Research On Automatic Cloud Detection Method For Remotely Sensed Satellite Imagery With High Resolution

With the rapid development of Earth observation technology,the acquisition of remote sensing images is becoming increasingly easier.A large number of new remote sensing image data are generated every day,and these images are now widely used in a large number of fields.The statistics of the International Satellite Cloud Climate Project(ISCCP)show that an average of 66.7% of the Earth's surface is covered by clouds.Under a clouded scenario,ground images captured by remote sensing satellites are subject to The effect of cloud cover.The obstruction of the remote sensing image capture area by the cloud layer not only results in the loss of available information in the image,but also affects the inversion of atmospheric parameters,target detection,classification and identification of features,atmospheric correction,aerosol inversion,stereo matching,and foliage.Area index(LAI,Leaf Area Index),FAPAR(The Fraction of Absorbed Photosynthetically Active Radiation,photosynthetic active radiation absorption coefficient),NPP(Net Primary Productivity,vegetation net primary productivity)and other organisms The estimation of physical parameters will also have many impacts on the subsequent processing of image registration and fusion,so we need to accurately identify and segment the clouds in remote sensing images.The traditional spectral detection algorithm based on spectral characteristics requires that the remote sensing image to be detected contains a large number of specific bands.The generality of the algorithm is not strong,and it cannot be applied to various remote sensing images.There are a wide variety of cloud and ground objects,and many types of underlying surfaces.The spectral characteristics are very similar to those of clouds.The textures of different types of clouds and various ground targets are also varied.Therefore,we need to study the characteristics of clouds and ground objects.There are many basic solutions to the issue of remote sensing image cloud detection.One of the typical ideas for highresolution remote sensing images is to treat the process of cloud detection as a semantic segmentation of the image.The effective automatic cloud detection method proposed in the third chapter of this paper firstly using color features to extracts the high-reflectivity target from the remote sensing image,and then combines the multi-scale image decomposition technique with the domain transform filter in the edge-preserving filtering technical field to extract the texture features,and then refine the segmentation of edges in results using color and texture features,remove regular-shaped buildings and other non-cloud areas,this paper designs an effective automatic cloud detection method.Based on the third chapter,the method is improved in the forth chapter by combining the U-Net,a fully convolution network for semantic segmentation.Through experimental comparison,it is proved that the two proposed algorithms have better detection performance.