测绘通报 ›› 2018, Vol. 0 ›› Issue (11): 30-35.doi: 10.13474/j.cnki.11-2246.2018.0345

• 学术研究 • 上一篇    下一篇

自适应聚类学习的道路网提取方法

陈光1,2, 薛梅1,2, 陈良超1,2, 眭海刚3   

  1. 1. 重庆市勘测院, 重庆 401121;
    2. 智慧城市时空大数据重庆市工程研究中心, 重庆 401121;
    3. 武汉大学测绘遥感信息工程国家重点实验室, 湖北 武汉 430079
  • 收稿日期:2018-02-24 出版日期:2018-11-25 发布日期:2018-11-29
  • 作者简介:陈光(1984-),男,博士,高级工程师,主要研究方向为遥感影像处理与信息提取、三维GIS应用。E-mail:teddycg@qq.com
  • 基金资助:

    国家重点研发计划(2018YFB0505400);重庆市社会事业与民生保障科技创新专项(cstc2017shmsA120008);2018年重庆市技术创新与应用示范(产业类)重大主题专项(cstc2018jszx-cyztzx0057)

Extraction Method of Road Network Based on Adaptive Clustering Learning

CHEN Guang1,2, XUE Mei1,2, CHEN Liangchao1,2, SUI Haigang3   

  1. 1. Chongqing Survey Institute, Chongqing 401121, China;
    2. Chongqing Engineering Research Center of Spatiotemporal Big Data in Smart City, Chongqing 401121, China;
    3. State Key Laboratory of Information Engineering in Surveying, Mapping and Remote Sensing, Wuhan University, Wuhan 430079, China
  • Received:2018-02-24 Online:2018-11-25 Published:2018-11-29

摘要:

针对复杂场景道路样本特征异质性导致的分类器过拟合问题,本文提出了一种基于自适应聚类学习的道路网自动提取方法。方法以高分辨率遥感影像和旧时相路网矢量数据为输入,在矢量数据引导下自动获取对象化的正负样本;提出了一种道路样本自适应聚类策略,根据集合内特征分布情况重组样本,并分别训练分类器进行道路提取;最后利用多数投票方法集成多组道路提取结果。基于大场景数据的试验结果表明,本文方法能够有效地顾及不同特征的道路对象,定量的试验比对结果进一步表明了方法的适用性。

关键词: 道路提取, 聚类, 遥感影像, 导航路网, 支撑向量机, 面向对象

Abstract:

Aiming at the problem of over fitting of classifiers caused by the heterogeneity of road samples in complex scenes,this paper proposes an automatic road-network extraction method based on adaptive clustering learning.Method takes high resolution remote sensing image and older road-network as input data.Positive and negative samples are automatically acquired by the guiding of road-network.Then authors present an adaptive clustering method for road samples,which is based on the feature distribution in sample-set.Road extraction results are integrated by the majority voting method.The experimental results based on large scene data show that the proposed method can effectively take into account the different features of road objects.Quantitative experimental comparison of the results further shows the applicability of the method.

Key words: road extraction, clustering, remote sensing image, navigation road network, SVM, object-oriented

中图分类号: