基于Sentinel-1A极化SAR数据与面向对象方法的山区地表覆被分类

地表覆被分类对国土资源调查评估及全球变化具有基础性和关键性意义,但山区由于地形和云雾等的影响,可利用光学遥感影像和其他资源十分稀缺。因此,论文以渝东南山区为研究区,基于Sentinel-1A极化合成孔径雷达（SAR）数据,通过系列预处理,得到后向散射系数值,同时对各类地物的VV/VH极化后向散射系数、纹理、高程和坡度等特征值统计分析,综合这些特征值运用面向对象分类方法对单时相与多时相SAR数据进行地表覆被分类,最后将这两种分类方法与Landsat 8 OLI数据分类作对比。研究表明：1）在同时运用面向对象分类方法的前提下,单时相SAR数据分类和Landsat 8 OLI数据分类精度相当,多时相SAR数据面向对象分类精度最高,总精度为85.65%,Kappa系数为0.829 9;2）与光学数据相比,SAR数据对阔叶林、人工建筑提取有优势,精度提高了10%以上,多时相特征有利于耕地和针阔混交林提取,分类精度比单时相提高了9%左右;3）研究区土地覆被类型以林地为主,占总面积的42.68%,耕地、草灌次之,人工建筑、草地与河流占地面积较少。

Land Cover Classification in Mountain Areas Based on Sentinel-1A Polarimetric SAR Data and Object Oriented Method

Land cover classification is fundamental and critical to the investigation and assessment of land resources and the global change. However, the cloud-prone and rainy weather in mountain areas make it difficult to obtain valid optical remote sensing images. In addition, the complexity of terrain also has a negative impact on the accuracy when classifying land covers using only optical remote sensing imagery. Synthetic aperture radar (SAR) can transmit energy at microwave frequencies that are unaffected by weather conditions. This advantage gives SAR all-day and all-weather imaging capability. In this paper, the research area is situated in the mountain areas of Southeast Chongqing. We obtained the backscattering coefficient through a series of preprocessing of the Sentinel-1A polarization data. Then, according to the statistical analyses of VV/VH polarization backscattering coefficients, textures, elevations and slopes of all kinds of land covers, we employed object oriented approach on single temporal and multi-temporal images to improve land cover classification accuracy by combing these features. Finally, we compared the classification results with the result of Landsat 8 OLI data. The research indicated that: 1) The object-oriented classification method with single temporal SAR data has about the same accuracy as the OLI Landsat 8 data, the object-oriented classification with multi-temporal SAR data has the best classification result with the total accuracy 85.65% and Kappa coefficient 0.829 9. 2) SAR data have the advantages in extracting broad-leaved forest and artificial building which improve the accuracy by more than 10%. The classification with multi-temporal data has the advantage in the extraction of coniferous and broad-leaved mixed forest and farmland whose accuracy is about 9% higher than that with single temporal data. 3) Forest is the main land cover type in the study area which accounts for 42.68% of the total area, followed by farmland and shrub, and the artificial construction, grassland and river are less.