Shallow Water Bathymetry Retrieving Of Optical Remote Sensing Combined With SVM Bottom Classification

Shallow water bathymetry is an important part of marine surveying and mapping,which is of great significance to the development and utilization of the ocean.Traditional ship-borne survey is mainly affected by the weather condition and other factors.Due to the vast ocean area of China,it is difficult for survey ships to reach some sea areas.Remote sensing bathymetry has become one of the main methods for shallow water bathymetry with large range,flexibility and economy.The bottom of shallow water is often complex and diverse,and different types of sea bottom have different effects on the bathymetry retrieving.One of the methods to improve the accuracy of bathymetry retrieving is to reduce the adverse effect of complex bottoms on bathymetry retrieving by bottom classification.Stumpf ratio model and Lyzenga logarithmic model are widely used for remote sensing bathymetry retrieving,with the characteristics of simplicity and fewer model parameters.The main purposes of this study include:(1)to explore the influence of bottom classification on the Stumpf ratio model and the Lyzenga logarithmic model;(2)to research the feasibility of bottom classification based on the difference of retrieved water depth.In this paper,GeoEye-1 multispectral high spatial resolution remote sensing image and in-situ water depth data are used for bathymetry retrieving in the shallow water around Robert Island,Xisha Islands,South China Sea.Main works include data preprocessing,SVM bottom classification,bathymetry retrieving using Stumpf ratio model combined with bottom classification,bathymetry retrieving using Lyzenga logarithmic model combined with bottom classification,and bottom classification by OTSU method based on the difference of retrieved water depth.Main conclusions include:(1)On the premise of unclassified mixed bottom,the MSE and MAE of Stumpf ratio model is higher than that of Lyzenga logarithmic model.Stumpf ratio model can weaken the adverse influence of complex bottoms on bathymetry retrieving,so it is more suitable for large area and complex bottom waters.(2)When thinking about bottom classification,in the four sub-areas and the total research area,the accuracy of sub-area models is higher than that of the whole-area model without bottom classification.Bottom classification can improve the retrieving accuracy of Stumpf ratio model and Lyzenga logarithmic model in Robert Island,but its improvement effect on the Stumpf ratio model is less than that of the Lyzenga logarithmic model.(3)The bottom classification accuracy based on the difference of retrieved water depth is between 55%and 70%.The optimal classification result is the classification based on mixed-sand model difference after histogram stretching,with the overall classification accuracy of 67.7085%and the Kappa coefficient of 0.3568.For the same image,the classification result and classification accuracy of OTSU method are affected by how to stretch the image.