Research On Oil Spill Detection Method Based On Multi-Polarization SAR Image

Marine oil spill disaster is one of the most concerned issues in the marine environment,and it has caused great harm to the marine ecological environment.Therefore,after the marine oil spill occurs,the oil spill should be detected accurately and promptly,and corresponding measures should be formulated based on the test results.Measures are not only important measures to minimize losses,they are also of great significance to marine environmental protection and ecological resource development.Due to the particularity of marine oil spill detection,large-scale inspections are required.Using remote sensing technology to conduct large-scale and large-area continuous observation of marine oil spills has become one of the most effective oil spill detection methods.All-weather synthetic aperture radar(Synthetic Aperture Radar,SAR)provides a rich data basis for marine oil spill detection,especially polarized SAR images provide irreplaceable data resources for marine oil spill detection.However,due to the SAR image imaging itself and the interference of the external environment,there will be a lot of noise that will interfere with the detection of marine oil spills and even misjudge.At the same time,the low-wind area and the biological oil film appear as black bars on the SAR image and the marine oil spill,which brings a huge challenge to the marine oil spill detection,and it is also a key issue that the marine oil spill detection urgently needs to solve.At the same time,traditional oil spill detection methods mostly rely on the setting of thresholds or model parameters,which have significant subjectivity and uncertainty,resulting in low accuracy of offshore oil spill detection and lack of generalization capabilities.In response to the above problems,this paper proposes an improved Fully Convolutional Networks(FCN)multi-polarization SAR image marine oil spill intelligent detection framework.The main research content of the paper is:(1)A FCN-based polarimetric SAR oil spill detection framework is proposed.First,in view of the widespread noise in polarimetric SAR images,Pauli decomposition and Refined-Lee filtering preprocessing are performed on polarimetric SAR images to ensure the polarization feature information of SAR images.At the same time,it reduces the impact of suspected oil spill noise on detection accuracy;secondly,in order to effectively distinguish between oil spill and suspected oil spill,four types of features are extracted,namely,intensity feature,texture feature,coherence feature,and phase difference feature,and FCN is used to generate spills.Oil detection probability map,and then identify the oil spill coverage area from the probability map.Experiments have shown that the oil spill detection framework can effectively identify oil spills with a detection accuracy of89.1%.On this basis,the oil spill detection results of FCN-32 s,FCN-16 s and FCN-8s were compared to verify the validity of the experiment.(2)An improved oil spill detection method based on multi-polarization SAR images is proposed.Aiming at the lack of consideration of spatial information in the FCN model,a fusion mechanism of different levels of convolutional layers is proposed to achieve the fusion of high-level semantic features and low-level spatial details,and then Improve the detection accuracy of offshore oil spill areas.The comparative analysis of experiments shows that the marine oil spill intelligent detection framework based on the improved FCN can effectively reduce the impact of suspected oil spill areas on the detection accuracy,while taking into account multi-polarization and edge feature information to achieve pixel-based oil spill area detection.The excellent detection accuracy can reach95.7%.Compared with the traditional FCN deep learning model,the experimental results show that the oil spill detection results of this method show significant advantages.