Hyperspectral remote sensing and water quality sequence-based permanganate index multi-point inversion method

Abstract

The invention discloses a permanganate index multipoint inversion method based on hyperspectral remote sensing and a water quality sequence. The method comprises the following steps: acquiring a hyperspectral remote sensing image and water quality time sequence data of a point location to be inverted; a remote sensing regression inversion model is constructed, and a spectral information training set is used for training; constructing a Gaussian hidden Markov model, and training by using water quality time sequence grouped data; inputting the spectral information in the test set into the remote sensing regression inversion model to obtain an inversion result of the point location to be inverted; and inputting the water quality time sequence grouped data in the test set into the Gaussian hidden Markov model, and inferring the current hidden state of the Gaussian hidden Markov model by using the inversion result of the point location to be inverted to obtain the final permanganate index prediction result of each point location. According to the method, the Gaussian hidden Markov model and the hyperspectral remote sensing regression model are combined, multi-dimensional data are utilized, the influence of sparse sampling points on precision is reduced, interference of the same-spectrum foreign matter phenomenon is inhibited, and the precision of permanganate index multi-point inversion is improved.

Description

Technical Field

[0001] The invention belongs to the field of water resource monitoring, and in particular relates to a multi-point inversion method of permanganate index based on hyperspectral remote sensing and water quality sequences. Background Art

[0002] Inland water bodies such as rivers and lakes play an important role in maintaining ecological balance. However, with the rapid development of social economy in recent years, the pollution and damage of water bodies have been increasing. Therefore, accurate and real-time monitoring of organic pollution at important points in the basin is the long-term basis for ecological governance and restoration. Mn ) refers to the amount of oxidant consumed when reducing substances in a unit of water are oxidized under specified conditions. It is an important indicator reflecting organic pollution in a river basin, and its monitoring is of great significance.

[0003] Traditional COD Mn The ground measurement of COD relies on laboratory sampling or automatic station measurement, which consumes a lot of manpower, material and financial resources and has a long measurement cycle. With the continuous development of remote sensing technology, the use of satellite images for water quality inversion has gradually become the mainstream, but COD Mn The remote sensing inversion model is difficult to capture the COD Mn The correlation with the spectral bands and the high density of water quality sampling points result in poor inversion accuracy and weak model mobility.

[0004] In addition to the above two mainstream methods, methods based on water quality time series for prediction are also gradually developing in water quality monitoring technology. Their prediction accuracy is high but their ability to cope with sudden environmental changes is weak.

[0005] Currently, the existing COD Mn Remote sensing inversion methods have high requirements for spectral correlation and distribution of water quality sampling points, which makes it difficult for conventional remote sensing inversion models to adapt to scenarios with sparse sampling points. Mn The prediction is also difficult to capture sudden changes in water quality and cannot be used in practice. Summary of the invention

[0006] The present invention provides a multi-point inversion method of permanganate index based on hyperspectral remote sensing and water quality sequence, combined with the COD Mn Water quality data and remote sensing spectral data are jointly predicted to reduce the impact of sparse sampling points on accuracy while capturing COD Mn The trend of change, thereby increasing COD Mn The inversion accuracy of the permanganate index is as follows:

[0007] A hyperspectral remote sensing image of the point to be inverted is obtained and preprocessed to obtain a preprocessed hyperspectral remote sensing image, and permanganate index data collected by an automatic water quality monitoring station within the range of the hyperspectral remote sensing image is obtained to obtain permanganate index water quality time series data.

[0008] The spectral information of the preprocessed hyperspectral remote sensing image is extracted and the bands are aligned. The permanganate index water quality time series data is extracted and time grouped to obtain the permanganate index water quality time series grouped data. The spectral information after band alignment and the permanganate index water quality time series grouped data are randomly sampled as a training set and a test set at a ratio of 2:1, respectively.

[0009] A remote sensing regression inversion model is constructed, wherein the remote sensing regression inversion model includes sensitive band screening, inversion band combination screening and final regression inversion model determination, and the remote sensing regression inversion model is trained using a spectral information training set.

[0010] A Gaussian hidden Markov model is constructed, wherein the Gaussian hidden Markov model includes three hidden states and an optimal step size selection, and the permanganate index water quality time series grouping data training set is used to input the Gaussian hidden Markov model for training.

[0011] The spectral information in the test set is input into the pre-trained remote sensing regression inversion model to obtain the inversion results of the points to be inverted; the permanganate index water quality time series grouped data in the test set is input into the pre-trained Gaussian hidden Markov model to obtain the prediction results under three different hidden states.

[0012] The inversion result of the point to be inverted is used to infer the current hidden state of the Gaussian hidden Markov model to obtain the final permanganate index prediction result of each point, specifically: the final regression inversion model is verified for accuracy to obtain the error mean as a threshold, and the previous day's data in the permanganate index water quality time series data is used as a benchmark value; the prediction results of the Gaussian hidden Markov model are sorted from small to large to obtain the first hidden state, the second hidden state, and the third hidden state; it is determined whether the regression inversion result fluctuates compared with the previous day's data. If there is no fluctuation or the upper and lower fluctuation range does not exceed the threshold, the second hidden state is selected as the final permanganate index prediction result; if the downward fluctuation range exceeds the threshold, the first hidden state is selected as the final permanganate index prediction result; if the upward fluctuation range exceeds the threshold, the third hidden state is selected as the final permanganate index prediction result.

[0013] Furthermore, the preprocessing is to perform radiation calibration, atmospheric correction and orthorectification on the hyperspectral remote sensing image. The calculation formula of the radiation calibration is: L = Gain·DN + Bias, where L is the pixel radiation brightness, the unit is W / (cm 2 ·μm·sr), DN is the brightness value of the remote sensing image pixel, Gain is the radiation calibration gain, and Bias is the radiation calibration offset, all in W / (cm 2 ·μm·sr); the calculation formula for the atmospheric correction is: Among them, C is the pixel radiation brightness after atmospheric correction, ρ is the pixel reflectivity, and ρ e is the average reflectivity of the area around the pixel, A and B are calculated based on the actual atmosphere and geometric conditions, S is the albedo of the large balloon surface, and C a It is the atmospheric radiation.

[0014] Furthermore, the permanganate index data collected by the automatic water quality monitoring station within the range of the hyperspectral remote sensing image is obtained, specifically, the permanganate index data collected 14 days before the shooting date of the hyperspectral remote sensing image is obtained, and the permanganate index water quality time series data is grouped in units of N days to obtain the permanganate index water quality time series grouped data, where 7≤N≤14.

[0015] Furthermore, the processing method for band alignment is: for hyperspectral remote sensing images taken by different hyperspectral satellites, select one of the images taken by the hyperspectral satellite as the reference image, and the rest as images to be registered; when the difference between the central bands of each band of the image to be registered and each band of the reference image is within a specified range, retain the band; when the difference exceeds the specified range, take the average of the band of the image to be registered and the front band or the back band as the new central band, so that the difference of the new central band is within the specified range, and take the average of the radiation brightness values ​​of the front and back bands as the radiation brightness value of the new central band.

[0016] Furthermore, the sensitive band screening and inversion band combination screening in the remote sensing regression inversion model use the Pearson correlation coefficient method to select the band or band combination with the highest correlation as the calculation formula from spectral information to remote sensing regression inversion independent variables. The formula of the Pearson correlation coefficient is: Among them, r is the Pearson correlation coefficient, X i is the remote sensing regression inversion independent variable, Y i For X i Corresponding to the measured permanganate index water quality data on the day the hyperspectral remote sensing image was taken, n is the number of samples, For X i and Y i The sample mean, S X , SY is the sample standard deviation; the final regression inversion model in the remote sensing regression inversion model is that after the linear regression model, logarithmic regression model, polynomial regression model and power regression model are established, R 2 To verify the accuracy, the regression inversion model with the highest coefficient of determination was selected, among which, Among them, R 2 is the coefficient of determination, SSR is the regression sum of squares, SST is the total sum of squares, Yes and Y i The corresponding inverted water quality.

[0017] Furthermore, the Gaussian Hidden Markov Model includes:

[0018] Construct a Gaussian hidden Markov model to describe the hidden state of water quality and initialize the model parameters, including the state transfer matrix, initial probability matrix, mean matrix and covariance matrix;

[0019] Using the permanganate index water quality time series grouping data, constructing the Gaussian hidden Markov observation sequence;

[0020] The number of hidden states of water quality is set to 3, and the parameters of the Gaussian hidden Markov model are estimated using the Baum-Welch algorithm to obtain a trained state transfer matrix, an initial probability matrix, a mean matrix, and a covariance matrix;

[0021] The weighted average of the expected observation values ​​of each state in the next stage is calculated through the state transfer matrix and the mean matrix as the predicted value of the hidden state.

[0022] Preferably, a test sample is used to verify the accuracy of the permanganate remote sensing inversion result.

[0023] Preferably, the error mean is taken from the mean absolute error of the final regression inversion model during the training process, and its formula is: Among them, MAE is the mean absolute error, Y i To measure the water quality data of permanganate index, Yes and Y i The corresponding inverted water quality.

[0024] Preferably, it is characterized in that the hyperspectral remote sensing images are taken from the Ziyuan-1 02D hyperspectral camera, the Ziyuan-1 02E hyperspectral camera and the GF-5 02 satellite hyperspectral camera.

[0025] The multi-point inversion method of permanganate index based on hyperspectral remote sensing and water quality series provided by the present invention combines hyperspectral remote sensing images and measured water quality time series data, utilizes multidimensional data, reduces the influence of sparse sampling points on accuracy, and suppresses the interference of the same-spectrum heterogeneous object phenomenon, thereby improving the accuracy of multi-point inversion of permanganate index. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 This is a flow chart of the multi-point inversion method of permanganate index based on hyperspectral remote sensing and water quality series.

[0027] Figure 2 This is the schematic diagram of the model of the multi-point inversion method of permanganate index.

[0028] Figure 3 4 is a comparison diagram of the inversion errors of the models in the embodiments. DETAILED DESCRIPTION

[0029] The embodiment of the present invention provides a multi-point inversion method of permanganate index based on hyperspectral remote sensing and water quality sequence. Figure 1 and attached Figure 2 As shown, the multi-point inversion method of permanganate index includes:

[0030] The first step is to obtain a hyperspectral remote sensing image of the point to be inverted and preprocess it to obtain a preprocessed hyperspectral remote sensing image, obtain permanganate index data collected by an automatic water quality monitoring station within the range of the hyperspectral remote sensing image, and obtain permanganate index water quality time series data.

[0031] In this step, the hyperspectral remote sensing images are taken from the Ziyuan-1 02D hyperspectral camera, the Ziyuan-1 02E hyperspectral camera and the GF-5 02 satellite hyperspectral camera, and the permanganate index water quality time series data are taken from the permanganate index data collected by the automatic water quality monitoring station set up within the remote sensing image range 14 days before the date of the hyperspectral remote sensing image shooting.

[0032] The preprocessing is to perform radiation calibration, atmospheric correction and orthorectification on the hyperspectral remote sensing image. The calculation formula of the radiation calibration is: L = Gain·DN + Bias, where L is the pixel radiation brightness, the unit is W / (cm 2 ·μm·sr); DN is the brightness value of the remote sensing image pixel, Gain is the radiation calibration gain, and Bias is the radiation calibration offset, all in units of W / (cm 2 ·μm·sr); the calculation formula for the atmospheric correction is: Among them, C is the pixel radiation brightness after atmospheric correction, ρ is the pixel reflectivity, and ρ eis the average reflectivity of the area around the pixel, A and B are calculated based on the actual atmosphere and geometric conditions; S is the albedo of the large balloon surface, and Ca is the atmospheric radiation.

[0033] In the second step, the spectral information of the preprocessed hyperspectral remote sensing image is extracted and the bands are aligned, the permanganate index water quality time series data is extracted and time grouped to obtain the permanganate index water quality time series grouped data, and the spectral information after band alignment and the permanganate index water quality time series grouped data are randomly sampled as a training set and a test set at a ratio of 2:1, respectively.

[0034] In this step, the processing method of the band alignment is: process the hyperspectral remote sensing images taken by different hyperspectral satellites respectively, select one of the images taken by the hyperspectral satellite as the reference image, and the rest as the images to be registered; when the central band difference between each band of the image to be registered and each band of the reference image is within the specified range, the band is retained; when the difference exceeds the specified range, the mean of the band of the image to be registered and the front or back band is taken as the new central band, so that the difference of the new central band is within the specified range, and the mean of the radiation brightness values ​​of the front and back bands is taken as the radiation brightness value of the new central band. The processing method of the time grouping is: according to the shooting date of the hyperspectral remote sensing image, the permanganate index data of the specific time period before the shooting time is extracted from the water quality time series data set to form grouped data, and the permanganate index water quality time series data is grouped in units of N days, where 7≤N≤14. In this embodiment, the hyperspectral image taken by the Resource No. 1 02E satellite is selected as the reference image, and the results of the band alignment part are compared as shown in Table 1 and Table 2.

[0035] Table 1 Central wavelengths of each band before band alignment

[0036]

[0037]

[0038] Table 2 Central wavelength of each band after band alignment

[0039] Band Name Resource No. 102E Resource No. 102E Resource No. 102E b1 395.44 395.86 395.60 b2 404.03 404.45 404.15 b3 412.61 413.04 412.70 b4 421.20 421.63 421.25 b5 429.79 430.23 429.80 b6 438.37 438.58 438.36 b7 446.96 447.10 446.91 b8 455.54 455.79 455.46 b9 464.13 464.37 464.01 b10 472.71 473.21 472.56 … … … …

[0040] After obtaining the above-mentioned band alignment results, the spectral information after the band alignment and the permanganate index water quality time series data can be randomly sampled as a training set and a test set at a ratio of 2:1, and the test set can be used to verify the accuracy of the permanganate remote sensing inversion results.

[0041] The third step is to construct a remote sensing regression inversion model, which includes sensitive band screening, inversion band combination screening and final regression inversion model determination, and use the spectral information training set to input the remote sensing regression inversion model for training.

[0042] The sensitive band screening and inversion band combination screening in the remote sensing regression inversion model adopt the Pearson correlation coefficient method to select the band or band combination with the highest correlation as the calculation formula from spectral band information to remote sensing regression inversion independent variables. The formula of the Pearson correlation coefficient is: Among them, r is the Pearson correlation coefficient, X i is the remote sensing regression inversion independent variable, Y i For X i Corresponding to the measured permanganate index water quality data on the day the hyperspectral remote sensing image was taken, n is the number of samples, For X i and Y i The sample mean, S X , S Y is the sample standard deviation; the final regression inversion model in the remote sensing regression inversion model is that after the linear regression model, logarithmic regression model, polynomial regression model and power regression model are established, R 2 To verify the accuracy, the regression inversion model with the highest coefficient of determination was selected, among which, Among them, R 2 is the coefficient of determination, SSR is the regression sum of squares, SST is the total sum of squares, Is the measured water quality Y i Corresponding inverted water quality. In this embodiment, the band and band combination with the highest correlation are b22, and the final regression inversion model is a polynomial model.

[0043] The fourth step is to construct a Gaussian hidden Markov model, which includes three hidden states and optimal step size selection, and use the permanganate index water quality time series data training set to train the Gaussian hidden Markov model.

[0044] In this step, the Gaussian Hidden Markov Model includes:

[0045] Construct a Gaussian hidden Markov model to describe the hidden state of water quality and initialize the model parameters, including the state transfer matrix, initial probability matrix, mean matrix and covariance matrix;

[0046] Using the permanganate index water quality time series grouping data, constructing the Gaussian hidden Markov observation sequence;

[0047] The number of hidden states of water quality is set to 3, and the parameters of the Gaussian hidden Markov model are estimated using the Baum-Welch algorithm to obtain a trained state transfer matrix, an initial probability matrix, a mean matrix, and a covariance matrix;

[0048] The weighted average of the expected observation values ​​of each state in the next stage is calculated through the state transfer matrix and the mean matrix as the predicted value of the hidden state.

[0049] The Gaussian hidden Markov model contains three hidden states, and the parameter estimation uses the Baum-Welch algorithm, and the state sequence inference algorithm is not used. After the permanganate index grouping data is input into the Gaussian hidden Markov model, three prediction results corresponding to three hidden states are generated. In this embodiment, the minimum step size of the Gaussian hidden Markov model training is 7, the maximum step size is 14, and the accuracy of different step sizes obtained using the training set is shown in Table 3. Finally, a step size of 10 is selected as the step size of the subsequent Gaussian hidden Markov model.

[0050] Table 3. Changes in model accuracy under different step lengths

[0051]

[0052]

[0053] In the fifth step, the spectral information in the test set is input into the pre-trained remote sensing regression inversion model to obtain the inversion results of the points to be inverted; the permanganate index water quality time series grouped data in the test set is input into the pre-trained Gaussian hidden Markov model to obtain the prediction results under three different hidden states.

[0054] In this embodiment, the pre-trained remote sensing regression inversion model refers to a polynomial regression model fitted using a spectral band information training set, and the pre-trained Gaussian hidden Markov model refers to a model step size of 10 determined using a permanganate index training set grouping data.

[0055] The sixth step is to use the inversion results of the points to be inverted to infer the current hidden state of the Gaussian Hidden Markov Model to obtain the final permanganate index prediction results of each point.

[0056] The above-mentioned method of using the inversion result to obtain the permanganate index prediction result includes the following steps: first, verifying the accuracy of the final regression inversion model to obtain the error mean as a threshold, and using the previous day's data in the permanganate index water quality time series data as a benchmark value; second, sorting the three prediction results generated by the Gaussian hidden Markov model from small to large, corresponding to the first, second, and third hidden states respectively; third, judging whether the regression inversion result fluctuates compared with the previous day's data. If it does not fluctuate or the upper and lower fluctuation ranges do not exceed the threshold, the second hidden state is selected as the final permanganate index prediction result; if the downward fluctuation range exceeds the threshold, the first hidden state is selected as the final permanganate index prediction result; if the upward fluctuation range exceeds the threshold, the third hidden state is selected as the final permanganate index prediction result.

[0057] In this embodiment, the error mean is taken from the mean absolute error of the final regression inversion model during the training process, and its formula is: Among them, MAE is the mean absolute error.

[0058] After obtaining the final permanganate index prediction results of each point, the test samples can also be used to verify the accuracy of the permanganate remote sensing inversion results. The results of the accuracy verification are shown in Table 4. The comparison of the inversion errors of each model is shown in the attached figure. Figure 3 As shown, Figure 3 (a) is the inversion error of the polynomial inversion model, Figure 3 (b) is the Gaussian hidden Markov model inversion error, Figure 3 (c) is the inversion error of the multi-point inversion method of permanganate index based on hyperspectral remote sensing and water quality series.

[0059] Table 4 Accuracy verification results

[0060]

[0061] The above-mentioned multi-point inversion method of permanganate index by combining Gaussian hidden Markov and hyperspectral remote sensing regression is constructed as follows: 1) The mean absolute error of the regression inversion model is used as the threshold, and the predicted value of the regression inversion model is input into the Gaussian hidden Markov model with a hidden state of 3 using the permanganate index grouping training set data with a length of 7 to 14 relative to the water quality of the previous day, and the optimal step length of the Gaussian hidden Markov model for predicting the permanganate index in this study area is obtained. 2) After fitting the regression inversion model using the training set, the mean absolute error of the regression inversion model is recorded, and then the test set is input to obtain the predicted value of the regression inversion model. 3) The water quality data of the previous day of the target prediction date in the permanganate index grouping data is used as the benchmark value, and the predicted value of the Gaussian hidden Markov model is intervened by the fluctuation amplitude of the data to obtain the final prediction result of each point.

[0062] The data length of the permanganate index group training set used above is 7 to 14, so that the Gaussian hidden Markov model can obtain sufficient historical permanganate index change information while retaining the model's sensitivity to short-term changes. The number of hidden states of the Gaussian hidden Markov model is set to 3, corresponding to the growth, stability and decrease trends of the permanganate index respectively. The mean absolute error of the regression inversion model is recorded, and combined with the data of the previous day, the fluctuation amplitude of the permanganate index can be obtained in the regression inversion model with lower accuracy, thereby capturing the fluctuation trend of the permanganate index on the predicted day. The hidden state prediction value corresponding to the Gaussian hidden Markov model is selected as the prediction result using the above fluctuation trend, which makes up for the lack of the slow response of the Gaussian hidden Markov model to environmental changes, thereby improving the prediction accuracy of the model.

[0063] In this example, the hyperspectral remote sensing data are the hyperspectral remote sensing images taken by the Resource-1 02D hyperspectral camera, Resource-1 02E hyperspectral camera and GF-5 02 satellite hyperspectral camera in the study area of ​​this example from December 2022 to December 2023. The permanganate index data comes from the daily water quality data collected by the automatic water quality detection station in the study area of ​​this example from December 2022 to December 2023. Taking into account the characteristics of traditional remote sensing water quality inversion and water quality prediction, in order to quantitatively evaluate the performance of the method of the present invention, the following accuracy evaluation indicators are used: mean absolute error (MAE), mean relative error (MAPE), determination coefficient (R 2 ).

[0064] Those skilled in the art should understand that the discussion of the above embodiments is merely illustrative and is not intended to imply that the scope of protection of the present application is limited to these examples; based on the concept of the present application, the technical features in the above embodiments may be combined, the steps may be implemented in any order, and there are many other variations of different aspects of one or more embodiments of the present application as described above, which are not provided in detail for the sake of simplicity.

[0065] One or more embodiments of the present application are intended to cover all such substitutions, modifications and variations that fall within the broad scope of the present application. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of one or more embodiments of the present application should be included in the protection scope of the present application.

Claims

1. A multi-point inversion method for permanganate index based on hyperspectral remote sensing and water quality series, characterized in that: The method comprises: Acquire a hyperspectral remote sensing image of the point to be inverted and perform preprocessing to obtain a preprocessed hyperspectral remote sensing image, obtain permanganate index data collected by an automatic water quality monitoring station within the range of the hyperspectral remote sensing image, and obtain permanganate index water quality time series data; Extracting spectral information of the preprocessed hyperspectral remote sensing image and performing band alignment, extracting permanganate index water quality time series data and performing time grouping to obtain permanganate index water quality time series grouping data, and randomly sampling the spectral information after band alignment and the permanganate index water quality time series grouping data at a ratio of 2:1 as a training set and a test set, respectively; Constructing a remote sensing regression inversion model, wherein the remote sensing regression inversion model includes sensitive band screening, inversion band combination screening and final regression inversion model determination, and training the remote sensing regression inversion model using a spectral information training set; Constructing a Gaussian hidden Markov model, wherein the Gaussian hidden Markov model includes three hidden states and an optimal step size selection, and using the permanganate index water quality time series grouping data training set as input into the Gaussian hidden Markov model for training; The spectral information in the test set is input into the pre-trained remote sensing regression inversion model to obtain the inversion results of the points to be inverted; the permanganate index water quality time series grouped data in the test set is input into the pre-trained Gaussian hidden Markov model to obtain the prediction results under three different hidden states; The inversion result of the point to be inverted is used to infer the current hidden state of the Gaussian hidden Markov model to obtain the final permanganate index prediction result of each point, specifically: the final regression inversion model is verified for accuracy to obtain the error mean as a threshold, and the previous day's data in the permanganate index water quality time series data is used as a benchmark value; the prediction results of the Gaussian hidden Markov model are sorted from small to large to obtain the first hidden state, the second hidden state, and the third hidden state; it is determined whether the regression inversion result fluctuates compared with the previous day's data. If there is no fluctuation or the upper and lower fluctuation range does not exceed the threshold, the second hidden state is selected as the final permanganate index prediction result; if the downward fluctuation range exceeds the threshold, the first hidden state is selected as the final permanganate index prediction result; if the upward fluctuation range exceeds the threshold, the third hidden state is selected as the final permanganate index prediction result.

2. The multi-point inversion method of permanganate index as claimed in claim 1, characterized in that: The preprocessing is to perform radiation calibration, atmospheric correction and orthorectification on the hyperspectral remote sensing image. The calculation formula of the radiation calibration is: L = Gain·DN + Bias, where L is the pixel radiation brightness, the unit is W / (cm 2 ·μm·sr), DN is the brightness value of the remote sensing image pixel, Gain is the radiation calibration gain, and Bias is the radiation calibration offset, all in W / (cm 2 ·μm·sr); the calculation formula of the atmospheric correction is: C= Among them, C is the pixel radiation brightness after atmospheric correction, ρ is the pixel reflectivity, and ρ e is the average reflectivity of the area around the pixel, A and B are calculated based on the actual atmosphere and geometric conditions, S is the albedo of the large balloon surface, and C a It is the atmospheric radiation.

3. The multi-point inversion method of permanganate index as claimed in claim 1, characterized in that: Obtain permanganate index data collected by automatic water quality monitoring stations within the range of the hyperspectral remote sensing image, specifically, obtain permanganate index data collected 14 days before the shooting date of the hyperspectral remote sensing image, group the permanganate index water quality time series data in units of N days, and obtain permanganate index water quality time series grouped data, wherein 7≤N≤14.

4. The multi-point inversion method of permanganate index as claimed in claim 1, characterized in that: The processing method for band alignment is as follows: for hyperspectral remote sensing images taken by different hyperspectral satellites, one of the images taken by the hyperspectral satellite is selected as the reference image, and the rest are used as images to be registered; when the central band difference between each band of the image to be registered and each band of the reference image is within a specified range, the band is retained; when the difference exceeds the specified range, the average of the band of the image to be registered and the front band or the back band is taken as the new central band, so that the new central band difference is within the specified range, and the average of the radiation brightness values ​​of the front and back bands is taken as the radiation brightness value of the new central band.

5. The multi-point inversion method of permanganate index as claimed in claim 1, characterized in that: The sensitive band screening and inversion band combination screening in the remote sensing regression inversion model adopt the Pearson correlation coefficient method to select the band or band combination with the highest correlation as the calculation formula from spectral information to remote sensing regression inversion independent variable. The formula of Pearson correlation coefficient is: Among them, r is the Pearson correlation coefficient, X i is the remote sensing regression inversion independent variable, Y i For X i Corresponding to the measured permanganate index water quality data on the day the hyperspectral remote sensing image was taken, n is the number of samples, For X i and Y i The sample mean, S X , S Y is the sample standard deviation; the final regression inversion model in the remote sensing regression inversion model is that after the linear regression model, logarithmic regression model, polynomial regression model and power regression model are established, R 2 To verify the accuracy, the regression inversion model with the highest coefficient of determination was selected, among which, Among them, R 2 is the coefficient of determination, SSR is the regression sum of squares, SST is the total sum of squares, Yes and Y i The corresponding inverted water quality.

6. The multi-point inversion method of permanganate index as claimed in claim 1, characterized in that: The Gaussian Hidden Markov Model includes: Construct a Gaussian hidden Markov model to describe the hidden state of water quality and initialize the model parameters, including the state transfer matrix, initial probability matrix, mean matrix and covariance matrix; Using the permanganate index water quality time series grouping data, constructing the Gaussian hidden Markov observation sequence; The number of hidden states of water quality is set to 3, and the parameters of the Gaussian hidden Markov model are estimated using the Baum-Welch algorithm to obtain a trained state transfer matrix, an initial probability matrix, a mean matrix, and a covariance matrix; The weighted average of the expected observation values ​​of each state in the next stage is calculated through the state transfer matrix and the mean matrix as the predicted value of the hidden state.

7. The multi-point inversion method for permanganate index according to any one of claims 1 to 6, characterized in that: The test samples were used to verify the accuracy of the permanganate remote sensing inversion results.

8. The multi-point inversion method for permanganate index according to any one of claims 1 to 6, characterized in that: The error mean is taken from the mean absolute error of the final regression inversion model during the training process, and its formula is: Among them, MAE is the mean absolute error, Y i To measure the water quality data of permanganate index, Yes and Y i The corresponding inverted water quality.

9. The multi-point inversion method for permanganate index according to any one of claims 1 to 6, characterized in that: The hyperspectral remote sensing images are taken from the Ziyuan-1 02D hyperspectral camera, the Ziyuan-1 02E hyperspectral camera and the GF-5 02 hyperspectral camera.

Images