Oil spill detection on sea surface with dual-polarimetric SAR imagery integrating polarization features and oil-seawater boundary information.

Oil Spill can cause a series of ecological disasters and irreversible damages to marine environment, making the timely detection of oil films on the sea surface critical. In this field, Polarimetric Synthetic Aperture Radar (PolSAR) is an effective tool to identify oil spills. However, two main limitations exist in current researches: on the one hand, they fail to take full advantage of the polarimetric features in SAR image; on the other hand, they often neglect the clear boundary information between oil spill and seawater. These result in unsatisfactory recognition accuracy and the inability to effectively differentiate the oil slicks from the "look-alikes". Starting from the above viewpoints, we design a deep learning-based oil spill detection method for dual-polarimetric SAR images using Polarization features and oil-seawater Boundary Information combined TransU-Net (PBITU-Net) model. To make full use of the polarization information, an oil spill index is proposed which combines polarimetric decomposition parameters, and is taken as an important component of the model's input. To utilize the oil-seawater boundary information, an improved superpixel segmentation method is developed, and the segmentation results are used to optimize the model's loss function during training phase. Experimental results indicate that the proposed model performs better than classical methods, obtaining satisfactory detection results with high precision; in addition, we deployed the PBITU-Net method to successfully monitor oil spill events occurred in the Yellow Sea off the coast of China. The proposed method shows its feasibility in effectively monitoring oil spill events and providing supports for decision making of oil spill recovery actions.