基于图神经网络的多尺度网状河系分类匹配方法  

A multi-scale mesh river system classification matching method based on graph neural network

作　　者：黄哲琨 钱海忠[1] 蔡中祥[1] 王骁[1] 王俊威 孔令辉 HUANG Zhekun;QIAN Haizhong;CAI Zhongxiang;WANG Xiao;WANG Junwei;KONG Linghui(Institute of Geospatial Information,University of Information Engineering,Zhengzhou 450001,China)

机构地区：[1]信息工程大学地理空间信息学院,河南郑州450001

出　　处：《测绘学报》2025年第2期371-384,共14页Acta Geodaetica et Cartographica Sinica

基　　金：国家自然科学基金(42271463;42101453;42371461);河南省杰出青年自然科学基金(212300410014)。

摘　　要：多尺度网状河系匹配是水系数据集成、融合与更新的重要组成部分。鉴于现有网状河系匹配方法未对匹配模式进行预先识别,并缺乏针对性的匹配策略,本文提出基于图神经网络的多尺度网状河系分类匹配方法。首先,将大比例尺网状河系构建为图结构,将其与小比例尺河系之间的匹配模式作为节点的标注,并计算节点特征;然后,利用图神经网络对节点特征进行采样和聚合,建立起河段特征与匹配模式之间的映射关系;最后,根据河系中各河段的匹配模式类别,对其采取相应的匹配策略。试验结果表明,本文方法有效提高了网状河系的匹配精度,具备较好的理论与应用价值。Multi-scale mesh river system matching is an important part of river system data integration,fusion and update.In view of the fact that the existing mesh river system matching methods do not pre-identify the matching patterns and lack a targeted matching strategy,this paper proposes a multi-scale mesh river system classification matching method based on graph neural networks.Firstly,constructing the large-scale mesh river system as a graph structure,label the matching patterns between it and the small-scale river system as nodes,and compute the node features;and then the graph neural network is used to sample and aggregate the node features to establish the mapping relationship between the river segment features and matching patterns;finally,according to the category of the matching patterns of each river segment in the river system,the matching strategy is adopted accordingly.The experimental results show that the method in this paper effectively improves the matching accuracy of the mesh river system,and has good theoretical and application value.

关 键 词：多尺度数据 匹配模式 匹配策略 网状河系 图神经网络 

分 类 号：P208[天文地球—地图制图学与地理信息工程]