Oil spill detection: SAR multiscale segmentation and object features evaluation

The use of image segmentation and object feature extraction in order to classify SAR image objects into oil spill or other features (oil slick look-alikes), it is widely acceptable in oil spill detection research. For this purpose, a number of features (geometric, surrounding, backscattering, etc.) are usually calculated and introduced in a decision support procedure. The aim of the present study is presentation, analysis and evaluation of the above features in order to produce general rules adequate to identify oil spills in any SAR image. SAR image processing is based on a new multi-segmentation technique. As a first step, image objects in different scales are extracted using the multi-segmentation procedure. Following segmentation, a hierarchical network of image objects is developed, which simultaneously presents object information and fuzzy rules for classification. In experiments implemented in SAR images, the method developed has successfully detected oil spills and look alikes. Texture behavior most contributes to detection (texture characteristics 80%), followed by physical behavior (actual backscatter characteristics 53%, spot surroundings 26 %) and finally geometry behavior (geometrical characteristics 2%).