Research On High-resolution Optical And SAR Remote Sensing Data Fusion Processing And Typical Land Surface Types Extraction Technology

Land cover information is an important data foundation for "well known land use / cover",Remote sensing image is one of the important means to obtain land cover elements and grasp land cover status.With the implementation of our country’s "High-resolution Earth Observation System Major Special Project",it provides a rich remote sensing data foundation for optical,SAR,hyperspectral and other remote sensing data for the research on the classification and extraction of land cover elements based on remote sensing images.GF-1 and GF-3,as the first high-resolution optical satellite and the first SAR satellite in the "High Score Special Project",have played an important role in the classification and extraction of land cover elements.Due to the different characteristics of the data obtained by optical and SAR sensors,it is of great significance to explore how to make full use of the advantages of the two data to achieve complementary advantages,so as to maximize the classification and extraction accuracy of typical land cover elements.In this study,two typical areas of plains with flat terrain and mountainous areas with large undulations were selected as the study area,GF-1 multispectral and GF-3 fully polarized SAR were used as data sources.First,in the two study areas,GF-1 and GF-3 single data sources were used to classify and extract land cover elements and analyze the classification results.On this basis,the pixel-level and feature-level fusion processing is performed on the GF-1 optical and GF-3 SAR data.Finally,typical land cover elements are extracted based on the fusion results.The main conclusions are as follows:(1)Compared with pixel-level fusion,the feature set constructed by using GF-1 and GF-3original band data,texture features,and polarization features in this study achieved better results in the classification and extraction of land cover elements.This also shows that when using heterogeneous remote sensing data fusion methods to extract land cover elements,heterogeneous remote sensing data fusion is more suitable for feature-level fusion.(2)In different scenarios,the performance of pixel-level fusion algorithms is inconsistent.In the pixel-level fusion of optical and SAR data,due to the large difference in SAR data quality under different terrain conditions,the same fusion algorithm performed inconsistent in the two study areas,so different fusion algorithms are applicable to different scenarios.(3)The results of using SAR remote sensing data to classify and extract land cover elements are not as good as using optical remote sensing data only.However,in areas with flat terrain,the SAR data quality is better,and the classification accuracy and optics are not much different.Therefore,under the premise of flat terrain and high SAR data quality,GF-3 SAR data can be used instead of GF-1/MS data for the classification and extraction of land cover elements.(4)When using GF-3 fully polarized SAR data instead of GF-1 multispectral data to classify and extract land cover elements,Yamaguchi decomposition and Pauli decomposition have better results,and both decomposition methods can more accurately reflect the land cover status.The decomposed and synthesized RGB false-color image is better than the singlepolarization SAR when it is used for the classification and extraction of land cover elements.