mainly relies on on-site sampling and laboratory testing. Although these methods offer high accuracy, they suffer from drawbacks such as complex operation, use of toxic reagents, potential for secondary pollution, and poor monitoring timeliness.

[0003] Traditional hyperspectral water quality monitoring methods use UAV hyperspectral imagers or spectrometers to observe and collect water samples from the water area under test, establish regression models between sensitive bands and water quality parameters, and realize inversion monitoring of water quality parameters. This method has the advantages of high detection efficiency and no secondary pollution.

[0004] However, traditional hyperspectral water quality monitoring methods still have significant limitations. On the one hand, the technical process of traditional methods relies on manual intervention and requires regular scheduling of drone flights and water sample collection, making it impossible to achieve unattended continuous monitoring and difficult to capture the dynamic changes of sudden pollution events or water quality parameters. On the other hand, the regression models of traditional methods are simple and difficult to solve complex nonlinear problems. For water bodies with complex optical properties, their inversion accuracy and generalization ability are weak. Summary of the Invention

[0005] The purpose of this invention is to provide a polarization hyperspectral water quality monitoring system and method to solve the problem that it is difficult to perform automated, continuous and rapid monitoring of large-area water bodies around the clock in the prior art.

[0006] To achieve the above objectives, the present invention provides a polarization hyperspectral water quality monitoring system, comprising: a tower-based polarization hyperspectral imaging module deployed on a building adjacent to the water body to be measured, which uses a polarization hyperspectral imager to perform area array staring scanning of the water body to acquire a polarization hyperspectral image signal of the water body; an image spectral data processing module, used to subtract dark current and stray signals from the water body polarization hyperspectral image signal, and calculate the water body spectral reflectance of the water body to be measured based on the relationship model between gray values ​​and the reflectance corresponding to the gray values ​​constructed using the gray-scale calibration plate hyperspectral image signal; and a water quality parameter spatial distribution inversion module, used to invert the water quality parameters to be measured in the water body to be measured based on a pre-constructed water quality parameter inversion model and the water body spectral reflectance.

[0007] Preferably, the tower-based polarization hyperspectral imaging module includes: a first information acquisition unit for acquiring dark current and stray signals when the amplitude modulation of the acousto-optic tunable filter is 0%; a second information acquisition unit for acquiring the polarization hyperspectral image signal of the grayscale calibration plate when the amplitude modulation of the acousto-optic tunable filter is 100%; and a third information acquisition unit for acquiring the water body polarization hyperspectral image signal of the water body to be measured when the amplitude modulation of the acousto-optic tunable filter is 100%; wherein the acousto-optic tunable filter is a component of the polarization hyperspectral imager.

[0008] Preferably, the image spectral data processing module includes: a first information processing unit, used to subtract the dark current and the stray signal from the grayscale calibration plate polarization hyperspectral image signal to obtain first processing information; a second information processing unit, used to subtract the dark current and the stray signal from the water body polarization hyperspectral image signal to obtain second processing information; a third information processing unit, used to extract the grayscale mean value of each standard grayscale plate on the grayscale calibration plate from the first processing information, and construct a relationship model between the grayscale mean value and the reflectance corresponding to the grayscale mean value; and a fourth information processing unit, used to calculate the spectral reflectance of the water body based on the relationship model and the second processing information.

[0009] Preferably, the polarization hyperspectral water quality monitoring system provided by the present invention further includes: a water sample spectral measurement module, which uses a field spectrometer to perform spectral measurements on the water body at the sampling point to obtain the spectral reflectance of the water sample; and a water quality parameter inversion model construction module, which constructs the water quality parameter inversion model through the following process: obtaining the equivalent spectral reflectance of each band of the polarization hyperspectral imager by band matching of the water sample spectral reflectance; forming sample information based on the equivalent spectral reflectance, each band corresponding to the equivalent spectral reflectance, and the pre-acquired water quality parameters of the water body at the sampling point, wherein the sample information includes model training... Training samples and model test samples; based on the model training samples, combined with the random frog leaping algorithm and the partial least squares algorithm, the selection probability of each band corresponding to the equivalent spectral reflectance is calculated, the selection probability of each band is sorted from high to low, and the corresponding bands with the highest selection probability α are selected as feature bands, where α is a positive integer; and for the equivalent spectral reflectance of the feature bands in the model training samples, the corresponding water quality parameters of the water samples, and the number of corresponding model training samples, the particle swarm optimization algorithm and the backpropagation neural network algorithm are combined to construct the water quality parameter inversion model.

[0010] Preferably, the step of combining particle swarm optimization (PSO) and backpropagation neural network (HNN) algorithms to construct the water quality parameter inversion model includes: constructing an initial network structure for the HNN algorithm based on the equivalent spectral reflectance of the model training samples in the characteristic band, the corresponding water quality parameters of the water samples, and the number of model training samples corresponding to the characteristic band; using the model training samples corresponding to the characteristic band and the network structure of the HNN algorithm, calculating the global optimal position and the number of optimized nodes in the hidden layer of the HNN algorithm through a finite number of iterations using PSO; decoding and transforming the global optimal position to obtain the optimized weights and optimized thresholds of each layer node in the HNN algorithm; and using the number of optimized nodes, the optimized weights, and the optimized thresholds as initialization parameters of the HNN algorithm, performing backpropagation training using the model training samples, and fine-tuning the parameters through gradient descent to generate the water quality parameter inversion model.

[0011] Accordingly, the present invention also provides a polarization hyperspectral water quality monitoring method, which includes: using a polarization hyperspectral imager to perform area array staring scanning of the water body to be measured, thereby obtaining a polarization hyperspectral image signal of the water body to be measured; subtracting dark current and stray signals from the polarization hyperspectral image signal of the water body, and calculating the spectral reflectance of the water body to be measured based on the polarization hyperspectral image signal of the water body, in conjunction with a relationship model between gray values ​​and the reflectance corresponding to the gray values ​​constructed using a gray-scale calibration plate hyperspectral image signal; and inverting the water quality parameters to be measured of the water body to be measured based on a pre-constructed water quality parameter inversion model and the spectral reflectance of the water body.

[0012] Preferably, acquiring the water body polarization hyperspectral image signal of the water body to be tested includes: acquiring dark current and stray signals when the amplitude modulation of the acousto-optic tunable filter is 0%; acquiring the grayscale calibration plate polarization hyperspectral image signal when the amplitude modulation of the acousto-optic tunable filter is 100%; and acquiring the water body polarization hyperspectral image signal of the water body to be tested when the amplitude modulation of the acousto-optic tunable filter is 100%; wherein, the acousto-optic tunable filter is a component of the polarization hyperspectral imager.

[0013] Preferably, the step of calculating the water spectral reflectance of the water body to be tested based on the water body polarization hyperspectral image signal includes: subtracting the dark current and the stray signal from the grayscale calibration plate polarization hyperspectral image signal to obtain first processing information; subtracting the dark current and the stray signal from the water body polarization hyperspectral image signal to obtain second processing information; extracting the grayscale mean value of each standard grayscale plate on the grayscale calibration plate from the first processing information, constructing a relationship model between the grayscale mean value and the reflectance corresponding to the grayscale mean value; and calculating the water body spectral reflectance based on the relationship model and the second processing information.

[0014] Preferably, the polarization hyperspectral water quality monitoring method provided by the present invention further includes: using a field spectrometer to perform spectral measurements on the water body at the sampling point to obtain the spectral reflectance of the water sample; and constructing the water quality parameter inversion model through the following process: obtaining the equivalent spectral reflectance of each band of the polarization hyperspectral imager by band matching of the spectral reflectance of the water sample; forming sample information based on the equivalent spectral reflectance, each band corresponding to the equivalent spectral reflectance, and the pre-acquired water quality parameters of the water body at the sampling point, wherein the sample information includes model training samples and model test samples; Based on the model training samples, the selection probability of each band corresponding to the equivalent spectral reflectance is calculated by combining the random frog leap algorithm and the partial least squares algorithm. The selection probabilities of each band are sorted from high to low, and the bands with the highest selection probabilities α are selected as feature bands, where α is a positive integer. Furthermore, for the equivalent spectral reflectance of the feature bands in the model training samples, the corresponding water quality parameters of the water samples, and the number of corresponding model training samples, the water quality parameter inversion model is constructed by combining the particle swarm optimization algorithm and the backpropagation neural network algorithm.

[0015] Preferably, the step of combining particle swarm optimization (PSO) and backpropagation neural network (HNN) algorithms to construct the water quality parameter inversion model includes: constructing an initial network structure for the HNN algorithm based on the equivalent spectral reflectance of the model training samples in the characteristic band, the corresponding water quality parameters of the water samples, and the number of model training samples corresponding to the characteristic band; using the model training samples corresponding to the characteristic band and the network structure of the HNN algorithm, calculating the global optimal position and the number of optimized nodes in the hidden layer of the HNN algorithm through a finite number of iterations using PSO; decoding and transforming the global optimal position to obtain the optimized weights and optimized thresholds of each layer node in the HNN algorithm; and using the number of optimized nodes, the optimized weights, and the optimized thresholds as initialization parameters of the HNN algorithm, performing backpropagation training using the model training samples, and fine-tuning the parameters through gradient descent to generate the water quality parameter inversion model.

[0016] This invention acquires polarized hyperspectral image signals of the water body by performing area array staring scanning of the water body using a polarization hyperspectral imager. After subtracting dark current and stray signals from the acquired polarized hyperspectral image signals, the spectral reflectance of the water body is calculated using a model relating the grayscale value of a grayscale target to its corresponding reflectance. Water quality parameters are then retrieved based on a pre-constructed water quality parameter inversion model and the spectral reflectance. The technical solution provided by this invention eliminates the influence of water surface glare by combining an acousto-optic tunable filter and a polarizer, ensuring the validity of observation data at various times and angles. The constructed model relating grayscale value to reflectance provides a high-precision data foundation for subsequent water quality parameter inversion. The selection of characteristic bands of the hyperspectral data is achieved through random frog-jumping algorithm and partial least squares algorithm, improving the accuracy of water quality parameter inversion for complex optical characteristics. The technical solution provided by this invention has high integration and automation, enabling rapid deployment and continuous unattended monitoring around the clock. Attached Figure Description

[0017] The accompanying drawings are provided to further illustrate embodiments of the present invention and form part of the specification. They are used together with the following detailed description to explain the embodiments of the present invention, but do not constitute a limitation thereof. In the drawings:

[0018] Figure 1 This is a block diagram of the polarization hyperspectral water quality monitoring system provided by the present invention;

[0019] Figure 2 This is a schematic diagram of the tower-based polarization hyperspectral imaging module provided by the present invention;

[0020] Figure 3 This is a block diagram of another polarization hyperspectral water quality monitoring system provided by the present invention;

[0021] Figure 4 This is a flowchart of the algorithm combining the random frog jumping algorithm and the partial least squares algorithm provided in a specific embodiment of the present invention;

[0022] Figure 5 This is a flowchart of the backpropagation neural network algorithm for particle swarm optimization provided in a specific embodiment of the present invention;

[0023] Figure 6 This is a flowchart of the polarization hyperspectral water quality monitoring method provided by the present invention;

[0024] Figure 7 This is a schematic diagram of deploying a polarization hyperspectral imaging system on a tower beside the water body to be measured, according to a specific embodiment of the present invention;

[0025] Figure 8This is a 760nm band image provided by a specific embodiment of the present invention, without deducting dark current and spurious signals;

[0026] Figure 9 This is a 760nm band image after deducting dark current and stray signals, provided in a specific embodiment of the present invention. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other. To achieve the above objectives, this invention adopts the following technical solution.

[0028] Figure 1 This is a block diagram of the polarization hyperspectral water quality monitoring system provided by the present invention, as shown below. Figure 1 As shown, the system includes a tower-based polarization hyperspectral imaging module 10, an image spectral data processing module 20, and a water quality parameter spatial distribution inversion module 30.

[0029] The tower-based polarization hyperspectral imaging module 10 is deployed on a building adjacent to the water body to be measured. It uses a polarization hyperspectral imager to perform area array staring scans of the water body to acquire polarization hyperspectral image signals. The tower-based polarization hyperspectral imaging module 10 is a highly integrated, miniaturized detection system that can be deployed on towers, rooftops, or other locations around the water body to be measured, allowing it to observe the water body within the movement range of the multi-functional support. The pitch and rotation angles of the multi-functional observation support 18 corresponding to each scanning field of view can be set according to the size of the water body, the observation distance, and the observation field of view.

[0030] Figure 2 This is a schematic diagram of the tower-based polarization hyperspectral imaging module 10 provided by the present invention, as shown below. Figure 2As shown, the tower-based polarization hyperspectral imaging module 10 includes a polarization hyperspectral imager 11, a GPS 12, a gyroscope 13, a host computer 14, a power supply submodule 15, an I / O submodule 16, a housing 17, a multi-functional observation support 18, and a grayscale calibration plate 19. The polarization hyperspectral imager 11 is used to perform area array staring scanning of the water body to obtain the water body polarization hyperspectral image signal. The Global Positioning System (GPS) 12 is used to locate its associated tower-based polarization hyperspectral imaging module, facilitating the system to calculate the solar altitude angle based on the location of the observation point and the observation date, thus ensuring the signal-to-noise ratio of the measurement signal. The gyroscope 13 is used to measure the rotation angle and angular velocity of the multi-angle observation support to ensure the accuracy of the attitude angle during the staring scanning process. The display screen (not shown in the figure) is connected via I / O submodule 16. The process of acquiring image signals by the polarization hyperspectral imager 11 is controlled by the host computer 14. The time points for area array staring scanning observations, the rotation parameters of the multi-functional observation bracket 18 for each scanning field of view, and the imaging parameters of the polarization hyperspectral imager 11 are set. The multi-functional observation bracket 18 is used to fix the polarization hyperspectral imaging system (which includes the polarization hyperspectral imager 11, GPS 12, gyroscope 13, host computer 14, power supply submodule 15, I / O submodule 16, and housing 17) and grayscale calibration plate 19. Area array staring scanning observations of the water area to be measured are achieved through pitch and rotation. The polarization hyperspectral imager 11 is the core component of the observation. It consists of an acousto-optic tunable filter (AOTF), a polarizer, an optical lens, and a detector. The combination of the AOTF and the polarizer can eliminate the influence of water surface glare and ensure the validity of observation data at each time period and angle. The grayscale calibration plate 19 is used to calibrate the obtained water body polarization hyperspectral image signal to obtain the water body spectral reflectance of the water body to be tested, which is then used for water quality parameter inversion.

[0031] The image spectral data processing module 20 is used to subtract dark current and stray signals from the water body polarization hyperspectral image signal, and combine the relationship model between the gray value constructed using the gray scale calibration plate 19 hyperspectral image signal and the reflectance corresponding to the gray value to calculate the water body spectral reflectance of the water body to be measured based on the water body polarization hyperspectral image signal.

[0032] The grayscale calibration plate 19 consists of multiple uniform regions with different reflectances, typically arranged from low to high, forming multiple brightness gradients (e.g., from 5% to 60% reflectance), with each region corresponding to a specific reflectance value. Based on the relationship model between grayscale values ​​constructed using the hyperspectral image signal from the grayscale calibration plate 19 and the corresponding reflectance, the corresponding water spectral reflectance can be obtained from the grayscale values ​​of the water body polarization hyperspectral image of the water body under test. The water body polarization hyperspectral image signal contains grayscale value (intensity) information.

[0033] The water quality parameter spatial distribution inversion module 30 is used to invert the water quality parameters of the target water area based on a pre-constructed water quality parameter inversion model and the spectral reflectance of the water body. The water quality parameter inversion model is pre-constructed, and its specific construction process will be described below. The water quality parameter inversion model is a model relating water body spectral reflectance to water quality parameters. After constructing the water quality parameter inversion model, the target water quality parameters of the target water area can be inverted based on the spectral reflectance of the water body, thus realizing the spatial distribution inversion of water quality parameters.

[0034] The tower-based polarization hyperspectral imaging module 10 includes a first information acquisition unit, a second information acquisition unit, and a third information acquisition unit. The first information acquisition unit is used to acquire dark current and stray signals when the amplitude modulation of the acousto-optic tunable filter is 0%. The second information acquisition unit is used to acquire the grayscale calibration plate polarization hyperspectral image signal when the amplitude modulation of the acousto-optic tunable filter is 100%. The third information acquisition unit is used to acquire the water body polarization hyperspectral image signal of the water body under test when the amplitude modulation of the acousto-optic tunable filter is 100%. The acousto-optic tunable filter is a component of the polarization hyperspectral imager.

[0035] After the polarization hyperspectral imaging system, including the polarization hyperspectral imager 11, is started, area array staring scan observations of the water area to be measured are performed at pre-set time points. For each area array staring scan observation, the AOTF amplitude is first modulated to 0% to acquire dark current and stray signal D. i,j,k This is used for background and zero-order light subtraction. Then, the AOTF amplitude is modulated to 100%, and the grayscale calibration plate 19 is observed first to obtain the grayscale calibration plate polarization hyperspectral image signal C. i,j,k Then, the water bodies within each field of view are scanned sequentially, and the observation illumination geometry parameters, imaging parameters, and other information for each scanned field of view are recorded, thereby acquiring the water body polarization hyperspectral image signal I of the water body to be measured. i,j,k Where i and j represent the row and column numbers of the hyperspectral image signal, and k represents the band number.

[0036] The image spectral data processing module 20 includes a first information processing unit, a second information processing unit, a third information processing unit, and a fourth information processing unit.

[0037] The first information processing unit is used to subtract dark current and stray signals from the grayscale calibration plate polarization hyperspectral image signal to obtain the first processed information.

[0038] Dark current and stray signals are subtracted from the grayscale calibration plate polarization hyperspectral image signal to obtain a grayscale calibration plate polarization hyperspectral image signal after dark current and stray signals are deducted. In this paper, the grayscale calibration plate polarization hyperspectral image signal after dark current and stray signals are deducted is referred to as the first processed information. The specific calculation formula is as follows:

[0039] (1)

[0040] In formula (1), This represents the grayscale calibration plate polarized hyperspectral image signal after deducting dark current and stray signals, i.e., the first processing information. This represents the original grayscale calibration plate polarization hyperspectral image signal, i.e., the grayscale calibration plate polarization hyperspectral image signal before dark current and stray signals are subtracted. This represents dark current and stray signals. i and j represent the image row and column numbers, and k represents the band number.

[0041] The second information processing unit is used to subtract dark current and stray signals from the water body polarization hyperspectral image signal to obtain second processed information.

[0042] Dark current and stray signals are subtracted from the water body polarization hyperspectral image signal of the water body to be tested, resulting in a water body polarization hyperspectral image signal with dark current and stray signals deducted. In this paper, the water body polarization hyperspectral image signal with dark current and stray signals deducted is referred to as the second processing information. The specific calculation formula is as follows:

[0043] (2)

[0044] In formula (2), This represents the water body polarization hyperspectral image signal after deducting dark current and stray signals, i.e., the second processing information. This represents the original water body polarization hyperspectral image signal, i.e., the water body polarization hyperspectral image signal before dark current and stray signals are subtracted. This represents dark current and stray signals.

[0045] The third information processing unit is used to extract the grayscale mean value of each standard grayscale plate on the grayscale calibration plate 19 from the first processed information, and to construct a relationship model between the grayscale mean value and the reflectance corresponding to the grayscale mean value.

[0046] The grayscale calibration plate polarization hyperspectral image signal after removing dark current and stray signals. In the process, the mean gray value T of each standard grayscale plate on the grayscale calibration plate 19 is extracted. k Construct the grayscale mean T k Its reflectivity Ref k The relationship model is shown in the following formula:

[0047] (3)

[0048] In formula (3), Ref k A represents reflectivity. k and B k T represents the model coefficients. k The gray mean of the k-th band is represented by the model coefficient A, which can be calculated using formula (3). k and B k .

[0049] Constructing a model of the relationship between grayscale mean and reflectance using a grayscale calibration plate 19 is an indispensable step. Essentially, it is a calibration of the output intensity signal of the polarization hyperspectral imaging system, which is the basis for ensuring the accuracy of subsequent measurements and analyses.

[0050] The fourth information processing unit calculates the water spectral reflectance based on the relational model and the second processed information.

[0051] The water spectral reflectance is calculated based on the relational model constructed by the third information processing unit and the water body polarization hyperspectral image signal obtained by the second information processing unit after deducting dark current and stray signals. The specific calculation formula is as follows:

[0052] (4)

[0053] In formula (4), ρ i,j,k A represents the spectral reflectance of water. k and B k This represents the model coefficients calculated using formula (3). This represents a water polarization hyperspectral image signal after removing dark current and stray signals.

[0054] Figure 3 This is a block diagram of another polarization hyperspectral water quality monitoring system provided by the present invention, such as... Figure 3 As shown, the system also includes a water sample spectral measurement module 40 and a water quality parameter inversion model construction module 50.

[0055] The water sample spectral measurement module 40 uses a field spectrometer to perform spectral measurements on the water at the sampling points to obtain the spectral reflectance of the water sample. Specifically, multiple sampling points are set at equal intervals within the water area to be tested, the location information of each sampling point is recorded, and the spectral measurements of the water at the sampling points are performed using a field spectrometer. The spectral reflectance of the water sample is obtained by preprocessing the spectra of the water at the sampling points.

[0056] The water quality parameter inversion model construction module 50 constructs the water quality parameter inversion model through the following process: The equivalent spectral reflectance of each band of the polarization hyperspectral imager is obtained by band matching of the water sample spectral reflectance; based on the equivalent spectral reflectance, the corresponding bands of the equivalent spectral reflectance, and the pre-acquired water quality parameters of the water sample at the sampling point, sample information is formed, including model training samples and model test samples; based on the model training samples, combined with the random frog jumping algorithm and the partial least squares algorithm, the selection probability of each band corresponding to the equivalent spectral reflectance is calculated; the selection probabilities of each band are sorted from high to low, and the bands with the highest selection probabilities are selected. The corresponding bands are used as characteristic bands, among which, The value is a positive integer. A water quality parameter inversion model is constructed by combining particle swarm optimization algorithm and backpropagation neural network algorithm, based on the equivalent spectral reflectance in the characteristic band of the model training samples, the corresponding water quality parameters of the water samples, and the number of corresponding model training samples.

[0057] The process of obtaining water quality parameters of water samples at sampling points in advance is as follows: At each sampling point, a water sample is collected from the surface layer to a depth of 0.5m using a water sample collection bottle. The water quality parameters of the water samples at the sampling points can be obtained by conducting water quality tests on the water samples at the sampling points in accordance with national standards.

[0058] The equivalent spectral reflectance of each band of the polarization hyperspectral imager is obtained by band matching of the water sample spectral reflectance. Specifically, this process includes: obtaining the equivalent spectral reflectance of each band of the polarization hyperspectral imager by band matching of the water sample spectral reflectance measured by the field spectrometer. To ensure consistency between the measured spectrum obtained from the spectral measurement of the water body at the sampling point and the spectral resolution of the polarization hyperspectral imager, the equivalent spectral reflectance of each band of the polarization hyperspectral imager is measured. The calculation method is as follows:

[0059] (5)

[0060] In formula (5), λ is the equivalent spectral reflectance of the k-th band, where λ1 is the starting wavelength of the corresponding band and λ2 is the ending wavelength of the corresponding band. The spectral reflectance of the water sample was measured by a field spectrometer. Let be the spectral response function of the k-th band polarization hyperspectral imager.

[0061] Water quality parameters can include total nitrogen and phosphorus, chlorophyll, dissolved oxygen, turbidity, suspended solids concentration, etc. However, this invention is not limited to these; all water quality parameters are within the scope of protection of this invention. In this invention, the water sample parameters and the parameters to be tested are merely used to distinguish between the water sample and the object being tested. Those skilled in the art should understand that the water sample parameters and the parameters to be tested essentially contain the same water quality parameters. Regarding the various bands corresponding to equivalent spectral reflectance, different water quality parameters correspond to different bands. For example, the band corresponding to chlorophyll is different from the band corresponding to dissolved oxygen.

[0062] The process of forming sample information based on the equivalent spectral reflectance, the corresponding bands of the equivalent spectral reflectance, and the pre-acquired water quality parameters of the water samples at the sampling points is as follows: The equivalent spectral reflectance of the polarization hyperspectral imager obtained by band matching of the spectral reflectance of the water samples at each sampling point is paired with the pre-acquired water quality parameters, that is, the equivalent spectral reflectance is paired with the water quality parameters of the water samples to form sample information. The specific implementation method provided by this invention is to extract three-quarters as model training samples and one-quarter as model test samples according to the distribution of water sample parameters.

[0063] Based on the model training samples, and combining the Random Frog (RF) and Partial Least Squares (PLS) algorithms, the selection probability of each band corresponding to the equivalent spectral reflectance is calculated, and the specific process of selecting characteristic bands will be explained through... Figure 4 To elaborate, Figure 4 This is a flowchart of the algorithm combining the random frog jumping algorithm and the partial least squares algorithm provided in a specific embodiment of the present invention, as shown below. Figure 4 As shown, the process includes:

[0064] Step S401: Formatted storage of model training samples. The model training samples include equivalent spectral reflectance and water quality parameters of water samples used for model training. Specifically, the N sample data (equivalent spectral reflectance and water quality parameters) in the model training samples are formatted and stored. Let X represent the sample data of equivalent spectral reflectance and Y represent the sample data of water quality parameters of water samples. Here, X is an N-row B-column matrix that stores the equivalent spectral reflectance of N samples in B bands, and Y is an N-row 1-column matrix that stores the water quality parameters of N samples. Here, B and N are both positive integers.

[0065] Step S402, data initialization, including the initialization of the iteration number M1, the number of target bands α, the number of initial bands Q, and the initialization subset V0, wherein the initialization subset V0 means: a subset of N samples of X containing Q bands, where M1, α, and Q are all positive integers.

[0066] Step S403: Determine whether the number of iterations is greater than M1. If the result is yes, proceed to step S410. If the result is no, proceed to step S404.

[0067] In step S404, if the result of step S403 is negative, i.e., the number of iterations is less than or equal to M1, then construct a PLS for the current subset V0 and calculate the importance of each band.

[0068] Step S405: Generate a new subset band number based on the current subset band number Q. Specifically, it involves randomly generating a new subset of band numbers according to a normal distribution. .

[0069] Step S406: Generate a new subset based on the importance of each band. That is, the number of new subset bands The corresponding new subset .

[0070] Step S407: Calculate the root mean square error of the new subset and the root mean square error of the old subset based on PLS, wherein the root mean square error of the new subset is used as... The root mean square error of the old subset is denoted by RMSE0.

[0071] Step S408: Determine the root mean square error of the new subset. Is it less than the root mean square error RMSE0 of the old subset? If the result is yes, proceed to step S409; if the result is no, proceed to step S403.

[0072] Step S409: If the result of step S408 is yes, then the root mean square error of the new subset is... If the root mean square error (RMSE) of the new subset is less than that of the old subset (RMSE0), then the new subset is accepted, i.e., V0 = .

[0073] In step S410, if the result of step S403 is yes, that is, if the number of iterations is greater than M1, the iteration ends, and the top α bands in the importance ranking and their importance are output, thus completing the selection of feature bands. These top α bands are the feature bands, and the importance is the selection probability. The feature bands are the bands with the highest selection probability selected from the highest to the lowest in the order of selection probability, where α is a positive integer.

[0074] By combining the above algorithm of random frog jumping and partial least squares, characteristic wavelengths are selected for the water quality parameters to be measured, and the dimensionality reduction of hyperspectral data is achieved for subsequent efficient modeling. It can also be used for the customization of AOTF bands, thereby improving the efficiency of effective data acquisition.

[0075] After calculating the characteristic bands, a water quality parameter inversion model is constructed based on the equivalent spectral reflectance of the characteristic bands and their corresponding water quality parameters, as well as the number of model training samples corresponding to the characteristic bands (that is, the number of model training samples for the equivalent spectral reflectance of the characteristic bands and their corresponding water quality parameters), combined with the particle swarm optimization algorithm and the backpropagation neural network algorithm.

[0076] The water quality parameter inversion model is constructed by combining particle swarm optimization (PSO) and backpropagation neural network (HNN) algorithms. This includes: constructing the initial network structure of the HNN algorithm based on the equivalent spectral reflectance of the model training samples in the characteristic bands, the corresponding water quality parameters, and the number of model training samples corresponding to the characteristic bands; using the model training samples corresponding to the characteristic bands and the network structure of the HNN algorithm, the global optimal position and the number of optimized nodes in the hidden layers of the HNN are calculated through a finite number of iterations using PSO; the global optimal position is decoded and transformed to obtain the optimized weights and thresholds of each layer node in the HNN; the number of optimized nodes, optimized weights, and optimized thresholds are used as the initial parameters of the HNN algorithm, and backpropagation training is performed using the model training samples. The parameters are then fine-tuned using gradient descent to generate the water quality parameter inversion model.

[0077] For the model training samples corresponding to the characteristic bands, the specific process of constructing the water quality parameter inversion model using the Particle Swarm Optimization (PSO) Backpropagation Neural Network (BPNN) algorithm will be explained below. Figure 5 To elaborate, Figure 5 This is a flowchart of the backpropagation neural network algorithm for particle swarm optimization provided in a specific embodiment of the present invention, as shown below. Figure 5 As shown, the process includes:

[0078] Step S501: Format and store the feature sample data. The feature sample data includes model training samples corresponding to the feature bands, specifically the equivalent spectral reflectance and water quality parameters of the water samples corresponding to the feature bands. This feature sample data includes feature training samples and feature test samples. Specifically, the feature sample data is normalized and formatted for storage. X1 is an N1-row, α-column matrix storing the equivalent spectral reflectance of α feature bands for N1 feature training samples; Y1 is an N1-row, 1-column matrix storing the water quality parameters of the water samples for N1 feature training samples (normalized to 0~1); X2 is an N2-row, α-column matrix storing the equivalent spectral reflectance of α feature bands for N2 feature test samples; and Y2 is an N2-row, 1-column matrix storing the water quality parameters of the water samples for N2 feature test samples (normalized to 0~1).

[0079] Step S502: Determine the BPNN network structure based on the model training samples (i.e., feature sample data) corresponding to the α feature bands, including the number of input layer nodes, the number of hidden layers, the number of hidden layer nodes, the number of output layer nodes, the activation function, and the loss function of the BPNN network.

[0080] Step S503: Set the PSO algorithm parameters, which include the number of iterations M2, the number of particles L, the particle dimension D (the total number of weights and thresholds in the BPNN network = the number of input layer nodes × the number of hidden layer nodes + the number of hidden layer nodes + the number of hidden layer nodes × the number of output layer nodes + the number of output layer nodes), and the particle velocity boundary V. max and V min Particle position boundary P max and P min Learning rates c1 and c2, and inertia weight w.

[0081] Step S504: Randomly initialize the position P of each particle. i0 and speed V i0 .

[0082] Step S505: Initialize the fitness F of each particle through forward propagation of BPNN. i (Root mean square error between the predicted result and the true value).

[0083] Step S506: Record the global optimal position P g_best Global optimal fitness F g_best The optimal position P of each particle p_best and optimal fitness F p_best The position of the particle with the lowest fitness among all particles (the globally optimal position) is the globally optimal position, denoted as P. g_best The corresponding fitness (globally optimal fitness) is denoted as F. g_best The optimal position of each particle is denoted as P. p_bestThe optimal fitness of each particle is denoted as F. p_best The particle position and particle velocity are recorded as follows:

[0084] (6)

[0085] In formula (6), i represents the particle number, which ranges from 1 to L, where L is a positive integer. Indicates the particle position. D represents the particle's position, and D represents the particle's dimension.

[0086] Step S507: Determine whether the number of iterations is greater than M2. If the result is yes, proceed to step S515. If the result is no, proceed to step S508.

[0087] Step S508, for each particle, based on its velocity V in its current state i0 Location P i0 Global optimal position P g_best The optimal position P of the particle p_best Calculate its new speed And perform a velocity boundary check, and assign a velocity boundary value if it exceeds the boundary.

[0088] New Particle Velocity The calculation formula is as follows:

[0089] (7)

[0090] In formula (7), V represents the new velocity of the particle. i0 P represents the velocity of the particle in its current state. i0 P represents the position of the particle in its current state. g_best P represents the global optimal position (i.e., the position of the particle with the lowest fitness). p_best Indicates the optimal position of the particle. Indicates inertial weight, and This represents the learning rate, where r1 and r2 are random numbers between 0 and 1.

[0091] Step S509, for each particle, based on its current position P i0 The new particle velocity calculated in step S508 Calculate its new position And perform position boundary checks, and assign position boundary values ​​if the value exceeds the limit.

[0092] New position of the particle The calculation formula is as follows:

[0093] (8)

[0094] In formula (8), P represents the new position of the particle. i0 This indicates the position of the particle in its current state. This represents the new particle velocity calculated in step S508.

[0095] Step S510: For each particle, calculate the fitness of each particle at its new position using BPNN forward propagation. .

[0096] Step S511: Determine the fitness of the particle at its new position. Is it less than the particle's optimal fitness F? p_best If the judgment result is yes, then proceed to step S512; if the judgment result is no, then proceed to step S507.

[0097] Step S512, optimal particle position P p_best Update to new location The optimal fitness F of the particle p_best Update to the fitness at the new position calculated in step S510. .

[0098] Step S513: Determine the fitness of the particle at its new position. Is it less than the global optimal fitness F? g_best If the judgment result is yes, then proceed to step S514; if the judgment result is no, then proceed to step S507.

[0099] Step S514, Global optimal position P g_best Update to new location Global optimal fitness F g_best Update to the fitness at the new position calculated in step S510. .

[0100] In step S515, if the result of step S507 is yes (i.e., the number of iterations is greater than M2), the iteration ends, and the global optimal position P is output. g_best And the number of hidden layer nodes.

[0101] Step S516, for the globally optimal position P output in step S515 g_best Decode and convert the data to obtain the weights and thresholds of each layer node in the BPNN.

[0102] Step S517: Use the weights and thresholds obtained in step S516 as the initialization parameters of the BPNN, and input the feature training samples into the BPNN for backpropagation training.

[0103] Step S518: Determine whether the loss function has converged. If the result is no, proceed to step S519. If the result is yes, proceed to step S520.

[0104] Step S519: Fine-tune the parameters of the water quality parameter inversion model using the gradient descent algorithm.

[0105] Step S520: Output the parameters of the water quality parameter inversion model.

[0106] pass Figure 5 The process of constructing a water quality parameter inversion model shown can generate a water quality parameter inversion model. After generating the water quality parameter inversion model, the model can be verified and its accuracy evaluated using feature test samples.

[0107] By using PSO to intelligently optimize the number of hidden layer nodes, the weights of each node, and the thresholds of BPNN, the BPNN algorithm can avoid getting trapped in local optima, solve the gradient vanishing / exploding problem, improve robustness and prediction accuracy, and play an important role in solving nonlinear problems.

[0108] Figure 6 This is a flowchart of the polarization hyperspectral water quality monitoring method provided by the present invention, as follows: Figure 6 As shown, the method includes:

[0109] Step S601: Use a polarization hyperspectral imager to perform a surface array staring scan of the water body to be measured in order to obtain the polarization hyperspectral image signal of the water body to be measured.

[0110] Step S602: Dark current and stray signals are subtracted from the water body polarization hyperspectral image signal. The water body spectral reflectance of the water body to be tested is calculated based on the relationship model between the gray value constructed using the gray scale calibration plate 19 hyperspectral image signal and the reflectance corresponding to the gray value.

[0111] Step S603: Invert the water quality parameters of the water body to be tested based on the pre-constructed water quality parameter inversion model and the water body spectral reflectance.

[0112] The process of acquiring the water body polarization hyperspectral image signal of the water body to be measured includes: acquiring dark current and stray signals when the amplitude modulation of the acousto-optic tunable filter is 0%; acquiring the grayscale calibration plate polarization hyperspectral image signal when the amplitude modulation of the acousto-optic tunable filter is 100%; and acquiring the water body polarization hyperspectral image signal of the water body to be measured when the amplitude modulation of the acousto-optic tunable filter is 100%. The acousto-optic tunable filter is a component of the polarization hyperspectral imager.

[0113] The calculation of the water spectral reflectance of the water body to be tested based on the water body polarization hyperspectral image signal includes: subtracting dark current and stray signals from the grayscale calibration plate polarization hyperspectral image signal to obtain first processing information; subtracting dark current and stray signals from the water body polarization hyperspectral image signal to obtain second processing information; extracting the grayscale mean value of each standard grayscale plate on the grayscale calibration plate 19 from the first processing information, and constructing a relationship model between the grayscale mean value and the reflectance corresponding to the grayscale mean value; and calculating the water body spectral reflectance based on the second processing information according to the relationship model.

[0114] The polarization hyperspectral water quality monitoring method provided by this invention further includes: using a field spectrometer to perform spectral measurements on the water body at the sampling point to obtain the spectral reflectance of the water sample; constructing a water quality parameter inversion model through the following process: obtaining the equivalent spectral reflectance of each band of the polarization hyperspectral imager by band matching of the water sample spectral reflectance; forming sample information based on the equivalent spectral reflectance, the corresponding bands of the equivalent spectral reflectance, and the pre-acquired water quality parameters of the water body at the sampling point, including model training samples and model test samples; calculating the selection probability of each band corresponding to the equivalent spectral reflectance based on the model training samples, combined with the random frog jumping algorithm and the partial least squares algorithm, sorting the selection probabilities of each band from high to low, and selecting the corresponding bands with the highest selection probabilities α as feature bands, where α is a positive integer; constructing a water quality parameter inversion model by combining the equivalent spectral reflectance of the feature bands in the model training samples, the corresponding water quality parameters of the water sample, and the number of corresponding model training samples, using the particle swarm optimization algorithm and the backpropagation neural network algorithm.

[0115] The water quality parameter inversion model is constructed by combining particle swarm optimization (PSO) and backpropagation neural network (HNN) algorithms. This includes: constructing the initial network structure of the HNN algorithm based on the equivalent spectral reflectance of the model training samples in the characteristic bands, the corresponding water quality parameters, and the number of model training samples corresponding to the characteristic bands; using the model training samples corresponding to the characteristic bands and the network structure of the HNN algorithm, the global optimal position and the number of optimized nodes in the hidden layers of the HNN are calculated through a finite number of iterations using PSO; the global optimal position is decoded and transformed to obtain the optimized weights and thresholds of each layer node in the HNN; the number of optimized nodes, optimized weights, and optimized thresholds are used as the initial parameters of the HNN algorithm, and backpropagation training is performed using the model training samples. The parameters are then fine-tuned using gradient descent to generate the water quality parameter inversion model.

[0116] It should be noted that the specific details and benefits of the polarization hyperspectral water quality monitoring method provided by this invention are similar to those of the polarization hyperspectral water quality monitoring system provided by this invention, and will not be repeated here.

[0117] Figure 7 This is a schematic diagram of a polarization hyperspectral imaging system installed on a tower beside the water body to be measured, provided by a specific embodiment of the present invention. Figure 7 As shown, the multi-functional observation bracket 18 fixes the polarization hyperspectral imaging system (the system includes a polarization hyperspectral imager 11, GPS 12, gyroscope 13, host computer 14, power supply submodule 15, I / O submodule 16 and housing 17). The multi-functional observation bracket 18 and the polarization hyperspectral imaging system are mounted on a tower on the edge of the water area to be measured.

[0118] To enable technicians to more clearly see the difference between images with and without dark current and spurious signals removed, specific embodiments of the present invention provide images of the 760nm band with and without dark current and spurious signals removed. Figure 8 This is a 760nm band image provided by a specific embodiment of the present invention, without deducting dark current and spurious signals. Figure 9 The image provided in the specific embodiment of the invention is a 760nm band image after deducting dark current and stray signals.

[0119] The optional embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the embodiments of the present invention are not limited to the specific details in the above embodiments. Within the scope of the technical concept of the embodiments of the present invention, various simple modifications can be made to the technical solutions of the embodiments of the present invention, and these simple modifications all fall within the protection scope of the embodiments of the present invention.

[0120] The technical solution provided by this invention enables automated, rapid, and continuous monitoring of water quality parameters through intelligent inversion. The monitoring results are stable and reliable, unaffected by periods of solar flare, and effectively save labor costs. Specifically, the technical solution provided by this invention has the following beneficial effects:

[0121] (1) A polarization hyperspectral imaging system is used to conduct unattended and continuous observation of large-area water bodies, and to collect polarization hyperspectral images of the water bodies to be measured for inversion of the water quality parameters to be measured. An acousto-optic tunable filter (AOTF) is used as the spectroscopic device, which has the advantages of fast band tuning speed and programmable customization. Combined with a polarizer, the influence of water surface glare can be eliminated, ensuring the validity of observation data at all times and angles. The device has a high degree of integration, high degree of automation, and low operation difficulty, and can realize rapid deployment and continuous unattended monitoring around the clock.

[0122] (2) A hyperspectral data dimensionality reduction algorithm combining random frog jumping (RF) and partial least squares (PLS) was proposed and applied to the field of water quality monitoring for the first time. It can accurately select the characteristic wavelengths of the water quality parameters to be measured, efficiently construct the water quality parameter inversion model, and play an important supporting role in the customization and efficient observation of the AOTF band of the polarization hyperspectral imaging system.

[0123] (3) The Particle Swarm Optimized Backpropagation Neural Network (PSO-BPNN) algorithm was used to construct a water quality parameter inversion model. This algorithm uses PSO to intelligently optimize the number of hidden layer nodes, weights, and thresholds of the BPNN, avoiding the BPNN algorithm from getting trapped in local optima, solving the gradient vanishing / exploding problem, and improving its robustness and prediction accuracy. This is of great significance for solving nonlinear problems. The constructed water quality parameter inversion model has high accuracy and strong generalization ability, and is suitable for intelligent inversion of various water quality parameters.

[0124] It should also be noted that the various specific technical features described in the above embodiments can be combined in any suitable manner without contradiction. To avoid unnecessary repetition, the embodiments of the present invention will not describe the various possible combinations separately.

[0125] Furthermore, various different implementations of the present invention can be combined arbitrarily, as long as they do not violate the spirit of the present invention, they should also be regarded as the content disclosed in the present invention.

Claims

1. A polarization hyperspectral water quality monitoring system, characterized in that, The system includes: The tower-based polarization hyperspectral imaging module is deployed on a building next to the water body to be measured. It uses a polarization hyperspectral imager to perform a staring scan of the water body to be measured in order to obtain the water body polarization hyperspectral image signal of the water body to be measured. The image spectral data processing module is used to subtract dark current and stray signals from the water body polarization hyperspectral image signal, and, in conjunction with the relationship model between gray values ​​constructed using the gray-scale calibration plate polarization hyperspectral image signal and the reflectance corresponding to those gray values, calculate the water spectral reflectance of the water body to be measured based on the water body polarization hyperspectral image signal; and, The water quality parameter spatial distribution inversion module is used to invert the water quality parameters to be measured in the water body under test based on the pre-constructed water quality parameter inversion model and the spectral reflectance of the water body. The system also includes: The water sample spectral measurement module uses a field spectrometer to perform spectral measurements on the water at the sampling point, obtaining the spectral reflectance of the water sample; and, The water quality parameter inversion model construction module constructs the water quality parameter inversion model through the following process: The equivalent spectral reflectance of each band of the polarization hyperspectral imager is obtained by band matching of the spectral reflectance of the water sample. Based on the equivalent spectral reflectance, the corresponding spectral reflectance bands, and the pre-acquired water quality parameters of the water samples at the sampling points, sample information is formed, which includes model training samples and model test samples. Based on the training samples of the aforementioned model, and combining the random frog-jump algorithm and the partial least squares algorithm, the selection probability of each band corresponding to the equivalent spectral reflectance is calculated. The selection probabilities of each band are then sorted from high to low, and the bands with the highest selection probabilities are selected. α The corresponding band is used as the characteristic band, wherein the α It is a positive integer; Based on the equivalent spectral reflectance of the model training samples in the characteristic band, the corresponding water quality parameters of the water samples, and the number of corresponding model training samples, a water quality parameter inversion model is constructed by combining particle swarm optimization algorithm and backpropagation neural network algorithm. The tower-based polarization hyperspectral imaging module integrates an acousto-optic tunable filter and its amplitude modulation controller.

2. The polarization hyperspectral water quality monitoring system according to claim 1, characterized in that, The tower-based polarization hyperspectral imaging module includes: The first information acquisition unit is used to acquire dark current and stray signals when the amplitude modulation of the acousto-optic tunable filter is 0%. The second information acquisition unit is used to acquire the polarization hyperspectral image signal of the grayscale calibration plate when the amplitude modulation of the acousto-optic tunable filter is 100%; and, The third information acquisition unit is used to acquire the water polarization hyperspectral image signal of the water body to be tested when the amplitude modulation of the acousto-optic tunable filter is 100%. The acousto-optic tunable filter is a component of the polarization hyperspectral imager.

3. The polarization hyperspectral water quality monitoring system according to claim 2, characterized in that, The image spectral data processing module includes: The first information processing unit is used to subtract the dark current and the stray signal from the grayscale calibration plate polarization hyperspectral image signal to obtain first processing information; The second information processing unit is used to subtract the dark current and the stray signal from the water body polarization hyperspectral image signal to obtain second processed information; The third information processing unit is used to extract the average grayscale value of each standard grayscale plate on the grayscale calibration plate from the first processed information, and to construct a relationship model between the average grayscale value and the reflectance corresponding to the average grayscale value; and... The fourth information processing unit calculates the spectral reflectance of the water body based on the relationship model and the second processed information.

4. The polarization hyperspectral water quality monitoring system according to claim 3, characterized in that, The method of constructing the water quality parameter inversion model by combining particle swarm optimization algorithm and backpropagation neural network algorithm includes: Based on the equivalent spectral reflectance of the model training samples in the characteristic band and the corresponding water quality parameters of the water samples, and the number of model training samples corresponding to the characteristic band, the initial network structure of the backpropagation neural network algorithm is constructed. Based on the model training samples corresponding to the characteristic bands and the network structure of the backpropagation neural network algorithm, the global optimal position and the number of optimized nodes in the hidden layer of the backpropagation neural network are obtained by using the particle swarm optimization algorithm through a finite number of iterations. Decoding and transforming the global optimal position yields the optimized weights and optimized thresholds of each layer node in the backpropagation neural network; and, The number of optimized nodes, the optimized weights, and the optimized thresholds are used as the initialization parameters of the backpropagation neural network algorithm. The model training samples are used for backpropagation training, and the parameters are fine-tuned using the gradient descent method to generate the water quality parameter inversion model.

5. A polarization hyperspectral water quality monitoring method, characterized in that, The method includes: A polarization hyperspectral imager was used to perform a staring scan of the water body under test to obtain the polarization hyperspectral image signal of the water body under test. Dark current and stray signals are subtracted from the water body polarization hyperspectral image signal. Combined with the relationship model between grayscale values ​​constructed using the grayscale calibration plate polarization hyperspectral image signal and the corresponding reflectance, the spectral reflectance of the water body to be measured is calculated based on the water body polarization hyperspectral image signal; and... The water quality parameters to be measured in the water body are inverted based on the pre-constructed water quality parameter inversion model and the spectral reflectance of the water body. The method also includes: The spectral reflectance of the water sample was obtained by performing spectral measurements on the water sample using a field spectrometer; and The water quality parameter inversion model is constructed using the following process: The equivalent spectral reflectance of each band of the polarization hyperspectral imager is obtained by band matching of the spectral reflectance of the water sample. Based on the equivalent spectral reflectance, the corresponding spectral reflectance bands, and the pre-acquired water quality parameters of the water samples at the sampling points, sample information is formed, which includes model training samples and model test samples. Based on the training samples of the aforementioned model, and combining the random frog-jump algorithm and the partial least squares algorithm, the selection probability of each band corresponding to the equivalent spectral reflectance is calculated. The selection probabilities of each band are then sorted from high to low, and the bands with the highest selection probabilities are selected. α The corresponding band is used as the characteristic band, wherein the α are positive integers; and, Based on the equivalent spectral reflectance of the characteristic band in the model training samples and the corresponding water quality parameters and the number of corresponding model training samples, the particle swarm optimization algorithm and the backpropagation neural network algorithm are combined to construct the water quality parameter inversion model; The tower-based polarization hyperspectral imaging module integrates an acousto-optic tunable filter and its amplitude modulation controller.

6. The polarization hyperspectral water quality monitoring method according to claim 5, characterized in that, The process of acquiring the water polarization hyperspectral image signal of the water body to be measured includes: Dark current and stray signals were obtained when the amplitude modulation of the acousto-optic tunable filter was 0%. With the amplitude modulation of the acousto-optic tunable filter at 100%, the polarization hyperspectral image signal of the grayscale calibration plate is acquired; and, With the amplitude modulation of the acousto-optic tunable filter set to 100%, the water body polarization hyperspectral image signal of the water body to be tested is acquired. The acousto-optic tunable filter is a component of the polarization hyperspectral imager.

7. The polarization hyperspectral water quality monitoring method according to claim 6, characterized in that, The calculation of the water spectral reflectance of the water body to be tested based on the water body polarization hyperspectral image signal includes: The dark current and the stray signal are subtracted from the grayscale calibration plate polarization hyperspectral image signal to obtain the first processing information; The dark current and stray signals are subtracted from the water body polarization hyperspectral image signal to obtain the second processed information; Extract the mean grayscale values ​​of each standard grayscale plate on the grayscale calibration plate from the first processed information, and construct a relationship model between the mean grayscale value and the reflectance corresponding to the mean grayscale value; and... Based on the relationship model, the spectral reflectance of the water body is calculated according to the second processing information.

8. The polarization hyperspectral water quality monitoring method according to claim 7, characterized in that, The method of constructing the water quality parameter inversion model by combining particle swarm optimization algorithm and backpropagation neural network algorithm includes: Based on the equivalent spectral reflectance of the model training samples in the characteristic band and the corresponding water quality parameters of the water samples, and the number of model training samples corresponding to the characteristic band, the initial network structure of the backpropagation neural network algorithm is constructed. Based on the model training samples corresponding to the characteristic bands and the network structure of the backpropagation neural network algorithm, the global optimal position and the number of optimized nodes in the hidden layer of the backpropagation neural network are obtained by using the particle swarm optimization algorithm through a finite number of iterations. Decoding and transforming the global optimal position yields the optimized weights and optimized thresholds of each layer node in the backpropagation neural network; and, The number of optimized nodes, the optimized weights, and the optimized thresholds are used as the initialization parameters of the backpropagation neural network algorithm. The model training samples are used for backpropagation training, and the parameters are fine-tuned using the gradient descent method to generate the water quality parameter inversion model.

Images