地貌形态类型的自动分类方法综述  

Review of Methods for Landform Automatic Classification

作　　者：王彦文[1,2] 秦承志[1,2,3] WANG Yan-wen;Q1N Cheng-zhi(StateKey Laboratory of Resources and Environmental 1 Tiformatuon System, Institute of Geographic Sciences and Natural Resources, CAS, Beijingl00101;College of Resources and Environment, University of Chinese A cademy of Sciences, Beijing 100049;Jiangsu Center for Collaborative Innovation in Geographical Inform.ation Resource Development and Application, School of Geograj) hic Science, Nanjing Normal Ihdversity, Nanjing 210023, China)

机构地区：[1]中国科学院地理科学与资源研究所资源与环境信息系统国家重点实验室,北京100101 [2]中国科学院大学资源与环境学院,北京100049 [3]南京师范大学地理科学学院/江苏省地理信息资源开发与利用协同创新中心,江苏南京210023

出　　处：《地理与地理信息科学》2017年第4期16-21,共6页Geography and Geo-Information Science

基　　金：国家自然科学基金项目(41422109;41431177);资源与环境信息系统国家重点实验室自主创新项目

摘　　要：地貌形态分类对于地貌研究、土壤制图、滑坡防治等诸多地学研究及应用领域都有着重要作用。数字地形分析领域的研究者将地貌形态分类知识与地理计算相结合,已开发出许多地貌形态自动分类方法。该文将现有的地貌形态自动分类方法分为3类进行讨论:基于聚类的方法、基于规则知识的自动分类方法和基于典型样点的自动分类方法。基于聚类的方法对地貌形态分类专家知识缺乏考虑,聚类结果常难以明确对应到目标地貌形态类型;基于规则知识的方法常需要显式给出分类规则,应用难度较大,扩展性较差;基于典型样点的方法具有较好发展前景,但还有待改善对隐含专家知识的利用程度。各类方法目前对复合形态类型均难以有效提取。对于可能的方法改进方向,该文从空间结构特征信息、隐性专家分类知识的来源两方面进行了讨论。Accurate map of landform classification is one of the important basic data for many geographical researches and appli-cations(such as geomorphological research,digital soil mapping,landslide treatment engineering,etc.).Currently many methods of landform automatic classification have been designed through ct^mbining geoct^mputation and domain knowTledge on landform classification.However,their application results(especially for composite landform types)are still unsatisfactory due to the character is tics of landform types,such as semantic vagueness,multi-scale and spat ial structure,w hich is hard to be quantified.In this paper,existing landform classification methods are classified to be three types,including clustering-based methods,rule-based methods and typica-sample-based methods.The clustering-based methods lack consideration on expert classification knowledge.This situation makes the results from this type of method often cannot well match the target taxonomy of landform type.The rule-based methods need users to formalize the expert knowledge as explicit rules,which makes them not only hard to apply in practice but also less extensible to other areas.The typical-sample-based methods,although having good potential in ap-plication,still need more improveiment on the utilization of implicit expert knowTledge.In general,current automatic classification methods of landform are difficult to effec:tively extract composite landform types.Possible improvements for landform classifica-tion methods are further discussed from two aspects,including consideration of information on spatial structure and the source of implicit knowledge on landform classification.

关 键 词：数字地形分析 地貌形态类型 地貌形态分类 数字高程模型 

分 类 号：P931[天文地球—自然地理学]