Research Of Cloud And Shadow Removal Algorithm For Landsat Satellite Image

The acquisition and interpretation of remote sensing images have been developing continuously with the advances in computer science,electronic information technology and image processing technology in the information period.A large number of remote sensing images have some features including high resolution and extensive coverage area and have been widely used in many different fields,such as agricultural production,environment monitoring,ecological protection,landscape ecological change detection,terrain modeling.However,because of the effect of clouds and cloud-shadows,the images acquired from the earth resource satellites may be interfered,which seriously affects the timeliness and completeness of image information.In addition,it is difficult to detect in the region of bright clouds and dark shadows in different backgrounds.If the clouds and shadows are only detected manually,the production efficiency will be greatly affected.For overcoming the conflict of development and efficiency,it has become a very important scientific field for studying remote sensing images to automatically remove cloud and cloud-shadow.Based on the information compensation between multiple remote sensing images at different time phase,the utilization rate and clarity to interpret the images is improved.A simple image processing software is designed by using C ++ programming language that calls the Opencv library function and used successfully in Guangdong and Dao county.The main research methods are summarized as follows:As thick clouds and thin clouds with different physical properties,the paper uses the traditional threshold,improved watershed and wavelet fusion algorithm to detect the thick cloud region.Firstly,the bright white area is marked using the improved watershed algorithm by the maximum interclass variance method and then we can obtain the complete thick cloud boundary region to avoid the excessive segmentation of the detected image.The way of wavelet fusion makes full use of the integrity of thick cloud region of the blue channel image and differences between thick cloud and background region of near-infrared channel image.The multi-temporal image is decomposed by 3-layer wavelet,and the appropriate wavelet decomposition coefficient is selected.It is obviously enhanced the contrast between the thick cloud region and the background of the image after wavelet reconstructed.The fusion images are segmented by the traditional threshold algorithm and are able to identify the thick cloud areas with lower gray levels.Finally,the above three algorithms of thick cloud detection obtain the area of the thin cloud by means of regional expansion.In order to exclude gray value which are similar to the bright white background objects of the cloud and to detect gray value which are similar to background objects of the thin cloud,the mask algorithm is proposed and can get the accurate cloud areas.Shadow is one of the major problems in remotely sensed imagery which hampers the accuracy of information extraction and change detection.In these images,the shadow,produced by different objects,namely,urban materials,mountains and cloud,is similar to the water in the spectral characteristics.In order to exclude water regions and detect cloud shadow,we successfully detect water regions based on the spectral difference.The darkening effected by the cloud-shadow in the Near Infrared Band is used to product a potential shadow layer by employing the flood-fill transformation.To reduce the number of iterations of cloud-shadow matching,the cloud temperature and spectral reflectance are used to determine the height of the cloud,which improves the efficiency and accuracy of the shadow detection algorithm in this paper.we can predict the cloud-shadow location based on the geometric relationship between a cloud and cloud-shadow.For removing the detected cloud and cloud shadows,the regression relationship is established according to the difference among different time images.At last,the cloud and cloud-shadow of the target images are shifted by the same location on reference images whose spectral information are matched with the target images by linear regression method.The results demonstrate that it is significant to discover cloud accurately based on above algorithms from the point of subjective visual perception which mask algorithm can get more accurate results.We can get the borderless difference to the images which can be suited for Landsat satellite images resources at different phases.