Oil Spill Detection in PolSAR Imagery Using Multi-Scale Hybrid Feature Fusion Network

The issue of marine oil spills is receiving increasing attention, and polarimetric synthetic aperture radar (PolSAR) is an effective tool for detecting oil spills by distinguishing the backscatter differences between seawater and oil slicks. However, the inaccurate delineation of oil-water boundaries remains a challenge when using polarization features for oil spill detection. To address these challenges, we propose an oil spill detection framework based on Multi-Scale Hybrid Feature Fusion Network (MSHFFNet). This method combines polarization intensity information with the mixing pool module and feature fusion module to establish short-range and long-range connections between features of various sizes. This method enriches edge and semantic information by effectively utilizing contextual and detail information, significantly improving the recognition accuracy of oil-water boundaries. The experimental verification of Radarsat-2 fully polarized SAR data confirms the effectiveness of our method in improving boundary accuracy. Evaluation metrics including F1 and mIoU indicate that MSHFFNet surpasses two state-of-the-art methods and significantly improves the accuracy of oil spill detection.