Oil Spill Detection in X-Band Marine Radar Imagery Based on an Improved U-Net Model

Marine oil spill pollution can lead to severe ecological disasters and economic losses. Achieving efficient, reliable, and real-time detection of offshore oil spills is therefore a key challenge. Navigational radar, with its all-weather, around-the-clock capabilities, holds significant promise for monitoring oil spills at sea. However, to date it has been used primarily to determine whether a spill has occurred, and research on precisely delineating the spill area remains limited. To address this gap, this paper proposes an improved U-Net model that integrates a spatial attention mechanism with residual modules. During the image preprocessing stage, our approach effectively suppresses co-frequency interference and speckle noise; in the segmentation stage, detection is performed on a radar oil-spill dataset enlarged via data augmentation. Experimental results demonstrate that, compared with other existing segmentation models, our method substantially enhances the accuracy of oil-spill region delineation, providing real-time, effective data support for offshore oil-spill monitoring.