An efficient dark spot detection method for offshore oil spill in SAR images based on edge-enhanced attention fusion

Due to the significant environmental fluctuations at sea, oil spill dark spots in synthetic aperture radar (SAR) images are easily disturbed by environmental changes and noise, resulting in blurred edges and making accurate extraction challenging. Although existing deep learning algorithms can improve accuracy, they often consume large computational resources, making it difficult to balance efficiency and performance. To address this issue, this paper proposes an edge-enhanced attention fusion network (EAF-Net) for oil spill detection in SAR images, aiming to improve edge detection accuracy while reducing detection time. First, a residual bottleneck attention module (RBAM) is designed in the encoder to emphasize the edge details of the oil spill dark spots by weighting the spatial and channel attention of the input features. A lightweight structure is used to reduce computational cost. Second, an enhanced channel fusion module (ECFM) is proposed in the decoder to effectively fuse shallow features containing detail and edge information with deep features containing global information. Finally, a deep supervision refinement module (DSRM) is added to the decoder output to further enhance edge detail extraction. Experimental results show that EAF-Net achieves higher detection accuracy on the PALSAR and Sentinel-1 datasets while significantly reducing detection time, providing effective support for rapid and precise oil spill detection at sea.