| Bathymetry is an important topographical element.The measurement of bathymetry helps to better understand the topography of shallow seas,and also plays an important role in the development of economic and military activities in the coastal waters.Bathymetry remote sensing inversion has become an important method for measuring bathymetry because of its large coverage area,fast update and low cost.It is also a supplement and improvement to traditional bathymetry measurement methods and technologies.The principle of remote sensing inversion of bathymetry is to use remote sensing image data to invert the bathymetry according to the parameters that can be measured and have a large correlation with thebathymetry.Bathymetry detection based on remote sensing technology is a supplement to traditional sounding methods,but the accuracy is relatively low due to the influence of water quality and remote sensing images.This paper selects Ganquan Island and Qilianyu in the Paracel Islands in the South my country Sea as the research area,and uses GF-1,GF-2,GF-6 and World View-2 as remote sensing data and point cloud data measured by lidar As the measured bathymetry,the study of bathymetry inversion is carried out.According to the principle of light propagation in water,the basic principle of retrieving water depth through remote sensing and the source of total radiance received by the sensor are introduced.Based on this,the theoretical analytical model,semi-empirical semi-theoretical model and statistical model are expounded.It mainly introduces machine learning BP neural network,random forest and extreme learning machine,and establishes three machine learning depth inversion models for the study area based on the evaluation of remote sensing image quality.The main research contents and conclusions of this paper are as follows:1)According to the basic principles of remote sensing inversion of water depth,starting from the radiation transmission of light in water,the formulas representing the models in various models are deduced.2)2)The remote sensing image data is preprocessed,and three methods of raster resampling are compared for geometric correction.The results show that the value of the image processed by bilinear interpolation in the study area is closer to the original image.3)The image quality of the domestic satellites GF-1,GF-2,and GF-6 and the foreign satellite World View-2 were evaluated,using four objective indicators for evaluation and analysis.As a result,the blue-green bands of GF-2 and World View-2 are more independent,World View-2 suffers the least interference from noise,and the separability of various types of features on GF-2 images is higher.Four The information carrying capacity of satellites is basically the same.4)Three machine learning methods,BP neural network,random forest and extreme learning machine,are used to construct water depth inversion models on Ganquan Island and Qilian Island.After many experiments,the optimal parameters were found.5)According to the water depth results retrieved by the three models,three indicators of R2,RMSE and MAE are used to evaluate the error of the water depth simulation results and compare and analyze them.According to the results,the extreme learning machine has the highest accuracy.6)Comparing the inversion results of the four satellites,World View-2 has the highest accuracy,the accuracy of GF-1 and GF-2 are basically the same,and the accuracy of GF-6 is the lowest,so it is not suitable to use GF-6 satellite for this research.Inversion of the water depth of the area.This paper uses machine learning to invert the water depths of GF-1,GF-2 and GF-6 and World View-2 Ganquan Island and Qilianyu.The inversion accuracy of World View-2 is higher.Among the satellites,GF-2 has the highest accuracy and can replace foreign satellites to a certain extent.In the inversion model,the inversion effect of the extreme learning machine is the best.It provides a certain way of thinking for using remote sensing to retrieve water depth. |
PDF Full Text Request