Research On Ocean Oil Spill Detection And Recognition Using SAR Data

Accidental oil spills happen frequently from offshore oil platforms, pipe laying,ship collision or illegal discharge, which impacts seriously on marine environment,marine biology and marine economy. With the rapid development of the satellite,Synthetic Aperture Radar (SAR) has become one of the important and effective toolsin oil spill detection. This dissertation discussed the method of the sea surface oil spilldetection using single polarized and multi-polarized SAR data, respectively. The mainresearches focus on oil spill detection, feature extraction and distinguishing betweenoil spills and oil-alikes phenomena in SAR images. The main results and conclusionsof this dissertation are summarized as follows:Recognize ocean oil spill based on single polarized SAR data. Firstly, weextracted the16features in total, based on their characteristic geometry, gray leveland texture features in the SAR images. Then, we analyzed a fuzzy logic algorithm toseparate oil spills features from look-alikes.5features were selected again based onthe fuzzy theory as input parameters. Then, it went through three major processesknown as fuzzification, fuzzy inference, and defuzzification in the fuzzy logic model.The last result is to classify objects as oil spills or look-like based on the probability tobe oil spills.Recognize ocean oil spill based on multi-polarized SAR data. In thispaper, two multi-polarized methods were respectively used to recognize ocean oilspill in SIR-C/X SAR oil spill experiment: co-polarized phase difference used todistinguish oil spills from look-alike; Muller filtering used to distinguish oil spillsfrom look-alike. Both two methods are using co-polarized correlation degree.However, when we used the L-band fully polarimetric Uninhabited AerialVehicle-synthetic aperture radar (UAVSAR) data acquired during the2010DeepwaterHorizon oil spill disaster event in the Gulf of Mexico, we found that the Muller filter could not detect oil spill because of low contrast between oil spill and sea backgroundin L band SAR images. A new algorithm is presented based on the co-polarizedcorrelation coefficient(ρ), and the scattering matix decomposition parameters,Cloude entropy(H), mean scattering angle(α) and anisotropy(A). While each of theseparameters has oil spill signature in it, we find that combing these parameters into anew parameter, F, is more effective way for oil slick detection. Another algorithmusing the total power of four polarimetric channels image (SPAN) is also to find theoptimal representation of the oil spill signature. We are able to detect oil spill in theGulf of Mexico event by F and SPAN image, but the combined polarimetricparameter (F) image could supply four dimension polarization feature classificationspace for distinguishing oil spills from look-alike.Using the NASA program, L-band fully polarimetric Uninhabited Aerial Vehicle-Synthetic Aperture Radar (UAVSAR) data acquired during the2010DeepwaterHorizon oil spill disaster event in the Gulf of Mexico, classify oil spills in BaratariaBay. Based on Cloude-Pottier(CP) decomposition and classification, we introducedaveraged intensity (I) and changed H/α/A into I/α/H unsupervised classification, anddeveloped Wishart supervised classification to map oil spills, sea surface and land inocean bay area. This method is able to provide simple and effective remote sensingmap and technology support for oil spill evolution and coastline respiration.