利用Sentinel-2A影像的面向对象特征湿地决策树分类方法

Object-Based Wetland Decision Tree Classification Method Using Sentinel-2A Image

苏州市湿地众多、类型多样化、周围环境复杂,使用传统的遥感分类方法很难得到精度较高的湿地分类结果。研究了面向对象特征的湿地决策树分类方法,以苏州市澄湖地区为研究区域,使用欧空局的Sentinel-2A影像,先将研究区域分为湿地水体、植被和非植被3大类型,再分别构建鱼塘、河流、湖泊、农田和裸地等面向对象特征,据此实现湿地遥感分类。研究结果表明,该方法能够有效利用遥感影像提供的光谱特征、几何特征和纹理特征等多种丰富信息,产生较高的分类精度,总体分类精度可达80.67%,Kappa系数为77.80%。与传统的基于中低分辨率遥感影像的分类方法相比,该方法可以有效提取湿地不同地物对象的几何结构和纹理等特征,在提高湿地分类精度的同时实现对大面积湿地的快速动态监测。

There are many wide regions and types of wetlands in Suzhou city, and the surrounding environment of wetlands is complex. It is difficult to generate high classification accuracy results by using traditional remote sensing classification methods. In this paper, object-based wetland decision tree classification method was developed. Firstly, with Chenghu region selected as study area and Sentinel-2 A image of ESA used, image covering study area was divided into wetland water body, vegetation and non-vegetation types. Then, object-based features of fish pond, river, lake, farmland and bare land were derived respectively. The results of this study show that the object-based wetland decision tree classification method can effectively utilize the rich information provided by remote sensing images(e.g. spectral features, geometric features and texture features) and produce high classification accuracy. The overall classification accuracy and the Kappa coefficient of the method are 80.67% and 77.80% respectively. Compared with the traditional classification method based on middle and low resolution remote sensing images, the method can effectively derive the geometric structure and image texture characteristics of wetland objects and types, improve the classification accuracy and perform rapid dynamic monitoring of large area of wetland.