多核学习与用户反馈结合的WMS图层检索方法

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

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。

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.