基于字典学习的遥感影像超分辨率融合方法

发布时间：2018-05-08 16:41

本文选题：QuickBird + 字典学习　；参考：《国土资源遥感》2017年01期

【摘要】：鉴于多源遥感影像融合受现有分辨率的限制,结合稀疏表示理论,提出了一种基于字典学习的遥感影像超分辨率融合方法,可将多光谱影像的空间分辨率提升到全色影像空间分辨率的1倍或2倍。在遥感影像融合框架下,首先建立学习字典,利用冗余字典对影像稀疏表示,重构超分辨率;然后采用Gram-Schmidt(GS)光谱锐化法,融合得到超分辨率多光谱影像。利用QuickBird数据对提出的方法进行3个实验,结果都表明本文方法相对传统融合方法、传统超分辨率方法和其他字典学习方案具有一定优势,适用于遥感影像超分辨率融合,可为多源遥感影像融合的超分辨率问题提供1种可行的解决方案,而且对其他融合方法也有借鉴意义。
[Abstract]:Considering that multi-source remote sensing image fusion is limited by the existing resolution, a new super-resolution fusion method based on dictionary learning is proposed based on sparse representation theory. The spatial resolution of multispectral images can be increased to one or two times the spatial resolution of panchromatic images. In the framework of remote sensing image fusion, a learning dictionary is first established, and a redundant dictionary is used to represent the image sparsely to reconstruct super-resolution, and then Gram-SchmidtGS-GS-based spectral sharpening method is used to obtain super-resolution multi-spectral image. The results of three experiments using QuickBird data show that the proposed method has some advantages over the traditional fusion method, traditional super-resolution method and other dictionary learning schemes, and is suitable for super-resolution fusion of remote sensing images. It can provide a feasible solution to the super-resolution problem of multi-source remote sensing image fusion, and it can also be used for reference to other fusion methods.
【作者单位】：中国地质大学(北京)地球科学与资源学院;
【基金】：中国地质调查局地质调查项目“京津地区矿产资源开发环境遥感监测”(编号:12120115060901)资助
【分类号】：TP751

【参考文献】