基于关系语义的空间场景信息理解

项目名称： 基于关系语义的空间场景信息理解

项目编号： No.41471315

项目类型： 面上项目

立项/批准年度： 2015

项目学科： 天文学、地球科学

项目作者： 李梅

作者单位： 北京大学

项目金额： 87万元

中文摘要： 由于地理要素在特征空间的不可分性，缺乏有效知识学习与转换机制和知识表达模型，不能对多种地理要素的关系和语义进行综合处理，不具备高层知识推理能力，因此特征分类方法对空间数据的理解能力有限。 项目以空间关系为核心，首先，基于大比例尺GIS数据进行学习和转化，把基于几何表现的GIS数据转化为基于关系的空间场景表示，解决地理知识获取和空间场景表示问题；其次，基于知识推理和场景表示，结合高分遥感数据的低层影像特征，完成地理要素几何提取和语义解释；最后，对空间场景中各种地理要素的空间关系和语义进行综合推理与分析，消除空间、语义和关系上的不一致性，达到场景信息理解的目的。 基于关系语义的场景信息理解将解决地理空间关系组织、空间场景知识表达及理解等关键问题，发展基于关系语义的空间信息理解新机制，具有重要科学意义。在城市生态评估、地理国情监测、城市环境评价、国土资源调查、灾害监测评估等领域具有重要实用价值。

中文关键词： 形式化建模；知识表示；语义网络；地理认知

英文摘要： The classification methods are mostly used to understand geospatial information, which utilizes various features of geographical objects to classify and recognize geographical objects. However, due to the indiscernibility of geographical objects in feature space, the lack of knowledge learning and transfer mechanisms, and the disability of integratedly handling the relationships and semantics between various types of objects, the classification methods have limited ability of understanding geospatial information. This project will develop a relation-oriented mechanism to improve the ability of existing classification methods to understand spatial-scene information. It has three goals: (1) to develop automatic mechanisms to learn knowledge from high resolution GIS data, to transform GIS geometric representation to spatial-scene representation based on relationships, and to resolve the issues of acquiring geographical knowledge and representing spatial scenes; (2) to extract geometrical information and interpret semantics of geographical features by fusing knowledge inference and knowledge representation of spatial scenes; (3) to synthetically infer and analyze spatial relationships and semantics of various geographical features in spatial scenes, as well as to remove the spatial, semantic, and relational inconsistencies. Supported by spatial relationships and semantics knowledge, understanding spatial-scene information will help to organize geographical knowledge, represent spatial scenes, and apply the knowledge and representation to understanding geospatial information. Therefore, the theories and methodologies developed in this project will greatly facilitate many key technical applications, such as urban landscape ecolgy assessment, geographic national condition monitoring, urban environmental assessment, land and resource survey, and natural disaster monitoring.

英文关键词： formal method;knowledge representation;semantic network;geo-cognition