Research And Realization Of Water Quality Parameter Retrieval Algorithm Based On High Resolution Remote Sensing Data

The traditional methods of water quality monitoring is time-consuming and inefficient, it is difficult to meet real-time, large-scale monitoring of lake.Focusing on these issue, this paper using remote sensing technology to predict the water quality indicators such as suspended matter, chlorophyll a and so on, so as to achieve the goal of water quality monitoring. This is of great significance to guarantee the normal life of people and promote the sustainable development of the economy.This paper chooses Longquan Lake in Chengdu as the research object, we collects water samples and hyper spectral data in Longquan Lake, analyzing the spectral characteristics of the water surface,and obtaining the best sensitive band combination of suspended matter and chlorophyll a, eventually establishing a quantitative inversion semi-empirical model of water quality parameters. The model algorithm uses the support vector regression algorithm (SVR) based on statistical learning theory. S VR has the advantages of small sample learning, strong pro motion ability, nonlinear fitting, it is well suited for inversion of water quality parameter concentrations.Parameters and kernel functions of SVR have a great influence on the accuracy of regression in SVR inversion model. This paper selects radial basis function as kernel function in SVR, using the grid search method and cross validation method to get the appropriate SVR parameters. Comparing with the traditional linear regression model,SVR inversion model achieves a better result in the application of water quality parameter inversion. The fitting degree, the mean square error and the relative error are improved in SVR inversion model. In order to further improve the performance of SVR inversion model, we uses genetic algorithm and particle swarm optimization to optimize SVR parameters,these algorithms can improve the accuracy and generalization ability of SVR. This study shows that, comparing with the SVR inversion model by using the grid search method and the cross validation method, the SVR inversion model optimized by GA or PSO has higher regression accuracy, in addition, the fitting degree and relative error of PSO-SVR model are better than GA-SVR model and SVR model.The relative error of PSO-SVR inversion model for suspended solids concentration reaches 8.580%, and the relative error of chlorophyll a inversion model is 12.9979%.Finally, this paper uses GF-1 high-resolution remote sensing image to inversion suspended matter and chlorophyll a concentration based on PSO-SVR model in Longquan lake.