Study Of Clouds Removal Methods On Remote Sensing Images

The application of satellite images becomes more and more extensive in many fields with the fast development of remote sensing technology.But the remote sensing image are highly susceptible to weather conditions in the process of imaging.Most of the remote sensing image we get would inevitably be covered by the cloud in some local area,so that to obtain a completely without cloud image is rather difficult.Clouds in remote sensing images not only affect the further processing of remote sensing image,but also can reduce the utilization and interpretation accuracy of the remote sensing image.So to explore an effective method of clouds removal on remote sensing image has very important practical significance and research value.In remote sensing image,clouds an be divided into thin clouds and thick clouds according to the different characteristics.There are already some conventional clouds removal methods of remote sensing image.Generally different types of clouds cover in the remote sensing image need Choose to use different clouds removal methods.The reason to deal with the clouds in remote sensing image,not only for purpose of improving the mapping or classification accuracy of remote sensing image,but also the important step for information Extraction of ground feature.So the ideal clouds removal methods need remove the influence of the cloud,at the same time need as far as possible restore the feature of spectral information witch covered by clouds.Eventually lay a good foundation for a better implement further processing of remote sense images.This paper studied relative basic theories of removing clouds cover,cloud imaging model and some clouds removal methods.The study puts an emphasis on thick cloud removal methods,and put forward a thick clouds removal method from remote sensing images based on inpainting technique and a method based on similar pixel substitution approach respectively.The work of this paper mainly embodies aspects as follows:1.Discussion and analysis was made about the common clouds removal methods.Then make an indepth study on homomorphic filtering method,substitution method time-sequenced filter(Hants).Through summarizes the advantages and disadvantages of these methods,puts forward some commonly problems of traditional cloud remove methods.The article studies later mainly aiming at these problems2.Criminisi image inpainting method was used for cloud removing in remote sensed image.This method could restore gray-level information of the cloud area,but the texture information of the image was lost.Substitution method could reserve texture information,however usually accompanied by "Patch" phenomenon.This study proposed a new thick clouds removal method based on inpainting and substitution method.1)Experiments were carried out to remove the clouds of the remote sensed image using the inpainting method and substitution method respectively.2)Wavelet-based fusion was done on the results of 1).3)post-processing on cloud regions' edges in order to get a smooth result image with cloud removed.Compared with the method of inpainting and substitution respectively,the combination of this two methods shows the better result on thick cloud removing.3.Cloud removing is one of the important steps in remote sensing image processing and mapping.Regular cloud removing methods could remove cloud affects,however usually accompanied by loss of spectral information.Based on similar pixel replacement,this study proposed a new thick clouds removal method.Composed of three steps.1)an improved single-temporal cloud detection method to detect the thick cloud in the remote sensing image and generate cloud mask image.2)using appropriate similar pixels within the remaining region of the target image to fill the masked pixels.3)post-processing on cloud regions' edges in order to get a smooth result image with cloud removed.The results show that this method is highly valid in thick cloud removing and spectral information restoring.