Water quality remote sensing inversion method based on spatial-spectral fusion and ensemble learning

Through the methods of fusion and ensemble learning, the image features of Sentinel-2 and Gaofen-2 are integrated, and the problems of low water quality inversion accuracy and difficulty in monitoring for long time series in small and medium-sized lakes are solved, and high-precision water quality inversion is achieved and dynamic monitoring.

CN119091313AActive Publication Date: 2024-12-06NINGBO UNIV
View PDF 6 Cites 0 Cited by

Patent Information

Application Number
CN202411013392.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-26
Publication Date
2024-12-06
Estimated Expiration
2044-07-26

AI Technical Summary

Technical Problem

The existing water quality inversion method based on multispectral remote sensing images has low inversion accuracy in small and medium-sized lakes and is difficult to meet the needs of dynamic monitoring of long-term time series changes in water quality.

Method used

Using the method based on null spectrum fusion and ensemble learning, an empty spectrum fusion model with multi-scale fusion function and deep residual learning function is constructed, and an integrated learning regression model including multiple basic models and meta models is constructed in combination with Stacking strategy, and the fusion of Sentinel II is integrated. and Gaofen2 satellite image features to improve water quality inversion accuracy and information mining capabilities.

Benefits of technology

It realizes high-precision water quality inversion in small and medium-sized lakes, and can dynamically monitor long-term series changes in water quality to meet practical application needs.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119091313A_ABST
    Figure CN119091313A_ABST
Patent Text Reader

Abstract

The invention discloses a water quality remote sensing inversion method based on spatial-spectral fusion and ensemble learning, and the method comprises the steps: constructing a spatial-spectral fusion model with a multi-scale fusion function and a deep residual learning function, and constructing an ensemble learning regression model which comprises a plurality of base models and a meta model based on a Stacking strategy; a space-spectrum fusion model is adopted to fuse the characteristics of a sentinel second satellite image and a high-resolution second satellite image of a target water area with low spatial resolution, a sentinel second satellite image with high spatial resolution is obtained, spectral band data is extracted, each base model is adopted to predict the spectral band data, and the spectral band data is predicted to obtain a high-resolution sentinel second satellite image. Multiple initial prediction results are obtained, the multiple initial prediction results are predicted through a meta-model, and water quality inversion data are obtained; the method has the advantages of being high in water quality inversion precision, suitable for small and medium-sized lakes and suitable for dynamic monitoring of water quality long-time sequence changes of the lakes, and meeting the current actual application requirements.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to a water quality remote sensing inversion method, in particular to a water quality remote sensing inversion method based on space-spectrum fusion and integrated learning. Background Art

[0002] Water quality monitoring is an important reference for water quality status and water body management. Water quality status is closely related to people's daily life, so it is necessary to study how to conduct water quality monitoring on a large scale, quickly and accurately. Remote sensing technology has become an important means of water quality monitoring in lakes and other places. The basic idea of ​​water quality remote sensing inversion is to first use water quality detection data and corresponding remote sensing satellite image data to establish a regression model, and then properly verify the regression model. Remote sensing satellite images can provide high-precision interpolation and more effectively generate spatially clear water quality maps, so this method consumes less time and labor costs. Water quality inversion methods based on multispectral remote sensing images have been widely used in the field of water quality monitoring.

[0003] The existing water quality inversion methods based on multispectral remote sensing images rely on the regression model established by water quality detection data and corresponding remote sensing satellite image data. At present, the spectral band value or band combination of the corresponding pixel of the sampling point in the multispectral remote sensing satellite image (i.e., remote sensing satellite image data) is used as the independent variable, and the water quality parameter concentration value of the sampling point (i.e., water quality detection data) is used as the dependent variable, which are used as the input and output of the regression model to train the regression model. However, some of the remote sensing satellite image data obtained by remote sensing technology have high spatial resolution but low spectral resolution, which makes the water quality inversion accuracy low; some have high spectral resolution but low spatial resolution, which limits the application in small and medium-sized lakes.

[0004] In addition, the concentration of water quality parameters in surface water is low. The current water quality inversion method based on multispectral remote sensing images only uses the regression model of a single machine learning machine, resulting in low information mining ability and generalization. There are great limitations in the dynamic monitoring of long-term series changes in lake water quality, which cannot meet the needs of current practical applications. Summary of the invention

[0005] The technical problem to be solved by the present invention is to provide a water quality remote sensing inversion method based on spatial-spectral fusion and ensemble learning which has high water quality inversion accuracy and is also applicable to small and medium-sized lakes and can be used for dynamic monitoring of long-term series changes in lake water quality, thus meeting the current practical application needs.

[0006] The technical solution adopted by the present invention to solve the above technical problems is: a water quality remote sensing inversion method based on spatial-spectral fusion and integrated learning, by constructing a spatial-spectral fusion model with multi-scale fusion function and deep residual learning function, and constructing an integrated learning regression model including multiple different base models and a meta-model based on the Stacking strategy, using the spatial-spectral fusion model to fuse the features of the Sentinel-2 satellite image and the Gaofen-2 satellite image of the target water area with low spatial resolution to obtain the Sentinel-2 satellite image with high spatial resolution, extracting the spectral band data of the water body in the Sentinel-2 satellite image with high spatial resolution to obtain the spectral band data, using each base model of the integrated learning regression model to predict the spectral band data respectively to obtain multiple initial prediction results, and then using the meta-model of the integrated learning regression model to predict the multiple initial prediction results to obtain water quality inversion data.

[0007] The spatial-spectral fusion model includes two feature extraction units, a multi-scale fusion unit and a stitching module, and the stitching module is called a first stitching module; the two feature extraction units are respectively called a first feature extraction unit and a second feature extraction unit; the first feature extraction unit is used to extract features of the Sentinel-2 satellite image of the target water area, and obtain the features of the Sentinel-2 satellite image and output them to the multi-scale fusion unit; the second feature extraction unit is used to extract features of the Gaofen-2 satellite image of the target water area, and obtain the features of the Gaofen-2 satellite image and output them to the multi-scale fusion unit; the multi-scale fusion unit is used to perform multi-scale fusion on the features of the Sentinel-2 satellite image and the Gaofen-2 satellite image output thereto, and obtain the fused high spatial resolution satellite image and output it to the stitching module; the first stitching module is used to stitch the fused high spatial resolution satellite image and the Sentinel-2 satellite image of the target water area to obtain a high spatial resolution Sentinel-2 satellite image.

[0008] Each of the feature extraction units includes N feature extraction modules, where N is an integer greater than or equal to 4; each of the feature extraction modules includes 4 first convolution modules, each of the first convolution modules includes a convolution layer and a PReLU activation function, the convolution layer is called the first convolution layer, the PReLU activation function is called the first PReLU activation function, the first convolution layer and the first PReLU activation function both have an input end and an output end, the output end of the first convolution layer is connected to the input end of the first PReLU activation function; in each first convolution module of each feature extraction module, The first convolution layer is used to extract local features of the image features connected to its input end, and the local image features are output to the input end of the first PReLU activation function through its output end. The first PReLU activation function is used to perform nonlinear transformation on the image features connected to its input end, and the corresponding image features are output through its output end. The input end of the first convolution layer is the input end of the first convolution module, and the output end of the first PReLU activation function is the output end of the first convolution module. In each feature extraction module, the output end of the pth first convolution module is connected to the pth The input end of the first convolution module is connected to the input end of the first convolution module, the input end of the first convolution module is connected to the output end of the fourth first convolution module, p=1,2,3, the input end of the first convolution module is used as the input end of the feature extraction module, and the output end of the fourth first convolution module is used as the output end of the feature extraction module; among the N feature extraction modules, the output end of the i-th feature extraction module is connected to the input end of the i+1-th feature extraction module, i=1,2,3,…,N-1; the input end of the first feature extraction module is the input end of the feature extraction unit, and the output end of the k-th feature extraction module is the output end of the feature extraction module. The kth output end of the feature extraction unit, k=1, 2, 3, ..., N; the input end of the first feature extraction unit is used to access the Sentinel-2 satellite image of the target waters, and the kth output end of the first feature extraction unit is used to output the Sentinel-2 satellite image features extracted by the kth feature extraction module therein to the multi-scale fusion unit; the input end of the second feature extraction unit is used to access the Gaofen-2 satellite image of the target waters, and the kth output end of the second feature extraction unit is used to output the Gaofen-2 satellite image features extracted by the kth feature extraction module therein to the multi-scale fusion unit.

[0009] The multiscale fusion unit includes N multiscale fusion modules and N-1 shallow fusion modules; each multiscale fusion module includes 1 computing module and 4 second convolution modules, each computing module has two input terminals and one output terminal, and the two input terminals are respectively referred to as the first input terminal and the second input terminal, the first input terminal of the computing module is used to access the image features of the Sentinel-2 satellite, and the second input terminal is used to access the image features of the Gaofen-2 satellite, and the computing module is used to perform a difference operation on the image features of the Sentinel-2 satellite and the Gaofen-2 satellite, and obtain the difference image features and output them through its output terminal; each second convolution module includes a convolution layer and a PReLU activation function, the convolution layer is referred to as the second convolution layer, the PReLU activation function is referred to as the second PReLU activation function, the second convolution layer and the second PReLU activation function both have an input terminal and an output terminal, and the output terminal of the second convolution layer is connected to the input terminal of the second PReLU activation function; in each second convolution module of each multiscale fusion module, the second convolution layer is used to connect its input terminal to the output terminal of the second convolution layer. The input image features are extracted locally, and the local image features are output to the input end of the second PReLU activation function through its output end. The second PReLU activation function is used to perform nonlinear transformation on the image features connected to its input end, and the corresponding image features are output through its output end. The input end of the second convolution layer is the input end of the second convolution module, and the output end of the second PReLU activation function is the output end of the second convolution module. In each multi-scale fusion module, the first input end of the calculation module is the first input end of the multi-scale fusion module, the second input end of the calculation module is the second input end of the multi-scale fusion module, the output end of the calculation module is respectively connected to the input ends of the four second convolution modules, the output end of the first second convolution module is respectively connected to the input end of the second second convolution module, the input end of the third second convolution module and the input end of the fourth second convolution module, the output end of the third second convolution module is connected to the input end of the fourth second convolution module, and the output end of the fourth second convolution module is the output end of the multi-scale fusion module.Each shallow fusion module includes a splicing module and a convolution layer. The splicing module is called the second splicing module, and the convolution layer is called the third convolution layer. The second splicing module has two input ends and one output end, and the third convolution layer has an input end and an output end. The output end of the second splicing module is connected to the input end of the third convolution layer. The two input ends of the second splicing module are the two input ends of the shallow fusion module, and the output end of the third convolution layer is the output end of the shallow fusion module. The output end of the first multi-scale fusion module and the output end of the second multi-scale fusion module are connected to the two input ends of the first shallow fusion module in a one-to-one correspondence. The jth shallow The output end of the layer fusion module and the output end of the j+2th multiscale fusion module are connected to the two input ends of the j+1th shallow fusion module in a one-to-one correspondence, j=1,2,…,N-2; the first input end of the kth multiscale fusion module is connected to the kth output end of the first feature extraction unit, the second input end of the kth multiscale fusion module is connected to the kth output end of the second feature extraction unit, the first splicing module has a first input end, a second input end and an output end, the output end of the N-1th shallow fusion module is connected to the first input end of the first splicing module, and the second input end of the first splicing module is used to access the Sentinel-2 satellite image of the target waters. ;

[0010] After training, the spatial-spectral fusion model fuses the features of the Sentinel-2 satellite image and the Gaofen-2 satellite image of the target water area with low spatial resolution. Before predicting the target water area, the integrated learning regression model also needs to be trained. The specific training process of the spatial-spectral fusion model and the integrated learning regression model is as follows:

[0011] Step 1: Select a certain water area as a reference water area, obtain N original Gaofen-2 satellite images of the reference water area from the Natural Resources Satellite Remote Sensing Cloud Service Platform, and obtain N original Sentinel-2 satellite images of the reference water area from the Copernicus open data platform of the European Space Agency, where N is an integer greater than or equal to 6, perform central cropping on the N original Gaofen-2 satellite images and the N original Sentinel-2 satellite images respectively to ensure that the sizes of the cropped N original Gaofen-2 satellite images and the N original Sentinel-2 satellite images are integer multiples of 256 pixels, and perform image registration to obtain N pre-processed Gaofen-2 satellite images and N pre-processed Sentinel-2 satellite images;

[0012] Step 2: According to the band characteristics of GF-2 satellite images and Sentinel-2 satellite images, each pre-processed GF-2 satellite image includes four bands, B1, B2, B3 and B4, and each pre-processed Sentinel-2 satellite image includes eight bands, B2, B3, B4, B5, B6, B7, B8 and B8A. Using the spectral response function of GF-2 satellite images, the N pre-processed GF-2 satellite images are divided into four bands, B1, B2, B3 and B4. The images are divided into four groups to obtain four groups of preprocessed Gaofen-2 satellite images. The first group includes N preprocessed Gaofen-2 satellite images with B1 band, the second group includes N preprocessed Gaofen-2 satellite images with B2 band, the third group includes N preprocessed Gaofen-2 satellite images with B3 band, and the fourth group includes N preprocessed Gaofen-2 satellite images with B4 band. The four groups of preprocessed Gaofen-2 satellite images total 4N single-band Gaofen-2 images. The spectral response function of the Sentinel-2 satellite image is used to divide the eight bands B2, B3, B4, B5, B6, B7, B8 and B8A of the N preprocessed Sentinel-2 satellite images into four groups, thereby obtaining four groups of preprocessed Sentinel-2 satellite images. The first group includes N preprocessed Sentinel-2 satellite images with the B2 band, the second group includes N preprocessed Sentinel-2 satellite images with the B3 band, and the third group includes N preprocessed Sentinel-2 satellite images with the B4 band. The fourth group of preprocessed Sentinel-2 satellite images includes N preprocessed Sentinel-2 satellite images with B5 band, B6 band, B7 band, B8 band and B8A band, and the four groups of preprocessed Sentinel-2 satellite images total 3N single-band Sentinel-2 satellite images and N conforming Sentinel-2 satellite images; at this time, 3N single-band Gaofen-2 satellite images and N composite-band Sentinel-2 satellite images are obtained, totaling 4N Sentinel-2 satellite images;

[0013] Step 3, constructing a sample set using 4N Gaofen-2 satellite images and 4N Sentinel-2 satellite images, to obtain a sample set including 4N Gaofen-2 satellite images and 4N Sentinel-2 satellite images;

[0014] Step 4, cropping the 4N Gaofen-2 satellite images and the 4N Sentinel-2 satellite images in the sample set into a plurality of small-block images of 256 pixels × 256 pixels with a step length of 128 pixels, to obtain a plurality of cropped small-block Gaofen-2 satellite images and a plurality of small-block Sentinel-2 satellite images;

[0015] Step 5: construct a data set using the cropped multiple small-block Gaofen-2 satellite images and multiple small-block Sentinel-2 satellite images to obtain a data set including multiple small-block Gaofen-2 satellite images and multiple small-block Sentinel-2 satellite images, which is a spatial-spectral fusion data set;

[0016] Step 6: Divide the spatial-spectral fusion dataset into a training set and a test set in proportion, where the training set contains 70% of the total dataset. If 70% of the total dataset is not an integer, it is rounded up and the rest is used as the test set.

[0017] Step 7: Input the training set into the spatial-spectral fusion model, and use the Adam optimization algorithm to train the spatial-spectral fusion model. Use the test set to verify the accuracy of the spatial-spectral fusion model until the accuracy meets the use requirements, and obtain the trained spatial-spectral fusion model.

[0018] Step 8: When it is necessary to invert the water quality parameters of the target water area, obtain the water quality data, geographic location data and Sentinel-2 satellite image data of the target area;

[0019] Step 9, using the same method as steps 1 to 5 to process the Sentinel-2 satellite image data of step 8, after obtaining a plurality of small-block Sentinel-2 satellite images, splicing the plurality of small-block Sentinel-2 satellite images into the trained spatial-spectral fusion model, using the trained spatial-spectral fusion model to generate a high-spatial-resolution Sentinel-2 satellite image, extracting the spectral band data of the generated high-spatial-resolution Sentinel-2 satellite image, and constructing a water quality inversion dataset using the extracted spectral band data and the water quality data and geographic location data in step S8 to obtain a water quality inversion dataset;

[0020] Step 10: Use the water quality inversion data set obtained in step 9 to train the ensemble learning regression model to obtain a trained ensemble learning regression model.

[0021] When performing remote sensing inversion of water quality in the target water area, after obtaining the Gaofen-2 satellite image of the target water area from the Natural Resources Satellite Remote Sensing Cloud Service Platform and the Sentinel-2 satellite image of the target water area from the Copernicus open data platform of the European Space Agency, the Gaofen-2 satellite image of the target water area and the Sentinel-2 satellite image of the target water area are first processed into a plurality of small-block Gaofen-2 satellite images and a plurality of small-block Sentinel-2 satellite images according to the methods of steps 1 to 5, and then the plurality of small-block Gaofen-2 satellite images and the plurality of small-block Sentinel-2 satellite images are matched one by one, and a corresponding small-block Gaofen-2 satellite image and a small-block Sentinel-2 satellite image form a combination to obtain a plurality of combinations, and the small-block Gaofen-2 satellite image and the small-block Sentinel-2 satellite image of each combination are input into the trained spatial-spectral fusion model for prediction to obtain the corresponding Sentinel-2 satellite image with high spatial resolution. Each combination obtains a high spatial resolution Sentinel-2 satellite image, and multiple combinations consisting of multiple small-block Gaofen-2 satellite images and multiple small-block Sentinel-2 satellite images obtain multiple high spatial resolution Sentinel-2 satellite images. The spectral band data of each high spatial resolution Sentinel-2 satellite image are extracted and input into the trained ensemble learning regression model, which outputs the corresponding water quality inversion data.

[0022] Compared with the prior art, the advantage of the present invention lies in that a spatial-spectral fusion model with multi-scale fusion function and deep residual learning function is constructed, and an integrated learning regression model including a plurality of different base models and a meta-model is constructed based on the Stacking strategy, and the spatial-spectral fusion model is used to fuse the features of the Sentinel-2 satellite image and the Gaofen-2 satellite image of the target water area with low spatial resolution to obtain the Sentinel-2 satellite image with high spatial resolution, and the spectral band data of the water body in the Sentinel-2 satellite image with high spatial resolution is extracted to obtain the spectral band data, and each base model of the integrated learning regression model is used to predict the spectral band data respectively to obtain multiple initial prediction results, and then the meta-model of the integrated learning regression model is used to predict the spectral band data. The model predicts multiple initial prediction results to obtain water quality inversion data. Since the spatial-spectral fusion model integrates multi-scale fusion function and deep residual learning function, it has high information mining ability and generalization. The spatial-spectral fusion model can fuse the Sentinel-2 satellite image and Gaofen-2 satellite image features of the target water area with low spatial resolution to obtain the Sentinel-2 satellite image with high spatial resolution, and then obtain spectral band data based on the Sentinel-2 satellite image with high spatial resolution as the input of the spatial-spectral fusion model for water quality inversion. Therefore, the present invention has high water quality inversion accuracy and is also suitable for small and medium-sized lakes, and can be used for dynamic monitoring of long-term series changes of lake water quality, meeting current practical application needs. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 It is a structural schematic diagram of an integrated learning regression model in a water quality remote sensing inversion method based on spatial-spectral fusion and integrated learning of the present invention;

[0024] Figure 2 It is a structural schematic diagram of a spatial-spectral fusion model in a water quality remote sensing inversion method based on spatial-spectral fusion and ensemble learning of the present invention;

[0025] Figure 3 It is a structural schematic diagram of a feature extraction module of a space-spectrum fusion model in a water quality remote sensing inversion method based on space-spectrum fusion and ensemble learning of the present invention;

[0026] Figure 4 It is a structural schematic diagram of the first convolution module of the space-spectrum fusion model in the water quality remote sensing inversion method based on space-spectrum fusion and ensemble learning of the present invention;

[0027] Figure 5 It is a structural schematic diagram of a multi-scale fusion module of a spatial-spectral fusion model in a water quality remote sensing inversion method based on spatial-spectral fusion and ensemble learning of the present invention;

[0028] Figure 6 It is a structural schematic diagram of the second convolution module of the spatial-spectral fusion model in the water quality remote sensing inversion method based on spatial-spectral fusion and ensemble learning of the present invention. DETAILED DESCRIPTION

[0029] The present invention is further described in detail below with reference to the accompanying drawings.

[0030] Embodiment 1: Figure 1 As shown, a water quality remote sensing inversion method based on spatial-spectral fusion and ensemble learning is disclosed. A spatial-spectral fusion model with multi-scale fusion function and deep residual learning function is constructed, and an ensemble learning regression model including a plurality of different base models and a meta-model is constructed based on the Stacking strategy. The spatial-spectral fusion model is used to fuse the features of the Sentinel-2 satellite image and the Gaofen-2 satellite image of the target water area with low spatial resolution to obtain the Sentinel-2 satellite image with high spatial resolution. The spectral band data of the water body in the Sentinel-2 satellite image with high spatial resolution is extracted to obtain the spectral band data. Each base model of the ensemble learning regression model is used to predict the spectral band data respectively to obtain multiple initial prediction results. Then the meta-model of the ensemble learning regression model is used to predict the multiple initial prediction results to obtain water quality inversion data.

[0031] In this embodiment, since the spatial-spectral fusion model integrates multi-scale fusion function and deep residual learning function, it has high information mining ability and generalization. The spatial-spectral fusion model can fuse the Sentinel-2 satellite image and Gaofen-2 satellite image features of the target water area with low spatial resolution to obtain the Sentinel-2 satellite image with high spatial resolution, and then obtain spectral band data based on the Sentinel-2 satellite image with high spatial resolution as the input of the spatial-spectral fusion model for water quality inversion. Therefore, the present invention can be applicable to small and medium-sized lakes while having high water quality inversion accuracy, and can be applicable to the dynamic monitoring of long-term series changes of lake water quality, meeting the current practical application needs.

[0032] Embodiment 2: This embodiment is basically the same as Embodiment 1, except that: Figure 2 As shown, in this embodiment, the spatial-spectral fusion model includes two feature extraction units, a multi-scale fusion unit and a stitching module, and the stitching module is referred to as a first stitching module; the two feature extraction units are respectively referred to as a first feature extraction unit and a second feature extraction unit; the first feature extraction unit is used to perform feature extraction on the Sentinel-2 satellite image of the target water area, and obtain the Sentinel-2 satellite image features to output to the multi-scale fusion unit; the second feature extraction unit is used to perform feature extraction on the Gaofen-2 satellite image of the target water area, and obtain the Gaofen-2 satellite image features to output to the multi-scale fusion unit; the multi-scale fusion unit is used to perform multi-scale fusion on the Sentinel-2 satellite image features and the Gaofen-2 satellite image features output thereto, and obtain the fused high spatial resolution satellite image to output to the stitching module; the first stitching module is used to stitch the fused high spatial resolution satellite image with the Sentinel-2 satellite image of the target water area to obtain a high spatial resolution Sentinel-2 satellite image.

[0033] like Figure 3 and Figure 4As shown, in this embodiment, each feature extraction unit includes N feature extraction modules, where N is an integer greater than or equal to 4; each feature extraction module includes 4 first convolution modules, each first convolution module includes a convolution layer and a PReLU activation function, the convolution layer is called the first convolution layer, the PReLU activation function is called the first PReLU activation function, the first convolution layer and the first PReLU activation function both have an input end and an output end, and the output end of the first convolution layer is connected to the input end of the first PReLU activation function; in each first convolution module of each feature extraction module, the first convolution layer is used to perform local feature extraction on the image features connected to its input end, and the obtained local image features are output to the input end of the first PReLU activation function through its output end, the first PReLU activation function is used to perform nonlinear transformation on the image features connected to its input end, and the obtained corresponding image features are output through its output end, the input end of the first convolution layer is the input end of the first convolution module, and the output end of the first PReLU activation function is the output end of the first convolution module; in each feature extraction module, the output end of the pth first convolution module is connected to the p+th first convolution module. The input end of the 1st first convolution module is connected, the input end of the 1st first convolution module is connected to the output end of the 4th first convolution module, p = 1, 2, 3, the input end of the 1st first convolution module is used as the input end of the feature extraction module, and the output end of the 4th first convolution module is used as the output end of the feature extraction module; among the N feature extraction modules, the output end of the i-th feature extraction module is connected to the input end of the i+1-th feature extraction module, i = 1, 2, 3, ..., N-1; the input end of the 1st feature extraction module is the input end of the feature extraction unit, the k-th feature extraction module The output end of is the kth output end of the feature extraction unit, k=1,2,3,…,N; the input end of the first feature extraction unit is used to access the Sentinel-2 satellite image of the target water area, and the kth output end of the first feature extraction unit is used to output the Sentinel-2 satellite image features extracted by the kth feature extraction module therein to the multi-scale fusion unit; the input end of the second feature extraction unit is used to access the Gaofen-2 satellite image of the target water area, and the kth output end of the second feature extraction unit is used to output the Gaofen-2 satellite image features extracted by the kth feature extraction module therein to the multi-scale fusion unit.

[0034] like Figure 5 and Figure 6As shown, in this embodiment, the multiscale fusion unit includes N multiscale fusion modules and N-1 shallow fusion modules; each multiscale fusion module includes 1 computing module and 4 second convolution modules, each computing module has two input terminals and one output terminal, and its two input terminals are respectively referred to as the first input terminal and the second input terminal, the first input terminal of the computing module is used to access the image features of the Sentinel-2 satellite, and the second input terminal is used to access the image features of the Gaofen-2 satellite, and the computing module is used to perform a difference operation on the image features of the Sentinel-2 satellite and the Gaofen-2 satellite, and obtain the difference image features and output them through its output terminal; each second convolution module includes a convolution layer and a PReLU activation function, and the convolution layer is referred to as the first input terminal. The second convolution layer, the PReLU activation function is called the second PReLU activation function, the second convolution layer and the second PReLU activation function both have an input end and an output end, the output end of the second convolution layer is connected to the input end of the second PReLU activation function; in each second convolution module of each multi-scale fusion module, the second convolution layer is used to extract local features of the image features connected to its input end, and the local image features are output to the input end of the second PReLU activation function through its output end, the second PReLU activation function is used to perform nonlinear transformation on the image features connected to its input end, and the corresponding image features are output through its output end, the input end of the second convolution layer is the input end of the second convolution module, and the second PReLU activation function The output end of the number is the output end of the second convolution module; in each multi-scale fusion module, the first input end of the calculation module is the first input end of the multi-scale fusion module, the second input end of the calculation module is the second input end of the multi-scale fusion module, the output end of the calculation module is connected to the input ends of the four second convolution modules respectively, the output end of the first second convolution module is connected to the input end of the second second convolution module, the input end of the third second convolution module and the input end of the fourth second convolution module respectively, the output end of the third second convolution module is connected to the input end of the fourth second convolution module, and the output end of the fourth second convolution module is the output end of the multi-scale fusion module; each shallow fusion module includes a splicing module and a convolution layer, the splicing module is connected to the convolution layer. The convolution module is called the second convolution module, the convolution layer is called the third convolution layer, the second convolution module has two input ends and one output end, the third convolution layer has an input end and an output end, the output end of the second convolution module is connected to the input end of the third convolution layer, the two input ends of the second convolution module are the two input ends of the shallow fusion module, the output end of the third convolution layer is the output end of the shallow fusion module, the output end of the first multi-scale fusion module and the output end of the second multi-scale fusion module are connected to the two input ends of the first shallow fusion module in one-to-one correspondence, the output end of the jth shallow fusion module and the output end of the j+2th multi-scale fusion module are connected to the two input ends of the j+1th shallow fusion module in one-to-one correspondence, j=1,2,…,N-2;The first input end of the k-th multi-scale fusion module is connected to the k-th output end of the first feature extraction unit, the second input end of the k-th multi-scale fusion module is connected to the k-th output end of the second feature extraction unit, the first splicing module has a first input end, a second input end and an output end, the output end of the N-1-th shallow fusion module is connected to the first input end of the first splicing module, and the second input end of the first splicing module is used to access the Sentinel-2 satellite image of the target waters. ;

[0035] In this embodiment, when performing water quality remote sensing inversion on the target water area, the Gaofen-2 satellite image of the target water area is obtained from the Natural Resources Satellite Remote Sensing Cloud Service Platform, and the Sentinel-2 satellite image of the target water area is obtained from the Copernicus open data platform of the European Space Agency. The Gaofen-2 satellite image of the target water area and the Sentinel-2 satellite image of the target water area are first processed into a plurality of small-block Gaofen-2 satellite images and a plurality of small-block Sentinel-2 satellite images, respectively. Then, after the plurality of small-block Gaofen-2 satellite images and the plurality of small-block Sentinel-2 satellite images are matched one by one, a corresponding small-block Gaofen-2 satellite image and a small-block Sentinel-2 satellite image are matched. The small Sentinel-2 satellite images of the target waters are combined into a combination to obtain a plurality of combinations, and the small blocks of the Gaofen-2 satellite images of each combination are input to the input end of the first feature extraction unit and the first input end of the first stitching module, and the small blocks of the Sentinel-2 satellite images of each combination are input to the input end of the second feature extraction unit. At this time, the first feature extraction unit performs image feature extraction on the small blocks of the Sentinel-2 satellite images of the target waters, and the second feature extraction unit performs feature extraction on the small blocks of the Gaofen-2 satellite images of the target waters. The first feature extraction unit performs image feature extraction on the small blocks of the Sentinel-2 satellite images of the target waters. The specific process of extraction is as follows: first, the first feature extraction module of the first feature extraction unit extracts features from the Sentinel-2 satellite image of the target waters, obtains the corresponding Sentinel-2 satellite image features and outputs them through its output end; from the second feature extraction module of the first feature extraction unit to the nth feature extraction module, when the corresponding Sentinel-2 satellite image features are input into the input end of a feature extraction module, the feature extraction module extracts features from the Sentinel-2 satellite image features input thereto, obtains the corresponding Sentinel-2 satellite image features and outputs them through its output end; the second feature extraction unit extracts features from the Sentinel-2 satellite image of the target waters, obtains the corresponding Sentinel-2 satellite image features and outputs them through its output end; The specific process of performing image feature extraction on a small block of Gaofen-2 satellite images is as follows: first, the first feature extraction module of the second feature extraction unit performs feature extraction on the Gaofen-2 satellite image of the target waters, obtains corresponding Gaofen-2 satellite image features and outputs them through its output end; from the second feature extraction module of the first feature extraction unit to the nth feature extraction module, when the Gaofen-2 satellite image features are input into the input end of a certain feature extraction module, the feature extraction module performs feature extraction on the Gaofen-2 satellite image features input thereto, obtains corresponding Gaofen-2 satellite image features and outputs them through its output end;When in the multiscale fusion unit, the first input end and the second input end of a multiscale fusion module are both connected to the Sentinel-2 satellite image features and the Gaofen-2 satellite image features, the multiscale fusion module performs feature fusion on the connected Sentinel-2 satellite image features and the Gaofen-2 satellite image features, and obtains the corresponding fused Gaofen-2 satellite image features and outputs them through its output end; when the first input end and the second input end of a shallow fusion module are both connected to the fused Gaofen-2 satellite image features, the shallow fusion module performs feature fusion on the two fused Gaofen-2 satellite image features and obtains the corresponding fused Gaofen-2 satellite image features and outputs them through its output end; in the multiscale fusion When the unit completes the multi-scale fusion of the input Sentinel-2 satellite image features and the Gaofen-2 satellite image features, the fused Gaofen-2 satellite image features output by the output end are the fused high spatial resolution satellite images; at this time, the first splicing module splices the fused high spatial resolution satellite images and the Sentinel-2 satellite images of the target waters to obtain the high spatial resolution Sentinel-2 satellite images of each combination, and then splices all the high spatial resolution Sentinel-2 satellite images of the combination to obtain the high spatial resolution Sentinel-2 satellite images of the target waters, which can be used for the subsequent extraction of spectral band data. ;

[0036] Embodiment 3: This embodiment is basically the same as Embodiment 1, except that: Figure 2 As shown, in this embodiment, after training, the spatial-spectral fusion model fuses the features of the Sentinel-2 satellite image and the Gaofen-2 satellite image of the target water area with low spatial resolution. Before predicting the target water area, the integrated learning regression model also needs to be trained. The specific training process of the spatial-spectral fusion model and the integrated learning regression model is as follows:

[0037] Step 1: Select a certain water area as a reference water area, obtain N original Gaofen-2 satellite images of the reference water area from the Natural Resources Satellite Remote Sensing Cloud Service Platform, and obtain N original Sentinel-2 satellite images of the reference water area from the Copernicus open data platform of the European Space Agency, where N is an integer greater than or equal to 6, perform central cropping on the N original Gaofen-2 satellite images and the N original Sentinel-2 satellite images respectively to ensure that the sizes of the cropped N original Gaofen-2 satellite images and the N original Sentinel-2 satellite images are integer multiples of 256 pixels, and perform image registration to obtain N pre-processed Gaofen-2 satellite images and N pre-processed Sentinel-2 satellite images;

[0038] Step 2: According to the band characteristics of GF-2 satellite images and Sentinel-2 satellite images, each pre-processed GF-2 satellite image includes four bands, B1, B2, B3 and B4, and each pre-processed Sentinel-2 satellite image includes eight bands, B2, B3, B4, B5, B6, B7, B8 and B8A. Using the spectral response function of GF-2 satellite images, the N pre-processed GF-2 satellite images are divided into four bands, B1, B2, B3 and B4. The images are divided into four groups to obtain four groups of preprocessed Gaofen-2 satellite images. The first group includes N preprocessed Gaofen-2 satellite images with B1 band, the second group includes N preprocessed Gaofen-2 satellite images with B2 band, the third group includes N preprocessed Gaofen-2 satellite images with B3 band, and the fourth group includes N preprocessed Gaofen-2 satellite images with B4 band. The four groups of preprocessed Gaofen-2 satellite images total 4N single-band Gaofen-2 images. The spectral response function of the Sentinel-2 satellite image is used to divide the eight bands B2, B3, B4, B5, B6, B7, B8 and B8A of the N preprocessed Sentinel-2 satellite images into four groups, thereby obtaining four groups of preprocessed Sentinel-2 satellite images. The first group includes N preprocessed Sentinel-2 satellite images with the B2 band, the second group includes N preprocessed Sentinel-2 satellite images with the B3 band, and the third group includes N preprocessed Sentinel-2 satellite images with the B4 band. The fourth group of preprocessed Sentinel-2 satellite images includes N preprocessed Sentinel-2 satellite images with B5 band, B6 band, B7 band, B8 band and B8A band, and the four groups of preprocessed Sentinel-2 satellite images total 3N single-band Sentinel-2 satellite images and N conforming Sentinel-2 satellite images; at this time, 3N single-band Gaofen-2 satellite images and N composite-band Sentinel-2 satellite images are obtained, totaling 4N Sentinel-2 satellite images;

[0039] Step 3, constructing a sample set using 4N Gaofen-2 satellite images and 4N Sentinel-2 satellite images, to obtain a sample set including 4N Gaofen-2 satellite images and 4N Sentinel-2 satellite images;

[0040] Step 4, cropping the 4N Gaofen-2 satellite images and the 4N Sentinel-2 satellite images in the sample set into a plurality of small-block images of 256 pixels × 256 pixels with a step length of 128 pixels, to obtain a plurality of cropped small-block Gaofen-2 satellite images and a plurality of small-block Sentinel-2 satellite images;

[0041] Step 5: construct a data set using the cropped multiple small-block Gaofen-2 satellite images and multiple small-block Sentinel-2 satellite images to obtain a data set including multiple small-block Gaofen-2 satellite images and multiple small-block Sentinel-2 satellite images, which is a spatial-spectral fusion data set;

[0042] Step 6: Divide the spatial-spectral fusion dataset into a training set and a test set in proportion, where the training set contains 70% of the total dataset. If 70% of the total dataset is not an integer, it is rounded up and the rest is used as the test set.

[0043] Step 7: Input the training set into the spatial-spectral fusion model, and use the Adam optimization algorithm to train the spatial-spectral fusion model. Use the test set to verify the accuracy of the spatial-spectral fusion model until the accuracy meets the use requirements, and obtain the trained spatial-spectral fusion model.

[0044] Step 8: When it is necessary to invert the water quality parameters of the target water area, obtain the water quality data, geographic location data and Sentinel-2 satellite image data of the target area;

[0045] Step 9, using the same method as steps 1 to 5 to process the Sentinel-2 satellite image data of step 8, after obtaining a plurality of small-block Sentinel-2 satellite images, splicing the plurality of small-block Sentinel-2 satellite images into the trained spatial-spectral fusion model, using the trained spatial-spectral fusion model to generate a high-spatial-resolution Sentinel-2 satellite image, extracting the spectral band data of the generated high-spatial-resolution Sentinel-2 satellite image, and constructing a water quality inversion dataset using the extracted spectral band data and the water quality data and geographic location data in step S8 to obtain a water quality inversion dataset;

[0046] Step 10: Use the water quality inversion data set obtained in step 9 to train the ensemble learning regression model to obtain a trained ensemble learning regression model.

[0047] In this embodiment, when performing water quality remote sensing inversion on the target water area, after obtaining the Gaofen-2 satellite image of the target water area from the Natural Resources Satellite Remote Sensing Cloud Service Platform and the Sentinel-2 satellite image of the target water area from the Copernicus open data platform of the European Space Agency, the Gaofen-2 satellite image of the target water area and the Sentinel-2 satellite image of the target water area are first processed into a plurality of small-block Gaofen-2 satellite images and a plurality of small-block Sentinel-2 satellite images according to the methods of steps 1 to 5, and then the plurality of small-block Gaofen-2 satellite images and the plurality of small-block Sentinel-2 satellite images are matched one by one, and a corresponding small-block Gaofen-2 satellite image and a small-block Sentinel-2 satellite image form a combination to obtain a plurality of combinations, and the small-block Gaofen-2 satellite image and the small-block Sentinel-2 satellite image of each combination are input into the trained spatial-spectral fusion model for prediction to obtain the corresponding high spatial resolution Sentinel-2 satellite image. Each combination obtains a high spatial resolution Sentinel-2 satellite image, and multiple combinations consisting of multiple small-block Gaofen-2 satellite images and multiple small-block Sentinel-2 satellite images obtain multiple high spatial resolution Sentinel-2 satellite images. The spectral band data of each high spatial resolution Sentinel-2 satellite image are extracted and input into the trained ensemble learning regression model, which outputs the corresponding water quality inversion data.

[0048] In this embodiment, when training the spatial-spectral fusion model, the multi-band data of the Sentinel-2 satellite images and the Gaofen-2 satellite images are fused, and centralized cropping, image registration and grouping processing are performed, and the images are cropped into small blocks of 256×256 pixels, so that a rich and diverse spatial-spectral fusion data set can be constructed. The spatial-spectral fusion model is then trained using the Adam optimization algorithm to improve image resolution, enhance the generalization ability of the spatial-spectral fusion model, and improve training efficiency, thereby ensuring that the spatial-spectral fusion model can accurately invert the actual situation of the target waters.

Claims

1. A water quality remote sensing inversion method based on spatial-spectral fusion and ensemble learning, characterized in that A spatial-spectral fusion model with multi-scale fusion and deep residual learning functions is constructed, and an integrated learning regression model including a plurality of different base models and a meta-model is constructed based on the Stacking strategy. The spatial-spectral fusion model is used to fuse the features of the Sentinel-2 satellite images and the Gaofen-2 satellite images of the target water area with low spatial resolution to obtain the Sentinel-2 satellite images with high spatial resolution. The spectral band data of the water body in the Sentinel-2 satellite images with high spatial resolution are extracted to obtain the spectral band data. Each base model of the integrated learning regression model is used to predict the spectral band data respectively to obtain multiple initial prediction results. Then the meta-model of the integrated learning regression model is used to predict the multiple initial prediction results to obtain water quality inversion data.

2. A water quality remote sensing inversion method based on spatial-spectral fusion and ensemble learning according to claim 1, characterized in that The spatial-spectral fusion model includes two feature extraction units, a multi-scale fusion unit and a stitching module, and the stitching module is called a first stitching module; the two feature extraction units are respectively called a first feature extraction unit and a second feature extraction unit; the first feature extraction unit is used to extract features of the Sentinel-2 satellite image of the target water area, and obtain the features of the Sentinel-2 satellite image and output them to the multi-scale fusion unit; the second feature extraction unit is used to extract features of the Gaofen-2 satellite image of the target water area, and obtain the features of the Gaofen-2 satellite image and output them to the multi-scale fusion unit; the multi-scale fusion unit is used to perform multi-scale fusion on the features of the Sentinel-2 satellite image and the Gaofen-2 satellite image output thereto, and obtain the fused high spatial resolution satellite image and output it to the stitching module; the first stitching module is used to stitch the fused high spatial resolution satellite image and the Sentinel-2 satellite image of the target water area to obtain a high spatial resolution Sentinel-2 satellite image.

3. A water quality remote sensing inversion method based on spatial-spectral fusion and ensemble learning according to claim 2, characterized in that Each of the feature extraction units includes N feature extraction modules, where N is an integer greater than or equal to 4; each feature extraction module includes 4 first convolution modules, each of the first convolution modules includes a convolution layer and a PReLU activation function, the convolution layer is called the first convolution layer, the PReLU activation function is called the first PReLU activation function, the first convolution layer and the first PReLU activation function both have an input end and an output end, and the output end of the first convolution layer is connected to the input end of the first PReLU activation function; in each first convolution module of each feature extraction module, the first convolution layer is used to perform local feature extraction on the image features connected to its input end, and the obtained local image features are output to the input end of the first PReLU activation function through its output end, the first PReLU activation function is used to perform nonlinear transformation on the image features connected to its input end, and the obtained corresponding image features are output through its output end, the input end of the first convolution layer is the input end of the first convolution module, and the output end of the first PReLU activation function is the output end of the first convolution module; In each feature extraction module, the output end of the pth first convolution module is connected to the input end of the p+1th first convolution module, and the input end of the 1st first convolution module is connected to the output end of the 4th first convolution module, p=1,2,3, the input end of the 1st first convolution module is used as the input end of the feature extraction module, and the output end of the 4th first convolution module is used as the output end of the feature extraction module; Among the N feature extraction modules, the output end of the i-th feature extraction module is connected to the input end of the i+1-th feature extraction module, i=1,2,3,…,N-1; the input end of the first feature extraction module is the input end of the feature extraction unit, and the output end of the k-th feature extraction module is the k-th output end of the feature extraction unit, k=1,2,3,…,N; The input end of the first feature extraction unit is used to access the Sentinel-2 satellite image of the target waters, and the kth output end of the first feature extraction unit is used to output the Sentinel-2 satellite image features extracted by the kth feature extraction module therein to the multi-scale fusion unit; the input end of the second feature extraction unit is used to access the Gaofen-2 satellite image of the target waters, and the kth output end of the second feature extraction unit is used to output the Gaofen-2 satellite image features extracted by the kth feature extraction module therein to the multi-scale fusion unit.

4. The water quality remote sensing inversion method based on spatial-spectral fusion and ensemble learning according to claim 2 is characterized in that The multiscale fusion unit includes N multiscale fusion modules and N-1 shallow fusion modules; each multiscale fusion module includes 1 computing module and 4 second convolution modules, each computing module has two input terminals and one output terminal, and the two input terminals are respectively referred to as the first input terminal and the second input terminal, the first input terminal of the computing module is used to access the image features of the Sentinel-2 satellite, and the second input terminal is used to access the image features of the Gaofen-2 satellite, and the computing module is used to perform a difference operation on the image features of the Sentinel-2 satellite and the Gaofen-2 satellite, and obtain the difference image features and output them through its output terminal; each second convolution module includes a convolution layer and a PReLU activation function, the convolution layer is referred to as the second convolution layer, and the PReLU activation function is referred to as the second PReLU activation function, the second convolution layer and the second PReLU activation function both have an input terminal and an output terminal, and the output terminal of the second convolution layer is connected to the input terminal of the second PReLU activation function; In each second convolution module of each multi-scale fusion module, the second convolution layer is used to perform local feature extraction on the image features connected to its input end, and the obtained local image features are output to the input end of the second PReLU activation function through its output end. The second PReLU activation function is used to perform nonlinear transformation on the image features connected to its input end, and the obtained corresponding image features are output through its output end. The input end of the second convolution layer is the input end of the second convolution module, and the output end of the second PReLU activation function is the output end of the second convolution module. In each multi-scale fusion module, the first input end of the calculation module is the first input end of the multi-scale fusion module, the second input end of the calculation module is the second input end of the multi-scale fusion module, the output end of the calculation module is respectively connected to the input ends of the four second convolution modules, the output end of the first second convolution module is respectively connected to the input end of the second second convolution module, the input end of the third second convolution module and the input end of the fourth second convolution module, the output end of the third second convolution module is connected to the input end of the fourth second convolution module, and the output end of the fourth second convolution module is the output end of the multi-scale fusion module; Each shallow fusion module includes a splicing module and a convolution layer. The splicing module is called the second splicing module, and the convolution layer is called the third convolution layer. The second splicing module has two input ends and one output end, and the third convolution layer has an input end and an output end. The output end of the second splicing module is connected to the input end of the third convolution layer. The two input ends of the second splicing module are the two input ends of the shallow fusion module, and the output end of the third convolution layer is the output end of the shallow fusion module. The output end of the first multi-scale fusion module and the output end of the second multi-scale fusion module are connected to the two input ends of the first shallow fusion module in a one-to-one correspondence. The jth shallow The output end of the layer fusion module and the output end of the j+2th multiscale fusion module are connected to the two input ends of the j+1th shallow layer fusion module in a one-to-one correspondence, j=1,2,…,N-2; the first input end of the kth multiscale fusion module is connected to the kth output end of the first feature extraction unit, the second input end of the kth multiscale fusion module is connected to the kth output end of the second feature extraction unit, the first splicing module has a first input end, a second input end and an output end, the output end of the N-1th shallow layer fusion module is connected to the first input end of the first splicing module, and the second input end of the first splicing module is used to access the Sentinel-2 satellite image of the target waters.

5. A water quality remote sensing inversion method based on spatial-spectral fusion and ensemble learning according to any one of claims 1 to 4, characterized in that After training, the spatial-spectral fusion model fuses the features of the Sentinel-2 satellite image and the Gaofen-2 satellite image of the target water area with low spatial resolution. Before predicting the target water area, the integrated learning regression model also needs to be trained. The specific training process of the spatial-spectral fusion model and the integrated learning regression model is as follows: Step 1: Select a certain water area as a reference water area, obtain N original Gaofen-2 satellite images of the reference water area from the Natural Resources Satellite Remote Sensing Cloud Service Platform, and obtain N original Sentinel-2 satellite images of the reference water area from the Copernicus open data platform of the European Space Agency, where N is an integer greater than or equal to 6, perform central cropping on the N original Gaofen-2 satellite images and the N original Sentinel-2 satellite images respectively to ensure that the sizes of the cropped N original Gaofen-2 satellite images and the N original Sentinel-2 satellite images are integer multiples of 256 pixels, and perform image registration to obtain N pre-processed Gaofen-2 satellite images and N pre-processed Sentinel-2 satellite images; Step 2: According to the band characteristics of GF-2 satellite images and Sentinel-2 satellite images, each pre-processed GF-2 satellite image includes four bands, B1, B2, B3 and B4, and each pre-processed Sentinel-2 satellite image includes eight bands, B2, B3, B4, B5, B6, B7, B8 and B8A; using the spectral response function of GF-2 satellite images, the N pre-processed GF-2 satellite images are divided into four bands according to the four bands of B1, B2, B3 and B4. groups, obtaining four groups of preprocessed Gaofen-2 satellite images, the first group including N preprocessed Gaofen-2 satellite images with B1 band, the second group including N preprocessed Gaofen-2 satellite images with B2 band, the third group including N preprocessed Gaofen-2 satellite images with B3 band, the fourth group including N preprocessed Gaofen-2 satellite images with B4 band, the four groups of preprocessed Gaofen-2 satellite images totaling 4N single-band Gaofen-2 satellite images; Using the spectral response function of the Sentinel-2 satellite image, the eight bands B2, B3, B4, B5, B6, B7, B8 and B8A of the N preprocessed Sentinel-2 satellite images are divided into four groups, and four groups of preprocessed Sentinel-2 satellite images are obtained. The first group includes N preprocessed Sentinel-2 satellite images with the B2 band, the second group includes N preprocessed Sentinel-2 satellite images with the B3 band, and the third group includes N preprocessed Sentinel-2 satellite images with the B4 band. The fourth group includes N pre-processed Sentinel-2 satellite images with B5 band, B6 band, B7 band, B8 band and B8A band. The four groups of pre-processed Sentinel-2 satellite images total 3N single-band Sentinel-2 satellite images and N conforming Sentinel-2 satellite images. At this time, 3N single-band Gaofen-2 satellite images and N composite-band Sentinel-2 satellite images are obtained, totaling 4N Sentinel-2 satellite images. Step 3, constructing a sample set using 4N Gaofen-2 satellite images and 4N Sentinel-2 satellite images, to obtain a sample set including 4N Gaofen-2 satellite images and 4N Sentinel-2 satellite images; Step 4, cropping the 4N Gaofen-2 satellite images and the 4N Sentinel-2 satellite images in the sample set into a plurality of small-block images of 256 pixels × 256 pixels with a step length of 128 pixels, to obtain a plurality of cropped small-block Gaofen-2 satellite images and a plurality of small-block Sentinel-2 satellite images; Step 5: construct a data set using the cropped multiple small-block Gaofen-2 satellite images and multiple small-block Sentinel-2 satellite images to obtain a data set including multiple small-block Gaofen-2 satellite images and multiple small-block Sentinel-2 satellite images, which is a spatial-spectral fusion data set; Step 6: Divide the spatial-spectral fusion dataset into a training set and a test set in proportion, where the training set contains 70% of the total dataset. If 70% of the total dataset is not an integer, it is rounded up and the rest is used as the test set. Step 7: Input the training set into the spatial-spectral fusion model, and use the Adam optimization algorithm to train the spatial-spectral fusion model. Use the test set to verify the accuracy of the spatial-spectral fusion model until the accuracy meets the use requirements, and obtain the trained spatial-spectral fusion model. Step 8: When it is necessary to invert the water quality parameters of the target water area, obtain the water quality data, geographic location data and Sentinel-2 satellite image data of the target area; Step 9, using the same method as steps 1 to 5 to process the Sentinel-2 satellite image data of step 8, after obtaining a plurality of small-block Sentinel-2 satellite images, splicing the plurality of small-block Sentinel-2 satellite images into the trained spatial-spectral fusion model, using the trained spatial-spectral fusion model to generate a high-spatial-resolution Sentinel-2 satellite image, extracting the spectral band data of the generated high-spatial-resolution Sentinel-2 satellite image, and constructing a water quality inversion dataset using the extracted spectral band data and the water quality data and geographic location data in step S8 to obtain a water quality inversion dataset; Step 10: Use the water quality inversion data set obtained in step 9 to train the ensemble learning regression model to obtain a trained ensemble learning regression model.

6. The water quality remote sensing inversion method based on spatial-spectral fusion and ensemble learning according to claim 5 is characterized in that When performing remote sensing inversion of water quality in the target water area, after obtaining the Gaofen-2 satellite image of the target water area from the Natural Resources Satellite Remote Sensing Cloud Service Platform and the Sentinel-2 satellite image of the target water area from the Copernicus open data platform of the European Space Agency, the Gaofen-2 satellite image of the target water area and the Sentinel-2 satellite image of the target water area are first processed into a plurality of small-block Gaofen-2 satellite images and a plurality of small-block Sentinel-2 satellite images according to the methods of steps 1 to 5, and then the plurality of small-block Gaofen-2 satellite images and the plurality of small-block Sentinel-2 satellite images are matched one by one, and a corresponding small-block Gaofen-2 satellite image and a small-block Sentinel-2 satellite image form a combination to obtain a plurality of combinations, and the small-block Gaofen-2 satellite image and the small-block Sentinel-2 satellite image of each combination are input into the trained spatial-spectral fusion model for prediction to obtain the corresponding Sentinel-2 satellite image with high spatial resolution. Each combination obtains a high spatial resolution Sentinel-2 satellite image, and multiple combinations consisting of multiple small-block Gaofen-2 satellite images and multiple small-block Sentinel-2 satellite images obtain multiple high spatial resolution Sentinel-2 satellite images. The spectral band data of each high spatial resolution Sentinel-2 satellite image are extracted and input into the trained ensemble learning regression model, which outputs the corresponding water quality inversion data.

Citation Information

Patent Citations

  • Satellite image and machine learning water quality monitoring method and system

    CN112464746A

  • Remote sensing water quality inversion method combining space-time fusion and deep learning

    CN113324923A

  • Chlorophyll concentration inversion method based on space-spectrum fusion and model learning coupling

    CN114330530A

  • Water quality inversion method based on feature adaptive operation and machine learning fusion

    CN115561176A

  • Water quality inversion method and device based on satellite remote sensing and integrated learning, and storage device

    CN117054436A