光学遥感影像不可避免会受到云的遮挡,导致影像应用性降低,遥感影像去云是近些年来的热门研究方向.插值算法利用一张或多张影像计算相似像素来重建缺失像素,但是最佳相似像素的确定和地物突变的高精度重建仍存在挑战.文章利用多时相影像时间相关性关系、空间关系和光谱关系,提出了基于时空谱约束的相似像素插值(SPISTS)的去云方法.该方法利用时域和谱域特征选择相似像素,并通过空间、时间相关性和光谱关系约束求得影像缺失值,最后通过正则化项对预测进行偏差改正.在3个不同地区进行实验并与加权线性回归(weighted linear regression,WLR)、时空加权回归(spatio-temporal weighted regression,STWR)、改进邻域相似像素插值(modified neighborhood similar pixel interpolation,MNSPI)结果进行比较.实验结果表明,该方法精度较高,能够减弱辐射差异带来的影响.
Cloud Removal Method Based on Spatiotemporal-spectral-constrained Similar Pixel Interpolation
Optical remote sensing images are inevitably affected by cloud cover,which reduces the applicability of the images.Removing clouds from remote sensing images has been a popular research direction in recent years.Interpolation algorithms use one or more images to calculate similar pixels to reconstruct missing pixels,but determining the best similar pixels and accurately reconstructing abrupt changes in terrain remains a challenge.In this paper,we propose a cloud removal method based on similar pixel interpolation constrained by spatio-temporal-spectral relationships(SPISTS),which uses the temporal and spectral features to select similar pixels and constraints of spatial,temporal,and spectral relationships to determine the missing values in the image.Finally,a regularization term is used to correct the bias in the prediction.The SPISTS method was compared with the results of WLR(weighted linear regression),STWR(spatio-temporal weighted regression),and MNSPI(modified neighborhood similar pixel interpolation)in three different regions.The experimental results show that the SPISTS method achieved high accuracy and reduced the impact of radiation differences.
NETL
NSTL
万方数据
