面向对象的安徽省基本地貌类型划分方法

Landform Classification Based on Object-Oriented Method in Anhui Province

以30mSTRM1-DEM为基本数据源,采用面向对象思想,从多元地形因子提取、多尺度对象分割和对象分类规则等方面构建了双层次多尺度地貌类型划分方法,实现了安徽省平原、台地、丘陵、小起伏山地、中起伏山地和大起伏山地6类基本地貌类型的自动划分。研究结果表明:1)双层次多尺度地貌类型划分法能够有效实现中尺度地貌类型自动划分,平原类用户精度和制图精度分别为92.53%和69.63%、台地类为55.17%和77.00%、丘陵类为41.48%和54.65%、山地类为60.63%和62.63%,总体精度达68.32%;2)SRTM1-DEM数据能够较好地表达地貌形态基本特征,安徽省宏观地形因子提取的最佳分析窗口为3.13km^2;3)面向对象的分类方法整体分类精度优于逐像元分类法,且面向对象的双层次分类法优于其单一分割尺度分类法。

Landform classification,one of the key points of digital geomorphologic mapping,can be used as an effective window for the study of landform spatial pattern.However,there are still certain one-sidedness and subjectivity for the traditional landform classification due to the use of fixed terrain factors analysis by per pixel.This paper constructs a two-level landform classification method with multi-scale control from the aspects of multi-factor extraction,multi-scale object segmentation and multi-object classification.Based on STRM1-DEM with 30 m resolution,the basic landform types of Anhui Province including plains,tableland,hills and mountains,are automatically classified respectively.The results show that:1)The two-level landform classification method can effectively achieve the medium-scale classification of geomorphological type.In the study area,the consumer′s accuracy and producer′s accuracy of plains are 92.53%and 69.63%,with 55.17%and 77.00%for tableland,41.48%and 54.65%for hills and 60.63%and 62.63%for mountains.What is more,the overall accuracy is 68.32%.2)The basic features of landform can be described better via SRTM1-DEM dataset.The method of mean change-point analysis is applied to get the optimum window size for the extraction of macroscopic terrain factors.The optimal size of the study area is 59×59 cells,about 3.13 km 2,to calculate multiple terrain factors such as mean,terrain relief and terrain cutting depth.3)The accuracy of object-oriented classification is higher than pixel-based classification.For another,the accuracy of two-level landform classification based on object-oriented method is superior to single-scale segmentation classification.