基于时变特征的多时相PolSAR农作物分类方法

Crop Classification Method with Differential Characteristics Based on Multi-temporal PolSAR Images

获取农作物分类信息是极化合成孔径雷达(Polarimetric synthetic aperture radar,PolSAR)的重要应用之一,然而单时相PolSAR数据能够提供的信息十分有限,而且单时相数据的获取时间也会影响农作物分类精度。随着技术的发展,出现了大量的机载和星载PolSAR系统,这些系统能够获取目标重复观测的PolSAR数据,这为多时相PolSAR的数据分析和应用提供了可能。本文以多时相PolSAR农作物分类为出发点,通过利用不同农作物的极化散射特性的变化特性来提高分类精度。首先,基于极化散射特性分解原理分析了不同农作物在生长过程不同时期所呈现的散射特性变化规律,在此基础上定义了一个新的参数描述其散射特性的变化特性。其次,基于这一新参数提出了一种多时相PolSAR农作物监督分类算法。最后,通过对欧洲空间局所提供的基于Radarsat-2实测仿真生成的Sentinel-1数据处理结果表明,相比于基于复Wishart分布的监督分类算法,农作物的整体分类精度提高了约4个百分点,当农作物种类合并为4类时,整体分类精度提高了约6个百分点。

Crop type classification is one of the most significant applications in polarimetric synthetic aperture radar（ PolSAR） imagery. As an advanced remote-sensing technique,PolSAR has been proved to provide high-resolution information,including the intensity and polarization of illustrated land surface.However, single-temporal PolSAR data are restricted to provide sufficient information for crop classification and identification. With the increase of number of airborne and spaceborne PolSAR systems,a large number of real PolSAR data are generated,and thus provides opportunities for multitemporal data analysis. The potential of improving crop classification accuracy by introducing the differential characteristics of H/α parameters for multi-temporal PolSAR images was investigated. Firstly,by analyzing the characteristics of several typical crops in different growing stages,a new parameter was defined for the first time to describe the differential characteristics of H/α distribution. Therefore,a new supervised classification method with the newly defined parameter was proposed to classify different crop types. The main idea of the proposed method was to apply various features of classical H/α parameters to improve the accuracy of crop classification. A validation test for the new approach was performed with Sentinel-1 data sets which were simulated by Radarsat-2 data sets and provided by ESA. The results showed that the mean accuracy of the proposed method was improved by 4 percentage points compared with the supervised complex Wishart classifier when the six kinds of crops were classified. Furthermore,the number of classes was reduced to 4 and the accuracy was almost improved by 6 percentage points.