The issue of marine oil spills has been receiving increasing attention. Polarimetric synthetic aperture radar (SAR) is effective in detectin
The issue of marine oil spills has been receiving increasing attention. Polarimetric synthetic aperture radar (SAR) is effective in detecting oil spills by distinguishing backscatter differences between seawater and oil slicks. However, traditional polarimetric SAR features struggle to accurately distinguish between oil slicks and oil slick analogs generated by natural phenomena. In addition, using polarimetric features for oil spill detection faces challenges of poor performance in detecting elongated and small-scale oil spills. To address the interference from look-alikes, we introduce a composite polarimetric feature, termed composite polarimetric scattering entropy, which effectively differentiates between oil spills and look-alikes. To refine oil–water boundary segmentation, we propose an oil spill detection framework based on a multiscale hybrid feature fusion network. The proposed method combines composite polarimetric scattering entropy with polarimetric intensity information and employs both square and strip pooling in the hybrid pooling module to capture a larger receptive field. This approach effectively captures information that helps distinguish elongated oil slicks. In addition, in the feature fusion module, the method efficiently utilizes shallow features that contain detailed information, such as edges and textures, as well as deep features that contain semantic information. This significantly enhances the detection accuracy of small-scale oil slicks. Experiments conducted on Radarsat-2 data demonstrate that our method exhibits outstanding performance in reducing similar interferences and improving boundary accuracy, surpassing four state-of-the-art methods in terms of detection accuracy.
