摘要
针对传统道路网重要度评价模型忽略邻居节点对道路重要度影响等问题,该文提出了基于图卷积神经网络的选取方法。该方法将道路选取问题作为道路图节点的二类分类问题,从谱域空间的角度构建了图卷积算子,采用邻居节点随机采样的方式来固定节点局部结构,将道路节点的度、接近中心性、中介中心性和stroke长度作为节点特征输入至图卷积神经网络,并输出道路节点的分类,最终完成道路网的选取。实验结果表明,该方法选取效果较好、效率较高。
According to the shortage of current research that ignoring the effect from the neighbor node and the road density when calculating the node importance degree,a method based on graph convolution network was proposed in the paper.The road selection problem was taken as a class classification problem of road graph nodes.Then the graph convolution network was built up by the view of spectral domain.The random sampling of neighbor node was used for fixing local structure.Next,the stroke length,degree,closeness centrality and betweenness centrality were selected as the input of the network.Finally,some road data were used for experimental verification,and the results showed that the proposed method could effectively classify the road note.
作者
马超
熊顺
蒋丹妮
MA Chao;XIONG Shun;JIANG Danni(Xi’an Institute of Surveying and Mapping,Xi’an 710054,China;National Key Laboratory of Geo-Information Engineering,Xi’an 710054,China)
出处
《测绘科学》
CSCD
北大核心
2022年第12期200-205,215,共7页
Science of Surveying and Mapping
关键词
图卷积神经网络
道路网选取
深度学习
图数据
graph convolution network
road network selection
deep learning
graph data
