Co-YOLOv7: An Efficient Oil Spill Identification Network Based on SAR Images

Oil spills cause significant damage to marine ecosystems, resulting in substantial ecological and economic losses. Synthetic aperture radar (SAR) is widely utilized in oil spill detection tasks due to its all-day and independence from conditions such as clouds and fog. However, SAR images contain a considerable amount of speckle noise, and oil spills are influenced by environmental factors in the marine context, exhibiting a highly irregular shape. So, the traditional deep learning network feature extraction efficiency is very low. Addressing the issues above, this paper proposes an efficient marine oil spill identification model (Co-YOLOv7). Co-YOLOv7 incorporates the efficient attention Co-attention module into the backbone network to obtain a novel feature extraction network, significantly enhancing the model’s efficiency in extracting oil spill features. FPNet is used to fuse the oil spill features extracted efficiently, and the oil spill recognition results are obtained through the idetect head. The experimental results show that Co-YOLOv7 can completely identify the oil spill area. Meanwhile, it performs well in addressing the issues of false positives and false negatives in complex regions with oil spills.