Research On The Application Of UAV Hyperspectral Technology In The Identification Of Inferior Class Ⅴ Water Body In Small And Medium River

In order to further improve the quality of the water environment in Shanghai,the Shanghai Municipal Water Affairs Bureau proposed a three-year action plan to eliminate water bodies inferior to Grade V in 2018,striving to completely eliminate water bodies inferior to Grade V that have lost their use by the end of 2020.Whether the water quality monitoring is timely and comprehensive will directly affect the water body prevention and control plan of the entire region.For the monitoring of many small and medium-sized rivers in the suburbs,the technical methods of manual monitoring and satellite remote sensing monitoring all have limitations such as low efficiency,poor timeliness,high cost,and low accuracy.Therefore,Shanghai urgently needs a new,fast and reliable water quality monitoring technology.This paper uses the characteristics of high resolution,easy operation,and flexible maneuverability of UAVs,and combines them with hyperspectral imaging technology.Starting from the application of UAV remote sensing technology,combined with the actual needs of water quality monitoring of small and medium rivers,it is Data acquisition,pre-processing,model establishment,and interpretation of the results of water body identification inferior to Class V application research.The main tasks completed are as follows:（1）The current status of water quality evaluation and monitoring is studied,and the development of UAVs in multiple fields and their applications and prospects in remote sensing water quality monitoring are introduced.The technical principle of water color remote sensing was explained,including the classification basis of five types of water bodies,the optical properties of water bodies and the principle of water color remote sensing radiation,and designed total phosphorus（TP）,total nitrogen（TN）,ammonia nitrogen（NH₃-N）,permanganate index(COD_Mn)indoor water color spectrum experiment of four water quality parameters.It verifies the feasibility of quantitative inversion of a single water quality parameter concentration based on empirical methods,and provides a basis for subsequent water quality parameter inversion.（2）The technical plan for acquiring UAV hyperspectral image,surface water spectrum data,and water quality sampling data was designed,and introduced the specific operation process of data acquisition in detail.The key processes such as radiometric calibration,geometric correction,reflectivity calculation,river extraction,image denoising of UAV hyperspectral data were processed,and 55 sets of ground-measured water spectrum data and water quality sampling data were analyzed and screened.（3）The characteristics of the reflection spectrum curve of the water body in the middle and small rivers were described.Through the removal of the spectral continuum,the spectral differentiation technique,and the correlation analysis,the characteristic bands of five water quality parameters such as total phosphorus,total nitrogen,ammonia nitrogen,permanganate index,and dissolved oxygen are selected.Used for subsequent modeling.（4）A quantitative inversion model for the concentration of five water quality parameters including the permanganate index was constructed,and multiple machine learning models such as simple linear regression model and random forest regression were analyzed and compared.The results show that the machine learning model is superior to the simple linear regression model and supports The vector regression model and the random forest regression model have higher inversion accuracy than other models.The test set and training set R² exceed 0.8;a qualitative identification model between the water spectrum curve and the water body category is established,and the result shows that the misclassification error is 6.0%,The missing score error is 3.3%,and the average recognition accuracy rate of various water bodies is as high as 79.3%,indicating that the recognition model combining principal component analysis（PCA）and random forest has high qualitative classification accuracy.The spatial distribution of the concentration of each water quality parameter and the spatial distribution of the water body categories in six medium and small rivers was showed.The main factor for the analysis of water bodies inferior to Category V is total nitrogen.Finally,five water pollution monitoring and prevention are put forward.There are 67 figures,38 tables and 89 reference in the thesis.