Methodology Of "Text-sketch-scene" Conversion For Geographical Scene

Natural language and virtual geographical scene, as two important representation types of geographic scene, have different advantages in the aspects such as the abundance of data, the intuitive expression, cognitive habits, and the popularity of applications. It has become a hot research issue in geographic information science on how to bridge the gap between natural language and virtual geographical scene. This paper supported by China National Natural Science Foundation "Reconstruction of Virtual Geographical Scene in Natural Language", aims to solve the expressing mechanism of geographical scene in natural language from the aspect of structured text, spatial sketch and virtual geographical scene, in order to reveal the natural language description habits of geographic scene, and achieve the qualitative to quantitative conversion of geographical scene. This study will further promote the universal application of virtual geographical scene in geography, smart city, film production, visualization of historical events etc. The main research topics and conclusions in this dissertation are discussed as follows:(1) Structured representation frame of geographic scene in natural languageBased on analysis of the expression methods of geographic scene features in virtual geographical scene and natural language, the structured model is presented with frame knowledge representation method on three levels i.e. geographical feature, spatial relation, geographical scene. Frame knowledge representation could not only describe the spatial structure features of geographic scene, but also identify the semantic relation between the features.(2)"Text to Sketch" conversion with geographical knowledge basesSpatial layout sketch of geographic scene could not only express the distribution of geographical features in the scene, but also reflect the qualitative semantic relations between geographic features in the scene. Hence it’s a more intuitive and visual way to represent geographical scene than natural language. The knowledge bases of geographic scene are explored including:geographical feature classification scheme, geographical common knowledge base, and geographical entity symbol base and spatial structure templates bases of geographical scene. The spatial relation semantic models are introduced to solve the problem of rough spatial layout of geographic scene. Then, the colony algorithm is developed to implement precise spatial layout, in order to realize conversion of text and sketch of geographic scene with human-computer assistance.(3)"Text to Scene" conversion of geographical scene bases spatial similarity mappingBased on the multi-level spatial similarity matching, a conversion method of virtual geographical scene and the geographical scene described in natural language is proposed, of which the similarity calculation of geographical concepts, spatial relation and spatial structure are proposed, and some cases are illustrated to evaluate the performance. The study demonstrates that there are great differences between natural language and virtual geographic scene, and their conversion is to find spatial semantic similarity essentially. Spatial layout sketch just provides distinct spatial structure features of geographical scene which are hidden described in natural language. Based on the spatial layout sketch,"Text to Scene" conversion can be more effectively implemented.