Based On Support Vector Machines Wei River Water Quality Quantitative Remote Sensing Research

At home,some water areas such as Yangtse River,Pearl River,Xiang River,Taihu Lake, Shiyanghe River are monitored based on remote sensing by many experts and scholars.Few people do quantitative remote sensing research about Weihe River's water quality based on SPOT data though it is an important river.In the past,because of the limitations of image's space resolution,we can only monitor large-scale region.Now we can extract water quality information for small area by high resolution remote sensing imagery such as SPOT-5.Based on this,after selection and analysis of monitoring data,using SPOT-5 satellite remote sensing data,we attempt to do quantitative remote sensing research about water quality of Weihe River by a new machine learning method-Support Vector Machine(SVM).As a new machine learning method,there are still some shortcomings such as difficulty of SVM mode parameters selection,and penalty coefficient C,kernel function and the parameter of the kernel functionσ² can greatly affected the accuracy of the model.Due to absence of theoretical guidance,parameters selection through repeated tests needs people's experience,and needs higher time cost.Firstly,radial basis function is as the kernel function of SVM in the paper, then a self-adaptive optimization algorithm for the selection of the Support Vector Classification(SVC) and the Support Vector Regression(SVR) models parameters-C,σ² using float Genetic Algorithm(GA),is proposed.The models-SVC and SVR are experimented with monitoring water quality data of Wei River and SPOT-5 data to evaluate water quality,and to inverse the water quality parameters and water quality evaluation grade.The main works in this thesis are as follows:(1)The encoding mechanism of GA,fitness function establishing,genetic operations including the selection,crossover and mutation,and the genetic algorithm steps,are given,and the SVC and SVR, models parameters of which a self-adaptive optimization algorithm can select using float Genetic Algorithm(GA),are stated in the paper.(2) we obtain the SPOT-5 images and simultaneous monitoring data in situ.Firstly,the remote sensing data are preprocessed including sensor calibration,geometric correction and radiation correction.Four kinds of representative water quality parameters:Permanganate Index(CODmn), Chemical Oxygen Demand(COD),ammonia(NH₃-N) and Dissolved Oxygen(DO) are selected. Then correlation analysis is done between the bands of remote sensing data,between the water quality parameters and between the bands of remote sensing data and water quality parameters. Principal Component Analysis(PCA) is done on remote sensing data bands for removing redundancy.Finally,univariate and multivariate linear regression models for water quality parameters and water quality evaluation grade are established based on the best correlation coefficients among the above-mentioned correlation analysis.And significant test of the regression models are all done.(3) The GA-SVC model is experimented with monitoring water quality data of Wei River,and is compared with water quality evaluation methods of single factor assessment,principal components analysis(PCA) and BP neural network.The results demonstrate that the proposed method can give a better quality comprehensive evaluation,and can reflect the water quality of rivers accurately and objectively from the overall.(4) The GA-SVR model is established for quantitative remote sensing research about Weihe River's water quality,and is compared with water quality inversion methods of traditional statistical regression method and BP neural network.The results show that water quality parameters can be predicted and water grade can be assessed quantificationally using SPOT-5 data according to the SVR models.What's more,at the same condition,the multivariate regression models are better than the univariate regression models.