Research And Application Of Water Quality Inversion Method Based On UAV Hyperspectral Data

With the rapid development of society,a large amount of industrial effluent is discharged illegally to aggravate the water pollution.Water eutrophication leads to the imbalance of the water bottom environment,so the monitoring and management of water quality has become one of the hot issues that society needs to solve.The content of total nitrogen and total phosphorus in water bodies is closely related to the environmental balance of water bodies.And them is an important indicator used to measure the water quality.The original water quality monitoring means is mainly to collect water samples for local waters.Then water samples are sent to the chemical laboratory to analyze the content of each component.Through the means of monitoring each component is very accurate,but only for local water monitoring,if the overall water monitoring will require a lot of human and financial resources.The gradual maturity of hyperspectral remote sensing technology provides a good technical means for the monitoring of the whole water body,and can achieve a wide range of water quality control.The new means is through the collection of hyperspectral data of the water body,and synchronous field data sampling,through the analysis of the relationship between the concentration of water quality parameters and spectral characteristics,the establishment of water quality inversion model,to achieve the purpose of monitoring a wide range of waters.However,the complex spectral information of hyperspectral remote sensing data also poses a challenge for the construction of the model.Taking the Liaohe estuary as the research area,this thesis aims to improve the water quality prediction capability of the Liaohe estuary.Based on the hyperspectral data from UAV,the inversion models of total nitrogen and total phosphorus concentrations are studied in two directions: spectral feature selection and nonlinear inversion model construction,as follows:(1)For the characteristics of hyperspectral data band redundancy,the original spectral features need to be selected,so a band selection method based on LASSO is proposed.The water quality inversion models of total phosphorus and total nitrogen concentrations are constructed separately based on this method.The key of the method is the sparse solution of the matrix equation consisting of the concentration vector of the measured water quality parameters and the spectral feature information with the help of the sparse property of LASSO.Through the coordinate descent algorithm to solve the LASSO problem.The insignificant spectral features are compressed to zero to achieve the purpose of band selection.On the basis of feature band selection,the relationship between the feature variables and total nitrogen and total phosphorus concentration is further analyzed to construct a linear water quality inversion model.(2)Considering the limited improvement of the linear inversion model on the accuracy,this thesis proposes a nonlinear water quality inversion model construction method from the perspective of significant band dimensional enhancement to explore the role of dimensional enhancement for accuracy improvement.In this method,nonlinear band expansion and polynomial kernel function band expansion are used to enhance the dimension of the characteristic band set obtained by LASSO-based band selection method.Then the relationship between the new spectral characteristics and the measured water quality parameter concentration is analyzed,so as to construct the nonlinear inversion model.(3)To prove the effectiveness of the proposed method,the band selection method based on Pearson correlation coefficient,the band selection method based on signal matching degree and the band selection method based on underdetermined system of equations solution were used as the comparison methods to construct models for total phosphorus and total nitrogen concentrations.Analyze the inversion accuracy of each model,and verify the applicability of the LASSO-based band selection method.The nonlinear models were constructed on the basis of the dimensional enhancement.The inversion accuracy and model construction time are compared to illustrate the feasibility of the dimensional enhancement in enhancing the model prediction capability.