Quantitative Monitoring And Analysis Of Surface Water Quality Based On UAV-borne Hyperspectral Imagery

Surface water resources represent the amount of fresh water that can be updated year by year in surface water,and are an important natural resource indispensable for the development of the national economy.With the development of society,population changes,and changes in natural conditions,in particular,pollutants emitted by human activities have caused surface water pollution.Water pollution treatment and prevention methods are important research directions for water environmental protection.However,the current monitoring of water quality still uses the traditional method of manual fixed-point sampling,which is not only time consuming and laborious,but also difficult to fully reflect the distribution of water quality conditions.As a new monitoring method,remote sensing technology makes up for the shortcomings of traditional water quality monitoring.With its high efficiency,real-time and wide observation range,it has played an important role in water quality monitoring of lakes,reservoirs and urban waters.With the successive launch of new sensors with small size,high detection accuracy and light weight,UAV-borne hyperspectral remote sensing technology brings new opportunities to quantitative remote sensing monitoring of surface water quality with its dual height characteristics.Based on the development prospects of UAV-borne hyperspectral remote sensing images for water environment monitoring and the ability of machine learning to predict automatically in regression modeling,in this paper,the DJI M600 Pro and a miniature hyperspectral resolution camera are selected as a monitoring platform to obtain hyperspectral image data,and the ground-based measured spectra are collected.Through two sets of experiments of quantitative inversion of suspended solids concentration(SSC)in Beigong Reservoir,Liuzhou City and quantitative inversion of nemerow comprehensive pollution index in Wuhan rivers,this paper expounds the validity and feasibility of quantitative monitoring of water quality under UAV-borne hyperspectral framework from two aspects:the inversion of single water quality parameter and comprehensive pollution index.Due to the complexity of fitting between the inversion target and the hyperspectral band information,this paper attempts to use the empirical or semi-empirical model and the more complex models such as particle swarm optimization-least squares support vector machine(PSO-LSSVM),gradient boosting decision tree(GBDT)and random forest(RF)for experiments.Then based on the trained model,a spatial distribution map of SSC in the entire reservoir and a spatial distribution map of urban river pollution index were produced.The paper analyzes the rationality of the spatial distribution of the inversion target from the natural environment around the study area and human social activities,and demonstrates the accuracy of the experimental results.The main research results of the paper are as follows:(1)The preprocessing method and process of the UAV-borne hyperspectral imagery are described.Because the initial data collected from the hyperspectral sensor storage module is a DN image,it must be subjected to a series of image preprocessing before it can be used for experiments.The preprocessing process includes sensor radiation calibration,geometric correction,site absolute radiation correction,water extraction and spectral extraction.Due to the low altitude of the drone,the radiation correction in flight can ignore complex atmospheric effects.The hyperspectral sensor Headwall Nano is a linear push-broom imaging sensor,so it is easy to shake during flight to cause serious distortion of the pushed-out image.The UAV integrates differential GPS technology and inertial measurement unit technology into a position and orientation system(POS)integrated with the sensor,which can provide the position and attitude parameters of the sensor to directly and quickly perform geographic positioning of the image.In addition,due to the extremely low remote sensing reflection ratio of the water body,the reflectivity of the standard board cannot meet the needs of the experiment.The radiation correction uses ground-based measured spectra to complete the linear relationship calibration of the radiance image.(2)The effects of various models on the inversion of the SSC in Beigong Reservoir were compared.Because the band ratio can eliminate the interference of background noise such as the surrounding environment that changes in different time and space,it is a contrast enhancement operation commonly used in quantitative inversion.In this paper,we try to use the exhaustive method to solve the band ratio and add a more relevant feature variable.Using the widely used particle swarm optimization algorithm to optimize the least squares support vector machine algorithm,compared with traditional exponential models,linear models,and currently widely used random forest algorithms,the estimation accuracy has improved significantly,and the test samples had a coefficient of determination(R~2)of 0.95,a root mean square error(RMSE)of 0.75 mg/L。The inversion results of the UAV-borne images are ideal.The distribution of the SSC is basically consistent with the actual situation.It shows the characteristics of high concentration on the shore and low concentration in the center of the reservoir.(3)In view of the cumulative error caused by the inversion of multiple single water quality parameters,this paper describes the process comprehensive pollution index to characterize the pollution levels of urban water body,and the process of fitting the pollution index based on the GBDT and other algorithms.Through the experiment,the accuracy of the comprehensive pollution index in the application of black-odor water identification and water pollution levels evaluation is verified.According to the spatial distribution of the overall pollution in the study area,Shahu Port is separated by the drainage outlet,showing the severe black-odor characteristic in the former part and non-black-odor characteristic in the latter part.As a whole,the Xunsi River belongs to severe black odor,and especially the flow velocity in the bend part slows down and the pollution degree is obviously increased.