基于U-NET的双分支海上SAR溢油检测模型

A two-branch marine oil spill detection model based on U-NET

为提高海上溢油SAR(Synthetic Aperture Radar)检测的准确率,本文提出一种基于U-NET和注意力门的海上溢油SAR检测模型(AW-net),该模型将U-NET中传统的单输入编码器替换为双分支编码器,分别输入纹理特征和SAR灰度特征,并进一步采用注意力门融合纹理信息和灰度信息。实验利用1景海丝一号(HISEA-1)SAR数据构建样本训练集进行AW-net模型训练,分别应用1景HISEA-1 SAR数据和1景Radarsat-2SAR数据开展模型测试,溢油检测准确率均优于U-NET、AttentionU-NET和FCN等语义分割模型,说明该模型具有较强的强鲁棒性和应用潜力。

To improve the accuracy of marine oil spill detection using synthetic aperture radar(SAR),an improved AW-net model based on U-NET and attention gate is proposed in this paper.In this model,the traditional single-input encoder in U-NET is replaced by a double-branch encoder and the multifeature input mode is changed from the previous feature stacking input to the texture feature and SAR gray image input in the double-branch encoder.This change is made to extract finer texture and gray information and improve the dimensionality caused by the multichannel overlay input.The multiscale texture information extracted by the double encoder is fused with the gray information using the attention gate.In the experiment,one piece of HISEA-1 SAR data was used for model training;furthermore,one piece each of HISEA-1 SAR data and Radarsat-2 SAR data was used for model testing.The oil spill detection accuracy of the two pieces of test data was better than that of other semantic segmentation models.These results demonstrate the robustness and application potential of the AW-net model.