稀疏认知下的遥感影像在轨变化检测与目标提取

从海量、动态、高维、异构的复杂空间数据中快速准确地提取目标信息，是当前信息科学领域的难题之一。本课题针对星载高分辨率、宽覆盖的多传感器产生的海量数据无法及时下传的瓶颈，借鉴生物视觉的稀疏认知机理，挖掘目标的稀疏性与变化有限等先验，借助稀疏编码、显著注意，在线学习、分布式协同等技术手段，探索遥感影像在轨稀疏表征与处理的新理论与新方法。期望通过稀疏序列表征、计算与理解等内容的研究，突破高分辨遥感卫星在轨几何定标与配准、在轨时空影像匹配与融合、在轨多时相影像变化检测等关键技术，实现遥感卫星海量数据的在轨智能化与高效处理，建立遥感影像变化检测与目标提取结果的质量评价体系，为有效支撑高分辨率对地观测等国家重大专项的发展奠定理论基础。

Extracting the target information accurately and rapidly from the complex remote sensing data that has the characteristic of large volume, dynamic, high-dimensional and heterogeneous, is one of the most difficult problems in the field of information science. In order to break through bottlenecks of the link transmission in space information network, in this project we explore the sparsity and limited variation of targets to develop new theories and methods for the efficient representation,computation and application of the massive spatial information,via sparse coding, saliency attention, online learning and cooperative learning technologies that are inspired by the sparse cognition of visual sensing systems. Based on these studies, we hope to achieve the in-orbit geometrical orbit determination and accurate allocation, in-orbit time-space registration and fusion, in-orbit change detection and targets extraction. By the above researches, we aim at realizing on-orbit,intelligent and effective processing of massive data from remote sensing satellite, and establish the quality evaluation system of change detection and target extraction of remote sensing images, thus laying theoretical foundations to the development of major special projects, such as high resolution earth observation and so on.