Mapping Shallow Water Bathymetry Using Hyperspectral Image And Sonar Data

As the frequent interaction between land and sea area, coastal and offshore continental shelf areas have the superior geographical position, many kinds of resource and intense human activity. From the view of resource development and utilization, the offshore continental shelf is the strategic development base of oil and gas resources, food resources, and maritime resources. For this purpose, the engineering services such as the harbor construction to coast protection, energy development, channel development, and cable laying, need different scale of the topography and bathymetry maps; from the view of military and national defense, aerial ability, continental ability, and sea ability have become three standards to measure the national modern power. The world’s major countries have actively to broaden the strategic space of the ocean as an important direction changing the timing of the world’s strategic situation. As an important part of the strategic guidance of the military action, mapping the bathymetry has received the national attention. Geographic and topographic mapping of the offshore seabed through the coverage of certain waters, the measurements to achieve a certain interval of the specific coordinates of the point of the sea deep to mean sea level as a benchmark to specific coordinates on the sea depth value as a measure to reflect the topography of the seabed message. In order to get accurate seabed topography to reflect the area of information, a comprehensive, accurate bathymetry is needed. Therefore the development of rapid, accurate bathymetry method has a great significance.When building the underwater bathymetry, ship-borne acoustic measurement can get high accuracy results, but the spatial distribution is uneven. For example, area sounding the routes course has a high sampling and measurement accuracy, but there is a big blank area between adjacent routes. In addition to the constraints of the environment, time, and funding, it is difficult to achieve normalization; the other hand, remote sensing methods get low accuracy, but can cover extensiveness, especially low-cost and short cycle, people can use the method to conduct dynamic monitoring easily. Based on these two points, this paper fusion Hyperion data and sonar data use semi-supervised manifold learning to reduce the dimension of Hyperion, then segmente the lower dimension images to get homogeneous terrain regions, finally interpolate the sonar data of each region by information diffusion, the research content are as follows:(1) Building the optimal depth of remote sensing space. Although hyperspectral remote sensing images provide detailed spectral information, at the same time, bring about the problems, such as "Dimension disaster" and "Huge phenomenon". It has become an important problem that how to get comprehensive and useful information which exactly related with water deep from complex hyperspectral data. By comparison to a variety of linear, nonlinear manifold learning dimension reduction methods, we select the semi-supervised Laplacian Eigenmaps method combine spatial distance metrics to build optimal depth inversion feature space.(2) Homogeneous terrain feature segmentation and bathymetry mapping. According to the optimal feature space of study area bathymetry, we use a multi-level segmentation method to get the homogeneous depth regions, and then interpolate the sonar data of each region by information diffusion.