Research On SAR Image Oil Spill Detection Algorithm Based On Deep Learning

The harm of environmental pollution caused by marine oil spills continues to deepen.The division and detection of oil spill areas play an important role in timely marine environmental governance and protection.SAR images have been widely used in agriculture,hydrology,geological surveying and mapping,resource surveying,environmental monitoring and other fields.This paper uses deep learning based neural networks to carry out oil spill detection from SAR images that contain oil spills.This paper mainly exploits the combination of the advantages of multiple models in order to make up for the potential shortcomings of a single network structure,and uses the idea of transfer learning to combine different neural networks to construct a segmentation model,which achieves a better segmentation performance than that of a single network.In this paper,three image segmentation models are proposed as Seg-Mobile Net,Deeplab-Mobile Net and Seg-Res Net.The Seg-Mobile Net model is obtained by combining some modules of Seg Net and Mobile Net through the idea of migration learning,and the proposed model is more accurate for the oil spill detection of SAR images.Deeplab-Mobile Net uses the Mobile Net module to greatly reduce the number of parameters,and the parallel atrous convolution blocks obtain context information from different scales.Seg-Res Net uses the residual structure to deepen the network to obtain higher accuracy.Building a deeper network can effectively improve the segmentation performance.Experiments show that these three models obtain better segmentation performances than that of using the single network model.Verify the feasibility of the combination of different model advantage modules to improve performance.The researches in this paper will bring promising theoretical and application values to the research of oil spill detection from SAR images.