单基线PolInSAR森林高度反演方法研究  

Comparison of Five Methods to Inverse Forest Height from Single-baseline PolInSAR Data

作　　者：张建双[1] 范文义[1] 毛学刚[1] 于颖[1] ZHANG Jianshuang;FAN Wenyi;MAO Xuegang;YU Ying(School of Forestry,Northeast Forestry University,Harbin 150040,China)

机构地区：[1]东北林业大学林学院,哈尔滨150040

出　　处：《农业机械学报》2018年第10期220-229,共10页Transactions of the Chinese Society for Agricultural Machinery

基　　金：国家重点研发计划项目(2017YFB0502700);中央高校基本科研业务费专项资金项目(2572018BA02)

摘　　要：极化干涉SAR(Polarimetric SAR interferometry,PolInSAR)森林高度反演是当前雷达遥感领域的研究热点,近年来出现了多种单基线PolInSAR森林高度反演方法。为了给单基线PolInSAR反演森林高度的算法提供基础,并探索和发展效果更优的反演方法,使用Pol SARpro软件模拟森林平均高度为18 m的L波段(L=23 cm)全极化干涉SAR数据,研究了森林高度反演算法中的DEM差分法、RVoG法、复相干幅度反演法、混合反演法,并基于相干优化法对混合反演法进行了改进;为了更准确地对算法的性能进行比较,给出方位向为48 bin时各算法的距离向剖面的对比图,并选取图像的中间区域,对森林高度位于3~30 m的1 104个样本点,应用均值和均方根误差RSME对5种方法模拟的18 m森林高度进行比较。结果表明:森林高度平均值反演结果由大到小依次为:复相干幅度反演法、混合反演法、改进的混合反演法、RVoG法、DEM差分法,分别为19.40、18.31、18.12、10.55、10.05 m,均方根误差(RMSE)由小到大依次为:改进的混合反演法、混合反演法、复相干幅度反演法、RVoG法、DEM差分法,分别为1.06、1.48、3.49、7.51、8.04 m;说明DEM差分法与RVoG法反演的森林高度存在明显低估,复相干反演法出现明显高估且其离散程度最大,混合反演法和改进的混合反演法与真实值的误差分别为0.31、0.12 m,改进的混合反演法与真实值的相差最小,离散程度最小,均方根误差最小,反演结果最优。改进的混合反演法综合了混合反演法与相干优化法的优点,使其估计的地形相位的均方根误差最小(0.045 rad),森林高度与真实值的误差最小,均方根误差最小,并且具有一定的鲁棒性。To provide the basis for the methods of vegetation height inversion by using single-baseline PolInSAR SAR data and explore a more effective inversion method,the European Space Agency(ESA)Toolbox PolSARPro was used to simulate L-band(L=23 cm)PolInSAR SAR data with an average vegetation height of 18 m.The DEM difference,RVoG,SINC,Hybrid,and Hybrid method were studied based on coherent optimization.The vegetation height ranged from 3m to 30m was analyzed with 1104 sample points in the middle region of the image.Giving a 3D image,range profile image with azimuth of 48 and statistical image of vegetation height and topographic phase were used to compare the performance of five methods.Compared with the real value of 18 m,the descending order of vegetation height means was SINC,Hybrid,improved Hybrid,RVoG and DEM difference method.The difference between the improved Hybrid inversion method and the real value was the smallest as 0.12 m,smaller than Hybrid of 0.31 m.RMSE of the improved Hybrid,Hybrid,SINC,RVoG and DEM difference was 1.06 m,1.48 m,3.49 m,7.51 m and 8.04 m,respectively.The vegetation height of the improved Hybrid method had the smallest difference and RMSE.The estimated topographic phase average value of the improved Hybrid,RVoG/Hybrid and DEM difference method was-0.018 rad,0.011 rad and 0.1 rad;RMSE was 0.045 rad,0.054 rad and 0.15 rad;and mean value of absolute value was 0.03 rad,0.04 rad,and 0.1 rad,respectively.The topographic phase of the improved Hybrid method was approximately the closest to the simulated and had the smallest RMSE and the mean of absolute value.Improved Hybrid inversion method produced the best result among the five methods,combining the merits of Hybrid with the coherent optimization,with the smallest difference between real value and RMSE of vegetation height and topographic phase.The Hybrid method was improved based on the coherent optimization and the accuracy of vegetation height was analyzed with the ground phase estimation results to compare the five methods.

关 键 词：单基线森林高度反演 DEM差分法 RVoG法 复相干幅度反演法 混合反演法 相干优化法 

分 类 号：S758.54[农业科学—森林经理学] TP79[农业科学—林学]