| Ocean oil spill is one of the main causes related with today’s marine environment pollution, and the ocean ecosystem damaged by oil spill always need a long time to be recovered, therefore much attention are paid to such problem worldwide. When oil spill accident happens, how to monitor it’s scale and influence scope have a vital effect in preventing and controlling the accident deteriorating.Being a kind of active microwave imaging radar, Synthetic Aperture Radar (SAR) has several advantages such as its capability of obtaining data all the time, independence of weather condition and its high resolution. Thus it can detect the oil spill quickly and accurately as soon as the ocean oil spill accident happens. As a result, SAR has become one of the major means of remote sensing to monitor ocean oil spill at the present time.Based on information fusion theory as analysis method, target identification information fusion as analysis angle, this thesis discusses the ocean oil spill monitoring process based SAR transforming from visual interpretation to quantitative multi-source information fusion. Using the combination of feature level fusion and decision-making level fusion, realizes multi-source information fusion and improves the effect on SAR ocean oil spill detection. Based on the method of adaptive threshold segmentation,2685 oil or look-like sample targets are extracted in China Seas and surrounding waters according to the existing data sets, and then the wind field, oil platform, shipping lane and oil pipeline data of each sample are collected based on its geographical location information. This thesis analyzes the influence of environmental factors such as wind field on the discrimination between oil spill and the look-like based on Subjective Bayesian Method, and integrates environmental factors so as to preliminarily discriminate sample targets and verifies the validity of the method by the sample targets. On the other hand,66 features, which include geometric, gray and texture features are extracted from each sample target. The feature dimensions are reduced by using the principal component analysis(PCA) and finally BP neural network is used to identify the oil spill and the look-like. In the end, the decision-making level fusion of D-S evidence theory is used to fuse the results obtained from fusing environmental factors with Subjective Bayesian Method and the results obtained from fusing SAR features with BP neural network, and the practicability and validity of the method are verified by the sample targets. |
