Object-oriented SAR Oil Spill Detection Algorithm And System Implementation

Sea oil spill accidents take place frequently, which raises wide public concern.However, traditional monitering approaches can hardly meet the requirement of largecoverage and real-time surveilance. Space-borne Synthetic Aperture Radar (SAR) canwork all-weather and day and night, so it has been considered to be the mostappropriate sea oil spill monitering sensor."Lookalike" phenomena often appear in SAR images. How to differentiatelookalike from real oil spill is of great importance to increasing oil spill detectionaccuracy. In this dissertation, Envisat ASAR images, capturing the Gulf of Mexico oilspill accident in2010, have been used to carry out oil spill feature analysis.145oilspill and134lookalike samples are derived from the data set, and their geometric,physical and texture features are analysed. The analitical result indicates thatgeometric and physical features outperform texture ones in terms of discrimiatinglookalike from oil spill.SAR images are speckled, thus, object-basd (or object-oriented) imageprocessing methods tend to have better performance than pixel-based ones do. An oilspill SAR image segmentation algorithm based on Hierarchical AgglomerativeClustering (HAC) is proposed. This algorithm can not only maintain shape and edgeproperty of oil spill, but also has good segmentation accuracy for oil spills of differentscales, due to its multi-resolution feature.Combined with oil spill feature analysis, an oil spill detection algorithm usingFuzzy Logic (FL) is introduced. The algorithm takes advantage of fuzzy mathematics,and it can effectively discriminate lookalike from oil spills. Besideds, the algorithmcan yield the probability of the segment being an oil spill. Through oil spill detectionexperiments by3SAR images, this method can derive satisfactory result.A SAR oil spill detection software, integrating the algorithms proposed in thispaper, is estabilished. The software incorporates three moduls, which are:1) basic functionalities,2) oil spill detection,3) comprehensive analysis. By using EnvisatASAR images in2010, the operational performance of this software is tested.