Based On Least Squares Support Vector Machine Quantitative Remote Sensing Monitoring Of The Weihe River Water Quality Study

Water is the necessary resource for the survival and development of human society, and inland water bodies are our mains of freshwater. Currently, the water pollution problems is becoming more and more serious, especially in inland rivers, lakes and other water bodies, which restricts social and economic development. Therefore, we need to strengthen the control of water pollution, and now, it is the key to solve water pollution problems to obtain the status and changes in trend of water pollution in time, accurately and comprehensively. Conventional water quality monitoring methods, always includes such steps as artificial sampling, laboratory analysis and evaluation etc, which have long period, high cost and heavy workload of monitoring, and the monitoring results only represent the local situation of water pollution. With the development of remote sensing technology and the enhancement of remote sensing image resolution, remote sensing technology is applied to monitoring water quality in more and more researches, and its application gradually shift from the qualitative monitoring to more accurate quantitative monitoring of water quality. Due to such advantages of remote monitoring of water quality as high speed, low cost, wide range, and convenient long-term dynamic monitoring, etc, many experts and scholars from home and abroad, have applied remote sensing technology to monitoring the water quality of the oceans, lakes, rivers and other waters, and have achieved some research results. But now, there are few researches which use remote sensing technology to monitor the water quality of the Weihe River.Terefore, in this thesis, the SPOT5 remote sensing data is utilized for the quantitative remote sensing study of the Weihe River basin in Shaanxi province. In this thesis, the main research contributions are as follows.(1) Firstly, the background and research status of remote monitoring of water quality is introduced. And then based on the preliminary study of, the reason why the remote sensing images of SPOT5 is select as the remote sensing data source of this thesis and such four representative water quality parameters as permanganate index (CODmn), ammonia (NH3-N), dissolved oxygen (DO), chemical oxygen demand (COD) is applied to the inversion research is given. In addition, the purpose and methods of the remote sensing images preprocessing, including the radiometric calibration, atmospheric correction and geometric correction, etc. Since early researches show that there is a high correlation between the four-band remote sensing data and the four water quality parameters, which provides some theoretical basis for the present study. Therefore, the multiple linear regressions based on traditional statistical theory is used to model the water quality parameters in the thesis, achieving the conclusion that the inversion results are not precise enough, although the regression equation has passed the significance level test.(2) Due to the instability of water pollution and the impact of various factors in the remote sensing imaging process, there is a nonlinear relationship between the measured data and its corresponding spectral information of the water quality parameters and the sample points obtained are very limited. All these show that the study in the thesis is a nonlinear regression estimation problem with small samples. So the improved Least Squares Support Vector Regression(LSSVR) algorithm is introduced into the thesis, which is modified from the Support Vector Regression (SVR) algorithm based on the statistical learning theory. In the algorithm, the Genetic Algorithm (GA) is used to optimize the hyper parametersγand the kernel function parametersσ2 of the LSSVR, and three standard data sets are used in the experiments to test the feasibility of the model. For the LSSVR doesn't have the sparse nature, this pruning algorithm is introduced and the experimental results show that the pruning algorithm can effectively remove a part of the sample points with few contributions, so that the model is sparse, and can speed up the processing.(3) In the thesis, to estimate the prediction error of the model prediction error effectively in small samples, based on the 4 - fold cross validation method, the GA-LSSVR is utilized for modeling the four water quality parameters, and the inversion results are compared with the results of the multiple linear regression and the LSSVR,which show that the GA-LSSVR can achieve better retrieval accuracy and has generalization error, compared with other two models. In conclusion, all these show that the inversion model base on GA-LSSVR has stronger generalization ability.(4) The established GA-LSSVR models of the four water quality parameters are used to invert the overall Weihe River basin in Shaanxi province, and by comparing the inversion results with the actual pollution situation, we can conclude that the inversion results can reflect the river pollution in remote sensing imaging better, and the model can be applied to practical applications.