Research On Intelligent Extraction Of Water Bodies And Retrieval Methods Of Water Quality Parameter CDOM In Tiande Lake Based On Remote Sensing Images

With the rapid development of social economy,some surface water bodies have been seriously polluted,so it is particularly important to monitor the water range and water pollution accurately.Remote sensing has obvious advantages in water pollution monitoring and environmental protection due to its wide monitoring range and strong timeliness,so it has high application value to obtain water range and water quality based on remote sensing images.Therefore,centering on the problems of water range identification and concentration inversion of inland water quality parameters in water environment monitoring,this study carried out large-scale intelligent extraction of water based on OHS-1 hyperspectral images and CDOM(Colored Dissolved Organic Matter,Dissolved Organic Matter,Dissolved Organic Matter)based on multi-source remote sensing images in urban small scale water bodies.CDOM)concentration inversion.The former mainly builds multi-temporal spectral library of training samples based on existing land cover products to study how to quickly and accurately extract large-scale water boundary from remote sensing images.Taking Tiande Lake in Zhengzhou city as an example,the latter proposed corresponding independent variable selection strategies for different remote sensing images,aiming at studying the inversion method of CDOM concentration of inland water quality parameters.The main research contents and achievements are as follows:(1)Establish multi-phase spectral library of water body training samples.The coordinates of training samples were extracted based on the existing land cover products,and the multi-temporal spectral database of water body and other ground objects was established combining the OHS-1 hyperspectral image.The spectral database included 5051 water body samples and 5000 other ground objects samples.(2)Research on large-scale water intelligent extraction method based on multi-temporal spectral library.Based on the multi-temporal spectral library,the water body intelligent extraction model is established by using the differential evolution algorithm of self-defined initialization population to select band of OHS-1 hyperspectral image and optimize the hyperparameter in XGBoost(e Xtreme Gradient Boosting,XGBoost)classifier.It is compared with decision tree algorithm and support vector machine algorithm.The results show that the proposed intelligent water extraction model has the highest Accuracy,with Accuracy of 0.93 and Kappa coefficient of 0.85.The accuracy and Kappa coefficient of support vector machine model were 0.91 and 0.81respectively.The accuracy of decision tree model is 0.89 and the Kapp coefficient is0.77.The optimal model was selected for water extraction from OHS-1 hyperspectral image,and its water extraction accuracy was high.(3)CDOM concentration inversion of Tiande Lake water quality parameters based on multi-source remote sensing images.In the inversion study of CDOM concentration of Tiande Lake water quality parameters,different independent variable selection strategies are proposed according to different remote sensing images.For multispectral images,two strategies are selected:using correlation coefficients to select independent variables and using all bands of remote sensing images as independent variables.Support vector machine regression models were established for Sentinel-2A and Landsat 8 images respectively,and the differential evolution algorithm was used to optimize the hyperparameters in the models,and the test set data were selected to evaluate the models.The results show that the Mean Absolute Error(MAE)and Root Mean Square Error(RMSE)of the model based on Landsat 8 data decrease by 3.33%and 2.86%respectively.Based on sentinel-2A data,MAE and RMSE decreased by 8.33%and 3.44%respectively.For OHS-1 hyperspectral images,three strategies are selected:independent variables selected by correlation coefficient,independent variables selected by band selection based on differential evolution algorithm,and all bands as independent variables.The support vector machine regression model was established,the differential evolution algorithm was used to optimize the hyperparameters in the model,and the test set data was selected to evaluate the model.It is concluded that the strategy modeling accuracy of independent variable of band selection based on differential evolution algorithm is as follows:determination coefficient R~2 is 0.78,MAE is 0.1mg/L,RMSE is 0.16mg/L;The modeling accuracy of all bands as independent variables is as follows:determination coefficient R~2 is 0.46,MAE is0.2mg/L,RMSE is 0.25mg/L;The accuracy of strategy modeling using correlation coefficient to select independent variables is as follows:determination coefficient R~2 is0.36,MAE 0.26mg/L,RMSE 0.28mg/L.The results show that for multispectral images,the strategy modeling accuracy of all bands as independent variables is the highest,while for hyperspectral images,the strategy modeling accuracy of band selection based on differential evolution algorithm is the highest.(4)The spatial distribution map of CDOM concentration of Tiande Lake water quality parameter was made based on THE OHS-1 hyperspectral inversion model.The results show that the overall distribution trend of CDOM concentration in Tiande Lake of Zhengzhou is as follows:the concentration in the west is higher,the concentration in the east is lower,the concentration in the inlet is higher than that in the outlet,and the concentration in the northwest is the highest.