CN-122024087-A - Automatic water body identification and water quality parameter inversion method and system based on satellite images

CN122024087ACN 122024087 ACN122024087 ACN 122024087ACN-122024087-A

Abstract

The invention discloses an automatic water body identification and water quality parameter inversion method and system based on satellite images, comprising the steps of acquiring multispectral satellite images covering a target water body; the method comprises the steps of automatically extracting a water body range from a multispectral satellite image based on a water body index, extracting spectral characteristic parameters from the water body range, carrying out water quality parameter inversion calculation by utilizing the spectral characteristic parameters based on a pre-established water quality parameter inversion model trained by a machine learning algorithm to obtain water quality parameter spatial distribution data, and outputting the water quality parameter spatial distribution data. According to the invention, the spatial distribution data of water quality is obtained by automatically extracting the water body range and spectral characteristics in the satellite image and inverting by utilizing a machine learning model. The full-flow automation of water quality parameter inversion is realized, the water quality spatial distribution information can be rapidly and widely acquired, and the monitoring efficiency and the coverage capacity are greatly improved.

Inventors

ZHU FENG
XUE LIHUI
LIN FEI
CHEN PENG
Kou Shuqi
LI XIAOMENG
WANG YUNZHUO
LI BOHAO
JIN ZHOU
HU YIMIN

Assignees

中科合肥智慧农业谷有限责任公司
长丰县农业技术推广中心

Dates

Publication Date: 20260512
Application Date: 20260205

Claims (10)

1. An automatic water body identification and water quality parameter inversion method based on satellite images is characterized by comprising the following steps: acquiring a multispectral satellite image covering a target water body; Automatically extracting a water body range from the multispectral satellite image based on a water body index; Extracting spectral characteristic parameters from the water body range; Based on a pre-established water quality parameter inversion model trained by a machine learning algorithm, carrying out water quality parameter inversion calculation by utilizing the spectral characteristic parameters to obtain water quality parameter spatial distribution data; And outputting the spatial distribution data of the water quality parameters.
2. The method for automated water body identification and water quality parameter inversion based on satellite images of claim 1, wherein the step of acquiring multispectral satellite images covering the target water body comprises: Acquiring multispectral image data, and performing pretreatment of radiation correction and atmospheric correction on the image data; Wherein, the MSI comprises 13 spectrum wave bands which cover the wavelength range from 443nm to 2190nm from visible light to short wave infrared, and the spatial resolution comprises 10 meters, 20 meters and 60 meters; The image data used is L2A-level data subjected to atmospheric correction or L1C-level data subjected to atmospheric correction.
3. The method for automatically identifying water and inverting water quality parameters based on satellite images according to claim 1, wherein the step of automatically extracting the water range based on the water index comprises the following steps: Based on a programming language, calculating a normalized water body index NDWI of an image, generating a water body mask by adopting a self-adaptive threshold segmentation method, and further obtaining a vector file of a water body range, wherein the NDWI has a calculation formula as follows: In the formula, The green wave band corresponds to the B3 wave band of the satellite; the near infrared band corresponds to the B8 band of the satellite.
4. The method for automatically identifying water and inverting water quality parameters based on satellite images according to claim 1, wherein the spectral characteristic parameters are constructed based on the spectral reflectivities of actual measurement sampling points extracted from the preprocessed images, the B10 wave band of the satellite is eliminated during construction, the rest 12 spectral wave bands are used for constructing various types of spectral indexes including a difference index DI, a ratio index RI, a normalized index NDI, a value index SI and a product index PI in a double-wave-band combination mode.
5. The method for automated water identification and water quality parameter inversion based on satellite images according to claim 3, further comprising a spectral feature screening step after extracting spectral feature parameters from the water volume: screening spectral characteristic parameters which are obviously related to the target water quality parameters through Pearson correlation analysis and double-tail significance test; The Pearson correlation coefficient r is calculated through a formula, a significance p value is calculated through t test, and when p is less than 0.05, the significance p value is judged to be significantly correlated, specifically: Through satellite band B (b= { B k ∣1≤k≤8,k∈Z}∪{B 8a }∪{B k |9. Ltoreq.k. Ltoreq.12, k e Z }) 342 spectral index combinations are obtained: wherein k is the number of image wave bands, j is the number of wave bands participating in constructing the wave band index; The Pearson correlation coefficient between the water quality parameter and each spectrum index is calculated, and the calculation formula is as follows: In the formula, A sample value representing each of the spectral parameters, Represents the water quality parameter sample value, n is the sample size, 、 Representing the observed value of the i-th corresponding sample, And (3) with Respectively mean values of the corresponding samples; the data of (1-1, 1), Approximately close to 1, the stronger the correlation; judging the correlation coefficient through the double-tail saliency test Whether the spectrum is obvious or not is screened, so that the spectrum characteristics strongly related to the water quality target elements are screened, and the calculation formula is as follows: wherein n is the number of samples, r is the Pearson correlation coefficient, and t statistics obey t distribution with the degree of freedom of df=n-2; Wherein T is the T distribution with degrees of freedom n-2, The probability is the right tail probability of t distribution, and in general, p <0.05 indicates that the original assumption is refused, the correlation coefficient is considered to be obvious, and when p is more than or equal to 0.05, the original assumption cannot be refused, and the correlation coefficient is considered to be not obvious.
6. The method for automatically identifying water and inverting water quality parameters based on satellite images according to claim 1, wherein the water quality parameter inversion model is a machine learning model constructed based on CatBoost algorithm; the CatBoost algorithm adopts a symmetrical tree as a base learner in the training process, and applies an ordered lifting algorithm to process characteristics, and the tree depth, the learning rate and the regularization parameters are dynamically adjusted through a self-adaptive optimizing mechanism.
7. The method for automated water body identification and water quality parameter inversion based on satellite images according to claim 6, further comprising a model evaluation step of evaluating the accuracy of said model using a determination coefficient R2 and a root mean square error RMSE index and selecting an optimal model for water quality parameter inversion calculation, wherein the calculation formula is: In the formula, Representing the sample value of the water quality parameter, For model predictive element values, n is the sample size, Representing the average value of the water quality parameter samples.
8. The method for automated water body identification and water quality parameter inversion based on satellite images of claim 6, wherein the sample dataset is divided into a training set and a validation set in an 8:2 ratio when constructing the CatBoost model.
9. The method for automatically identifying water and inverting water quality parameters based on satellite images according to claim 1, wherein the output spatial distribution data of the water quality parameters is a grid-type water quality parameter distribution map.
10. An inversion system employing an automated satellite image-based water body identification and water quality parameter inversion method according to any one of claims 1 to 9, comprising: the image acquisition and processing module is used for acquiring multispectral satellite images covering the target water body and preprocessing the multispectral satellite images; the water body range extraction module is used for automatically extracting a water body range from the preprocessed image based on the water body index; the spectral feature extraction module is used for extracting and screening spectral feature parameters related to the target water quality parameters from the water body range; The water quality inversion model module is embedded with a water quality parameter inversion model which is obtained through training of a machine learning algorithm and is used for calculating according to input spectral characteristic parameters to obtain water quality parameter space distribution data; And the result output module is used for outputting spatial distribution data of the water quality parameters.

Description

Automatic water body identification and water quality parameter inversion method and system based on satellite images Technical Field The invention relates to the technical field of remote sensing monitoring, in particular to an automatic water body identification and water quality parameter inversion method and system based on satellite images. Background With global water resource pollution problems, water quality monitoring is becoming increasingly important. Traditional water quality monitoring methods rely primarily on-site sampling and laboratory analysis. While this approach can ensure high accuracy, it suffers from a number of drawbacks. First, this method is costly because it takes a lot of manpower and resources to sample and laboratory analysis. Secondly, because the sampling and analysis require time, the whole process period is long, and the water quality condition cannot be reflected in real time. Finally, this method has a limited monitoring range, since it is not possible to set a sampling point at every point that may be contaminated, which makes the water quality monitoring coverage of the area not wide enough. In recent years, remote sensing technology has shown great potential in the field of water quality monitoring. The sentinel No. 2 satellite is an important component in European space agency Goinbeir project, can perform multispectral imaging (comprising 13 wave bands), has spatial resolution of 10-60 meters, can cover a monitoring area once every 5 days, and provides an ideal satellite data source for water quality remote sensing monitoring. However, some technical bottlenecks exist in the current water quality inversion method based on the sentinel satellite No. 2. The prior art has the problems that the process is scattered, a plurality of links are needed to be manually interfered from the image acquisition to the final result output, and the processing efficiency is low. The method not only increases the workload, but also can delay data processing or inaccurate results due to human misoperation, and is difficult to meet the daily business monitoring requirement. In order to solve the problems, a technical method capable of realizing automatic water body identification, accurate inversion of water quality parameters and having automatic processing capability is urgently needed. Disclosure of Invention The invention aims to overcome the defects in the prior art, and aims to solve the problems in the background art by adopting an automatic water body identification and water quality parameter inversion method and system based on satellite images. An automatic water body identification and water quality parameter inversion method based on satellite images comprises the following steps: acquiring a multispectral satellite image covering a target water body; Automatically extracting a water body range from the multispectral satellite image based on a water body index; Extracting spectral characteristic parameters from the water body range; Based on a pre-established water quality parameter inversion model trained by a machine learning algorithm, carrying out water quality parameter inversion calculation by utilizing the spectral characteristic parameters to obtain water quality parameter spatial distribution data; And outputting the spatial distribution data of the water quality parameters. As a further proposal of the invention, the step of acquiring the multispectral satellite image covering the target water body comprises the following steps: Acquiring multispectral image data, and performing pretreatment of radiation correction and atmospheric correction on the image data; Wherein, the MSI comprises 13 spectrum wave bands which cover the wavelength range from 443nm to 2190nm from visible light to short wave infrared, and the spatial resolution comprises 10 meters, 20 meters and 60 meters; The image data used is L2A-level data subjected to atmospheric correction or L1C-level data subjected to atmospheric correction. As a further scheme of the invention, the method for automatically extracting the water body range based on the water body index comprises the following steps: Based on a programming language, calculating a normalized water body index NDWI of an image, generating a water body mask by adopting a self-adaptive threshold segmentation method, and further obtaining a vector file of a water body range, wherein the NDWI has a calculation formula as follows: In the formula, The green wave band corresponds to the B3 wave band of the satellite; the near infrared band corresponds to the B8 band of the satellite. The method is characterized in that the spectral characteristic parameters are constructed based on the spectral reflectivity of the actual measurement sampling points extracted from the preprocessed images, the B10 wave band of the satellite is eliminated during construction, and the rest 12 spectral wave bands are used for constructing various forms of spectral indexes