Study Of The Hyperspectral Remote Sensing Of Shallow Waters Bathymetry With Artificial Neural Network Technology

It is important to develop the fast and accurate methods of bathymetry for economic development, military affairs and national defence. A method of hyperspectral remote sensing of shallow water bathymetry based on artificial neural network (ANN) technique is studied in this thesis.Based on the inherent optical properties of the relevant waters, hyperspectral remote sensing reflectctance (R_rs) simulation is carried out by semi-analytical numerical simulation technique of the radiative transfer equation. One three-layer artificial neural network is developed to derive depth of shallow waters based on the simulated hyperspectral remote sensing reflectctance data. A method for distinguishing optical shallow waters and optical deep waters is proposed based the inherent optical properties of the waters and the bottom features to guarantee the validity of the simulated data. In the process of modeling ANN, in order to accelerate the back-propagation learning process, the inputs of the ANN are processed in advance by principal component analysis . Compared with the result of the optimization analysis model, the ANN shows good performance in generalization.Besides, a preliminary study of the bottom classifying of shallow water bathymetry by spectral derivative based on simulation data is carried out in this thesis.