Research On Water Depth Inversion Method Based On Remote Sensing Image

Ocean is a strategic resource for national development and an important indicator to measure national comprehensive strength.As the basic data,water depth data plays a vital role in the development of marine resources,marine shipping and other fields.Because the traditional bathymetric survey is mostly through ship borne sounding instruments,the cost is high,and it can’t well complete the survey of complex terrain sea areas or disputed sea areas,so it has certain limitations.In recent years,the research of water depth inversion using remote sensing images has attracted more and more attention.Remote sensing technology has the advantages of large area,real-time and fast,and there is a large amount of potential information between different bands in remote sensing images.The method of bathymetric inversion measurement through remote sensing images has been used.However,affected by the imaging quality of remote sensing sensors and water composition,the existing remote sensing water depth inversion methods generally have problems such as low accuracy and poor stability of the model in the process of water depth inversion.Therefore,it is necessary to study the above problems to improve the accuracy of water depth inversion and the stability of the model.The main research contents and conclusions of this paper are as follows:(1)Firstly,according to the propagation process of light in water and air,this paper analyzes the composition of light energy sources received by satellite sensors,and establishes single band model,double band ratio model and multi band model.The experimental results show that in the semi empirical and semi theoretical model,the multi band model can weaken the influence of underwater sediment on water depth inversion to a certain extent,and has higher model inversion accuracy than single band and dual band models.(2)The water depth inversion experiment is carried out by using the highresolution World View-2 and GF-6 remote sensing images,and the geometric registration is carried out for the geometric error on the gf-6 remote sensing image,so as to reduce the influence caused by the geometric error to a certain extent,and make the image value of the study area closer to the value of the real image.NDWI(Normalized Difference Water Index)index is introduced to separate water and land from remote sensing images,enhance water information,and extract water boundary information through NDWI.The algorithm is used to suppress the solar flare on the remote sensing image and eliminate the flare noise on the image to a certain extent.(3)The GBDT(Gradient Boosting Decision Tree)algorithm is creatively introduced to construct the GBDT water depth inversion model,optimize the model parameters,explore the importance and reasons of different characteristics of the input model,and compare the accuracy with the common machine learning model.The results show that when the eigenvalue m of GBDT model is 6 and the number of iterations n is 600,the inversion performance of the model is the best,and B2(Green Band)is more important than other bands in the inversion process.The inversion accuracy of random forest model and GBDT model,which are both integrated learning algorithms,is better than that of BP(Back Propagation)neural network model and SVM(Support Vector Machine)model.The RMSE and MRE values of GBDT model are0.83 M and 18% respectively in Qilianyu area and 1.36 M and 20% respectively in Nanshan Bay area.They are the best of the four models and have high model stability and inversion accuracy,It provides a more convenient and fast method for large-area water depth inversion.