基于改进邻域聚合的路网图嵌入匹配方法

A Method of Road Network Matching Using Graph Embedding via Improved Neighbor Aggregations

道路网作为一种重要的交通基础设施,路网数据的及时更新对交通管理、应急救援和城市规划等领域有重要应用意义。通过路网匹配来确定不同来源的路网数据中要素间的对应关系,既是实现路网更新的重要技术途径,也为众源路网数据质量评估等任务提供技术支撑,因而备受地理信息领域学者的关注。传统的路网匹配方法主要通过路网数据的几何和拓扑属性来度量路网结构的相似性,以此确定路网要素的匹配关系。但人工设计的特征和阈值易受专家经验局限,使其在复杂路网结构下性能下降。近年来,基于图神经网络的路网数据建模成为研究热点,已在多个路网建模任务中取得优异性能。但现有方法多采用在图拓扑结构上直接进行邻域聚合的方式,学习路网结构的嵌入表示,未在这一关键步骤中考虑路网要素的空间关系,没能充分利用图神经网络的表示学习能力。为此,本研究面向路网匹配任务,采用空间显式建模的思想,提出一种基于改进的邻域聚合图嵌入学习方法。首先,构建路网数据的道路图模型并提取几何、语义和位置特征。然后,基于GraphSAGE框架,提出空间、分类和混合3种邻域聚合算子,在邻域聚合操作中引入路网要素空间关系、属性类型的计算。最后,利用图节点嵌入的相似度确定路网要素的匹配关系。为验证本文方法的有效性,利用真实路网数据开展了充分实验,本文方法在实验区数据上的各项指标取得最优表现,比基线图神经网络方法的匹配正确率提升11%以上、召回率提升6.8%以上。并对路网图嵌入特征进行分析,从图嵌入结构和嵌入路网结构两方面,探讨了改进邻域聚合对图嵌入表示能力的作用,为进一步提升图神经网络路网建模提供了新视角。

As an important transportation infrastructure,the timely updating of road network data is of great significance in the fields of traffic management,emergency response,and urban planning.Road network matching that determines the correspondence between the features of road network data from different sources serves this purpose.It also provides technical support for tasks such as the quality assessment of crowdsourced road network data,which has attracted a lot of attention in the field of geographic information.However,traditional road network matching methods mainly measure the similarity of road network structure through the geometric and topological attributes of road network data to determine the matching relationship of road network elements.Such methods with manually designed features and thresholds are easily limited by experts'experience,which degrades their performance under complex road network structures.In recent years,road network data modeling based on graph neural networks has become a research hotspot and has achieved excellent performance in several road network modeling tasks.However,most of the existing methods use direct neighborhood aggregation on the graph topology to learn the embedded representation of the road network structure,without considering the spatial relationship of road network features in this key step,and failing to make full use of the representation learning capability of graph neural networks.For this reason,this study proposes an improved neighborhood aggregation that performs a spatially explicit graph-based embedding learning method for road network matching.First,a road graph model of the road network data is constructed,and geometric,semantic,and location features are extracted.Then,based on the GraphSAGE framework,three kinds of neighborhood aggregation operators(i.e.,spatial,classified,and hybrid)are proposed,and the computation of spatial relationships and attribute types of road network features is introduced in the neighborhood aggregation operations.Finally,the similarity of graph node embedding is utilized to determine the matching relationship of road network features.To verify the effectiveness of the proposed method,extensive experiments are carried out using real-world road network data.The proposed method achieves the optimal performance in all metrics on the test data of the study region,which improves the matching correctness rate by more than 11%and the recall rate by more than 6.8%compared to the baseline graph neural network method.Furthermore,the road network graph embedding features are analyzed from the aspects of graph embedding structure and embedded road network structure,which helps explore the role of improved neighborhood aggregation on the graph embedding representation capability and provides a new perspective for further improving the graph neural network road network modeling.