CWCM-Net: A Novel Feature Fusion Method for Oil Spill Detection Using Single- and Quad-Polarization SAR Data

Abstract

This study proposes a feature fusion framework that jointly takes advantage of singlepolarization (SP) and quadpolarization (QP) synthetic aperture radar (SAR) imagery for reliable detection of marine oil spills and distinguishing it from lookalike phenomena. By blending complementary polarimetric properties and textural cues at the feature level, the framework markedly reduces false alarm rates. To facilitate fair, comparable assessments, the work introduces the quadpolarization oil spill detection (QPOSD) dataset, a fully annotated suite of QP SAR scenes, this also alleviates the scene selection bias that has long constrained QP investigations. For demonstration, the feature-fusion method is instantiated with a lightweight channelwise concatenated multilayer network and trained in a parallel way on the SP and QP data. The resulting model achieves an overall accuracy of 99.02% while substantially reducing false positives, underscoring the effectiveness of the proposed feature fusion approach and the practical value of the new QPOSD dataset for largescale, operational oil spill monitoring.

Keywords

• Deep learning (DL)
• environmental monitoring
• feature fusion
• image segmentation
• multipolarization analysis
• oil spill detection