Research On Water Body Extraction And Water Quality Inversion Technology Based On Sentinel-2 Images

River Basin water environment is closely related to human life and social development,and its advantages and disadvantages are directly related to human survival and development.Since the 21 st century,with the increasing level of China’s economic development,people pay more attention to water environment protection.Water quality monitoring is the premise and foundation of water environment protection,and its demand is more and more urgent.However,the traditional water quality monitoring methods are expensive,time-consuming and limited in scope,which can not meet the needs of comprehensive monitoring of large-scale waters.In recent years,the spatial and spectral resolution of remote sensing earth observation has been greatly improved.In addition,machine learning theory provides technical support for remote sensing image interpretation in water quality remote sensing monitoring.Remote sensing monitoring has become an important means of large-scale water information dynamic monitoring.In this paper,Changsha section of Xiangjiang River is taken as the research area.Based on the analysis of the current research work of water body extraction and water quality inversion,the water body extraction method guided by prior knowledge of water body index and the water quality parameter inversion method combined with multi characteristic bands are proposed in this paper.The specific research contents are as follows:1)Aiming at the problem that the water body index model is difficult to effectively extract the water body information of complex regions and the extraction accuracy of the supervised method is too dependent on manual annotation,a method of water extraction guided by prior knowledge of water index based on Sentinel-2 images is proposed in this paper.Firstly,the characteristics of water body are extracted according to various water body index models.Then,the water features are fused to get the pre classification probability map.Finally,high confidence samples are selected as tags and input into U-shaped codec network for learning,and the water area of Changsha section of Xiangjiang River is extracted.The experimental results show that the accuracy of this method is better than other unsupervised extraction methods and traditional supervised extraction methods.2)Aiming at the problem that the existing water quality monitoring methods can not fully reflect the water environment of Changsha section of Xiangjiang River,a water quality parameter inversion method based on Sentinel-2 images is proposed in this paper.Firstly,a multi band combination model is constructed,and its correlation with water quality parameters is calculated.According to the correlation analysis results,a variety of characteristic bands are selected.Then,the feature bands with strong correlation are fused and input into a variety of machine learning regression models for training.Finally,the best regression model of water quality parameters is adaptively selected based on the accuracy evaluation index,and the inversion is carried out on Sentinel-2 image.The experimental results show that the method can effectively realize the large-scale water quality monitoring in Changsha section of Xiangjiang River.3)According to the actual application requirements of water quality remote sensing monitoring,the water quality remote sensing monitoring software of Changsha section of Xiangjiang River is designed and implemented in this paper.Based on Py Qt5.10 and QT Designer,the software front-end interface is designed in the environment of Python 3.6.Based on Python 3.6 and Matlab r2016 b,the functions of water extraction and water quality parameter inversion are realized.Through the application of sentinel-2 image,the function of the software is tested,and the test results show that the software can effectively realize the remote sensing monitoring of water quality.