互联网级地理媒体搜索模型与方法研究

The Internet geographical media bring opportunities and challenges to GIS, by its magnanimity of complication, multiple heterogeneous sources, and the spatiotemporal correlation. How to establish the uniform description and high dimension indexing for the unstructured and dynamic distributed geographical media, as well to precede efficiently analysis and mining, and then lead to acquiring the real-time and complete search results, becomes the crucial and fundamental problem for the project. Therefore, we firstly study on spatial attributes perception, correlation modeling and consistent representation methods, and faced global establish the geographical media representation and organizational structure; by the intermedia, give the geographical media uniform description, and establish the extension type of geographical media modeling methods and multi-dimensional indexing structure; and then, based on data driven study the pattern discovery of geographical media, by combining the geography calculation and cognition calculation, approach to the valuable mining of geographical media. It is innovatively put forward: 1) by adopting the relevance and complementarity between physical space and network space, establish the complex application environment adapted definitions and description methods of geographical media objects; 2) by pre-studying middle-level spatial attributes, implement the real-time mapping to high-level concept space and also establish the extension type of geographical media modeling frame; 3) the method that base on hidden machine learning and hidden data mining, by combining the geography calculation and cognition calculation, approach to the further mining of geography valued. The aim of this project is to make a breakthrough to the models and methods of spatial media big data, and laying the foundation of the ubiquitous Geographic Information Services.

海量庞杂、异质多源、时空关联的互联网地理媒体为GIS带来了机遇与挑战。其关键在于：如何为非结构化、动态分布的地理媒体建立统一描述与索引，并进行高效的分析与挖掘，从而实时、完备地返回搜索结果？为此，首先研究空间属性感知、关联建模和一致性表达方法，建立面向全局的地理媒体表达与组织框架；并通过中间媒体，对地理媒体统一描述，建立可伸缩地理媒体建模方法与索引结构；进而研究基于数据驱动的地理媒体模式发现，将地理计算与认知计算相结合，实现地理媒体的价值挖掘。创新性地提出：1) 利用物理空间与网络空间的关联与互补，建立可适应复杂应用的地理媒体对象定义和描述方法；2) 通过中间属性的预学习，实现向高层概念空间的即时映射，建立可伸缩的地理媒体建模框架；3) 基于隐机器学习和隐数据挖掘方法，将地理计算与情感认知计算结合，挖掘地理价值。本课题预期在空间媒体大数据模型和方法上取得突破，为泛在地理信息服务奠定基础。

随着传感器技术和移动互联网的发展，采集非测绘空间数据日益便捷。不断涌现的多样终端、异构网络、海量数据等为GIS带来了机遇和挑战。针对海量庞杂、异质多源、时空关联的网络地理媒体，如何进行高效处理与语义理解，是本项目的基本出发点。为此，首先通过研究地理媒体的跨模态特征抽取、基于视觉空间认知的地理媒体学习、以及图像场景语义识别等，实现地理媒体的空间属性感知、关联建模和一致性表达，建立地理媒体表达与组织框架；并通过地理标签特定特征的多目标回归集成算法、多类别不平衡数据分类方法、树形动态聚类算法等的研究，建立可伸缩地理媒体建模方法与索引结构；进而研究地理媒体关联规则与情感计算等，将地理计算与认知计算相结合，实现地理媒体的价值挖掘。在此基础上，通过大规模地理影像检索、特定领域的目标检测等应用实践，进行模型和方法的有效性验证。其特色在于：1）利用地理空间感知与网络空间感知的关联与互补，建立可适应复杂应用的地理媒体对象定义和描述方法；2）通过媒体中间件的预学习，以及多标签特征分类、多目标树形聚类等，实现向高层概念空间的映射，建立可伸缩的地理媒体建模框架；3）将空间认知与情感认知计算结合，挖掘认知行为的地理价值。其科学意义在于：将复杂的互联网地理媒体数据化繁为简，高效提炼出满足GIS和其他行业需求的信息和知识，形成位置驱动的非测绘数据采集与应用模型和方法框架，构筑大数据时代的地理信息服务新模式。本课题发表论文22篇(其中SCI/EI检索18篇)，授权发明专利4项，获得科技奖励1项，地方标准1项，培养硕士生10人。

内容获取失败，请点击重试