稀疏表达和跨领域学习的高光谱遥感图像亚像元目标探测研究

项目名称： 稀疏表达和跨领域学习的高光谱遥感图像亚像元目标探测研究

项目编号： No.61471274

项目类型： 面上项目

立项/批准年度： 2015

项目学科： 无线电电子学、电信技术

项目作者： 杜博

作者单位： 武汉大学

项目金额： 82万元

中文摘要： 高光谱遥感图像的亚像元目标探测是学术界普遍关注的研究方向。其中的关键问题在于：1）目标特征的有效表达，既能保持目标判别信息的同时又能避免背景信息干扰；2）目标探测方法的适应性，使目标探测器可以适用于各种不同数据；3）目标亚像元空间信息的准确还原，克服空间分辨率低导致的目标形状信息缺失。针对这三个问题，本项目从特征表达-目标探测-亚像元解译的思路出发，围绕稀疏表达的目标探测模型、目标探测方法的跨领域学习和目标地物的亚像元定位开展深入研究，形成稀疏表达和跨领域学习的亚像元目标探测理论。本项研究成果不仅可以提升高光谱遥感图像亚像元目标探测能力，也对促进我国高光谱遥感在农业监测、军事侦察、城市规划、矿产勘查等领域的广泛应用具有重大的实际意义。

中文关键词： 地物目标；光学遥感；海面目标

英文摘要： Sub-pixel targe detection from hyperspectral remote sensing images draws great interest in the academic fields. The key problems include: 1) Effectively represent the targets, by maintaining the target discriminative information and avoiding the background disturbing informatin; 2) Target detectors' adaptation ability, so as to perform well on different datasets; 3) Accurately restore the sub-pixel information of the targets, in order to avoid the loss of spatial information of targets due to the spatial resolution. To solve the above three problems, this project would research on the following respects: sparse representation based target detection models, domain adaptation of target detectors, and sub-pixel mapping of targets. In this way, a sparse representation and domain adaptation based sub-pixel target detection theory can be constructed. The research findings can not only remarkably increase the ability to detect sub-pixel targets from hyperspectral images, but can also promote the applications of hyperspectral remote sensing in many fields, such as agriculture monitoring, military reconnaissance, city plan, mineral exploration, and so on.

英文关键词： land ground objects;optical remote sensing;sea surface objects