基于Hyperion高光谱数据的土地退化制图研究--以陕西省横山县为例

以陕西省横山县境内的Hyperion数据为数据源，提出一种针对土地退化的制图方法:土地退化指数法(LDI: Land Degradation Index)。在影像分类的基础上，利用在分类过程中提取的训练样本进行线性波谱分离，得到各个端元的分离影像和RMS误差影像，再通过设置新的端元和反复运行线性波谱分离算法得到最终的训练样本，然后利用神经网络法对影像进行二次分类，最后在掩膜处理的基础之上把土地分为:轻度退化，中度退化和高度退化三种类型(Kappa=0.90)。文中分别采用了三种分类方法:监督分类与非监督分类相结合的混合分类方法、光谱角制图(SAM)方法、混合调制匹配滤波(MTMF)方法。结果显示混合分类方法(Kappa=0.71)具有比光谱角制图方法(Kappa=0.54)和混合调制匹配滤波方法(Kappa=0.60)更高的分类精度,所以选择在混合分类的基础上进行土地退化指数LDI的分析。

Study On Land Degradation Mapping by Using Hyperion Data in HengShan Region of China

Land degradation, defined as the loss or the reduction of the potential utility or productivity of the land, is a major environmental problem in the world today. The land degradation process is generally divided into three classes: (1) physical degradation; (2) biological degradation, and (3)chemical degradation. The assessment of land degradation requires the identification of indicators such as soil vulnerability to erosion. Generally, the assessment of the state of land degradation can be carried out by using the Global Assessment of Soil Degradation (GLASOD) method. Hoosbeek et al. recommended this qualitative method to classify soil degradation by using remote sensing data. Degradation features can be detected directly or indirectly by using image data. Based on the Hyperion images, this paper brings forward a new mapping algorithm, called Land Degradation Index, aimed at land degradation in Hengshan region of China. It is based on the classified process. We applied the linear spectral unmixing algorithm with the training samples derived from the formerly classified process so as to find out new endmembers in the RMS error imagine. After that, by using neutral net mapping with new training samples, the classified result was gained. In addition, after applying mask processing, the soils were grouped to 3 types (Kappa = 0.90): highly degraded soils; moderately degraded soils; and slightly degraded soils. By analyzing 3 mapping methods, i.e. mixture-classification, the spectral angle mapper and mixture-tuned matched filtering, the results suggest that the mixture-classification has the higher accuracy (Kappa=0.7075) than the spectral angle mapper (Kappa=0.5418) and the mixture-tuned matched filter (Kappa =0.6039). As a result, the mixture-classification is selected to carry out Land Degradation Index analysis.