Research On Automatic Cloud Detection Algorithm Of Satellite Remote Sensing Image

With the rapid development of satellite remote sensing technology and the increase in the number of satellites,more and more remote sensing images have been taken,and the application range of satellite remote sensing images is increasingly broad,involving more and more fields,including land supervision,urban planning,terrain navigation,etc.However,the captured remote sensing image will inevitably be interfered by cloud cover,resulting in the ground objects in the remote sensing image being blocked by cloud cover,which has a great impact on the subsequent remote sensing image target recognition,terrain matching and navigation.Cloud detection,as an important link in remote sensing image processing,should be able to accurately extract clouds in remote sensing images,as the basis for cloud removal,and provide convenience for subsequent image processing.Deep learning method can extract shallow and deep features of targets from large-scale data sets.When applied to cloud detection,cloud detection has higher accuracy and wider application range than traditional method.In this paper,the method based on deep learning is used to accurately detect clouds in remote sensing images,and the cloud removal algorithm is designed according to the detection results.The main research contents are as follows:(1)To solve the problems of scarce and poor quality of open source remote sensing image cloud detection datasets,this paper uses the remote sensing images shot by “GF”series satellites and “ZY” series satellites to outline cloud boundaries by using the super-pixel segmentation method,and then manually marks the outlined cloud boundaries to generate corresponding label maps,and independently produces high-quality remote sensing image cloud detection data sets.(2)To solve the problems of small scope of application and low accuracy of traditional cloud detection methods,this paper proposes a cloud detection method for remote sensing images based on U-net.VGG16 network is used to replace the original U-net feature extraction network to extract deeper cloud feature information,and Gaussian progressive fuzzy processing is carried out on the input image to separate the background image,reduce the impact on the cloud target,and improve the accuracy of cloud detection.(3)To solve the problem that the interference of white features such as snow and saline land leads to the decrease of remote sensing image cloud detection accuracy,this paper proposes an anti-jamming cloud detection algorithm based on U-net3+.By improving the U-net3+ network structure,a new loss function is proposed to strengthen feature fusion,solve the problem of unbalanced sample quantity,effectively eliminate all kinds of interference information in remote sensing images of different landforms,and accurately detect clouds in remote sensing images.The algorithm in this paper is used to detect clouds in remote sensing images,and remote sensing images of different phases in the same area are used to replace cloud regions,and local filtering is carried out to make the replaced images smoother,reduce the distortion rate to the greatest extent,and complete the cloud removal of remote sensing images.