An integrated intelligent inversion method for total phosphorus in water body based on feature map layer screening

Abstract

The present application relates to the field of water total phosphorus integrated intelligent inversion, in particular to a water total phosphorus integrated intelligent inversion method based on feature layer screening.The present application performs principal component analysis on Sentinel 2B satellite image data, filters out five feature layers through the thought of dimension reduction, extracts five groups of principal component data at the positions of the sampling points, performs correlation analysis on the single principal component and the multiple principal component combination and the measured total phosphorus concentration, and performs correlation analysis on the measured total phosphorus concentration and the remote sensing reflectivity of each band of the image.The principal component analysis method reduces and deletes the original redundant data, so that the selected principal components are not correlated with each other, can greatly represent the original variable data, and improves the efficiency of model establishment.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of water total phosphorus integrated intelligent inversion, in particular to a water total phosphorus integrated intelligent inversion method based on feature layer screening. BACKGROUND

[0002] Water total phosphorus concentration is an important factor of water eutrophication and the main reason for the massive reproduction of algae. With the continuous development of economy, a large amount of urban sewage, industrial wastewater and other polluted water bodies are discharged into rivers and lakes, causing the general increase of water total phosphorus concentration, and there is an urgent need for new technologies and means to quickly and efficiently monitor water total phosphorus concentration.

[0003] Traditional methods for obtaining water total phosphorus concentration mostly use artificial periodic field measurement or automatic sampling at water quality monitoring sites. This method is not only time-consuming, labor-intensive and costly, but also cannot cover the entire areal scale of water bodies, resulting in discontinuous water quality sampling and inability to reflect the overall situation of the water area. Remote sensing, as a macroscopic means for large-scale monitoring, can obtain remote sensing images of large-scale space range in a short time, and through various water quality parameter remote sensing estimation models, it can quickly obtain water area areal scale total phosphorus concentration data. At present, the establishment of water quality parameter remote sensing estimation model is mostly obtained by correlation analysis of measured water quality parameters and remote sensing reflectivity, that is, a semi-empirical / semi-analytical method. In the process of finding the optimal inversion model, the measured data and the remote sensing reflectivity of single band and multi-band combination are correlated one by one, the number of bands is determined by the band number of the image itself, and there is a certain correlation between the data. The band combination method is also more, and it takes a lot of time to establish the optimal inversion model.

[0004] Therefore, the technical scheme uses the idea of dimension reduction to analyze the band data of the remote sensing image, selects several principal components with higher contribution rate and less number than the band number, and models with the measured total phosphorus concentration, reduces the redundancy of the data, and realizes the double improvement of model precision and model establishment efficiency. SUMMARY

[0005] The purpose of the present application is to provide a water total phosphorus integrated intelligent inversion method based on feature layer screening to solve the problems raised in the background art.

[0006] In order to solve the above technical problems, the present application provides the following technical scheme:

[0007] A water total phosphorus integrated intelligent inversion method based on feature layer screening, characterized in that the method comprises the following steps:

[0008] S1, acquiring an image by satellite;

[0009] S2, image pre-processing is performed on the image data according to the acquired image;

[0010] S3, feature layer screening is performed according to the pre-processed image;

[0011] S4, a total phosphorus inversion model is established in combination with the screened layer;

[0012] S5, precision verification is performed according to the total phosphorus inversion model established according to the total phosphorus concentration and the principal component combination data after the feature layer screening.

[0013] Further, the method of image pre-processing on the image data according to the acquired image in S2 comprises the following steps:

[0014] Step 1001, acquire an image by a satellite and extract image metadata, the present application acquires image data by means of Sentinel-2B satellite, Sentinel-2B satellite is launched by European Space Agency on March 7, 2017 with Vega rocket, carrying high-resolution multi-spectral imaging device (MSI), ground spatial resolution is up to 10 meters, revisit period is 10 days, it is the same satellite group as Sentinel-2A, the two stars complement each other to achieve 5-day revisit period.

[0015] Sentinel-2 satellite image includes five levels of products, Level-0 raw data, Level-1A geometric rough correction product, Level-1B radiance product, Level-1C atmospheric apparent reflectance product and Level-2A product. The present technical solution adopts Level-2A (L2A) product, which has been corrected for atmospheric correction and orthographic correction.

[0016] Compared with Landsat satellite image, Sentinel-2B satellite image has higher time and spatial resolution, compared with OLCI data, it has higher spatial resolution, compared with domestic high-resolution satellite, 10m resolution image data can be obtained free of charge. In addition, the image covers 13 spectral bands, and contains three bands in the red edge range, which is sensitive to the detection of ground objects and is beneficial to the inversion of water quality parameters;

[0017] Step 1002, resample the acquired metadata by using SNAP software;

[0018] Step 1003, fuse the resampled bands by using the Build Layer Stack tool of ENVI5.6 to obtain a multi-spectral image data containing 12 bands.

[0019] The application re-samples the acquired metadata through the SNAP software, fuses the re-sampled bands according to the BuildLayer Stack tool of ENVI 5.6, and obtains a multispectral image data containing 12 bands.

[0020] Further, the method for re-sampling the acquired metadata through the SNAP software comprises the following steps:

[0021] Step 1002-1, acquiring a Graph Builder tool in the SNAP software;

[0022] Step 1002-2, establishing a re-sampling batch processing command through the Graph Builder tool;

[0023] Step 1002-3, selecting bands B2, B3, B4 and B8, taking the B2 band with a 10-meter spatial resolution as a reference band, wherein the B2, B3, B4 and B8 bands contain a large amount of spectral information;

[0024] Step 1002-4, calling the band selection command through the Batch Processing tool, and uniformly re-sampling metadata from a multi-spatial resolution data set with a 10-meter, 20-meter and 60-meter spatial resolution into a data set with a 10-meter spatial resolution.

[0025] The application establishes a re-sampling batch processing command through the Graph Builder tool in the SNAP software, selects bands B2, B3, B4 and B8, takes the B2 band with a 10-meter spatial resolution as a reference band, calls the band selection command through the Batch Processing tool, and uniformly re-samples metadata from a multi-spatial resolution data set with a 10-meter, 20-meter and 60-meter spatial resolution into a data set with a 10-meter spatial resolution, thereby providing a data reference for subsequent feature layer screening.

[0026] Further, the method for screening feature layers according to the preprocessed image in the S3 comprises the following steps:

[0027] Step 2001, performing principal component analysis on the image data by using a multivariate statistical method based on the dimension reduction idea;

[0028] Step 2002, combining the principal component analysis result of step 2001 and performing data extraction, wherein the image processed through the principal component analysis contains five principal components, namely PCA1, PCA2, PCA3, PCA4 and PCA5, and the five principal component data of the longitude and latitude of 14 measured points are extracted by using the multi-value extraction to point function of Arcgis 10.6, and are used for subsequent total phosphorus inversion model modeling and verification with the measured total phosphorus concentration.

[0029] This invention utilizes a multivariate statistical method based on dimensionality reduction to perform principal component analysis on image data. By extracting the principal components, the image after principal component analysis contains five principal components: PCA1, PCA2, PCA3, PCA4, and PCA5. Using ArcGIS 10.6's multi-value extraction to point function, the latitude and longitude data of 14 measured points are extracted, providing data reference for subsequent modeling and verification of total phosphorus inversion model based on measured total phosphorus concentration.

[0030] Furthermore, the method of performing principal component analysis on image data using multivariate statistical methods based on dimensionality reduction includes the following steps:

[0031] Step 2001-1: Construct matrix M, where matrix M = (m1, m2, ..., m... p ) T It consists of p standardized indicators, each with p samples, and the matrix M is expressed as follows:

[0032]

[0033] Where m pp This represents the p-th indicator of the p-th sample;

[0034] Step 2001-2: Randomly select a p*p matrix A = (a1, a2, a3, ..., a... p ) is the coefficient matrix of matrix M, where the p*1 matrix Q = (Q1, Q2, Q3, ..., Q...). p ), Q1, Q2, Q3, ..., Q p If we represent a linear combination of p indicators, then the expression for each indicator is Q = AM, that is...

[0035]

[0036] Where a1, a2, a3, ..., a p All are real numbers;

[0037] Step 2001-3: Construct the covariance matrix of matrix M as V(M) = v, according to the formula... The variance of the linear combination of the i-th index is obtained according to the formula. The covariance of the linear combination of the i-th and j-th indicators is obtained, where Var represents the variance, COV represents the covariance, and a i For the elements in matrix A,

[0038] Step 2001-4, if satisfy:

[0039]

[0040] The first principal component is Q1, which is a linear combination of original variables, and has the maximum variance in all unit linear combinations of original variables, and can best represent the original variables.

[0041] Step 2001-5, if Satisfies:

[0042]

[0043] Q i When the i-th principal component is i=1, 2,..., p; when the cumulative variance contribution rate of the first n principal components in the p-th principal component exceeds 85%, the first n principal components are extracted as the final screening index value.

[0044] The technical scheme utilizes the Forward PCA Rotation New Satistics and Rotate tool of ENVI5.6 to process the image after band synthesis, that is, to perform principal component analysis on 12 bands of the image, and convert the 12 bands into 5 mutually irrelevant principal components for output.

[0045] The present application sets Satisfies And When the condition of the first principal component is Q1, which is a linear combination of original variables, has the maximum variance in all unit linear combinations of original variables, and can best represent the original variables, by setting Satisfies When i>1, And The i-th principal component Q i When the cumulative variance contribution rate of the first n principal components in the p-th principal component exceeds the set value, the first n principal components are extracted as the final screening index value, and data reference is provided for subsequent total phosphorus inversion model establishment.

[0046] Further, the method for establishing a total phosphorus inversion model in combination with the screening layer in S4 comprises the following steps:

[0047] Step 3001, correlation analysis;

[0048] Step 3002, total phosphorus inversion model establishment.

[0049] Further, the method for establishing a total phosphorus inversion model comprises the following steps:

[0050] Step 3002-1, randomly selecting 10 sampling point main component data to establish a total phosphorus inversion model, and the remaining 4 are used for total phosphorus inversion model accuracy verification, and the correlation between the single main component or multiple main component combined data and the measured total phosphorus concentration is used, wherein the multiple main component combination includes 2 main component combinations, 3 main component combinations, 4 main component combinations and 5 main component combinations;

[0051] Step 3002-2, comparing the size of the correlation coefficient, obtaining the highest correlation value of the combination of the main components PCA3 and PCA5 and the measured total phosphorus concentration, and establishing a total phosphorus concentration inversion model using the combination of PCA3 and PCA5 and the measured total phosphorus concentration, the expression is

[0052]

[0053] Wherein Y represents the total phosphorus concentration, X represents the nonlinear combination of PCA3 and PCA5, PCA3 represents the third principal component data, and PCA5 represents the fifth principal component data.

[0054] The application compares the size of the correlation coefficient, obtains the highest correlation value of the combination of the main components PCA3 and PCA5 and the measured total phosphorus concentration, and establishes a total phosphorus concentration inversion model using the combination of PCA3 and PCA5 and the measured total phosphorus concentration, which provides data reference for subsequent model progress verification.

[0055] Further, the method for verifying the accuracy of the total phosphorus inversion model established by the main component combination data of the total phosphorus concentration and the characteristic map in S5 comprises the following steps:

[0056] Step 4001, obtaining the total phosphorus inversion model established by the measured total phosphorus concentration and the main component combination data screened by the characteristic map in step 3002-2;

[0057] Step 4002, by average absolute percentage error and root mean square error, the measured total phosphorus concentration and the main component combination data screened by the characteristic map are obtained, and the total phosphorus inversion model accuracy verification is carried out, the expression is

[0058]

[0059] Wherein X true,i represents the measured value of the sample, X estimation,i represents the estimated value of the sample, n represents the number of samples, MAPE represents the average absolute percentage error, and RMSE represents the root mean square error;

[0060] Step 4003, the principal component data of the remaining 4 sampling points are brought into the expression in step 3002-2, the estimated data of the total phosphorus concentration of the sampling points is calculated, the estimated data is taken as the sample estimated value, the measured total phosphorus concentration is taken as the sample measured value, and the four groups of data are respectively brought into the expression in step 4002, and the total phosphorus concentration inversion model accuracy is verified by calculation.

[0061] The purpose of the application is to screen out 5 irrelevant feature layers by using the principal component analysis method, and obtain 5 groups of principal component data of the longitude and latitude of the sampling points, and the correlation between the principal component data and the measured total phosphorus concentration is analyzed, and the best total phosphorus concentration inversion model based on the feature layer screening is established, and the operation software involved in the scheme is: SentinelApplication Platform (hereinafter referred to as SNAP), ENVI5.6, Arcgis10.6, Excel.

[0062] The application screens out 5 feature layers by dimension reduction, extracts 5 groups of principal component data of the sampling points, and analyzes the correlation between the single principal component and the multiple principal component combination and the measured total phosphorus concentration.

[0063] The water body total phosphorus concentration inversion model based on the feature layer screening has high correlation coefficient and good model accuracy, and can more accurately reflect the change of the total phosphorus concentration in the research area. BRIEF DESCRIPTION OF DRAWINGS

[0064] Figure 1 It is a water body total phosphorus integrated intelligent inversion method flow chart based on feature layer screening;

[0065] Figure 2 It is a water body total phosphorus integrated intelligent inversion method based on feature layer screening, and the correlation between the measured total phosphorus concentration and (PCA3-PCA5) / (PCA3 / PCA5) is analyzed;

[0066] Figure 3 It is a water body total phosphorus integrated intelligent inversion method based on feature layer screening, and the sampling point distribution map is shown in the figure;

[0067] Figure 4 It is a water body total phosphorus integrated intelligent inversion method based on feature layer screening, and the resolution of each band of Sentinel-2MSI remote sensing image is shown in the figure;

[0068] Figure 5 is measured total phosphorus concentration of the water body total phosphorus integrated intelligent inversion method based on feature layer screening of the application;

[0069] Figure 6 is the cumulative variance contribution rate of the water body total phosphorus integrated intelligent inversion method based on feature layer screening of the application;

[0070] Figure 7 is the principal component data of the sampling point of the water body total phosphorus integrated intelligent inversion method based on feature layer screening of the application. DETAILED DESCRIPTION

[0071] The technical solutions in the embodiments of the application will be clearly and completely described below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, not all the embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the application.

[0072] Please refer to Figures 1-7 , in the embodiments of the application: a water body total phosphorus integrated intelligent inversion method based on feature layer screening, the intelligent optimization method comprises the following steps:

[0073] S1, acquiring an image image by satellite;

[0074] In the first embodiment, the Sentinel-2B satellite image has higher temporal and spatial resolution compared with the Landsat satellite image, and has higher spatial resolution compared with the OLCI data. Compared with the domestic high-resolution satellite, 10m resolution image data can be obtained for free. In addition, the image covers 13 spectral bands, and contains three bands in the red edge range, which is sensitive to the detection of ground objects and is conducive to the inversion of water quality parameters. Among them, the introduction of 13 spectral bands is as shown in Figure 4 .

[0075] This technical solution selects the measured total phosphorus concentration of 14 sampling points, and the points are located in the Wumahu water area of Changzhou City, which includes Changzhou Zhonglou District, Tianning District and part of Xinbei District and Wujin District. The distribution of the sampling points is as shown in Figure 3 , and the measured total phosphorus concentration is as shown in Figure 5 .

[0076] S2, according to the acquired image image, image preprocessing is performed on the image data;

[0077] S3, feature layer screening is performed according to the preprocessed image;

[0078] S4, total phosphorus inversion model is established in combination with the screening layer;

[0079] S5, the total phosphorus inversion model established according to the total phosphorus concentration and the principal component combination data screened from the feature layer is used for precision verification.

[0080] The method for image preprocessing of the image data according to the acquired image in S2 comprises the following steps:

[0081] Step 1001, acquiring an image by a satellite and extracting image metadata;

[0082] Step 1002, resampling the acquired metadata by using SNAP software;

[0083] Step 1003, fusing the resampled bands by using the Build Layer Stack tool of ENVI5.6 to obtain multispectral image data containing 12 bands.

[0084] The method for resampling the acquired metadata by using SNAP software comprises the following steps:

[0085] Step 1002-1, acquiring the Graph Builder tool in SNAP software;

[0086] Step 1002-2, establishing a resampling batch processing command by using the Graph Builder tool;

[0087] Step 1002-3, selecting bands B2, B3, B4 and B8, and taking the B2 band with a spatial resolution of 10 meters as a reference band, wherein the bands B2, B3, B4 and B8 are spectral bands;

[0088] Step 1002-4, calling the band selection command by using the Batch Processing tool, and uniformly resampling the metadata from the multispectral data sets with spatial resolutions of 10 meters, 20 meters and 60 meters into a data set with a spatial resolution of 10 meters.

[0089] The method for feature layer screening according to the preprocessed image in S3 comprises the following steps:

[0090] Step 2001, performing principal component analysis on the image data by using a multivariate statistical method based on the dimension reduction idea;

[0091] Step 2002: Combining the principal component analysis results from Step 2001, data extraction is performed. The image after principal component analysis contains five principal components: PCA1, PCA2, PCA3, PCA4, and PCA5. Using ArcGIS 10.6's multi-value extraction to point function, the latitude and longitude data of 14 measured points are extracted, representing the five principal components. This data will be used for subsequent modeling and validation of the total phosphorus inversion model with the measured total phosphorus concentration. The extracted principal component data are as follows: Figure 7 As shown.

[0092] The method for principal component analysis of image data using multivariate statistical methods based on dimensionality reduction includes the following steps:

[0093] Step 2001-1: Construct matrix M, where matrix M = (m1, m2, ..., m... p ) T It consists of p standardized indicators, each with p samples, and the matrix M is expressed as follows:

[0094]

[0095] Where m pp This represents the p-th indicator of the p-th sample;

[0096] Step 2001-2: Randomly select a p*p matrix A = (a1, a2, a3, ..., a... p ) is the coefficient matrix of matrix M, where the p*1 matrix Q = (Q1, Q2, Q3, ..., Q...). p ), Q1, Q2, Q3, ..., Q p If we represent a linear combination of p indicators, then the expression for each indicator is Q = AM, that is...

[0097]

[0098] Where a1, a2, a3, ..., a p All are real numbers;

[0099] Step 2001-3: Construct the covariance matrix of matrix M as V(M) = v, according to the formula... The variance of the linear combination of the i-th index is obtained according to the formula. The covariance of the linear combination of the i-th and j-th indicators is obtained, where Var represents the variance, COV represents the covariance, and a i For the elements in matrix A,

[0100] Step 2001-4, if satisfy:

[0101]

[0102] Q1 is the first principal component, which is a linear combination of original variables, and is the maximum variance value in all unit linear combinations of original variables, and can best represent the original variables;

[0103] Step 2001-5, if satisfies:

[0104]

[0105] Q1 is the first principal component, which is a linear combination of original variables, and is the maximum variance value in all unit linear combinations of original variables, and can best represent the original variables; i When the i-th principal component is the i-th principal component, i = 1, 2,..., p; when the cumulative variance contribution rate of the first n principal components in the p-th principal component exceeds 85%, the first n principal components are extracted as the final screening index value.

[0106] The technical scheme utilizes the Forward PCA Rotation New Satistics and Rotate tool of ENVI5.6 to process the image after band synthesis, that is, to perform principal component analysis on 12 bands of the image, convert into 5 mutually unrelated principal components for output, and the cumulative variance contribution rate reaches 98.24%. The cumulative contribution rate of each principal component is as shown in Figure 6 .

[0107] The method for establishing the total phosphorus inversion model in combination with the screening layer in S4 comprises the following steps:

[0108] Step 3001, correlation analysis;

[0109] Step 3002, total phosphorus inversion model model establishment.

[0110] The method for establishing the total phosphorus inversion model model comprises the following steps:

[0111] Step 3002-1, randomly selecting principal component data of 10 sampling points for total phosphorus inversion model establishment, and the remaining 4 for total phosphorus inversion model accuracy verification, and utilizing single principal component or multiple principal component combination data and measured total phosphorus concentration for correlation analysis, wherein the multiple principal component combination includes 2 principal component combinations, 3 principal component combinations, 4 principal component combinations and 5 principal component combinations;

[0112] Step 3002-2, comparing the size of the correlation coefficient, obtaining the highest correlation value of the combination of principal components PCA3 and PCA5 and the measured total phosphorus concentration, and establishing a total phosphorus concentration inversion model using the combination of PCA3 and PCA5 and the measured total phosphorus concentration, and the chart of correlation analysis is as shown in Figure 2 , and the expression is

[0113]

[0114] Wherein, Y represents the total phosphorus concentration, X represents the nonlinear combination of PCA3, PCA5, PCA3 represents the 3rd principal component data, and PCA5 represents the 5th principal component data.

[0115] In this embodiment two, by comparing the size of the correlation coefficient, the combination of principal components PCA3, PCA5 and the highest correlation of the measured total phosphorus concentration is obtained, which is 0.9695, so the combination of PCA3, PCA5 and the measured total phosphorus concentration is used to establish the total phosphorus concentration inversion model.

[0116] The method for verifying the accuracy of the total phosphorus inversion model established according to the total phosphorus concentration and the principal component combination data screened by the characteristic map in S5 comprises the following steps:

[0117] Step 4001, obtaining the total phosphorus inversion model established by the measured total phosphorus concentration and the principal component combination data screened by the characteristic map in step 3002-2;

[0118] Step 4002, verifying the accuracy of the total phosphorus inversion model by the average absolute percentage error and the root mean square error of the obtained measured total phosphorus concentration and the principal component combination data screened by the characteristic map, and the expression is

[0119]

[0120] Wherein X true,i represents the measured value of the sample, X estimation,i represents the estimated value of the sample, n represents the number of samples, MAPE represents the average absolute percentage error, and RMSE represents the root mean square error.

[0121] Step 4003, the principal component data of the remaining 4 sampling points is brought into the expression in step 3002-2, the estimated data of the total phosphorus concentration of the sampling points is obtained by calculation, the estimated data is taken as the sample estimated value, the measured total phosphorus concentration is taken as the sample measured value, and the 4 groups of data are respectively brought into the expression in step 4002, and the accuracy of the total phosphorus concentration inversion model is verified by calculation.

[0122] In this embodiment three, the principal component data of the remaining 4 sampling points is brought into the expression in step 3002-2, the estimated data of the total phosphorus concentration of the sampling points is obtained by calculation, the estimated data is taken as the sample estimated value, the measured total phosphorus concentration is taken as the sample measured value, and the 4 groups of data are respectively brought into the expression in step 4002, and the MAPE is 5.37%, and the RMSE is 0.1146. By comparing the value range of MAPE and RMSE, it is judged that the accuracy of the total phosphorus concentration inversion model is good.

[0123] It will be obvious to those of ordinary skill in the art that the present application is not limited to the details of the above-exemplified embodiments, and that the present application can be implemented in other particular forms without departing from the spirit or essential characteristics of the present application. The embodiments are therefore to be considered in all respects as illustrative and not restrictive, the scope of the present application being indicated by the appended claims rather than by the above description, and all changes which come within the meaning and range of equivalency of the claims are therefore intended to be embraced therein. No reference signs in the claims shall be construed as limiting the scope of the claims to the features or aspects to which the reference signs are attached.

[0124] It has to be understood that any reference to both a technical feature of one aspect of the application and a reference to a technical feature of another aspect of the application does not mean that these technical features belong to the same aspect of the application. It is noted that, as used in the specification and in the claims, the singular forms "a", "an" and "the" include their plural recitations unless the context clearly dictates otherwise. Thus, for example, reference to "a component" includes a plurality of such components, and so forth.

[0125] In conclusion, it has to be noted that the technical features of the application are not limited to the above described preferred embodiments, and that the technical solutions described above in relation to the preferred embodiments can be modified without departing from the scope of the present application. Any modification, equivalent replacement or improvement made in accordance with the principles and technical features of the present application shall fall within the scope of the present application.

Claims

1. An integrated intelligent inversion method for total phosphorus in water bodies based on feature map layer screening, characterized in that, The method comprises the following steps: S1, acquiring an image image through a satellite; S2, performing image preprocessing on the image data according to the acquired image image; S3, performing feature layer screening according to the preprocessed image; The method for performing feature layer screening according to the preprocessed image in S3 comprises the following steps: Step 2001, performing principal component analysis on the image data by using a multivariate statistical method based on the dimension reduction idea; The method for performing principal component analysis on the image data by using the multivariate statistical method based on the dimension reduction idea comprises the following steps: Step 2001-1, constructing a matrix M, wherein the matrix M = (m1, m2,..., m p ) T The matrix M is expressed by p standardized indexes, each of which has p samples. wherein m pp represents the pth indicator of the pth sample; Step 2001-2: Randomly select a p*p matrix A = (a1, a2, a3, ..., a... p ) is the coefficient matrix of matrix M, where the p*1 matrix Q = (Q1, Q2, Q3, ..., Q...). p ), Q1, Q2, Q3, ..., Q p If we represent a linear combination of p indicators, then the expression for each indicator is Q = AM, that is... where a1, a2, a3,..., a p are real numbers; Step 2001-3, construct the covariance matrix of matrix M as V(M) = v, according to the formula get the variance of the i-th index linear combination, according to the formula get the covariance of the i-th index linear combination and the j-th index linear combination, where Var represents variance, COV represents covariance, a i is an element in matrix A, Step 2001-4, if satisfies: Then the first principal component Q1 is a linear combination of the original variables, which is the maximum variance value in all unit linear combinations of the original variables and can best represent the original variables; Step 2001-5, if satisfies: Q i when the cumulative variance contribution rate of the first p principal components exceeds 85%, the first n principal components are extracted as the final screening index value; Step 2002, combining the principal component analysis result of step 2001 and performing data extraction, the image after the principal component analysis processing contains five principal components, namely PCA1, PCA2, PCA3, PCA4 and PCA5, the five principal component data of the longitude and latitude of the 14 measured points are extracted by using the multi-value extraction to point function of Arcgis10.6, which are used for subsequent total phosphorus inversion model modeling and verification with the measured total phosphorus concentration; S4, establishing a total phosphorus inversion model in combination with the screened layer; S5, verifying the accuracy according to the total phosphorus inversion model established according to the total phosphorus concentration and the principal component combination data after the feature layer screening.

2. The water body total phosphorus integrated intelligent inversion method based on feature map layer screening according to claim 1, characterized in that, The method for performing image preprocessing on the image data according to the acquired image image in S2 comprises the following steps: Step 1001, acquiring an image image through a satellite and extracting image metadata; Step 1002, performing resampling processing on the acquired metadata by using SNAP software; Step 1003, fusing the resampled bands by using the Build Layer Stack tool of ENVI5.6 to obtain a multispectral image data containing 12 bands.

3. The water body total phosphorus integrated intelligent inversion method based on feature map layer screening according to claim 2, characterized in that, The method for performing resampling processing on the acquired metadata by using SNAP software comprises the following steps: Step 1002-1, acquiring a Graph Builder tool in SNAP software; Step 1002-2, establishing a resampling batch processing command through the Graph Builder tool; Step 1002-3, selecting bands B2, B3, B4 and B8, taking the B2 band with a spatial resolution of 10 meters as a reference band, wherein B2, B3 and B4 are visible light bands, and B8 is a near-infrared band; Step 1002-4, calling the band selection command through the Batch Processing tool, and uniformly resampling the metadata from the multispatial resolution data sets of 10 meters, 20 meters and 60 meters into a data set with a spatial resolution of 10 meters.

4. The water body total phosphorus integrated intelligent inversion method based on feature map layer screening according to claim 3, characterized in that, The method for establishing a total phosphorus inversion model in combination with the screened layer in S4 comprises the following steps: Step 3001, correlation analysis; Step 3002, total phosphorus inversion model modeling.

5. The total phosphorus integrated intelligent inversion method based on feature map layer screening according to claim 4, characterized in that, The method for total phosphorus inversion model modeling comprises the following steps: Step 3002-1, randomly select 10 sampling point principal component data to establish total phosphorus inversion model, the remaining 4 are used for total phosphorus inversion model accuracy verification, using a single principal component or multiple principal component combination data and measured total phosphorus concentration correlation, wherein the multiple principal component combination includes 2 principal component combination, 3 principal component combination, 4 principal component combination and 5 principal component combination; Step 3002-2, compare the size of the correlation coefficient, the combination of principal component PCA3, PCA5 and the measured total phosphorus concentration has the highest correlation value, and the combination of PCA3, PCA5 and the measured total phosphorus concentration is used to establish the total phosphorus concentration inversion model, the expression is Wherein, Y represents the total phosphorus concentration, X represents the nonlinear combination of PCA3, PCA5, PCA3 represents the third principal component data, and PCA5 represents the fifth principal component data.

6. The total phosphorus integrated intelligent inversion method based on feature map layer screening according to claim 5, characterized in that, The method for verifying the accuracy of the total phosphorus inversion model established according to the total phosphorus concentration and the principal component combination data screened from the characteristic map in S5 comprises the following steps: Step 4001, obtaining the total phosphorus inversion model established by the measured total phosphorus concentration and the principal component combination data screened from the characteristic map in step 3002-2; Step 4002, by average absolute percentage error and root mean square error, the measured total phosphorus concentration and the principal component combination data screened from the characteristic map are obtained, and the total phosphorus inversion model accuracy verification, the expression is where X true,i represents the measured value of the sample, X estimation,i represents the estimated value of the sample, n represents the number of samples, MAPE represents the mean absolute percentage error, and RMSE represents the root mean square error; Step 4003, the remaining 4 sampling point principal component data is brought into the expression in step 3002-2, and the estimated data of the sampling point total phosphorus concentration is obtained by calculation, the estimated data is used as the sample estimated value, the measured total phosphorus concentration is used as the sample measured value, and the 4 groups of data are respectively brought into the expression in step 4002, and the total phosphorus concentration inversion model accuracy is verified by calculation.

Images