利用引导图像块排序的遥感影像厚云去除算法

摘要

针对现有厚云去除算法存在的辐射精确度不足和光谱不一致等问题,提出了一种利用引导图像块平滑排序的厚云去除算法.通过线性回归分析减轻多时相影像之间的光谱差异并增强时相相关性,然后从多时相影像中提取排序矩阵并引导云污染影像中的图像块进行排序,将相似性特征加入到排序后所得的平滑信号中,提高了信息重建的精确度,最后利用影像自身信息进行插值修复云污染区域信息,保证了光谱一致性.实验结果表明,与现有的2种基于辐射校正方法和图像信息克隆方法相比,该算法适用于复杂环境下遥感影像厚云去除问题,处理结果具有更好的辐射精确度和光谱一致性.%In order to improve the radiometric accuracy and spectral consistence in cloud removal,a novel method based on smooth patch ordering via smooth patch ordering guided with a multi-temporal image is proposed.Linear regression analysis is used for correction of spectral differences and enhancing the temporal coherence between multi-temporal images.The missing pixels of cloud-contaminated region are reconstructed by applying one-dimensional cubic spline interpolation on the reordered set of pixels.A multi-temporal image is used to settle for permutation metrics which is applied to reorder the pixels of cloudcontaminated image as regular signals.The proposed algorithm is tested on GF-1 multispectral images and a quantitative analysis of the simulated data with various cloud contamination conditions is conducted to evaluate the proposed method.The experimental results show that,compared with the two algorithms based on radiation correction and information cloning method,this algorithm can be applied to cloud removal in remotely sensed image under complex conditions,and the reconstructed image has better accuracy and spectrum consistency.