Study On Marine Oil Spill Extraction Method Based On The Image Characteristics Of Sentinel-1A

With the continuous development of marine trade,shipping,and oil exploitation,correspondingly,marine oil spill accidents have also occurred more and more frequently.The marine ecological environment crisis and direct and indirect economic losses caused by oil spill have attracted the attention of many governments,especially countries with long coastline.Therefore,after the oil spill incident,how to monitor the oil spill timely and accurately has become the research focus of marine environmental protection.In recent years,with the development and progress of the aerospace industry,satellite remote sensing technology has advanced by leaps and bounds,and has replaced traditional methods as an important way to monitor marine oil spill.Compared with optical Remote Sensing,which is easily affected by rain and snow and cannot be monitored at night,Synthetic Aperture Radar(SAR)can work all day and all-weather,making up for the deficiency of optical Remote Sensing that is easily affected by weather.At the same time,the longer wavelength makes it possible to penetrate clouds and fog,high-quality image data can be obtained.How to identify marine oil spills in SAR images more accurately has become a research hotspot.Taking Sentinel-1A SAR image as the research data source,this paper compares and analyzes the methods of marine oil spill recognition and classification based on different polarization and texture feature vectors,and focuses on the advantages and feasibility of different oil film recognition methods based on the polarization and texture features of oil spill SAR image.The main work includes the following aspects:1)The preprocessing of SAR images in the study area mainly includes radiometric correction,terrain correction and filtering processing to eliminate or correct image distortion caused by radiometric errors.In the removal of speckle noise,the Refined Lee method window 3*3 is selected for filtering,which not only reduces noise but also retains edge information.2)For SAR images,use the H/ɑ polarization decomposition method to extract the polarization feature vectors and use the Gray Level Cooccurrence Matrix(GLCM)algorithm to extract the texture feature vectors,and then use correlation and normalization to extract 23 feature vectors.A mean value analysis was performed twice to evaluate its ability to identify relevant information such as oil slicks.Finally,nine feature vectors were selected as the fusion data set for oil slick identification and classification for subsequent oil slick identification.3)Using the Support Vector Machine(SVM)and Random Rorest classifier(RF)to identify the oil film of a single feature image and the oil film of a fusion feature image on the filtered polarization feature vector and texture feature vector.By comparing and analyzing the classification results of different characteristic images,the following conclusions are obtained: 1)The overall accuracy of the oil film classification of the three kinds of polarization characteristic parameter images reaches 75%,which can distinguish the oil film from the other two types of ground objects.So the extraction of oil film by polarization feature is feasible.The six texture feature images under the two polarization modes have a classification accuracy of more than 90% for land,which can accurately extract land information and easily distinguish land from other two types of features.So it can be used for the identification and classification of land.At the same time,the images under the VV polarization method have more information and can roughly identify oil slicks.So they are more suitable for oil slick recognition than the VH polarization method.2)The classification accuracy of the five feature fusion images for oil spills is more than 80%,proving that polarization fusion can improve the accuracy of oil film classification.The fusion images of polarization feature vector and texture feature vector has a higher accuracy of oil spill monitoring than polarization fusion image and texture fusion image,and the classification accuracy reaches 96%.The classification accuracy of the best fusion feature image is about 15% higher than that of a single feature image.3)Using SVM and RF classification methods for oil film extraction,the best accuracy rates are 96.1530% and97.0840%,respectively,and RF classification method is better than SVM.It has a higher classification accuracy,and the overall accuracy has increased by 0.931%.It can better separate seawater,oil film and land area,and it is more feasible for oil film extraction.This shows that the Random Forest classifier has a great application prospects in the monitoring and identification of marine oil spills.