Research On Shallow Sea Water Bathymetry Inversion Method Based On Worldview-3 Multispectral Image

The strategy of strengthening the country by the sea is one of the important strategies for the overall goal of China's development in the 21st century,and it is also an important political content of our country.If you want to build a maritime power,you need to have a deeper understanding of ocean information.Water depth information is an important ocean information.The water depth information can reflect the marine environment and plays a vital role in rational planning and management of marine island resources,ensuring maritime safety,and in-depth study of marine science.The method used to obtain more accurate water depth information is an important part of our research.In particular,with the introduction of the national global development strategy,the support of global geographic information is urgently needed,and higher requirements are imposed on the measurement of water depth in offshore ships and aircraft inaccessible areas.Compared with traditional measurement techniques,it is time-consuming and labor-intensive to use a remote sensing technology to perform a wide range of water depth measurements in a short period of time.In this paper,the combination of Worldview3 multi-spectral high-resolution data and measured water depth data for shallow water bathymetry inversion is mainly divided into three major aspects:data preprocessing,establishment of water bathymetry inversion model,verification of water depth inversion model and comparison and analysis of water bathymetry inversion models' accuracy.The contents of data preprocessing mainly include radiometric calibration,atmospheric correction,tidal correction,water and land separation,data registration,data grouping,pixel value extraction and correlation analysis.The water bathymetry inversion models established by the research include support vector regression model,BP neural network model and statistical regression model(including single factor model and multi-factor model).The single factor water depth inversion model is established by using four commonly used regression equations:linear function,exponential function,logarithmic function and quadratic polynomial..The multi-factor model is established by stepwise regression method.The model is validated and the accuracy of the model is evaluated using the decision coefficient,standard error and mean absolute error.The study area is a shallow sea area near the North Island of the South China Sea.There are more than 8,000 measured water depth points in the study area.Considering whether the number of sample points used for modeling is higher,the inversion accuracy is higher.Two sets of data of different sample points are selected for model establishment,and the two sets of data are combined as the third group data for model establishment.For each set of data,the support vector regression model,BP neural network model and statistical regression model(including single factor model and multi-factor model)were used to perform water bathymetry inversion and the model was verified and the precision was compared.The inversion accuracy of each model between the data was compared and analyzed.The determination coefficients of support vector regression model,BP neural network model,single factor model and multi-factor model are all above 0.920,the standard errors are all below 1.380,the mean absolute errors are all below 1.000,and the deviation between the water bathymetry inversion data and the measured data is small.The water bathymetry inversion accuracy is high.For any set of data in the three sets of data,the support vector regression model has the highest accuracy.The same model is applied to the three groups of data respectively,by analyzing the variance between the inversion values and combining the corresponding standard error and the average absolute error,it is found that the accuracy of the water bathymetry inversion of the three groups of data is not much different and when the water depth is inverted.