一种生成式神经网络的道路简化方法  

A generative neural network method for road simplification

作　　者：罗飘 许俊奎 武芳[4] 吕亚坤 庄清文 LUO Piao;XU Junkui;WU Fang;L Yakun;ZHUANG Qingwen(College of Geography and Environmental Science,Henan University,Kaifeng 475001,China;Henan Industrial Technology Academy of Spatio-Temporal Big Data,Henan University,Kaifeng 475001,China;Key Laboratory of Geospatial Technology for the Middle and Lower Yellow River Regions(Henan University),Ministry of Education,Kaifeng 475001,China;Institute of Geographical Spatial Information,Information Engineering University,Zhengzhou 450001,China)

机构地区：[1]河南大学地理与环境学院,河南开封475001 [2]河南省时空大数据产业技术研究院,河南开封475001 [3]黄河中下游数字地理技术教育部重点实验室(河南大学),河南开封475001 [4]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》2024年第10期2007-2020,共14页Acta Geodaetica et Cartographica Sinica

基　　金：国家自然科学基金(U21A2014,42271463,42301504)。

摘　　要：道路数据具有数量大、变化频率高的特点,是地理空间数据的重要组成部分,道路要素化简也是地图制图综合和空间数据更新的核心技术环节之一。传统方法基于数据点压缩、弯曲识别和现有机器学习算法在道路化简中存在稳定性差、可控性弱,自动化程度低等问题,本文在视觉思维和句法模式相结合的理论基础上,利用深度学习算法的特征挖掘能力将生成式人工神经网络模型引入道路化简领域。首先,将需化简道路数据转化为序列数据,提取其序列特征,以此构造特征数据集;然后,利用门控循环单元(gated recurrent unit,GRU)神经网络构建的Seq2Seq编码模型,将大比例尺道路数据进行嵌入形成高维语义编码,通过对语义编码的解码生成化简后的小比例尺道路数据;最后,根据弧段压缩率、长度变化率、曲线折度、缓冲区限差4个指标评估模型有效性和适用性。通过试验与传统算法对比试验表明,本文模型可应用到道路形状化简中,丰富道路化简方法,促进地图制图综合智能化发展。Road data is an important part of geospatial data due to its large quantity and high frequency of change.Road feature simplification is also one of the core technical steps of cartographic generalization and spatial data updating.Traditional methods based on data point compression,bend recognition and existing machine learning algorithms have problems of poor stability,weak controllability and low degree of automation in road simplification.Based on the theory of combining visual thinking and syntactic pattern.This paper uses the feature mining ability of deep learning algorithm to introduce generative artificial neural network model into the field of road simplification.Firstly,the road data to be simplified is transformed into sequence data,and the sequence features are extracted to construct the feature data set.Secondly,the Seq2Seq coding model constructed by the gated recurrent unit(GRU)neural network is used to embed large-scale road data to form high-dimensional semantic coding.The simplified small-scale road data is generated by decoding the semantic coding.Finally,the effectiveness and applicability of the model were evaluated according to four indexes:compression rate of arc segment,change rate of length,bend of curve and buffer limit difference.The experimental results show that the proposed model can be applied to road shape simplification,enrich road simplification methods,and promote the comprehensive and intelligent development of map drawing.

关 键 词：生成式神经网络 道路简化 深度学习 句法模式识别 Seq2Seq编码模型 地图制图综合 

分 类 号：P283[天文地球—地图制图学与地理信息工程]