Depth Inversion Of Shallow Sea Based On GF-6 Multi-Spectral Remote Sensing Image

The strategy of maritime power is one of the important strategies for China’s overall development goals in the 21 st century,and is also an important political content of our country.In the report of the 20 th National Congress of the Communist Party of China,it was proposed to’accelerate the construction of a maritime power’,and the construction requires the support of a large number of accurate marine basic data,especially marine depth information,which can accurately reflect the seabed topography and marine environment,and plays a vital role in the rational planning and management of oceans,island resources,navigation safety and marine scientific research.Compared with the traditional measurement technology with high time and material cost,the application of remote sensing technology can be used to perform large-scale,safe and efficient water depth inversion in a relatively short time.With the progress and development of China’s marine satellite technology,the ability to observe water depth information useing satellite remote sensing images is also increasing.Based on the principle of water depth inversion using remote sensing,,this paper briefly describes the theoretical model,semi-empirical and semi-theoretical model and statistical model are expounded.Taking the sea area of Nanshan Port in Sanya and of Gouqi Island in Shengsi as the research objects,using the corresponding GF-6multi-spectral data and multi-beam measured data of the sea area,and on the basis of various preprocessing of remote sensing images,three classical semi-theoretical and semi-empirical models and four machine learning models were introduced to carry out water depth inversion experiments.Based on the accuracy evaluation of the inversion results,a comparative study was conducted with Sentinel-2,a mainstream foreign multispectral remote sensing data,to explore the ability and application prospects of shallow water depth inversion based on the high-resolution 6 multispectral image.The main research contents and conclusions of this paper are as follows:(1)The remote sensing image datas are preprocessed,including radiometric calibration,atmospheric correction,flare elimination,and land and water separation.The formula is used to remove the solar flares appearing in the image.In order to enhance the water body information,water body index and regional growth method are combined to separate water and land information from remote sensing images.(2)Four machine learning models(RF,GBDT,XGBoost,SVR)and three semi-theoretical and semi-empirical models(single-band model,dual-band ratio model,multi-band model)are used to construct water depth inversion models in the waters of Nanshan Port,Sanya,and of Gouqi Island,Zhejiang,and the inversion results are obtained.(3)Three semi-theoretical and semi-empirical models and four machine learning models are compared and analyzed by using multiple evaluation indicators(RMSE,ARE,R~2)between inversion water depth and measured water depth.The results showed that the green light band had the highest water penetration ability among the eight bands of GF-6 image,which is suitable for water depth inversion,while the near-infrared light band had the lowest inversion accuracy and is insensitive to water depth information.In the dual-band ratio model,the ratio model of blue light and green light has the highest inversion accuracy.In general,the accuracy of multi-band water depth inversion model is higher than that of single-band regression model and of dual-band ratio model,and the inversion effect obtained by the combination of blue light,green light and red light is the best.(4)Compared with the traditional water depth inversion models,the machine learning models have stronger advantages in the study of water depth inversion.The random forest model has significant effects on solving the nonlinear and highly complex problems between water depth data and radiance value in water depth inversion.The root mean square error of the inversion result is 1.49 m,and the average relative error is 20%,and R~2is 0.74,which is the best inversion model in this experiment.(5)The inversion effect of GF-6 remote sensing image in Nanshangang sea area is better than that of Gouqi Island sea area.Due to the high sediment content,turbid water body and frequent human activities,Gouqi Island sea area is not suitable for remote sensing water depth inversion research.(6)The results show that the water depth inversion effect of domestic GF-6multispectral data is consistent with that of Sentinel-2 data from foreign mainstream high-resolution satellites.It can effectively replace similar foreign satellite data and be applied to the extraction of shallow water depth below 20 m in the clear waters of the global offshore.With the increasing demand for shallow water surveying and mapping in China,it is of great scientific value and military significance to carry out research on shallow water depth inversion method using domestic satellite remote sensing data.The research methods and results of this paper provide references and ideas for the application of water depth inversion from domestic remote sensing data.