CN112630189B - Inland water body water quality inversion method based on improved QAA algorithm - Google Patents

Abstract

The invention provides an inland water body water quality inversion method based on an improved QAA algorithm. Firstly, based on the assumption that the inherent optical characteristics of the water bodies are similar between the adjacent points of the water bodies and the absorption coefficient and the backscattering coefficient are both greater than 0, an optimization objective function of the original QAA algorithm is defined. Secondly, combining the geographical position of the water body and the shape characteristics of the water body, selecting a part or all of remote sensing images of the water body as an interested area set, obtaining initial inherent optical quantity of the water body by using a QAA algorithm, and continuously adjusting parameters of the QAA algorithm until the value of a target function is minimum by approximating an optimization target. And then, calculating the inherent optical quantity of the water body by using the improved QAA algorithm, and constructing a water quality parameter inversion model by using the inherent optical quantity of the water body. And finally, obtaining a water quality parameter spatial distribution map. The invention can improve the QAA algorithm suitable for the ocean water body under the condition of no actually measured water body optical quantity data, so that the method can invert the inland water body, and solves the problems of high labor cost and economic cost, time and labor consumption, and incapability of quick and large-area inversion of the traditional water quality monitoring method.

Description

An inversion method of inland water quality based on improved QAA algorithm

technical field

The invention belongs to the field of measurement, and more particularly, relates to a method for obtaining water quality parameters of inland water bodies by using optical remote sensing data.

Background technique

Water is the blood of the ecosystem and an indispensable natural resource for human survival and development. Sufficient and high-quality water resources are the material basis for the healthy development of ecosystems, as well as the material guarantee for national economic and social development. In recent decades, in the process of my country's industrialization and urbanization, the ecological environment of inland water bodies has been constantly changing, which has also spawned many contradictions between ecological environmental protection and economic and social development. Domestic waste water pollution, agricultural pollution, industrial pollution, etc. have caused great damage to the water environment, which makes the development of my country's water environment face great challenges. Water quality monitoring is the main basis for water quality evaluation and water pollution prevention and control. Fast and accurate water quality monitoring is crucial to social and economic development. The traditional water quality monitoring method needs to set up sampling points in the water body for sampling, and obtain the corresponding water quality indicators through laboratory analysis. The emergence and development of remote sensing water quality inversion technology provides new options for water quality monitoring. Remote sensing water quality inversion uses empirical, semi-empirical or physical analysis methods to select appropriate remote sensing data and establish a remote sensing inversion model for water quality parameters to invert water quality parameters in water bodies, which is fast, accurate, low-cost, and scalable. Area batch inversion of water quality and other advantages.

There are three types of water quality inversion methods: empirical method, analytical method and semi-analytical method. Among them, the empirical method uses data mining and machine learning methods to find the relationship between water quality data and remote sensing reflectance data, and establishes an inversion model, which has the advantages of high inversion accuracy, strong universality, and can invert any water body. The disadvantage is that it requires a large amount of water quality measured data and remote sensing data, and the analysis method has no physical basis and is not universal; the analysis method requires a detailed and accurate analysis of the physical and optical characteristics of the water body to be studied, and the establishment of a physical-optical model of the water body, its advantages In order to have strong physical mechanism support, the disadvantage is that it is difficult to establish the model; the semi-analytical method can be regarded as a combination of empirical method and analytical method, and its advantages are that it has a certain physical basis and the sample demand is not large, but it is not suitable for inland water bodies. In other words, the semi-analytical method is weak in generality and needs to establish different models for different water bodies.

The water quality inversion algorithm based on QAA is a kind of semi-analytical inversion method. The full name of QAA algorithm is Quasi-Analytical Algorithm. The QAA algorithm can use the original remote sensing reflectivity data to invert to obtain the inherent optical quantities of the water body such as absorption coefficient and backscattering coefficient, and use the inherent optical quantities of the water body to invert the corresponding water quality parameters. The QAA algorithm provides a direct and efficient method for retrieving the intrinsic optical quantities of water bodies. According to the report of the International Ocean Color Remote Sensing Organization, the QAA algorithm has high applicability for clear seawater and coastal waters; but for nearshore and inland waters In terms of water bodies, since inland water bodies are affected by nearshore structures and human activities, their applicability is poor, and the original QAA algorithm needs to be adjusted before it can be applied to inland water bodies. At present, the inland water quality inversion based on improved QAA needs to measure the remote sensing reflectance and intrinsic optical quantity of the water body first, and based on the analysis of the remote sensing reflectance and intrinsic optical quantity data of the water body, set the reference wavelength in the QAA algorithm, adjust the QAA The parameters in the algorithm, and then use the modified QAA algorithm to invert the water quality parameters. At present, the shortcomings of the existing inland water quality inversion algorithms based on the improved QAA algorithm mainly include:

(1) In the existing inland water quality inversion algorithm based on the improved QAA algorithm, in order to obtain the inversion model of the inherent optical quantity of the water body, it is first necessary to measure the inherent optical quantities such as the absorption coefficient and the backscattering coefficient of the water body. The inherent optical quantity of water body needs to be measured quasi-simultaneously and quasi-situ with remote sensing data, and it needs the support of sufficient measuring devices, labor cost, economic cost and time cost. Therefore, it cannot meet the requirements of the convenience and rapidity of model establishment.

(2) In the existing inland water quality inversion algorithm based on the improved QAA algorithm, in order to realize the inversion of water quality parameters, it is first necessary to establish an inversion model of intrinsic optical quantities, and for different water bodies, it is necessary to establish different Intrinsic optical quantity inversion model. That is to say, the existing algorithms have the disadvantages of weak universality, poor transferability of the inversion model, and large repetitive workload when constructing inversion models for multiple water bodies.

SUMMARY OF THE INVENTION

In view of the fact that the existing technology needs to manually measure the inherent optical quantities such as water absorption coefficient and backscattering coefficient when inverting the water quality parameters of inland water, the model building process is complicated, and the cost of labor, equipment and time is high; In the inversion model of water body, it is necessary to measure the inherent optical quantity of each water body, and it is necessary to adjust the QAA model for each water body, which is difficult to popularize the model, and has a large amount of repetitive work. Inland water quality inversion method.

The QAA algorithm, the full name of Quasi-Analytical Algorithm, is a relatively mature semi-analytical model method based on the principle of radiative transfer to estimate the backscattering coefficient of water bodies. In the present invention, the fifth edition of the QAA algorithm issued by the International Ocean Water Color Remote Sensing Organization is used, the full name is Quasi-Analytical Algorithm Version 5, that is, QAA_v5". The Adam algorithm, the full name of which is Adaptive Moment Estimation, that is, adaptive moment estimation, is an alternative A first-order optimization algorithm for the traditional stochastic gradient descent process that iteratively updates model weights based on training data.

The improved QAA algorithm is an improved method for the QAA algorithm suitable for clear seawater and coastal waters. By defining the objective function and adjusting the parameters of the QAA using the Adam algorithm, the approximation of the objective function and the QAA are realized. Algorithm improvements. This method can realize the adjustment of the parameters in the QAA algorithm without actually measuring the water absorption coefficient and backscattering coefficient data, which provides a new idea for the water quality inversion of inland water using the semi-analytical method.

The technical solution adopted in the present invention is that the inversion method for inland water body water quality based on the improved QAA algorithm includes the steps:

Step 1: Obtain remote sensing image data of inland water bodies, and perform image preprocessing on the remote sensing images. The preprocessing steps generally include: geometric correction, radiation correction, and atmospheric correction;

Step 2, use the improved normalized difference water index, namely NDWI, the full name is Normalized Difference Water Index, extract the water body, and use the morphological processing method to post-process the extracted water body, filter the non-water body area, and smooth the water body boundary;

Step 3: Select remote sensing images of part or all of the water body as a collection of regions of interest in combination with the geographical location of the water body and the shape characteristics of the water body;

Step 4, define the objective function, and use Adam algorithm to adjust the parameters in the QAA algorithm to realize the optimization of the objective function, and further, realize the improvement of the QAA algorithm;

Step 5, use the improved QAA algorithm to invert the water absorption coefficient and the backscattering coefficient;

Step 6: Complete the training of the water quality parameter inversion model by using the inherent optical quantity of the water body and the measured data of water quality parameters, and use the trained water quality parameter inversion model to invert the water quality parameters of the water body, and obtain the spatial distribution of the water quality parameters.

All in all, compared with the prior art, the technical solution conceived by the present invention has the following advantages:

(1) The inland water quality inversion method based on the improved QAA algorithm proposed by the present invention does not need to obtain the inherent optical quantities such as the absorption coefficient and backscattering coefficient of the water body in advance, but utilizes the inherent optical properties between adjacent positions of the same water body. The properties are homogenous and reasonable, and the water absorption coefficient and backscattering coefficient should be greater than 0, which realizes the unsupervised adjustment of the QAA algorithm.

(2) The method for improving the original QAA algorithm proposed by the present invention is not limited by regions, and can be easily extended to the water quality inversion of other inland water bodies.

(3) The present invention belongs to one of the semi-analytical methods. Compared with the empirical method, the present invention does not require a large amount of measured water quality data and remote sensing data, and only limited data can be used to achieve a certain inversion accuracy.

Description of drawings

FIG. 1 is a flowchart of a method for inversion of inland water quality based on an improved QAA algorithm provided by an embodiment of the present invention.

Fig. 2 is the location of the research area and the distribution map of sampling points provided by the embodiment of the present invention

FIG. 3 is an RGB band diagram of a remote sensing image provided by an embodiment of the present invention.

FIG. 4 is a candidate water body image provided by an embodiment of the present invention.

FIG. 5 is a rectangular convolution operator provided by an embodiment of the present invention.

FIG. 6 is an effect diagram of water and land segmentation provided by an embodiment of the present invention.

FIG. 7 is a candidate region set provided by an embodiment of the present invention.

FIG. 8 is a fitting effect diagram of a first-order linear model provided by an embodiment of the present invention.

FIG. 9 is a spatial distribution diagram of the concentration of suspended solids in a first-order linear model provided by an embodiment of the present invention.

Detailed ways

In order to make the objectives, technical solutions and advantages of the present invention clearer, the following describes the inland water quality inversion method based on the improved QAA algorithm provided by the present invention in detail with reference to the accompanying drawings and embodiments. It should be noted that the specific embodiments described herein are only used to explain the present invention, but not to limit the present invention.

The embodiment of the present invention, based on the improved QAA algorithm inversion method for inland water quality, as shown in Figure 1, includes:

Step 1: Obtain remote sensing image data of inland water bodies, and perform image preprocessing on the remote sensing images: including one or more of radiometric calibration, geometric correction, and atmospheric correction.

Step 2, using the improved normalized differential water index NDWI to extract the water body, and use the morphological processing method to post-process the extracted water body, filter out the non-water body area, and smooth the water body boundary.

Step 3: Select remote sensing images of part or all of the water body as a collection of regions of interest in combination with the geographical location of the water body and the shape characteristics of the water body.

Step 4, define the objective function, and use the Adam algorithm to adjust the parameters in the original QAA algorithm, so as to realize the optimization of the objective function, and further, realize the improvement of the QAA algorithm.

Step 5, use the improved QAA algorithm to invert the water absorption coefficient and the backscattering coefficient.

Step 6: Complete the training of the water quality parameter inversion model by using the inherent optical quantity of the water body and the measured data of water quality parameters, and use the trained water quality parameter inversion model to invert the water quality parameters of the water body, and obtain the spatial distribution of the water quality parameters.

Wherein, in step 1, the preprocessing step of the remote sensing image includes one or more of radiometric calibration, geometric correction, and atmospheric correction. The embodiment of the present invention provides a remote sensing image with a length of 2394 pixels, a width of 460 pixels, and a total of 270 bands. As shown in Figure 2, it is a map of the location of the research area and the distribution of sampling points. As shown in Figure 3, it is the preprocessed RGB band remote sensing images.

Further, step 2 includes:

Step 2-1, use the improved normalized difference water index NDWI to achieve water and land segmentation, where the full name of NDWI is Normalized Difference Water Index, and the water body area is obtained:

In the formula, R ₇₅₀ and R ₅₆₀ are the apparent reflectivity of the water body at the wavelength of 750 and 560, respectively, the NDWI is between -1 and 1, and when the NDWI is greater than or equal to the threshold 0, the pixel is a candidate water body. As shown in Figure 4, the candidate water body is obtained by using the improved normalized differential water body index NDWI. The candidate water body has the problems of unclean water body segmentation, uneven land and water boundary, and some water bodies are disconnected from the main area. Therefore, morphological post-processing is required. .

In step 2-2, the specific implementation method of post-processing the extracted water body by using the morphological processing method is as follows:

Select the 5*5 convolution kernel as shown in Figure 5, and use the morphological opening operation to eliminate the connection between some non-water areas and water areas; mark the connected areas of candidate water bodies, and filter the connected areas of non-water bodies according to the size of the area; The 5*5 convolution kernel shown in 5 uses morphological closing operation to smooth the water-land boundary. Figure 4 and Figure 6 show the effect of water and land segmentation before and after morphological processing, respectively, where the white part is the water body area, and the black part is the land area. As shown in Figure 4, the water body area obtained before morphological processing contains many misclassified non-water body areas, and the water body edge is not smooth enough; Figure 6 shows the water body after morphological processing, the water body extraction is more accurate and Smoother edges.

In step 3, it is necessary to select R regions of suitable size to form a set of regions of interest in combination with the geographical location and shape characteristics of the water body. When selecting the area, the geographical location and shape characteristics of the water body should be combined, and the area far from the shore and less affected by human activities should be selected, and the selected area of interest should cover all areas of the water body. For example, lakes should include lakes. The east, south, west, north, and middle areas of the river should include the upstream and downstream of the river; theoretically, under the premise of satisfying the first two principles, the area of the area of interest should be as large as possible. Considering the amount of subsequent calculations, the area of the water body covered by the regional collection can be moderately reduced. Figure 7 shows a set of regions of interest selected by the embodiment of the present invention. In the embodiment of the present invention, four regions including upstream and downstream, and near the north and south banks are selected as regions of interest, that is, R=4, the size of the four regions All are 100*50, and the center coordinates of the 0-3 areas are (185,400), (355,870), (215, 1350), (135,1770) respectively.

Step 4, improve the QAA algorithm, including:

Step 4-1, define the objective function, the specific expression of the objective function is as follows:

重要性的权重，ρ的取值范围为0<ρ<1,本实施例中，ρ＝0.005。

为λ波长，w参数时， (row_i,col_i)坐标处的吸收系数，

为在λ波长，w参数时，(row_i,col_i) 坐标处的后向散射系数，

为在参数w波长λ下，(row_i,col_i) 坐标处的水体吸收系数与其8邻域内其他坐标处水体吸收系数差值平方的加权和，

为在参数w波长λ下，(row_i,col_i)坐标处的水体后向散射系数与其8邻域内其他坐标处水体后向散射系数差值平方的加权和，以水体吸收系数为例，

的具体表现形式如下：where R is the size of the area of interest set, in this embodiment, R=4; λ is the wavelength, (row _i , col _i ) is the pixel coordinate of the row-th row and the col-th column in the i-th area, and w is w ₀ , w ₁ , w ₂ , w ₃ , w ₄ , that is, w=[w ₀ , w ₁ , w ₂ , w ₃ , w ₄ ], the initial value of w is w=[0.39, 1.14, 2.0, -2.4,-0.9]; ρ is the last item in the control objective function

The weight of importance, the value range of ρ is 0<ρ<1, in this embodiment, ρ=0.005.

is the wavelength of λ, the absorption coefficient at (row _i , col _i ) coordinates when the w parameter,

is the backscattering coefficient at (row _i , col _i ) coordinates at λ wavelength, w parameter,

is the weighted sum of the square of the difference between the absorption coefficient of the water body at the coordinate (row _i , col _i ) and the absorption coefficient of the water body at other coordinates in the 8 neighborhood under the parameter w wavelength λ,

is the weighted sum of the squares of the difference between the backscattering coefficient of the water body at the coordinate (row _i , col _i ) and the backscattering coefficient of the water body at other coordinates in the 8 neighborhood under the parameter w wavelength λ, taking the water body absorption coefficient as an example,

The specific manifestations are as follows:

Among them, a(λ) and b _bp (λ) are the water absorption coefficient and backscattering coefficient at the wavelength of λ, respectively, a(λ) and b _bp (λ) are calculated by the QAA algorithm, and the calculation steps are as follows:

As shown in formula (4), the remote sensing reflectivity R _rs (λ) of the lower surface of the water body is obtained by using the remote sensing reflectivity R _rs (λ) of the upper surface of the water body.

Further, as shown in formula (5), using the remote sensing reflectivity r _rs (λ) of the lower surface of the water body, the ratio μ(λ) of the backscattering coefficient to the sum of the backscattering coefficient and the absorption coefficient is obtained.

Further, as shown in formula (6), the reference wavelength λ ₀ is specified to be 670 nm.

λ ₀ =670nm (6)

Further, as shown in formula (7), the water absorption coefficient at the reference wavelength λ ₀ is obtained, where a _w (λ ₀ ) is the water absorption coefficient of pure water at the reference wavelength λ ₀ , w ₀ , w ₁ are The initial values of the two parameters that need to be adjusted are w ₀ =0.39 and w ₁ =1.14.

Further, as shown in formula (8), the backscattering coefficient of the water body at the reference wavelength λ ₀ is obtained, where b _bw (λ ₀ ) is the backscattering coefficient of the water body of pure water at the reference wavelength λ ₀ .

Further, as shown in formula (9), the slope Y of the backscattering spectrum of the particulate matter is obtained, wherein w ₂ , w ₃ and w ₄ are three parameters that need to be adjusted, and the initial values are w ₂ =2.0, w ₃ = -2.4, w ₄ =-0.9.

Further, as shown in formula (10), the backscattering coefficient b _bp (λ) of water body at all wavelengths is obtained, and as shown in formula (11), the absorption coefficient a(λ) of water body at all wavelengths is obtained.

Step 4-2, set the maximum number of iterations to 500, and adjust the value of the parameter w after 500 iterations to improve the QAA algorithm. Finally, the result of obtaining the parameter w is w = [-0.64367, 0.40317, -0.85593, 241.39779, -5.5001].

Among them, the training samples come from the sample points in the region of interest set obtained in step 3. In each iteration, the absorption coefficient and backscattering coefficient of each sample point are calculated, and the target defined in step 4-1 is obtained. function value. In each iteration, the Adam algorithm is used to adjust the value of the parameter w to optimize the objective function. The parameter configuration of Adam algorithm is: learning rate=0.01, beta1=0.9, beta2=0.999, eps=1e-08, weight_decay=0.

Step 5, use the improved QAA algorithm to invert the absorption coefficient and backscattering coefficient of each point in the study area.

Step 6: Complete the training of water quality parameter inversion model using the water absorption coefficient, backscattering coefficient and water quality parameter measured data, use the trained water quality parameter inversion model to invert the water quality parameters of the water body, and obtain the spatial distribution of water quality parameters. Among them, water quality parameters include chlorophyll, total suspended solids concentration and turbidity. The specific implementation method for completing the training of the water quality parameter inversion model by using the inherent optical quantity of the water body and the measured data of water quality parameters is as follows:

The normalized difference index of absorption coefficient, NDAI, is used to expand the band. The full name of NDAI is Normalized Difference Absorption Index. The correlation between the inherent optical quantity of the water body and the water quality parameters is calculated, and the band is selected according to the correlation. The optimal band is obtained at the wavelength of 451 nm. The water absorption coefficient and the normalized difference index of the water absorption coefficient at the wavelength of 449 nm are shown in formula (12).

Further, an inversion model of water quality parameters was constructed using linear regression. Among them, the model training is realized by linear regression algorithm, the sample label is the water quality data measured at the water quality sampling point, and the training data is the water absorption coefficient and data in the optimal band. Select 20% of the data as the test set. Figure 8 shows the fitting effect of the model on the test set, where the horizontal axis is the actual value of the suspended solids concentration, and the vertical axis is the predicted value of the suspended solids concentration. The distribution of the actual and predicted values of the suspended solids concentration of each sample. The straight line in the figure is the ideal distribution of suspended solids concentration. The closer the point is to the straight line, the better the model effect is. It can be seen from the figure that the suspended solids obtained in this example is obtained. The overall accuracy of the inversion model is high, which can meet the requirements of inversion of suspended matter in inland water.

Further, the water quality parameters in the study area are inverted using the obtained water quality parameter inversion model.

Further, the spatial distribution of water quality parameters is obtained by image mosaicking. Figure 9 shows the spatial distribution of the suspended solids concentration of the first-order linear model. The more blue it is, the lower the suspended solids concentration is; after the color map is converted to black and white, the lighter the water color is, the higher the suspended solids concentration is, and the darker the water color is, the lower the suspended solids concentration is.

Compared with the prior art, the method of the embodiment of the present invention does not need to obtain the inherent optical quantities such as the absorption coefficient and backscattering coefficient of the water body in advance, but utilizes the similar inherent optical characteristics between adjacent positions of the same water body, and the absorption coefficient of the water body is the same as the backscattering coefficient. The property that the scattering coefficient should be greater than 0 realizes the unsupervised adjustment of the QAA algorithm; the method for improving the original QAA algorithm proposed in the embodiment of the present invention is not limited by the region, and can be easily extended to other inland water bodies. In water quality inversion; the method of the embodiment of the present invention does not require a large amount of measured water quality data and remote sensing data, and only limited data can be used to achieve a certain inversion accuracy.

The description of the drawings shown in the embodiments of the present invention can make the purpose, technical solutions and advantages of the present invention more clearly introduced. It should be noted that the specific embodiments described herein are only used to explain the present invention, but not to limit the present invention. All equivalent replacements, improvements, etc. made within the method ideas and principles provided by the present invention shall be included within the protection scope of the present invention.

Claims (7)

1. an inland water quality inversion method based on improving QAA algorithm, is characterized in that, comprises: (1) Obtain remote sensing image data of inland water bodies; (2) Use the improved normalized differential water index NDWI to extract the water body, and use the morphological processing method to post-process the extracted water body; (3) Select remote sensing images of some or all water bodies as a collection of regions of interest in combination with the geographical location of the water body and the shape of the water body; (4) Define the optimization objective function of the original QAA algorithm, use the QAA algorithm to obtain the initial water intrinsic optical quantity in the region of interest, and use the Adam algorithm to continuously adjust the parameters of the QAA algorithm until the objective function value is the smallest; (5) Using the improved QAA algorithm to invert the inherent optical quantity of the water body; (6) Complete the training of the water quality parameter inversion model by using the inherent optical quantity of the water body and the measured data of water quality parameters, and use the trained water quality parameter inversion model to invert the water quality parameters of the water body, and obtain the spatial distribution of water quality parameters; The normalized difference water index, namely NDWI, is calculated as follows:

Among them, R ₇₅₀ and R ₅₆₀ refer to the remote sensing reflectivity of 750nm wavelength and 560nm wavelength respectively, and the NDWI is between -1 and 1. When the NDWI is greater than the threshold value of 0, the pixel is a candidate water body; The QAA algorithm, namely QAA_V5, has a specific form of expression:

λ ₀ =550nm (4)

where λ is the wavelength, λ ₀ is the reference wavelength, 550 nm is taken in QAA_V5, R _rs (λ) is the remote sensing reflectance of the upper surface of the water body at λ wavelength, r _rs (λ) is the remote sensing reflectivity of the lower surface of the water body at λ wavelength, Y is the backscattering spectral slope of particulate matter, μ(λ) is the ratio of the backscattering coefficient to the sum of the backscattering coefficient and the absorption coefficient at λ wavelength, a(λ) is the water absorption coefficient at λ wavelength, b _bp ( λ) is the backscattering coefficient of water at λ wavelength, a _w (λ ₀ ) is the absorption coefficient of pure water at the reference wavelength λ ₀ , b _bw (λ ₀ ) is the backscattering coefficient of pure water at the reference wavelength λ ₀ Scattering coefficient, w ₀ , w ₁ , w ₂ , w ₃ , and w ₄ are parameters that need to be adjusted. In the QAA_V5 algorithm, w ₀ =0.39, w ₁ =1.14, w ₂ =2.0, w ₃ =-2.4, w ₄ =-0.9; The specific implementation mode of the optimization objective function of the QAA algorithm is:

Among them, R is the size of the area of interest set, λ is the wavelength, (row _i , col _i ) is the pixel coordinates of the row row and col column in the i area, and w is w ₀ , w ₁ , w ₂ , w ₃ A vector composed of , w ₄ , namely w=[w ₀ , w ₁ , w ₂ , w ₃ , w ₄ ], ρ is the last item in the control objective function

The weight of importance, the value range of ρ is 0<ρ<1;

is the wavelength of λ, when the parameter w, the absorption coefficient at the coordinates of (row _i , col _i ),

is the backscattering coefficient at (row _i , col _i ) coordinates at λ wavelength, w parameter,

is the weighted sum of the square of the difference between the absorption coefficient of the water body at the coordinates (row _i , col _i ) and the absorption coefficients of the water body at other coordinates in the 8 neighborhood under the parameter w wavelength λ,

is the weighted sum of the square of the difference between the backscattering coefficient of the water body at the coordinates (row _i , col _i ) and the backscattering coefficient of the water body at other coordinates in the 8 neighborhood under the parameter w wavelength λ; The improvement point of the improved QAA algorithm is the adjustment of five parameters in the original QAA algorithm, namely w ₀ , w ₁ in formula (5) and w ₂ , w ₃ , w in formula (7) ₄ ; The Adam algorithm is used to continuously adjust the parameters of the QAA algorithm until the objective function value is the smallest; its specific form is: define the total number of iterations, and use the Adam algorithm to update w in each iteration, so that the target represented by formula (10) is The function gradually gets smaller. 2. a kind of inland water quality inversion method based on improved QAA algorithm as claimed in claim 1, is characterized in that, the concrete realization that described utilizing morphological processing method to carry out post-processing to the water body that extracts is: The candidate water bodies are post-processed by morphological processing methods to filter out non-water body areas and smooth the water body boundaries. 3. a kind of inland water quality inversion method based on improved QAA algorithm as claimed in claim 1, is characterized in that, described morphological processing method is: utilize open operation to eliminate the connection of part non-water body area and water body area; Mark the connected domain of candidate water bodies, filter the connected domain of non-water bodies according to the size of the area; use the closing operation to smooth the water-land boundary. 4. a kind of inland water quality inversion method based on improved QAA algorithm as claimed in claim 1, is characterized in that, the selection of described interest area set, should combine the geographical position of water body, shape characteristic to select suitable size. area; note the size of the area of interest set as R. 5. The inland water quality inversion method based on an improved QAA algorithm as claimed in claim 1, wherein the number of iterations should be set to more than 1000, and the parameters of Adam are learning rate=0.01, beta1=0.9 , beta2=0.999, eps=1e-08, weight_decay=0. 6. a kind of inland water quality inversion method based on improved QAA algorithm as claimed in claim 1, is characterized in that, described utilizing water body inherent optical quantity and water quality parameter measured data to complete the concrete performance of water quality parameter inversion model training The way is: Calculate the correlation between the inherent optical quantity of the water body and the water quality parameters, and select the band according to the correlation; use the linear regression to obtain the model of the optimal band to invert the water quality parameters. 7. An inland water quality inversion method based on an improved QAA algorithm according to any one of claims 1 to 6, wherein the water quality parameters include chlorophyll, total suspended solids concentration and turbidity.

Images