典型深度学习算法在地图图像特征点匹配上的适用性分析  

Analysis of the applicability of typical deep learning algorithms for map feature point matching

作　　者：马瑞苑 王勇[1] 车向红 杜凯旋 MA Ruiyuan;WANG Yong;CHE Xianghong;DU Kaixuan(Chinese Academy of Surveying and Mapping,Beijing 100036,China;School of Resource and Environmental Sciences,Wuhan University,Wuhan 430072,China)

机构地区：[1]中国测绘科学研究院,北京100036 [2]武汉大学资源与环境科学学院,武汉430072

出　　处：《测绘科学》2024年第8期173-180,共8页Science of Surveying and Mapping

基　　金：国家重点研发计划项目(2022YFC3005700);自然资源部业务专项(121136000000180004,102121241360000000984)。

摘　　要：为探究不同地图要素对典型深度学习图像特征点匹配算法的影响,分析不同算法在地图图像特征点匹配任务中的适用性,该文基于标准地图数据,选择地理要素、地图比例尺、地图符号、制图区域为研究变量,使用单一变量原则构建地图特征点匹配样本库,分别应用典型深度学习算法SuperGlue、GlueStick、Patch2Pix进行地图图像特征点匹配,使用地图匹配率、平均正确匹配对数、平均匹配正确率3个指标进行结果评价。结果表明,SuperGlue算法适用于较小制图区域中具有地图符号差异的地图图像特征点匹配;GlueStick算法适用于中等和较大制图区域中具有地图比例尺差异的地图图像特征点匹配,以及具有制图区域差异的地图图像特征点匹配;Patch2Pix算法适用于较小制图区域中具有地图比例尺差异的地图图像特征点匹配,中等和较大制图区域中具有地图符号差异的地图图像特征点匹配,以及具有地理要素差异的地图图像特征点匹配。To explore the impact of different map features on typical deep learning image feature point matching algorithms and analyze the suitability of different algorithms for map image feature point matching tasks,this paper is based on standard map data and focuses on geographical elements,map scale,map symbols and cartographic area as the research variables,constructs a map feature point matching sample library based on the principle of a single variable.Applying typical deep learning algorithms SuperGlue,GlueStick,and Patch2Pix for map image feature point matching.The results are evaluated using three metrics:map matching rate,average logarithm of correct matches,and average matching accuracy rate.The results show that the SuperGlue algorithm is suitable for map feature point matching with map symbols differences in smaller cartographic area.The GlueStick algorithm is suitable for map feature point matching with map scale differences in medium and large cartographic area,and map feature point matching with cartographic area differences.The Patch2Pix algorithm is suitable for map feature point matching with map scale differences in smaller cartographic area,map feature point matching with map symbols differences in medium and large cartographic area,and map feature point matching with geographical elements differences.

关 键 词：地图特征点匹配 SuperGlue Patch2Pix GlueStick 

分 类 号：P208[天文地球—地图制图学与地理信息工程]