A Study Of Oil Spill Detection In SAR Images By Level Set Method

In recent years, both Chinese maritime career and offshore oil business have obtained a rapid development. The development can bring economic benefits but at the same time also lead to the frequent happening of oil spill accidents, which have caused certain effect on and destruction to the ocean ecological environment and world economy. As a result, it is particularly important to timely detection, recognition and management of oil spills for the nature and the human. It is possible to monitor and identify oil spill with the rapid development of remote sensing technology, and Synthetic Aperture Radar(SAR) has been widely used in oil spill detection because of its all-day, all-weather capability, wide coverage and good real-time performance. When there are oil spills on the sea, oil film spread to sea surface produces a damping effect to gravity capillary waves of sea surface. It makes surface which spread by oil film smoother than the usual,then the backscattering radar echoes became smaller, so the oil spills display as the dark area in the SAR image.The thesis mainly studies on the SAR oil spill image segmentation. During the past few years, a variety of image segmentation methods have been proposed, and as a new method based on partial differential equation, the level set method possesses the advantage of free topology changes and has achieved a widely attention. In this thesis,we use the level set method to segment oil spill images. Experimental data are the oil spill images in the Gulf of Mexico in May, 2010. The images are provided by European Space Agency(ESA) online image database system. Noise and intensity inhomogeneity are common problems in these oil spill images, so we put forward a new multiplicative model to solve the problem of the multiplicative speckle noise and a bias filed to solve the problem of intensity inhomogeneity. As a result, we build a new model to modeling the oil spill images. The segmentation results of the proposed algorithm have great improvement compared with the classical active contour models, such as CV model,LBF model and so on. Besides, the proposed algorithm needn’t any prepossessing.