Research On Recognition Of Sea Surface Oil Spill Area In SAR Image Based On Residual Network

Both oil film and oil-like film(biological oil and emulsified oil)show dark spots on the fully polarized SAR data,and the two have a high degree of consistency,which is easy to cause confusion during identification.For SAR images,different features have different distinguishing effects for each type of oil film.Therefore,selecting appropriate features is particularly important for distinguishing oil films and oil-like films on SAR images.In this paper,we focus on using a combination of multiple features and a deep residual network(ResNet)to distinguish oil films and oil-like films on fully polarized SAR data.The main research contents of this article can be summarized as follows:First,extract and analyze 21 kinds of features(including oil film and oil-like film)on the dark spot area of the fully polarized SAR image,and then select the distinguishing features as the basis for network input.After experiments and analysis,three polarization characteristics,Entropy,Polarization Scattering Angle(Alpha)and Relative Difference of Single Reflection Characteristic Value(SERD),were selected as the basis for network input.Secondly,multiple interest regions are selected as the training set and test set of the ResNet network on the feature maps corresponding to the determined three polarization features.Finally,the network model of this paper is trained by 10800 samples(3600 crude oil film samples,3600 biological oil film samples and 3600 emulsified oil film samples).The test set is composed of 600 crude oil samples,600 bio-oil samples and 600 emulsified oil samples(a total of 1800),and the final classification accuracy is 97.56%.Finally,use the same experimental data to use the same deep learning VGG and AlexNet classification algorithms for oil film and oil-like film classification,and compare with the ResNet classification algorithm classification results.Considering that the over-fitting phenomenon of the network model should be avoided to obtain more reliable experimental results,K-cross validation and ROC curve experiments were also carried out in this paper.The results show that the algorithm selected in this paper is effective,and it proves that the algorithm proposed in this paper can accurately identify the dark spots on the fully polarized SAR image,and can also have the unstructured characteristics on the fully polarized SAR image.To distinguish between oil film and oil-like film phenomena.But when the experimental data is not enough,the experimental model trained in this paper has overfitting phenomenon,so it is necessary to enrich and expand the experimental data in the subsequent research.