Selective Omission of Road Networks in Multi-scale Representation.

摘要

Spatial data can be represented at different scales, and multi-scale spatial representation has found wide applications. So far, there are still two main limitations with current methodologies. First, continuous transformation of spatial representation to any arbitrary scale is still not available; second, spatial data are represented at different scales for different applications and are updated separately. An ideal solution is to automatically transform the spatial representation at the largest scale to that at any smaller scale. This transformation may involve a series of operations. This study focuses on the selective omission in a road network. More specifically, three issues are addressed:;First, a road network in the database is normally stored in the form of intersections and segments. However, recognition is normally performed on roads. Thus it is very desirable to build road segments into long individual roads (called strokes). In stroke building, a total of seventeen strategies were investigated. In this investigation, a measure was proposed and statistical tests were carried out to detect any significance differences of performance using these strategies.;Second, for selective omission in a road network, two typical existing approaches were first evaluated by both quantitative analysis and visual inspection. It was found that the existing approach either performs better in a road network with linear patterns, or only performs better in a road network with areal patterns. This inspired us to develop an integrated approach. In this integrated approach, a hierarchical structure for both meshes and linear roads were first built and these two hierarchies were integrated into a single structure.;Third, there also is a need for determining the percentage of roads for selection. This process involves either some scale-related parameters or empirical models to express the relationship between a map scale and the number of roads to be selected. However, these parameters may vary between cases, and these models are not suitable for all possible cases. It is therefore very desirable to adaptively determine the percentage required for a representation at a specific scale. The back propagation neural network was adopted to give such a solution.