测绘学报 ›› 2019, Vol. 48 ›› Issue (10): 1320-1330.doi: 10.11947/j.AGCS.2019.20180410

• 地图学与地理信息 • 上一篇    下一篇

多核学习与用户反馈结合的WMS图层检索方法

李牧闲1, 桂志鹏1,2,3, 成晓强4, 吴华意2,3, 秦昆1   

  1. 1. 武汉大学遥感信息工程学院, 湖北 武汉 430079;
    2. 武汉大学测绘遥感信息工程国家重点实验室, 湖北 武汉 430079;
    3. 地球空间信息技术协同创新中心, 湖北 武汉 430079;
    4. 湖北大学资源环境学院, 湖北 武汉 430062
  • 收稿日期:2018-08-31 修回日期:2019-01-04 出版日期:2019-10-20 发布日期:2019-10-24
  • 通讯作者: 桂志鹏 E-mail:zhipeng.gui@whu.edu.cn
  • 作者简介:李牧闲(1996-),女,硕士生,研究方向为网络地图服务检索。E-mail:limuxian@whu.edu.cn
  • 基金资助:
    国家自然科学基金(41501434;41501443);深圳大学空间信息智能感知与服务深圳市重点实验室开放基金

A content-based WMS layer retrieval method combining multiple kernel learning and user feedback

LI Muxian1, GUI Zhipeng1,2,3, CHENG Xiaoqiang4, WU Huayi2,3, QIN Kun1   

  1. 1. School of Remote Sensing and Information Engineering, Wuhan University, Wuhan 430079, China;
    2. State Key Laboratory of Information Engineering in Surveying, Mapping and Remote Sensing, Wuhan University, Wuhan 430079, China;
    3. Collaborative Innovation Center of Geospatial Technology, Wuhan 430079, China;
    4. Faculty of Resources and Environmental Science, Hubei University, Wuhan 430062, China
  • Received:2018-08-31 Revised:2019-01-04 Online:2019-10-20 Published:2019-10-24
  • Supported by:
    The National Natural Science Foundation of China (Nos. 41501434;41501443);The Open Foundation of Shenzhen Key Laboratory of Spatial Smart Sensing and Service of Shenzhen University

摘要: 现有WMS检索方法多基于服务元数据文本匹配,缺乏对地图内容的“感知”,无法应对元数据缺失或图文不符的情境。本文设计了一种多特征多核学习和用户反馈结合的WMS图层检索方法,利用多核学习算法融合颜色、形状与纹理特征,实现图层分类和相似度排序,并通过采集检索结果展示页面中的兴趣图层标记进行用户反馈,以优化分类模型和提高检索精度。试验结果表明,该方法查准率高且检索用时较短,能够与现有基于文本检索的地理信息资源门户集成,实现WMS的快速检索与有效发现。

关键词: 地理信息资源检索, 多核学习, 多特征融合, 用户反馈, 网络地图服务

Abstract: To facilitate the discovery and use of geographic information, it is necessary to design an effective retrieval strategy to locate the map layers that customers want from massive WMS resources. Existing text-based WMS retrieval strategies are unable to solve the problems of metadata loss and inconsistency between pictures and metadata text, without considering map content. The visual similarity between maps is used to design a WMS layer retrieval method that combines multi-feature multiple kernel learning and user feedback to help users search for desired WMS layers. Color, shape and texture features are fused by multiple kernel learning to classify and rank layers according to similarity. A feedback mechanism is also established in the retrieval strategy, which is an effective guarantee that improves accuracy by collecting user-marked layers. Various kinds of WMS layers are selected to calculate the precision ration, analyze the time cost, and validate the retrieval feedback mechanism. The experimental results of selected WMS layers verified that the proposed algorithm is fast and highly precise. It can be integrated with existing text-based retrieval and discovery portals of geographic information.

Key words: geographic information resource retrieval, multiple kernel learning, multi-feature fusion, user feedback, web map service

中图分类号: