基于多参考影像信息融合的遥感影像厚云去除

Thick cloud removal of remote sensing images based on multi-reference image information fusion

遥感影像去云是遥感影像处理与分析的重要领域,对影像后续的信息提取等操作起到至关重要的作用。针对多时相遥感影像融合去云中对待重建图像的质量要求较高以及适用性较低的问题,提出了一种基于一幅或多幅参考影像信息进行多时相遥感影像融合的厚云去除算法,包括参考影像的选取、辐射归一化、多时相影像融合以及泊松图像编辑4个主要步骤。首先根据影像掩模及主成分信息选取参考影像,并且进行多源遥感影像辐射归一化保留地物信息的变化情况;然后基于选择性多源全变分模型对影像进行融合处理,并通过泊松图像编辑技术改善影像融合后的边界梯度不连续问题。实验结果表明,所提方法可以对带有厚云且质量不一的多源遥感影像进行有效去云处理,并在整体上获得比传统方法更高的影像细节精度。

The cloud removal of remote sensing images is very important in the processing and analysis of remote sensing images and plays a crucial role in the subsequent image information extraction and other operations.Aiming at the high-quality requirements and low applicability of the reconstructed images in the cloud removal of multi-temporal remote sensing image fusion,a thick cloud removal algorithm based on one or more reference images was proposed,mainly including a selection of reference image,radiometric normalization,multi-temporal image fusion,and Poisson image editing.Firstly,the reference image was selected according to the image masking and the principal component information,and the radiometric normalization of the multi-source remote sensing image was carried out to preserve the change of ground feature information.Then,the image was fused based on the selective multi-source total variation model,and the boundary gradient discontinuity after image fusion was reduced by Poisson image editing.The experimental results show that the proposed method can effectively remove clouds from multi-source remote sensing images with thick clouds and different quality,and obtain higher image detail precision than traditional methods.