线性渐进滤光成像高光谱数据的波段配准方法  

Band registration method of hyperspectral data based on linear progressive filter imaging

作　　者：于春瑶 方俊永 王潇 张晓红 刘学 YU Chunyao;FANG Junyong;WANG Xiao;ZHANG Xiaohong;LIU Xue(Aerospace Information Research Institute,Chinese Academy of Sciences,Beijing 100094,China;University of Chinese Academy of Sciences,Beijing 100049,China)

机构地区：[1]中国科学院空天信息创新研究院,北京100094 [2]中国科学院大学,北京100049

出　　处：《中国科学院大学学报（中英文）》2023年第3期406-414,共9页Journal of University of Chinese Academy of Sciences

基　　金：中国科学院仪器设备功能开发技术创新项目资助。

摘　　要：线性渐进滤光成像传感器作为一种新型传感器,其高光谱数据处理原理与传统线阵高光谱数据处理原理存在差异,研究较为匮乏。在图像配准方面,利用传统方法配准精度不高。基于几何校正后的线性渐进滤光高光谱图像,提出一种分组逐层返回配准的波段配准策略和双充分性SIFT算法(DS-SIFT算法)。DS-SIFT算法包括粗配准及精配准,其中精配准使用分块的改进SIFT算法,通过引入信息熵和结构相似性进行图像子块的查找,有效提高不同波段之间配准精度及效率。通过飞行数据进行算法验证,结果表明,利用本文所提的处理算法及流程,可以得到质量较好的高光谱配准数据。As a new type of sensor,the hyperspectral data processing principle of linear progressive filter imaging sensor is different from that of traditional linear array hyperspectral data,and the related research is scarce.When the traditional method is used in image registration of this new type data,the accuracy of image registration is not high.Aiming at the problem of geometric correction of linear progressive filtering hyperspectral images,this paper proposes a band registration strategy of grouping return registration and an image registration algorithm based on improved SIFT algorithm(DS-SIFT algorithm).DS-SIFT algorithm includes rough registration and precise registration.The precise registration is based on the improved SIFT algorithm of block,which can improve the registration accuracy and efficiency between different bands by introducing information entropy and structural similarity for the search of image blocks.The flight data were used to verify the algorithm,and the results show that the proposed process can obtain high quality hyperspectral registration data.

关 键 词：线性渐进滤光成像 几何校正 波段配准 DS-SIFT算法 

分 类 号：P237[天文地球—摄影测量与遥感]