利用辅助数据的荒漠区高分辨率遥感分类研究

High-resolution Remote Sensing Classification Using Auxiliary Data in Desert-mountain Area

使用美国NAIP高分辨率航空遥感影像,在多尺度、多变量影像分割的基础上,采用决策树方法建立干旱区半干旱区的荒漠分类规则,并结合水系、道路等辅助地理数据进行干旱区半干旱区面向对象遥感分类。选择位于美国亚利桑那州菲尼克斯大都市区的周边典型荒漠地区为实验区,利用河流、道路等辅助数据进行面向对象遥感分类效果要优于单纯依靠遥感影像的分类,能够有效地提取季节性河流和简易道路。研究对美国亚利桑那州菲尼克斯都市区周边的同一荒漠地区进行了实验,利用决策规则有效提取植被和荒地,以及提取简易道路和土壤,分类总精度从常规面向对象分类方法的82.85%提高到92.45%。研究结果表明:本文提出的分类方法对荒漠地区的泥土路和灌木及其整体分类精度有较大提高。利用辅助数据进行遥感分类可以改善特定研究区的高分辨率遥感影像分类精度。

Because of the presence of seasonal rivers and unpaved roads in semi-arid area,it is difficult to perform classification with high-resolution remote sensing imagery especially in desert areas.This study improves the results of high-resolution remote sensing overall classification by using ancillary data.In this paper,object-oriented remote sensing classification in semi-arid region was performed with the National Agriculture Imagery Program(NAIP)High resolution aerial remote sensing images and geographic auxiliary data such as drainage and road data.After using the multi-resolution and multi-variable image segmentation,decision tree was used to establish classification rules in object-objected classification.A typical desert-mountain area surrounding Phoenix Metropolitan Areas,Arizona,USA was selected as the experimental region.Experimental results showed that the classification results of object-oriented remote sensing with the use of geographic auxiliary data were better than that only using the remote sensing images,and the seasonal rivers or unpaved roads were efficiently classified from desert or vegetation areas.In the experiment,vegetation and barren were effectively extracted with the decision rule,and the same to unpaved road and soil,with the overall classification accuracy increased from 82.85%to 92.45%.The results show that the classification accuracy of unpaved road and shrub as well as overall classification accuracy were greatly improved.High-resolution remote sensing classification using auxiliary data can improve the classification accuracy in specific research area.