Methods And Applications Of Cloud Detection For Optical Remote Sensing Satellite Imagery

Optical remote sensing satellite images,such as QuickBird,WorldView,PlanetScope and Chinese ZY3,GF1,GF2 satellite images,are widely used in the fields of navigation,resource investigation,environment protection,disaster prevention,ocean exploitation and urbanization studies,etc.The presence of clouds not only covers the ground information,but also changes the spatial and spectral characteristics of the images,therefore brings additional troubles to the subsequent processes and analysis.Under these conditions,cloud detection methods for optical remote sensing imagery are proposed in this paper.Experimental results verified the feasibility,accuracy,efficiency and automaticity of the proposed methods.The applications of cloud detection results in optical remote sensing image processing were also discussed.The ultimate aim of this paper is to extract the available information from massive optical remote sensing satellite images and to improve the value of these images.The main contents and conclusions of this paper are as follows:1.A fast and automatic cloud detection approach for single optical image.In this work,the grey-level histogram of the image is fitted approximately by a Gaussian mixture model(GMM),and the threshold between cloud and clear sky is chosen based on the distribution of the Gaussian components.Cloud pixels were segmented by the threshold and integrated by using mathematical morphology operations.Experimental result verified the feasibility of this approach in both extracting cloudy areas and identifying cloudless images.This approach is fast and automatic,which can be used in quality control of satellite images,and can provide initial results for more accurate detection subsequently.2.An automatic man-made objects extraction approach by using visual saliency.Firstly,linear features,which are peculiar to man-made objects in high-resolution remote sensing images,were selected as the measure of visual saliency,and extracted by using the 2D Gabor filter.Then,man-made objects,which were regarded as visual salient areas of the Gabor feature map,were recognized and excluded from the clouds.In this way,the robustness of the cloud detection approach was enhanced.3.An improved cloud detection approach by using open geospatial data.Geographic information is peculiar to remote sensing images,which can be associated with the open geospatial data.Open thematic maps,such as glacier maps,can be used in recognizing specific objects.Existing satellite images,such as Landsat-8 images,can be used in recognizing unchanged objects.The robustness of the cloud detection approach was enhanced by excluding these specific or unchanged objects.4.A cloud detection approach for multi-view optical images by using height displacement.Clouds are hundreds or thousands of meters above the ground,which means in the multi-view images,clear sky areas can be regarded as constant background and clouds can be regarded as moving objects,attribute to the presence of height displacement.Therefore,the clouds can be precisely detected by using differential analysis with the supplement of the brightness analysis.Experimental results verified the superiority of this approach in detecting all kinds of clouds and avoiding misrecognitions.5.The production methods of optical remote sensing images considering the clouds.The impact of clouds on radiometric calibration,geometric rectification,image composition and DEM production,etc.in the processing of remote sensing images are analyzed.Related cloud detection approaches are applied to different applications and production methods of optical remote sensing images considering the disturbance of clouds are discussed to extract useful and efficient information from massive images and thus improve the utilization of images.