测绘学报 ›› 2016, Vol. 45 ›› Issue (5): 581-591.doi: 10.11947/j.AGCS.2016.20150500

• 摄影测量学与遥感 • 上一篇    下一篇

国产高分辨率遥感卫星影像自动云检测

谭凯, 张永军, 童心, 康一飞   

  1. 武汉大学遥感信息工程学院, 湖北 武汉 430079
  • 收稿日期:2015-10-14 修回日期:2016-02-17 出版日期:2016-05-20 发布日期:2016-05-30
  • 通讯作者: 张永军 E-mail:zhangyj@whu.edu.cn
  • 作者简介:谭凯(1991-),男,硕士生,主要从事遥感影像云检测方面的研究。E-mail: kai_tan@whu.edu.cn
  • 基金资助:
    国家自然科学基金(41322010;41571434);国家863计划(2013AA12A401)

Automatic Cloud Detection for Chinese High Resolution Remote Sensing Satellite Imagery

TAN Kai, ZHANG Yongjun, TONG Xin, KANG Yifei   

  1. School of Remote Sensing and Information Engineering, Wuhan University, Wuhan 430079, ChinaAbstract
  • Received:2015-10-14 Revised:2016-02-17 Online:2016-05-20 Published:2016-05-30
  • Supported by:
    The National Natural Science Foundation of China (Nos.41322010;41571434);The National High-teh Research and Development Program of China (863 Program) (No.2013AA12A401).

摘要: 云检测一直是卫星影像处理的难题,特别是混有地物光谱特性的薄云长期成为影像产品生产的阻碍。本文所介绍的国产高分辨率遥感卫星影像自动云检测方法能够有效克服这一难题。首先采用改进的颜色转换模型,将影像由RGB转换至HIS颜色空间,利用影像强度信息与饱和度信息生成基底图,并使用影像近红外与色调信息对其进行优化,生成修正图。然后利用直方图均衡化与双边滤波结合带限定条件的Otsu阈值分割提取纹理信息,并对修正图进行误差剔除生成云种子图。最后以HIS颜色空间的强度信息为向导,结合云种子图进行云精确提取。与不同自动、人工交互式云检测方法相比,总体精度提高了10%左右,并且能够较好地提升云检测效率。

关键词: 国产卫星影像, 云检测, 改进HIS模型, 双边滤波, Otsu阈值分割

Abstract: Cloud detection is always an arduous problem in satellite imagery processing, especially the thin cloud which has the similar spectral characteristics as ground surfacehas long been the obstacle of the production of imagery product. In this paper, an automatic cloud detection method for Chinese high resolution remote sensing satellite imagery is introduced to overcome this problem.Firstly, the image is transformed from RGB to HIS color space by an improved color transformation model. The basic cloud coverage figure is obtained by using the information of intensity and saturation,followed by getting the modified figure with the information of near-infrared band and hue. Methods of histogram equalization and bilateral filtering, combined with conditioned Otsu thresholding are adopted to generate texture information. Then the cloud seed figureis obtained by using texture information to eliminate the existed errors in the modified figure. Finally, cloud covered areas are accurately extracted by integration of intensity information from the HIS color space and cloud seed figure. Compared to the detection results of other automatic and interactive methods, the overall accuracy of our proposed method achieves nearly 10% improvement, and it is capable of improving the efficiency of cloud detection significantly.

Key words: Chinese satellite imagery, cloud detection, improved HIS model, bilateral filtering, Otsu thresholding

中图分类号: