Study Of Oil Spill Detection Technology In SAR Imagery

With the rapid development of economy and society, the marine petroleum exploitation and marine transportation become significantly prosperous. However, various types of oil spill accidents happen frequently, such as ran aground blowouts and explosions caused by inappropriate exploiting marine petroleum, and colliding and strike on rocks during the sailing of the ships or tankers. These accidents make massive oil spill into the sea, which causes severe damage and loss to the ecology, environment and economy of the accident spot and its neighbor countries. In addition, oil spill monitoring provides powerful decision information to maritime military reconnaissance and the crashed aircraft searching. Therefore, oil spill pollution monitoring receives significant attention in each country. Currently, one of the research hotspots in each country is how to effectively monitor the oil spill, timely obtain valuable ocean surface information for the succeeding works. Synthetic aperture radar(SAR), as an active microwave sensor, obtains the target information through emitting an electromagnetic wave to the target and receiving echo signals. Since it is independent of the sunlight and illumination, insensitive to adverse weathers like the rain, fog or night, and able to continuously obtain data and provide strong penetrability, it is becoming the mainstreaming technique of monitoring oil spill.SAR-based oil spill monitoringis the foundation of the further monitoring. The mainstreaming method of existing works is to remove noise of oil spill image through pretreatment, to split the seawater and oil film through segmentation in oil spill area, and obtain the oil film features through extraction and classification of the features of the oil spill area. However, The works on selecting filters, improving segmentation performance, designing classifiers remain in a early stage of exploration and development. In this paper, we firstly introduce the background and importance of the oil spill monitoring, and point out the necessity of developing satellite remote sensing technology. Secondly, we conclude the existing oil spill remote monitoring tools around the world. Next, considering the program of SAR image processing, we focus on three aspects: the pretreatment of oil spill SAR image monitoring, segmentation of oil film and background seawater, and feature extraction of oil spill category.Pretreatment is the basis for oil spill monitoring research. We firstly summarize the basic concepts of SAR, imaging principle, oil spill monitoring mechanism and main affecting the oil spill monitoring. Secondly, based on the theorem of radiometric correction and geometric correction of oil spill SAR image, we mainly analyze the features of a number of typical filters, which can efficiently remove the noise of the oil spill SAR images. Thirdly, in the experiment, we compare and analyze six filters with three filtering windows to quantificationally evaluate the filtering performance of oil spill SAR image.Segmentation is the key of SAR oil spill monitoring. Fristly, according to the analyzing of the image segmentation theory we mainly study data-driven threshold segmentation and model-driven MRF segmentation. Secondly, to solve the problem of low quality and high latency of traditional segmentation algorithm, we propose an improved 2D-Otsu algorithm for SAR sea oil spill. This algorithm can efficiently obtain the context information of the spill area accroding to the uniform backscatter of the oil spill area by injection of radar. Compared with the maximum entropy algorithm and the original 2D-Otsu algorithm, our algorithm provides better uniformity, higher contrast gradient and lower segmentation latency. Thirdly, to solve the problem of low accuracy of splitting the target and the background using a single model, we propose a new MRF segmentation method, which combines both models of Gamma distribution and Log-Normal distribution. Through 2? validating criterion, our algorithm splits the target and the background by MRF. The experimental result shows that the combined model has lower error rate of segmentation and better segmentation effects.The classification is the objective of the oil spill monitoring. Firstly, we analyze the common phenomenon of class oil films. Secondly, we carefully analyze the characteristics of oil spill geometry, gray feature, texture features, compare the correlation of various feature parametersand class oil films, which improves the efficiency of the classification, reduces the classification latency and refresh the feature parameters. Thirdly, most existing works of classification of oil spill are based on one single classifier. To this end, we propose a new comparing method which uses four classifers: K-neighbors, Bayes, BP neural network and SVM. We train the sampling datasets through four different classification methods, and test the feature values of the marked areas through these classification methods. According to the known results, we compare the accuracy of each classifiers, and lay a strong foundation to the future works.