| Maritime oil spill accidents have taken place in many regions all over the world, which have caused huge environmental and economical losses. Airborne and spaceborne SAR sensors have both proved their potential in early warning and mapping of oil slicks since they have all day and all weather observation capability. However, the biggest challenge for single polarimetric SAR oil spill detection is to separate weak damping look-alikes (e.g. biogenic films) from mineral oil. In recent years, a lot of studies take advantage of polarimetric information in SAR images to improve the performance of oil spill classification. However, fully polarimetric (FP) SAR has only half the swath width for each channel compared with single polarimetric SAR systems, due to the doubled pulse repletion frequency (PRF). As the result, in several recent studies compact polarimetric (CP) SAR has been considered for maritime oil spill detection since it could maintain the swath width of single polarimetric SAR sensors while obtaining partial polarimetric information of the targets, which is especially beneficial for the task of large area marine oil spill surveillance.;This dissertation concentrates on the investigation of oil spill detection based on FP and CP SAR data. Firstly the basic principles of SAR oil spill detection and sea surface scattering models are briefly introduced. Then using UAVSAR L-band fully polarimetric SAR data, tilted Bragg scattering model is implemented to estimate the sea surface dielectric constant in the Deepwater Horizon (DWH) oil spill accident case.;Secondly, a variety of FP SAR features are derived and a framework of analyzing these features based on statistical distances and normalized intensity moments (NIM) is proposed. Based on Radarsat-2 fully polarimetric SAR data, statistical analysis is carried on the very special DWH case to analyze the capability of FP SAR features in distinguishing different damping status.;Thirdly, the two ways to derive features from CP SAR data: via pseudo quad-pol reconstruction and by extracting the features directly, are introduced in this thesis. An improved pseudo quad-pol reconstruction algorithm is proposed and tested on UAVSAR data. Moreover, based on UAVSAR data, features extracted by different methods from pi/2 (circular transmit and linear receive) CP SAR data are compared and analyzed, and they are also compared with those derived from FP SAR on their capability of distinguishing different damping status.;Finally, based on SIR-C C-band fully polarimetric SAR data, oil spill classification algorithms by using features extracted from FP and CP SAR data are investigated. For unsupervised classifier, K-means clustering is considered, and for supervised classifier, Support Vector Machine (SVM), Artificial Neural Network (ANN) and Maximum Likelihood Classifier (ML) are considered. During the study oil spill classification performance obtained by different features and classifiers are analyzed, and classification results based on pi/2 and pi/4 (45° transmit and linear receive) CP SAR modes are compared with those obtained by FP SAR data.;The proposed methods and experimental results obtained in this thesis could benefit the applications of SAR marine and coastal environment monitoring, and provide guidelines for the development of future spaceborne SAR earth observation methods. |
