基于地形信息的Landsat与Radarsat-2遥感数据协同分类研究

针对不同成像机理的光学与雷达遥感数据协同应用于地表信息提取瓶颈问题，提出了一种基于地形信息的光学与雷达数据协同分类方法。首先利用InSAR测量技术从Radarsat-2数据中提取DEM地形信息，然后构建基于地形信息的Landsat光学数据和Radarsat-2雷达数据的不同特征集输入模型，最后通过随机样本选取构建随机森林（Random Forest,RF）、支持向量机（Support Vector Machine, SVM）和决策树（Decision Tree,DT）分类算法模型提取地表信息。结果表明：①针对不同特征协同策略，在随机选取10%训练样本时，Radarsat-2干涉提取DEM与Landsat数据集提取精度优于ASTER GDEM与光学影像协同策略；②针对不同地表信息提取算法模型，通过50次随机选取训练样本构建模型评价分类精度，验证RF算法的鲁棒性和提取精度都要优于DT算法和SVM算法。研究充分利用光学和雷达遥感的优势信息，为光学和雷达遥感协同地表信息提取提供新的思路。

Remotely Sensed Data Classification by Collaborative Processing of Landsat,Radarsat-2 and Topography Information

Remote Sensing Data (RSD) and corresponding information extraction technologies are widely used in urban planning, ecological environment modeling, change detection, etc. Optical and radar remote sensing data, due to different imaging mechanisms, have complementary advantages for information extraction and applications. In order to improve the limitations of combining optical and SAR data for land surface information extraction. Strategies for Land Use and Land Cover (LUCC) classification based on collaborative processing optical, SAR RSD and topography information was proposed. Firstly, Digital Elevation Model (DEM) information was extracted from multi-temporal Radarsat-2 SAR images using InSAR technique. Then integrated model was constructed for information extraction based on inputs from topography data, Landsat optical RSD, Radarsat-2 SAR RSD. Finally, LULC information was extracted by random sample selection and machine learning algorithms (e.g. Random Forest (RF) ensemble learning method, Support Vector Machine (SVM) and Decision Tree (DT)). The results demonstrate that (1) when 10% training sample was selected, advantages come from combination of DEM extracted from Radarsat-2 SAR and Landsat data compared with combination of ASTER GDEM and the corresponding optical dataset; (2) Comparison results among different algorithm models by averaging 50 times classification accuracy of each model, demonstrate the robust and advantage of RF than DT and SVM. In this research, the combination advantage of optical and SAR remotely sensed data are explored, which can provide a new approach for making full use of optical and SAR data in the process of LULC classification.