中国空间科学技术

中国空间科学技术 ›› 2022, Vol. 42 ›› Issue (3): 132-140.doi: 10.16708/j.cnki.1000-758X.2022.0044

• 钱学森空间技术实验室十周年专栏 • 上一篇    下一篇

基于高分四号数据的多时相多通道云检测算法

王纪辉，李峰，鹿明， 马骏，郭毅   

  1. 1河南大学 软件学院，开封475100
    2中国空间技术研究院 钱学森空间技术实验室，北京100094
    3澳大利亚西悉尼大学，悉尼NSW2150

  • 出版日期:2022-06-25 发布日期:2022-06-22

Multi-temporal and multi-channel cloud detection algorithm based on GF-4 data

WANG Jihui，LI Feng，LU Ming，MA Jun，GUO Yi   

  1. 1College of Software, Henan University，Kaifeng 475100,China
    2Qian Xuesen Space Technology Laboratory, China Academy of Space Technology，Beijing 100094，China
    3Western Sydney University, Sydney, NSW 2150, Australia
  • Published:2022-06-25 Online:2022-06-22

PDF (PC)

119

摘要： 针对高分四号（GF-4）卫星影像波段较少导致传统云检测算法难以区分云与冰雪像元的问题，提出一种多时相多通道云检测算法。该算法首先对GF-4卫星影像进行辐射定标和配准，然后利用云与典型地表的光谱差异得到潜在云像元，之后利用序列GF-4卫星影像之间的差异识别出移动的云像元，最后利用中红外波段反演地表亮度温度来去除冰雪像元。该算法在海南、辽宁和安徽3个研究区域进行验证，并将检测结果与传统单时相云检测算法、支持向量机（SVM）云检测算法和实时差分（RTD）云检测算法的检测结果进行对比。结果表明，该算法优于其他3种云检测算法，准确识别率均达到90%以上，误检率均低于5%，有利于GF-4卫星影像的进一步利用。

关键词: 高分四号卫星影像, 云检测, 多时相, 光谱差异, 中红外波段

Abstract: Aiming at the problem that the traditional cloud detection algorithm is difficult to distinguish between clouds and ice pixels due to the lack of bands of the Gaofen-4 (GF-4) satellite imagery, a multi-temporal and multi-channel cloud detection algorithm was proposed. The algorithm first carried out the radiation calibration and registration of the GF-4 satellite image, used the spectral difference between the cloud and the typical ground surface to obtain potential cloud pixels, and then used the difference between the sequence GF-4 satellite image to identify cloud pixels. Finally, the mid-infrared band was used to retrieve the brightness temperature of the surface to remove the ice and snow pixels. The algorithm was verified in three research areas of Hainan, Liaoning and Anhui, and the detection results were compared with the detection results of traditional single-phase cloud detection algorithm, SVM cloud detection algorithm, and RTD cloud detection algorithm. The results show that the algorithm is better than the others. The proposed algorithm has an accurate recognition rate of more than 90%, and the false detection rate is less than 5%, which is conducive to further use of GF-4 satellite imagery.

Key words: GF-4 satellite image, cloud detection, multi-temporal, spectral differences, mid-infrared band