建筑物形状特征分析表达与自适应化简方法

An adaptive building simplification approach based on shape analysis and representation

建筑物化简是地图制图领域关注的热点问题之一。集成不同算法构建形状特征自适应的化简模型是应对建筑物多样化形态的有效策略,但当前相关研究主要从局部结构模式或化简结果评价展开,缺乏对形状结构的整体分析视角和深层次认知。本文提出一种深度学习支持下的形状自适应建筑物化简方法。首先,利用图卷积自编码网络对建筑物形状进行深度认知,提取隐含在边界节点分布中的形状特征并进行编码表达;然后,通过监督学习方法建立形状编码与化简算法之间的映射关系,从而实现依据输入建筑物的形状特征选择适宜化简算法的自适应机制。试验表明,本文方法的化简结果在位置、方向、面积和形状保持指标上总体优于单一算法,具备较好的理论与应用价值。

Building simplification is one of the long-standing challenges in cartography.Establishing a hybrid simplification mechanism based on shape characteristics is an effective strategy to adapt to the diversity and complexity of building shapes.However,existing studies mainly focus on local structure analysis or simplified result evaluation,lacking analytical perspective and deep understanding of the overall shapes.This study proposed a shape-adaptive building simplification approach using deep learning.First,a graph convolutional autoencoder was designed to encode the shape features implicated in the boundary of each building.Then,the mapping relationship between the shape encodings and four candidate simplification algorithms was established using a supervised learning model,so as to realize an adaptive mechanism of selecting the appropriate simplification algorithm according to the shape characteristics of the input building.Experimental results show that our approach performs better than the standalone application of existing algorithms in measuring the changes of position,orientation,area,and shape,and have good theoretical and practical significance.