Active Contour Segmentation For Hyperspectral Oil Spill Remote Sensing

The oil spill accident will be serious pollution to the Marine environment, destroy the ecological health. Remote sensing technology is an important irreplaceable means of oil spill monitoring, hyperspectral remote sensing technology relied on own significant characteristics quickly become a highlight of oil spill detection yields. In the field of marine oil spill monitoring, the outline of oil spill area is the very important information, and the basis of the contaminated area estimate. However, with the continuous development of hyperspectral remote sensing technology and the improved of spatial resolution and spectral resolution, the traditional image segmentation algorithms can not make full use of its abundant information, cannot satisfy the oil spill area precise segmentation in hyperspectral image.With the introduction of the level set, active contour model has been an unprecedented development. Regional active contour model has free topology changes, is not sensitive to initial contour and is not dependent on the image gradient, can restrain the noise, is easy to expand application. The process of minimize the energy function is continuous approaching the boundary of the object contour curve, so you can always get the smooth and continuous boundary.The main content of this study is how to apply the hyperspectral images abundant spatial information and spectral information to the ACM model, drive curve evolution, and make integral effect more accurately. In this paper, expanding the active contour model applied to the actual hyperspectral oil spill image segmentation, mainly carried out the following two areas:(1) In order to take full advantage of spatial information for the ACM model in hyperspectral images, vector image gradient function is introduced to structures improvements length term of energy function, the outline of the curve is constrained by spatial gradient information in a process of evolution, stops at the boundary of the image. Experiment shows that the improved model of evolution can avoid curves excessively smooth, stops at more accurate boundary.(2) In order to make full use of spectral information for the ACM model in hyperspectral images, spectral angle distance measure replaces gray distance metric in original model. Since spectral angle is insensitive to light intensity, the improved data fitting item has some spectral recognition ability. Experiments show that the improved model can curb the influence of wave reflection and shadow, enhance the ability to distinguish between the oil spill and the sea.