数字地形分析应用适配性知识的案例表达与推理方法

Case-based formalization and inference method of application-matching knowledge on digital terrain analysis

数字地形分析(Digital Terrain Analysis,DTA)在应用时依赖于建模知识,尤其是关于所建的应用模型是否与研究区特点、数据等条件相适配的知识(称为"应用适配性知识");由于这类知识难以形式化表达,现有的数字地形分析工具对此类知识缺乏利用,从而导致普通用户在应用数字地形分析时建模困难。针对该问题,设计了一套数字地形分析领域应用适配性知识的案例表达与相应的推理方法。以美国32个河网提取案例为例,通过交叉验证,初步表明案例及其推理应用方法适合于数字地形分析领域应用适配性知识的形式化表达与应用,该方法通过与建模环境的集成,可大幅降低数字地形分析应用建模难度。

Application of digital terrain analysis（DTA） relies heavily on the DTA- domain knowledge on the match between the chosen algorithm（and its parameter-settings） and the application context（such as target task,terrain condition of the study area, and DEM resolution） —the so- called application- matching knowledge. This type of knowledge has a direct impact on the quality of DTA modelling when users of DTA, especially nonexpert users, do not have sufficient amount of such knowledge to support their DTA applications. Existing DTAassisted tools often cannot use application- matching knowledge because this type of knowledge has not been formalized in DTA to be available for inference in these tools. This is mainly because this type of DTA knowledge is currently inaccurate and non-systematic, and often exists in documents for specific case studies, or as personal knowledge of domain experts. This situation makes the DTA modelling process difficult for users,especially for non- expert users. Case- based reasoning method that originated from artificial intelligence is appropriate for formalization and inference of non- systematic knowledge. In this article, we propose a casebased formalization and inference method for the application-matching knowledge in DTA. The specific design of the proposed case- based method can be divided into two parts： formalization of the application- matching knowledge, and inference method. The case of this knowledge consists of a series of indices to formalize the DTA application-matching knowledge and the corresponding similarity calculation methods for inference based on the case. To evaluate the performance of the proposed method, we implemented it in a software prototype of DTA modelling environment and then applied it to a DTA application of river network extraction. In the experiment we prepared 32 cases of river network extraction in the USA. The results of cross validation preliminarily show that the proposed case- based method is suitable for using the application- matching knowledge in DTA. It reduced the modelling burden greatly for users.