Water quality parameter inversion method based on multi-source data
By combining data from drones and satellite images, and using the ratio correction method and Bayesian optimization algorithm to construct a water quality parameter inversion model, the problem of insufficient resolution of satellite remote sensing images was solved, and efficient water quality monitoring of urban river waters was achieved.
Patent Information
- Application Number
- CN202510754991.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-06
- Publication Date
- 2025-10-10
AI Technical Summary
The spatial resolution of satellite remote sensing images is low, making it difficult to effectively monitor water quality parameters in urban rivers.
Combining UAV images and satellite images, data normalization was performed through band matching and reflectivity correction ratio correction method, and the Pearson correlation coefficient method and Bayesian optimization algorithm were used to construct a water quality parameter inversion model, and the SVR, RF, GBDT, and XGBoost models were used for accuracy evaluation.
It achieves high spatial resolution and timely monitoring of water quality parameters, makes up for the shortage of ground observation points, and improves the accuracy and stability, adaptability and interpretability of the model.
Smart Images
Figure CN120761304A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of water quality remote sensing, and in particular relates to a water quality parameter inversion method based on multi-source data. Background Art
[0002] There are many methods for water quality monitoring, with traditional experimental testing and satellite remote sensing technology being the most commonly used. The development of satellite remote sensing technology has opened up new avenues for water monitoring research. Satellite remote sensing water quality monitoring technology offers the advantages of high dynamics, low cost, and strong temporal and spatial data availability, making it irreplaceable for studying water pollution in rivers and lakes. It can meet the needs of large-scale water quality monitoring and reveal the spatial and temporal distribution and variations of water quality, thus overcoming the shortcomings of surface sampling alone. It can also reveal the distribution of pollution sources, as well as the migration characteristics and impact ranges of pollutants, which are difficult to reveal using conventional methods, providing a basis for the scientific deployment of surface sampling points.
[0003] However, due to their low spatial resolution, most satellite remote sensing images can only identify large water bodies such as lakes and oceans, and there are many defects in monitoring urban river waters. Summary of the Invention
[0004] The present invention provides a water quality parameter inversion method based on multi-source data to solve the problem that satellite remote sensing images have low spatial resolution and are difficult to use for urban river water monitoring.
[0005] In order to solve the above technical problems, the technical solution adopted by the present invention is: A water quality parameter inversion method based on multi-source data includes the following steps: Step 1: Perform band matching on UAV images and satellite images, verify the feasibility of reflectivity correction, and normalize the reflectivity of satellite images based on the ratio correction method; Step 2: Use the Pearson correlation coefficient method to select model input features, use the Bayesian optimization algorithm to optimize the model hyperparameters, and construct a water quality parameter inversion model based on satellite images and drone images before reflectivity normalization. Furthermore, in step 1, when performing band matching on drone images and satellite images: If using drone multispectral imagery, directly select the band data that is consistent between the Sentinel-2 imagery and the drone imagery; If drone hyperspectral imagery is used, the average wavelength method is used to fit the narrow bands of the hyperspectral data into the wide bands of multispectral data. That is, the average reflectance of the hyperspectral data in the wavelength range corresponding to each band of the Sentinel-2 image is selected as the reflectance after fitting the corresponding band.
[0006] Furthermore, the calculation formula for reflectivity is ,in, The wavelength of the multispectral image is The reflectivity of the band, is the reflectance of the hyperspectral image at band i.
[0007] Furthermore, in step 1, the satellite image reflectivity correction specifically includes: Calculate the ratio of the reflectance of the green band of the Sentinel-2 image to the reflectance of the green band of the UAV image; The average value of all ratios is used as the reflectance normalization coefficient of the green light band, and similar processing is performed on other bands to obtain the reflectance normalization coefficients of the corresponding bands; The reflectance of each band of the Sentinel-2 image is divided by the corresponding reflectance normalization coefficient to obtain the normalized Sentinel-2 image.
[0008] Furthermore, in step 2, the Bayesian algorithm is used to optimize the water quality parameter inversion model, specifically including: Correlation analysis was performed by selecting the Pearson correlation coefficient in statistics; The remote sensing reflectance values of a single band or a band combination are used as independent variables, and the measured values of water quality parameters are used as dependent variables for correlation calculation. The band or band combination with the larger correlation coefficient is selected as the optimal inversion band or band combination to construct the water quality parameter inversion model.
[0009] Furthermore, in step 2, before constructing the inversion model, suitable sensitive feature bands or band combinations are first screened out as input data of the model to improve the model accuracy.
[0010] Furthermore, in step 2, the Bayesian optimization algorithm is used to optimize the model hyperparameters. The specific process includes the following steps: S21: Random Initialization point ,in is the hyperparameter space of the model; S22: Get the corresponding function value , the initial point set ; make , , loop steps 3 to 5 until the maximum number N is reached; S23: Based on the currently obtained point set , build a proxy model ; S24: Proxy-based model , maximize the acquisition function , get the next evaluation point:
[0011] S25: Obtaining Assessment Points The function value of , add it to the current evaluation point set:
[0012] S26: Output the optimal candidate evaluation point: , which is the optimal hyperparameter set of the model.
[0013] Furthermore, in step 2, a water quality parameter inversion model is constructed based on the satellite images and UAV images before reflectivity normalization, specifically including: based on the Bayesian optimization algorithm, based on the Sentinel-2 images and UAV images before reflectivity normalization, the SVR, RF, GBDT, and XGBoost models are used to construct the inversion models of water quality parameters, and the three indicators of R2, RMSE, and MRE are used to evaluate the model accuracy. The inversion results of the four models are analyzed, and the inversion models with the highest accuracy for Sentinel-2 images and UAV images are screened out respectively. Then, the reflectivity normalization is verified based on the constructed optimal inversion model.
[0014] Furthermore, after step 2, the reflectivity normalization is verified based on the constructed optimal inversion model.
[0015] Furthermore, the verification of reflectivity normalization includes the following steps: (a) Directly apply the optimal inversion model based on Sentinel-2 images to Sentinel-2 images; (b) Application of the optimal inversion model based on UAV imagery to Sentinel-2 imagery before reflectivity normalization. (c) Applying the optimal inversion model based on UAV imagery to Sentinel-2 imagery after reflectivity normalization; (d) Compare the results obtained in steps (b) and (c).
[0016] The present invention can achieve the following beneficial effects: 1. The present invention makes full use of the wide coverage of satellite image data and the high resolution of drone image data to construct a remote sensing inversion model. The water quality parameter concentration data of the study area are obtained through satellite and drone images, which are used as input data for the water quality inversion model, making up for the defects of insufficient and uneven distribution of ground observation points. Compared with the limitations of ground monitoring points, satellite image data can comprehensively reflect the conditions of large-scale water bodies and river basins, and cover a wide area on the same time scale. UAV image data has the advantages of high spatial and temporal resolution, small differences in the obtained water body information and high accuracy.
[0017] 2. Through the ratio correction method, satellite image data and UAV image data are combined to give full play to the advantages of the two image data. While ensuring that the research results can meet the needs of water quality inversion in large areas, they also have timeliness and high spatial resolution.
[0018] 3. In the construction of the inversion model, a variety of machine learning algorithms were introduced and horizontal comparative analysis was conducted to explore the accuracy differences of machine learning models with different structures when inverting different water quality parameters. A Bayesian optimization algorithm optimization modeling method was proposed to optimize the machine learning model, improve the model accuracy and stability, and effectively solve the problem of local optimality. It has the advantages of high efficiency, adaptability, and easy interpretability. BRIEF DESCRIPTION OF THE DRAWINGS
[0019] The present invention will be further described below with reference to the accompanying drawings and embodiments: Figure 1 This is a flow chart of the water quality parameter inversion method based on multi-source data of the present invention. DETAILED DESCRIPTION
[0020] To facilitate understanding of the present application, the present application will be described more fully below with reference to the accompanying drawings. The accompanying drawings provide embodiments of the present application. However, the present application may be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to make the disclosure of the present application more thorough and comprehensive.
[0021] like Figure 1 As shown in FIG, a water quality parameter inversion method based on multi-source data includes the following steps: Step 1: Perform band matching on UAV images and satellite images, verify the feasibility of reflectivity correction, and normalize the reflectivity of satellite images based on the ratio correction method.
[0022] Image band matching specifically includes: Sentinel-2 is a high-resolution multispectral imaging satellite that carries a multispectral imager for land monitoring. It can provide information on vegetation, soil and water cover, inland waterways, and coastal areas. Since Sentinel-2 images have 13 bands of data, UAV multispectral images contain four bands of data: green, red, red edge, and near-infrared, and UAV hyperspectral images contain thousands of bands. Considering the consistency between Sentinel-2 images and UAV images and the need to subsequently build a water quality inversion model, the satellite images and UAV images are first band-matched. If UAV multispectral images are used, the band data that is consistent between Sentinel-2 images and UAV images is directly selected, as shown in Table 1: Table 1: Sentinel-2 and UAV band matching selection
[0023] If drone hyperspectral images are used, the average wavelength method is used to fit the narrow bands of the hyperspectral data into the wide bands of the multispectral data. That is, the average reflectance of the hyperspectral data in the wavelength range corresponding to each band of the Sentinel-2 image is selected as the reflectance after fitting the corresponding band. The calculation formula is: , where The wavelength of the multispectral image is The reflectivity of the band, is the reflectance of the hyperspectral image at band i.
[0024] The satellite image reflectivity correction specifically includes: To determine the feasibility of correcting the reflectivity of Sentinel-2 imagery based on UAV imagery, we calculated the average reflectivity for each band across all sampling points in both the UAV and Sentinel-2 imagery and compared the trends in the average reflectivity for each band. Furthermore, we generated a scatter plot of the average reflectivity for the corresponding bands in the UAV and Sentinel-2 imagery to demonstrate the strong correlation between the reflectivity of the two imagery types.
[0025] To normalize the reflectance of Sentinel-2 images, a ratio correction method is used. First, the ratio of the reflectance of the green band of the Sentinel-2 image to the reflectance of the green band of the UAV image is calculated, such as B3 (green). The average of all ratios is then used as the reflectance normalization coefficient for the green band. Similar processing is performed for the other bands to obtain the reflectance normalization coefficients for the corresponding bands. Finally, the reflectance of each band of the Sentinel-2 image is divided by the corresponding reflectance normalization coefficient to obtain the normalized Sentinel-2 image.
[0026] Step 2: The Pearson correlation coefficient method is used to select model input features, the Bayesian optimization algorithm is used to optimize the model hyperparameters, and a water quality parameter inversion model is constructed based on satellite images and drone images before reflectivity normalization.
[0027] Among them, the Bayesian algorithm is used to optimize the water quality parameter inversion model, including: Before building the inversion model, we first screened suitable sensitive characteristic bands or band combinations as input data to ensure model accuracy. Correlation analysis was performed using the Pearson correlation coefficient, a statistical technique used in statistics. Correlations were calculated using the remote sensing reflectance values of a single band or band combination as the independent variable and the measured water quality parameter values as the dependent variable. The band or band combination with the largest correlation coefficient was selected as the optimal inversion band or band combination for building the water quality parameter inversion model.
[0028] The Bayesian optimization algorithm is used to optimize the model parameters. The specific process includes the following steps: S21: Random Initialization point ,in is the hyperparameter space of the model; S22: Get the corresponding function value , the initial point set ; make , , loop steps 3 to 5 until the maximum number N is reached; S23: Based on the currently obtained point set , build a proxy model ; S24: Proxy-based model , maximize the acquisition function , get the next evaluation point:
[0029] S25: Obtaining Assessment Points The function value of , add it to the current evaluation point set:
[0030] S26: Output the optimal candidate evaluation point: , which is the optimal hyperparameter set of the model.
[0031] Based on the Bayesian optimization algorithm, water quality parameter inversion models were constructed using Sentinel-2 and UAV imagery before reflectivity normalization using the SVR, RF, GBDT, and XGBoost models. Model accuracy was evaluated using R², RMSE, and MRE metrics. The inversion results of the four models were analyzed, and the highest-accuracy inversion models for Sentinel-2 and UAV imagery were selected. Reflectivity normalization was then verified using the optimal inversion model.
[0032] Verifying reflectance normalization involves the following steps: (a) Directly apply the optimal inversion model based on Sentinel-2 images to Sentinel-2 images; (b) Application of the optimal inversion model based on UAV imagery to Sentinel-2 imagery before reflectivity normalization. (c) Applying the optimal inversion model based on UAV imagery to Sentinel-2 imagery after reflectivity normalization; (d) Compare the results obtained in steps (b) and (c).
[0033] The above comparison shows that the accuracy of the model based on satellite imagery is lower than that of the model based on drone imagery. Furthermore, the inversion accuracy of the model based on drone imagery, when applied directly to satellite imagery without reflectivity normalization, is lower. Therefore, it is feasible to apply the optimal inversion model based on drone imagery to Sentinel-2 imagery after reflectivity normalization, and this can yield the best results.
[0034] After applying reflectance normalization, the study area was zoomed in and inverted based on the optimal inversion model and Sentinel-2 imagery to obtain the spatial distribution map of mass concentration of water quality parameters.
[0035] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application should be included in the scope of protection of the present application.
Claims
1. A water quality parameter inversion method based on multi-source data, characterized in that: The following steps are involved: Step 1: Perform band matching on UAV images and satellite images, verify the feasibility of reflectivity correction, and normalize the reflectivity of satellite images based on the ratio correction method; Step 2: Use the Pearson correlation coefficient method to select model input features, use the Bayesian optimization algorithm to optimize the model hyperparameters, and construct a water quality parameter inversion model based on satellite images and drone images before reflectivity normalization.
2. The water quality parameter inversion method based on multi-source data according to claim 1, characterized in that: In step 1, when performing band matching between drone imagery and satellite imagery: If using drone multispectral imagery, directly select the band data that is consistent between the Sentinel-2 imagery and the drone imagery; If drone hyperspectral imagery is used, the average wavelength method is used to fit the narrow bands of the hyperspectral data into the wide bands of multispectral data. That is, the average reflectance of the hyperspectral data in the wavelength range corresponding to each band of the Sentinel-2 image is selected as the reflectance after fitting the corresponding band.
3. The water quality parameter inversion method based on multi-source data according to claim 2, characterized in that: The formula for calculating reflectivity is ,in, The wavelength of the multispectral image is The reflectivity of the band, is the reflectance of the hyperspectral image at band i.
4. The water quality parameter inversion method based on multi-source data according to claim 1, characterized in that: In step 1, satellite image reflectivity correction specifically includes: Calculate the ratio of the reflectance of the green band of the Sentinel-2 image to the reflectance of the green band of the UAV image; The average value of all ratios is used as the reflectance normalization coefficient of the green light band, and similar processing is performed on other bands to obtain the reflectance normalization coefficients of the corresponding bands; The reflectance of each band of the Sentinel-2 image is divided by the corresponding reflectance normalization coefficient to obtain the normalized Sentinel-2 image.
5. The water quality parameter inversion method based on multi-source data according to claim 1, characterized in that: In step 2, the Bayesian algorithm is used to optimize the water quality parameter inversion model, specifically including: Correlation analysis was performed by selecting the Pearson correlation coefficient in statistics; The remote sensing reflectance values of a single band or a band combination are used as independent variables, and the measured values of water quality parameters are used as dependent variables for correlation calculation. The band or band combination with the larger correlation coefficient is selected as the optimal inversion band or band combination to construct the water quality parameter inversion model.
6. The water quality parameter inversion method based on multi-source data according to claim 5, characterized in that: In step 2, before building the inversion model, first select suitable sensitive feature bands or band combinations as the input data of the model to improve the model accuracy.
7. The water quality parameter inversion method based on multi-source data according to claim 1, characterized in that: In step 2, the Bayesian optimization algorithm is used to optimize the model hyperparameters. The specific process includes the following steps: S21: Random Initialization point ,in is the hyperparameter space of the model; S22: Get the corresponding function value , the initial point set ; make , , loop steps 3 to 5 until the maximum number N is reached; S23: Based on the currently obtained point set , build a proxy model ; S24: Proxy-based model , maximize the acquisition function , get the next evaluation point: S25: Obtaining Assessment Points The function value of , add it to the current evaluation point set: S26: Output the optimal candidate evaluation point: , which is the optimal hyperparameter set of the model.
8. The water quality parameter inversion method based on multi-source data according to claim 1, characterized in that: In step 2, a water quality parameter inversion model is constructed based on satellite images and UAV images before reflectivity normalization. Specifically, based on the Bayesian optimization algorithm, SVR, RF, GBDT, and XGBoost models are used to construct inversion models of water quality parameters based on Sentinel-2 images and UAV images before reflectivity normalization. The model accuracy is evaluated using the three indicators of R2, RMSE, and MRE. The inversion results of the four models are analyzed, and the inversion models with the highest accuracy for Sentinel-2 images and UAV images are selected respectively. Then, the reflectivity normalization is verified based on the constructed optimal inversion model.
9. The water quality parameter inversion method based on multi-source data according to claim 7, characterized in that: After step 2, the reflectivity normalization is verified based on the constructed optimal inversion model.
10. The water quality parameter inversion method based on multi-source data according to claim 9, characterized in that: Verifying reflectance normalization involves the following steps: (a) Directly apply the optimal inversion model based on Sentinel-2 images to Sentinel-2 images; (b) Application of the optimal inversion model based on UAV imagery to Sentinel-2 imagery before reflectivity normalization. (c) Applying the optimal inversion model based on UAV imagery to Sentinel-2 imagery after reflectivity normalization; (d) Compare the results obtained in steps (b) and (c).