面向建筑物数据的密度聚类算法研究  

Comparative study on clustering adaptability of DBSCAN extended algorithmin planar buildings

作　　者：孟妮娜[1] 高晨博 王正阳 李金秋 MENG Nina;GAO Chenbo;WANG Zhengyang;LI Jinqiu(School of Geological Engineering andMapping,Chang’an University,Xi’an 710061,China)

机构地区：[1]长安大学地质工程与测绘学院,西安710061

出　　处：《测绘科学》2022年第11期204-214,共11页Science of Surveying and Mapping

基　　金：国家自然科学基金项目(41501498);陕西省自然科学研究计划项目(2021JM-155)

摘　　要：针对DBSCAN算法需要人为输入参数Eps和MinPts以及该算法在聚类过程中很少顾及到建筑物本身的结构和分布特征等问题,该文从经典的DBSCAN的3种扩展算法入手,引入加权质心距离的概念代替原算法中的欧氏距离,在降低算法敏感度的同时使得聚类分析的结果更加符合人的视觉认知。实验结果表明:OPTICS算法可以通过可达距离图展示数据集的分布结构,以便我们进行参数设置,但算法运行效率低,同时不能识别密度差异较大的建筑物群;HDBSCAN算法可以发现不同密度的建筑物群,但对于边界区域的建筑物识别效果并不理想;DDBSCAN算法在识别不同密度且相邻分布的建筑物群上表现出一定的优越性。In view of the problems that DBSCAN algorithm requires artificial input parameters Eps and MinPts,and that the algorithm seldom takes into account the structure and distribution characteristics of buildings themselves in the clustering process,this paper starts with three classical DBSCAN extended algorithms,introduces the concept of weighted centroid distance to replace the euclidean distance in the original algorithm,which reduces the sensitivity of the algorithm and makes the results of cluster analysis more consistent with human visual cognition.Our results show that the OPTICS algorithm can show the distributed structure of the dataset through the reachable distance graph for parameter setting,but the algorithm is inefficient and cannot identify buildings with large density differences.HDBSCAN algorithm can find buildings of different densities,but it is not good for building recognition in boundary region.DDBSCAN algorithm has some advantages in clustering effect of different density and adjacent building groups.

关 键 词：制图综合 空间聚类 建筑物分布 相似性分析 

分 类 号：P208[天文地球—地图制图学与地理信息工程]