Cloud Removal Of Optical Remote Sensing Imageries

Optical remote sensing data have a wide range of applications,but they are often adversely affected by clouds.To improve the quality and utilization of optical remote sensing images,one develops cloud removal algorithms.This thesis study focuses the cloud-removal algorithm development.First,the principle of an empirical and radiative transfer model（RTM）-based algorithm to remove thin clouds is reviewed.The algorithm can remove thin clouds in visible bands.Unfortunately,the RTM-based algorithm requires a near-infrared（NIR）band and cloud-covered water pixels.Then,a revised RTM algorithm is developed to remove thin clouds.The algorithm estimates the thin cloud reflectivity in each optical band,and the estimation is further improved through the minimum operation and smoothing filtering.The cloud removal is obtained by subtracting the corrected thin cloud reflectivity,band-by-band.After analyzing multiple Landsat-8 operational land imager（OLI）datasets,the algorithm effectively removes thin clouds in the visible bands.Also,the spatial correlation coefficients of cloud-free areas before and after the algorithm is above 0.985,and it has almost no effect on the cloud-free regions.The algorithm can remove thin clouds over different land use and land cover（LULC）types since it only requires cloud-free areas.Unfortunately,the simplified algorithm can only remove thin clouds in visible bands.A second cloud-removal algorithm based on the linear discriminant analysis（LDA）is proposed.With multi-band remote sensing input data,the method sets the category label data as the cloud detection band,uses the output cloud components and conversion matrix to calculate the reflectivity of thin clouds in each band,and finally achieves cloud removal.The algorithm is qualitatively and quantitatively evaluated with simulated and acquired Landsat-8 OLI datasets.Coefficient of determination（R~2）values of the cloud-covered and cloud-free reference images ranged from 0.189-0.739,whereas the values changed between 0.729 and 0.870 after the algorithm.A significant overall increase was achieved.In the cloud-free areas,R~2 values before and after the algorithm are≥0.985.Thus,the algorithm can effectively remove thin clouds in the visible,near-infrared,and shortwave infrared bands.Finally,the algorithm is suitable to remove thin clouds over different LULC types.