长江科学院院报 ›› 2014, Vol. 31 ›› Issue (8): 98-102.DOI: 10.3969/j.issn.1001-5485.2014.08.0192014,31(08):98-102,121

• 信息技术应用 • 上一篇    下一篇

一种基于LDA的高分辨率遥感影像检索方法

沈盛彧1,刘哲2,张平仓1,张彤3,吴华意3,陈小平1   

  1. 1.长江科学院 水土保持研究所,武汉 430010;
    2.长江水利委员会 网络与信息中心,武汉 430010;
    3.武汉大学 测绘遥感信息工程国家重点实验室,武汉 430079
  • 收稿日期:2013-06-13 修回日期:2014-08-12 出版日期:2014-08-01 发布日期:2014-08-12
  • 作者简介:沈盛彧(1984- ),男,湖北武汉人,工程师,博士,主要从事高分辨遥感影像处理与水土保持研究,(电话)027-82926365(电子信箱)shshy.whu@gmail.com。
  • 基金资助:
    国家自然科学基金资助项目(41271400);国家973计划资助项目(2012CB719906);中央级公益性科研院所基本科研业务费(CKSF2014024/TB,CKSF2012055/TB,CKSF2012044/TB)

A Method of High-resolution Remote Sensing ImageRetrieval Based on LDA

SHEN Sheng-yu1, LIU Zhe2, ZHANG Ping-cang1, ZHANG Tong3, WU Hua-yi3, CHEN Xiao-ping1   

  1. 1.Department of Soil and Water Conservation, Yangtze River Scientific Research Institute,Wuhan 430010, China;
    2.Network and Information Center, Changjiang Water Resources Commission,Wuhan 430010, China;
    3.State Key Laboratory of Information Engineering in Survey,Mapping and Remote Sensing, Wuhan University, Wuhan 430079, China
  • Received:2013-06-13 Revised:2014-08-12 Online:2014-08-01 Published:2014-08-12

摘要: 传统遥感影像检索存在时间和人工成本高、内容信息描述太粗略等问题,更未充分考虑语义信息,难以应对高分辨率遥感影像的海量地物类型及其复杂关系。借鉴信息检索的思想,引入计算机视觉领域的视觉特征和自然语言处理领域的概率主题模型,提出了一种基于LDA(Latent Dirichlet Allocation)的高分辨率遥感影像检索方法。通过一组多主题个数的高分辨率遥感影像检索实验证明,该方法在主题个数较少时,能达到较好的检索效果,较高的查准率,而且在主题个数继续增加时,能使查准率保持在0.9左右。

关键词: LDA, 高分辨率遥感影像, 遥感影像检索, 视觉特征

Abstract: Traditional methods of remote sensing image retrieval cannot handle high-resolution images with huge amounts of surface feature types and complex relations. It requires high cost of time and labor, gives sketchy content description, and cannot adequately consider the semantic information. Inspired by text information retrieval, the visual features in computer vision and the probabilistic topic model in natural language processing are introduced to present a method of retrieving high-resolution remote sensing image based on LDA (Latent Dirichlet Allocation). Retrieval experiments on high resolution remote sensing images with multiple topic numbers suggest that even when the number of topic is small, good retrieval results and high precision can be achieved. As the topic number increases, the precision rate remains at about 0.9.

Key words: LDA, high-resolution remote sensing image, remote sensing image retrieval, visual features

中图分类号: