摘要
针对如何在时间序列尺度上利用多源时空融合方法高精度地重构高分辨率遥感影像的问题,该文提出了一种基于增强字典学习样本空间的单数据对稀疏学习融合算法,并利用现有稀疏学习算法、STARFM算法以及半物理模型对Landsat与MODIS卫星数据进行双向融合实验。结果表明:随着样本尺寸及空间的拓展,改进后的稀疏学习算法能够获得比原始算法、STARFM、半物理模型等算法更优的融合结果,其中ERGAS可达15.0以内、SSIM可达84%以上,并且融合质量对高、低分辨率图像间的空间尺度差异性不敏感。通过采用更高效的在线字典学习算法,该融合方法的处理效率与应用价值有望得到极大提升。
To accurately reconstruct high resolution remotely sensed data on the time series dimension with multi-source Spatiotemporal fusion techniques,a sparse-learning fusion algorithm using single-pair images is proposed on a basis of an enhanced sample space in the dictionary learning process,and the original algorithm,the STARFM and the semi-physical fusion model are employed for a bi-directional fusion test with Landsat and MODIS satellite data.The results show that the proposed algorithm can obtain better performance than the original version,STARFM and the semi-physical model with the extending of the training sample size and training space.In detail,the ERGAS can be under 15.0 and the SSIM can be above 84%,and the correlation lacks between fusion quality and spatial resolution.The efficiency and application value of the proposed method can be significantly promoted by using the effective online dictionary learning algorithm.
作者
葛艳琴
李彦荣
孙康
李大成
陈永红
李瑄
GE Yanqin;LI Yanrong;SUN Kang;LI Dacheng;CHEN Yonghong;LI Xuan(Taiyuan University of Technology,Taiyuan 030024,China;The 54th Research Institute of China Electronics Technology Group Corporation,Shijiazhuang 050081,China;Shanxi Center for Data and Application of High Resolution Earth Observation System,Jinzhong,Shanxi 030600,China)
出处
《测绘科学》
CSCD
北大核心
2019年第9期107-114,共8页
Science of Surveying and Mapping
基金
国家自然科学基金项目(41501372)
山西省高等学校科技创新项目(2016144)
关键词
稀疏学习
字典训练
陆地卫星数据
地表反射率
时空融合
sparse learning
dictionary training
Landsat data
surface reflectance
spatiotemporal fusion
