多源双时相光学遥感影像的厚云检测

Cloud Detection Using Multi-source and Two-date Optical Remote Sensing Images

遥感影像中厚云的存在完全遮蔽了下垫面,造成了厚云区域地物信息的丢失。针对厚云检测算法不够精准和适用性不够的问题,提出一种简单高效的多源双时相光学遥感影像的厚云检测方法。该方法不仅适用于中等空间分辨率和高空间分辨率的影像,而且也能进行多源影像的云检测。该方法需要2个时相影像:目标影像和参考影像,先进行影像配准和DN值归一化的预处理,然后基于云在蓝光波段的反射特征得到初始云掩膜。针对初始云掩膜中错检的高亮物体,该文提出了云边缘指数进行优化。最后使用区域生长法以得到最终的云掩膜。利用多幅中高分辨率影像,并且与PRC法和SVM法进行比较,实验结果证明了所提出方法的精确性和时效性。

Land surfaces are obstructed by thick clouds,which lose the land cover information.Aiming at the insufficient accuracy and applicability of thick cloud detection algorithm,this paper presents a simple and practical method for cloud detection using multi-source and two-date optical remote sensing images.This method is not only suitable for medium spatial resolution and high spatial resolution images,but also can be used to multi-source images.The proposed method requires two-date images:a target image and a reference image.After the image registration and DN normalization,the coarse cloud mask is extracted through threshold method based on the blue band information.The coarse cloud mask is further refined using cloud edge index.In order to extract final cloud mask,the seeded region growing method is adopted.Medium spatial resolution and high spatial resolution images are used in the experiments.The results compared with PRC method and SVM method demonstrate that the proposed method is precise and effectiveness.