数据驱动的海量遥感影像高效信息挖掘

项目名称： 数据驱动的海量遥感影像高效信息挖掘

项目编号： No.91338113

项目类型： 重大研究计划

立项/批准年度： 2014

项目学科： 无线电电子学、电信技术

项目作者： 夏桂松

作者单位： 武汉大学

项目金额： 80万元

中文摘要： 针对空间信息网络重大研究计划的需求，紧跟空间信息处理自动化和实时化的发展趋势，本项目从海量遥感数据的相关性和数据空间的冗余性入手，利用数据驱动的方法，研究海量遥感的高效信息挖掘问题。首先，从遥感影像的低层特征入手，结合遥感的稀疏特性，针对遥感影像的语义挖掘问题，采用基于深度学习的方法，建立遥感影像的语义层次挖掘模型，揭示遥感影像从观测数据到场景语义的认知规律；其次，在语义描述基础上，针对遥感影像场景更新的实际和在轨处理的潜在需要，建立 “高效检索-快速匹配-场景语义迁移”的信息挖掘模型，实现顾及增量数据的场景信息挖掘。 本项目的研究成果能为高分辨率对地观测环境下的高效数据挖掘提供理论和算法支持，同时有望为动态的遥感影像信息挖掘和在轨遥感应用算法的研究提供算法和技术支撑。

中文关键词： 场景分类；场景数据集；语义挖掘；深度学习；半监督/非监督特征学习

英文摘要： Aiming at the demand of the significant research plan in space information network, following the development trend of the automatic and real-time spatial information processing ,this program begin from the correlation of the project from massive remote sensing data and the redundancy of data space, and will do research on efficient information mining problems of mass remote sense images by using the method of data driven. Firstly,starting from the low-level features of remote sensing image, combined with the sparse characteristics of remote sensing,and aiming at the semantic information mining problems of remote sensing images , we propose a hierarchical semantic mining model of remote sensing images by adopting the method of deep learning , in order to reveal the cognitive principle of remote sensing image from the observed data to the semantic scene; Secondly, based on semantic description, in view of the needs of remote sensing image scene updating and potential requirements of the on-boarding processing, we propose an efficient information mining model of "retrieval - quick match - scene semantic transfer ", by taking the scene information mining of incremental data into account. This study can provide efficient data mining algorithms support for high resolution earth observation environment, as well as the

英文关键词： scene classification；scene datasets；semantic mining；deep learning；semi-supervised/unsupervised feature learning