PolSAR Sea Ice Types Extraction Method Based On Feature Fusion

Accurate and efficient extraction of the type and distribution of sea ice is of great significance to the monitoring and prevention of sea ice disasters.Polarized SAR data has the advantage of all-day,all-weather,and is not affected by clouds and fog,and has become an important technical means for extraction of sea ice types.The main problems of sea ice type extraction are as follows:first,the selection of classification feature parameters has an important impact on the results of sea ice type extraction;second,it is difficult to achieve accurate sea ice type monitoring based on a single attribute feature;Third,the common method for extracting sea ice types from dual-polarized SAR data is poorly adaptable,and a new type of sea-ice type extraction algorithm for dual-polarized SAR data is urgently needed.In view of the above problems,this article uses polarized SAR data to carry out sea ice in Liaodong Bay.The type extraction research has mainly carried out three aspects of research.(1)Feature Extraction and analysis of polarimetric SAR images.For GF-3 polarimetric SAR data,based on the characteristic of polarimetric data,the backscatter intensity feature parameters of sea ice SAR image were extracted;based on Pauli decomposition,H/?/A decomposition and Anyang decomposition,the polarimetric target decomposition feature parameters of sea ice SAR image are extracted;based on the gray level co-generation matrix,the texture feature parameters of sea ice SAR image were extracted,and a total of 22 sea ice type extraction feature parameters were obtained.The effectiveness of different attribute feature parameters for sea ice type recognition was analyzed.The results show that different feature parameters contained different ground object information and the combination of backscatter intensity feature parameters,polarization target decomposition feature parameters and texture feature parameters can improve the difference of sea ice types.(2)Sea ice type extraction algorithm for fully polarized SAR data.In this paper,taking Liaodong Bay of Bohai Sea as the experimental area,a sea ice classification method combining target decomposition features and texture features of polarimetric SAR was proposed by using the data of GF-3 full polarimetric SAR.Firstly,the effective feature parameters of sea ice classification were selected by feature analysis.Secondly,a vector was constructed by combining effective feature parameters.Finally,the sea ice types were extracted accurately based on support vector machine(SVM)classifier and polarimetric SAR data.The experimental results show that the effect of sea ice classification is greatly improved by combining multiple feature parameters.Compared with the maximum likelihood classifier(MLC)results,the SVM classifier is proved to be effective in the sea ice type extraction on fuse multiple features,and the total classification accuracy is up to 93%.(3)Sea ice type extraction algorithm for dual-polarized SAR data.In this paper,taking Liaodong Bay of Bohai Sea as the experimental area,a sea ice classification method based on multi-feature fusion was proposed from dual-polarized SAR data of GF-3 and Sentinel-1.Firstly,the multi-features of sea ice images were extracted,including backscattering features,polarimetric target decomposition features,derived features and texture features.Secondly,the feature vector was constructed by selecting the feature parameters which were effective to the sea ice type extraction.Finally,SVM classifier was used to realize large area monitoring of sea ice type.The results show that the method of sea ice type extraction based on multi-feature fusion are more accurate than other methods.